• Windows 11 just got a revamped internet speed test in preview
• This feature was shuttled swiftly through development using agentic AI in combination with human oversight
• Microsoft boasts that this is about: "Getting from idea to customer value faster. Without AI slop." Others are more cynical, though

Windows 11 just got a useful change in preview, with its built-in internet speed test being revamped, and the new version was produced using AI (in conjunction with human oversight, I should add).

Windows Latest spotted that Marcus Ash, Microsoft's VP of Windows Design and Research, posted on X to explain that the "team moved pretty fast on this one", adding that following feedback asking for the speed test to be improved, "within a day, Diego had a working prototype." (Diego Baca is VP of Design for Windows).

Ash tells us: "From there, our connectivity team worked with partners to bring the experience into Windows. The design and build work was driven by an agentic AI workflow, with a person guiding the process in their spare time."

The Microsoft exec further explains that this is where he sees "a lot of potential in AI" and using agentic workflows, in terms of "Getting from idea to customer value faster. Without AI slop. That is the way."

Analysis: AI, oversight, bugs, and jobs

Of course, this is a controversial topic. Mention AI being involved in the coding of Windows 11 and some tempers will flare online, even though we know this is happening anyway. Microsoft has made no secret of that fact.

Indeed, among the AI hate and flames on social media, a lot of people are point out that these days pretty much all software is being developed in conjunction with Large Language Models (LLM agents) — even Linux.

As one Redditor makes clear: "Newsflash — every single piece of software you use, except for open source projects that are making a statement, you should expect to have AI written code in it. It is pervasive, it is everywhere."

The key aspect here is exactly how the AI agent in question is employed.

It's one thing for such an agent to be given a task to write a piece of code for a certain feature, and then just to slap that into your software or operating system as it is (or maybe after a cursory glance through the work). It's another thing to have a human programmer working on a feature and using an agent to perform grunt-work (easy but perhaps time-consuming) coding therein — and then to properly check the AI's output.

Theoretically, there's nothing wrong with the latter approach, at least in terms of the soundness of the coding work. And as Microsoft's Ash points out, this is really accelerating development times — meaning Windows 11 users will get new features (hopefully capabilities they want, like this improved internet speed test) much more swiftly. Which is exactly why almost every developer is now utilizing this tech.

There are problems here, though, and they're bound up in the public perception of Microsoft. Do Windows 11 users trust the company to be using AI in the right — i.e., properly checked and vetted — way? Not really, in many cases, as there remains a lot of doubt over the overall quality of Windows 11 code, and rightly so given the regularity, and severity, of the bugs we've seen (with more prevalence since 2024, and that 24H2 update).

As another Redditor observes: "People aren't calling Windows slop because it's being made using AI these days. They're calling it slop because it's being made using AI with woefully insufficient human oversight and even more woefully insufficient QA, leading to repeated disastrous bugs making it to production and end users being treated as beta testers. You want people to stop calling your product slop? Stop shipping updates that take down people's computers or break existing features."

The other issue here is that of AI stealing coding jobs from people. Agents may make things happen a lot more quickly, but as Ash notes, this revamped speed test was implemented mainly by an agent — including its design — with the software engineer "guiding the process in their spare time", making it sound like a rather casual amount of effort on the human side of the equation.

Is it a particularly sensitive approach to be touting the benefits of AI — as real as they may be for an operating system's ongoing development and maintenance — in this way at a time when Microsoft has been swinging the axe with various job cuts and pushing staff towards voluntary redundancy in a big way? I'd suggest it isn't, and the timing of this particular push of the value of agentic AI is dubious in that respect.

At least in this case Microsoft is being upfront and transparent about what it's doing — and we can expect a good deal more AI acceleration of Windows 11's development going forward.

Generic AI tools promise to solve every problem under the sun, from running enterprise CRM to analysing why your cat isn't drinking enough water during a heatwave. And when you try to use one tool with no defined role for every task, results are disappointing and erode trust in technology.

Multiply that pattern across hundreds of employees, and you get tokenmaxxing: prompt after prompt, refining AI output until something usable comes out.

Some organizations, by instinct, default to blaming the employees for tokenmaxxing, or double down on training on ‘responsible use of AI’.

Tokenmaxxing is a symptom of capable people having to create solutions in an environment that lacks clear shape, scope, guardrails - and ways to analyze spend and outcomes.

Locking it down is not the solution

Traditional procurement systems were designed for traditional business software cycles. If building an enterprise system takes years, it’s not a big deal if procuring the right vendor takes years, too.

Routing AI through procurement to evaluate all options and keep the spend predictable initially felt to organizations like a safe and responsible choice. But software creation is very different today than it was even 5 years ago. Standard IT procurement still takes months, while opening a public LLM chatbot and asking it to generate a landing page is a matter of five seconds.

Policies tend to live their happy lives somewhere in the corporate portals, unbothered and forgotten unless an incident calls for a formal investigation. And when the approved systems are slow or restrictive, employees find workarounds. They turn to unvetted shadow AI tools for daily tasks and put proprietary corporate data into public models.

Look at the app prototypes that come out of it: the bad code is one problem, but the interfaces are off-brand, ignore corporate design systems and WCAG 2.2 accessibility rules, and lack basic authentication patterns.

Restricting AI use for employees in reality means they pay the £18 monthly subscription out of pocket and still use it. It moves the spend somewhere finance can't see it, at least not immediately and not until it’s too late, adding a data governance problem on top.

The other consequence is harder to put on a balance sheet. A company where only engineers can create software moves at the speed of engineering's queue, but opening options up to a wider group - product management, analysts, marketers - means a company can speed up innovation while having greater visibility into it.

The cost of ungoverned building

The pressure is already on. Teneo's Vision 2026 survey of more than 350 public company CEOs and 400 institutional investors found that 53% of investors expect to see a positive ROI on new AI initiatives in six months or less, while only 16% of large-cap CEOs believe they can deliver on that timeline. The same CEOs report that fewer than half of their current AI projects are ROI-positive. Investors want returns on a clock that executives don't think they can meet.

Volume is not the problem. Goldman Sachs Research forecasts that token consumption will multiply 24 times, to 120 quadrillion tokens per month, between 2026 and 2030, while semiconductor providers are already delivering 60% to 70% annual declines in cost per token. Tokens are getting cheaper, and we are going to use far more of them.

But who - or what - uses tokens is going to change. Organizations are building and deploying AI agents who act on their own. And agentic AI burns far more tokens than good old human prompting, because the queries repeat in sequence: a chatbot request multiplied by 20, or even 50, running unattended. Send an agent to work on an unscoped problem, and it will keep going long after a human would have stopped to ask whether this was worth it.

A lot of those tokens go into prototypes that never get to see the real world. Prototyping and sifting through ideas is how software evolves, but building with no eye on production drains resources. Gartner found that at least 50% of generative AI projects were abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs, or unclear business value. If a prototype works in isolation but has to be scrapped because engineering would need to rebuild it from scratch, that's pure waste.

Then organizations have to pay the bill twice - for inflated tokens and lost engineering hours. When teams build with generic LLMs and no guardrails, senior engineers spend hours cleaning up the output. That deepens the trust gap in technical teams. Stack Overflow's 2025 developer survey found that 46% of developers distrust the accuracy of AI output. "Almost right" (but not quite) AI answers are the biggest frustration for 66% of developers, while 45% report that debugging AI-generated code is more time-consuming.

Access for everyone, accountability from everyone

The speed of code generation isn't the problem. Lowering the technology barrier for non-engineers produces better software, software shaped by the people who understand the business need. AI building is an edge, and an edge works best when everyone has it.

But alongside democratized access to technology, we need new ways to think about accountability for how you use it and the associated costs. Modern corporate culture can promote unnecessary complexity in how we think and speak about our work. Being able to explain in simple words how you spent your day is a useful exercise, and we should apply the same thinking to AI spend.

So, stated plainly: tokens should be used for a job somebody actually needed. In turn, the user needs to be able to explain, beyond the buzzwords, what the purpose was behind a particular token spend.

To work, rather than become another source of corporate frustration, this accountability needs to be part of the environment offered. One way to start is to retire the do-everything chat box as the default. Instead, organizations should start thinking about how to give people task-specific agent workflows tailored to their roles, and build execution limits into the software environment.

This way, non-technical staff get brand guidelines, fonts, security rules, and components pre-coded into the environment, so the apps they build are compliant and production-ready by default. They don’t need to file separate requests because the shape of the environment already includes the right guardrails on security and spend.

What it looks like when it works

Bounded workflows pay off in ways you can measure. In a recent mid-sized company implementation, a 3-person data analytics team built an agent that analyses data across all their systems; it saved each of them 12 to 13 hours of manual work a week.

Three analysts now produce what once took a team ten times their size, and their daily job also got way more interesting. They spend their time on high-value strategy instead of gathering data by hand.

That agent has a clearly scoped job and executes it in line with pre-defined parameters - it is not a chat box with a company credit card.

Bringing together abundance and accounting

Uncontrolled AI costs and falling trust in technology have the same solution, and it is not rationing of tokens. Tokens will be getting ever cheaper, and their abundance is the point of this technology.

The answer is in defining and understanding the scope of AI. Give everyone the tool but make it purpose-specific from the start rather than a generic catch-all.

Every company I talk to can tell me what it spent on AI last quarter, but when I ask what they wanted to get in return, the answer is murky.

This lack of understanding of what job AI should do for an organization is the problem, and the surprising bill is only its symptom. The first step is for the organization to articulate, in the plainest words possible, what the tools are for before the tokens start moving.

Store your data in the cloud with the best cloud storage.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

It's the largest cryptocurrency theft in history: hackers from North Korea allegedly swiped $6.36 billion in funds from Binance in the space of just 20 minutes, which has left customers reeling and the financial industry in shock.

The heist raises all kinds of questions about how this was allowed to happen, what might happen next, and whether Binance customers are going to get any of their stolen money back.

Even if you never owned any cryptocurrency before, there are bigger questions too — like what this means for global financial stability, and what the aftershocks might be for everyone outside the Binance world.

Looking for a quick five-minute primer on the Binance hack? We've been speaking to industry experts to get some answers...

1. What actually happened?

The basics are now pretty clear — $6.36 billion in cryptocurrency was stolen from the Binance exchange in the space of just 20 minutes. Security analysts are pointing the blame at North Korea's TraderTraitor hacking group, which has a history of this kind of heist. What's less clear is exactly how the hackers got access. Early reports identified a Binance employee's compromised laptop as a potential way in, though it now seems there's more to the story.

"A compromised laptop may explain where an attack started, but it does not, on its own, explain how billions of dollars could leave an exchange," Pieter Arntz, researcher at cybersecurity firm Malwarebytes, told TechRadar. "The real question is what allowed that initial foothold to become access to withdrawal systems — and which safeguards failed to stop unauthorized transactions."

"According to blockchain analysts, the attacker spent at least a day and a half preparing, including testing their tools on public blockchains where anyone could see," Ben Bernstein, manager of cybersecurity advisors at Huntress, told TechRadar. "Then most of the $6.3 billion left Binance's wallets in about 20 minutes. That's common in attacks we investigate. The attacker does most of the work long before anything is stolen."

2. Will anyone get their money back?

Binance has promised to reimburse customer losses through its backup and emergency funds, but in the meantime we're seeing deposits and withdrawals frozen on the platform while the hack is investigated. It's a warning that even the largest and most sophisticated exchanges can still be vulnerable.

"Binance says it'll cover every cent," Nigel Green, CEO of the deVere Group, said in a press statement. "I hope so. The emergency fund is around $1.2 billion and the gap is several times bigger, so the rest has to come from the company’s own balance sheet. The market will want to see the numbers, and the new proof of reserves can’t come soon enough."

"Long term holders have just been reminded, very expensively, why keys kept offline matter. Crypto has spent years pushing back against heavy regulation, and plenty of those arguments were fair. This one's harder to argue with. Platforms holding billions of other people's money need enforceable security standards. I expect regulators in the US, Europe and Asia will likely move faster after this week," he added.

In the meantime, the US Secret Service has said in a statement that it's "working around the clock" with partners in the private sector "to identify, freeze, trace, and recover funds" that have been stolen from Binance.

3. What follow-up scams might appear?

Whether you're a Binance customer or not, you need to keep your wits about you over the next few days. It's likely that scams trying to literally cash in on the hack will start appearing — perhaps warning you that you need to quickly claim refunds, lock your account against a particular threat, or go through a bogus verification process.

"Criminals routinely use major breaches as a hook for phishing campaigns," Dray Agha, senior manager for tactical response at Huntress, told TechRadar. "Consumers should be highly suspicious of any emails or messages claiming to be from Binance offering urgent account recovery steps. You must only check your account status through the official application or website rather than clicking links in unsolicited messages."

According to the team at Huntress, you should watch out for social media posts, emails, and text messages that use the Binance hack as a hook — perhaps a fake news story suggesting that now's the time to buy crypto, while prices are down.

As always, think twice about following links or entering sensitive information into online forms and portals you're not familiar with.

4. How does this affect the average consumer?

If you don't own any cryptocurrency, the good news is that you're largely going to be unaffected by what's happened here. There's not going to be a stock market crash, or a run on the banks, or a global recession — not because of this hack, at least. However, it's always worth weighing up your own personal privacy and security setup.

"If you don't own crypto, the direct impact on you is small," Ben Bernstein from Huntress told us. "Your bank account and your card are not affected by this. The bigger takeaway for consumers is that attackers often get in by targeting people rather than technology, through a convincing email, a fake login page or a phone call from someone posing as IT support," he said.

"That makes security awareness everyone's job, at home and at work. If you pause before you click a link, approve a login or share a code, you're making it harder for an attacker, whether they're after your savings or your employer's systems," he added.

5. What does this mean for the future of money?

This clearly raises a lot of questions about the safety of cryptocurrency and the exchanges that it sits on — and of course it's far from the first crypto heist we've seen. In the longer term, it may lead to less confidence in crypto as a whole, but the experts we spoke to don't think it will bring the industry down. What will be important is making sure the risks of this happening again are minimized as much as possible.

"Despite the scale of the theft, this is not a threat to global financial stability," Feike Hacquebord, Principal Threat Researcher at TrendAI, told TechRadar. "Around half of the stolen funds have already been frozen, while the regulated banking system is well equipped to absorb a shock of this size. This is a major crisis for Binance, not a crisis for the global financial system."

"The change I’d like to see, for crypto firms, banks and every other business, is working on the assumption that someone will get in eventually," Ben Bernstein told us. "Keeping attackers out matters, but it isn't enough on its own. Companies also need someone watching around the clock, especially for logins and account activity that don't look right, so they can find and remove an intruder before the money starts moving," he added.

Long before anyone trained a model on customer data, the marketing data business worked out the terms for pooling it. Catalog retailers in the 1990s joined data cooperatives that combined their transaction histories so each member could find new buyers, and the arrangements held together because the rules were blunt.

A retailer contributed data to receive modeled prospects; no member got a rival's customer file; and a company that stopped contributing stopped receiving. Card fraud consortia run on similar give-to-get terms. Nobody would call these systems elegant, but every participant understood the trade.

Software vendors building AI features are now running into the same question, usually without the rules. Traditional software sat on top of a customer's data and operated on it. AI products learn from data, and on a multi-tenant platform, that learning rarely stays within a single account.

One company's records can sharpen the enrichment, lead scoring, and recommendations delivered to thousands of others. A vendor can improve its product for every customer by relying on the work of a few, and most contracts say nothing about whether that trade is fair. The customer still owns its data in any legal sense.

Ownership alone doesn't answer who benefits.

Users push back on sharing hard-earned data

HubSpot learned how touchy the question has become when it announced plans to pool enrichment data across customer accounts, including business contact details, employer information, and email deliverability signals, to improve prospect records platform-wide.

Every account was in by default. Within days, the company dropped the plan and promised that any future enrichment feature built on customer data would be opt-in. HubSpot had stated clearly that core CRM records would never be shared without permission, so privacy in the usual sense wasn't what upset the sales and RevOps teams, who pushed back. Their concern was harder to answer. Years of work spent cleaning and enriching their own contact data were about to improve the prospecting tools available to every other customer, competitors included, and nobody had asked what they would get for it. The product terms that allowed the change had been updated nearly two years before most customers heard about them.

AI creates accidental data companies

Plenty of software companies adding AI tools have never had to think this way. They are becoming data companies one model at a time, often without deciding to. Whether a customer's information feeds a shared model is often determined by a product team working toward a launch date rather than by anyone who owns the commercial relationship, and the answer is expressed in terms few buyers read.

McKinsey's 2026 survey on AI trust found that only about 30% of organizations have reached a mature level in AI strategy, governance, and controls. Pooled training data tends to fall between legal, product, and sales, with no single team feeling responsible for explaining it.

The need for strong exit clauses

Buyers will force the issue long before regulators do. Security certifications and uptime guarantees became standard contract language over the past decade because enough customers asked for them that vendors stopped arguing, and AI data terms are heading down the same road. Procurement teams are starting to ask whether their data trains or enriches models used by other customers, whether participation is opt-in, and whether it can be switched off later without losing core features. The more demanding buyers want to know what a contributing customer receives that a non-contributor doesn't.

Exit terms will be the hardest part to write. A customer can cancel a contract and demand its records back, but data that has already shaped a production model can't be cleanly pulled out again. Retraining is expensive, and removing one customer's influence from a live model remains an unsolved problem for most vendors. A vendor can honestly say it deleted every record a customer ever uploaded and still be running a model those records helped build. Contracts written for storage assumed that deleting data ended its use. Contracts written for AI will have to say plainly what persists after a customer leaves, and whether the customer agreed to it when it signed.

Opt-in consent answers part of the question. A checkbox tells a vendor it may use the data, but it says nothing about the return, and a customer who opts in for nothing in particular has made a poor bargain, even if it's a legal one. The old cooperatives signed up members because members could see exactly what they were getting.

Vendors that explain their data economics before customers ask will have an easier time at every renewal. The explanation belongs in the product and in the contract, and it should cover which customer data feeds shared models, what those models produce, and what contributors get back. Security questionnaires were once a formality, until buyers made them a condition of the sale. Pooled data terms are on the same schedule, and vendors that write clear ones now will set the standard that the rest of the market will follow.

Ask a room of executives about agentic AI and within about ninety seconds someone will ask how many agents they ought to be running. It has become the default measure of progress: a number you can put on a slide, benchmark against a competitor, and report upwards.

It is also the wrong question.

Across the 200-plus workshops my team and I have run with enterprises in Europe and the Middle East, the organizations getting real value from AI tools are not the ones with the longest agent inventory. They are the ones who never framed the problem that way to begin with. They started by identifying a business problem or point of friction and worked backwards. Everyone else started from the technology and went looking for somewhere to put it.

There are four questions I hear constantly that are quietly steering good companies into Pilot Purgatory. Here they are, and here is what I would ask instead.

Don't ask how many agents. Ask what outcome you are improving.

Much of the last two years of enterprise AI has been about helping people work faster. Copilots that answer questions, assistants that summarize the meeting you have just sat through, tools that get you past the blank page. These are genuinely useful and I would encourage everyone to switch them on.

But speed is not the same as value, and agent count is not the same as either. No enterprise ever created value by accumulating agents. Value shows up in real business outcomes such as fewer payroll errors, faster shift scheduling, lower supplier sourcing costs, better win rates, cash collected sooner.

That is the difference between an agent and an agentic application. An agent help with an individual task. An agentic application coordinates teams of specialized agents to deliver a business outcome. If you cannot name the KPI an agent is moving, you do not have a use case. You have a demo.

Don't ask whether AI can recommend. Ask whether it can finish the job.

The most persistent confusion I encounter is between advice and execution. Plenty of systems can now sense a situation, spot the pattern and tell you what should happen next. That is real progress. It also leaves almost all of the work exactly where it was.

Take sourcing. An assistant can flag that a category is overdue for renegotiation and suggest you run an event. Someone still has to create that event, chase the responses, score them against commercial criteria, route the approvals correctly, update the contract and shepherd the whole thing to a close.

An agentic application does that work, under policy, with human oversight and approval where required, and with a record of what it did and why.

The test is simple. At the end of the interaction, is the process further forward, or do you just have a better-informed human facing the same queue?

Don't ask which model tops the leaderboard. Ask whether it understands your business.

Benchmark scores dominate the AI conversation and matter far less in the enterprise than most leaders assume. Raw model intelligence is rarely the binding constraint.

A frontier model understands procurement as a discipline. It does not understand your supplier list, your delegation-of-authority matrix, your contractual obligations, your current inventory position, or the three exceptions your finance director signed off in March. Without that, sophistication amounts to a well-argued guess.

Enterprise context is the differentiator: trusted enterprise data and business context, including the policies, approvals, transaction history, organizational structure and live process state that already sit inside your core systems. The closer agentic capabilities are to the systems where business processes, data, security, and approvals already live, the more effectively agents can understand context and act within establish controls.

Don't ask for one system that does everything. Ask how the specialists coordinate.

There is a persistent hope that enterprise AI will eventually resolve into a single, omniscient digital assistant. Businesses do not work that way, so their AI will not either.

Organizations are ecosystems of specialists: finance, HR, supply chain, service, sales, each with its own expertise, obligations and objectives. The equivalent in AI is not one agent that knows everything, but coordinated teams of agents that know their own domain and work towards a shared goal.

Hiring is a good illustration. One agent surfaces and screens candidates. Another prepares onboarding. A third confirms the budget is genuinely available. A fourth keeps the hiring manager's approvals moving. Individually, each is modestly useful. Together they move a business process from vacancy to first day. The value sits in the coordination, not the components.

Systems of outcomes

What connects all four questions is a shift in what enterprise software is actually for. For decades our applications recorded what had happened. We are now entering the era of systems of outcomes: software that reads the state of the business, reasons over it, and acts to keep work moving towards a defined objective.

That shift demands different questions from leadership. Not how many agents, but which outcome. Not can it advise, but can it execute. Not how clever is the model, but how well does it know us. Not one brain, but one team.

Get the questions right and the technology decisions become considerably easier. Get them wrong and you will assemble an impressive collection of agents, tools, and models that nobody can tie to a number that matters.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Transparency duties under Article 50 have begun to apply, even as the EU AI Omnibus implementation for many high-risk systems has moved to December 2027 and, for certain AI-enabled products, to August 2028. For organizations building and deploying AI in Europe, the message is clear: regulators extended the timeline on complexity, but not on transparency.

While some organizations may view the extension as welcome breathing room, the time would be far better used preparing for the future of AI. To do that, we need to frame the challenge around creating the conditions for long-term innovation, rather than around compliance deadlines.

As AI becomes embedded in products, services and decisions that affect people's lives, organizations need confidence in the systems they are deploying, and users need confidence in the organizations deploying them. That work cannot begin in 2027. It needs to start now.

Compliance is achieved at a point in time; governance is built as a capability. A deadline can be moved, but an institutional capability cannot simply be created at short notice.

Start with the AI already in use

The starting point for every organization should be an honest assessment of whether it has a reliable picture of the tools, outputs and interactions that Article 50 covers. As we have seen in recent years, AI systems can quickly become more deeply embedded in business over time as employees explore new tools and suppliers innovate and ship new updates.

This natural expansion is a normal part of adopting a new technology, but it can result in AI becoming a core part of business processes before the proper governance is in place. Governance is most effective when it shapes those decisions from the beginning.

Attempting to retrofit governance after AI has become embedded is significantly more difficult, more expensive and ultimately less effective, and the resulting loss of trust can be more costly than the upfront work required to prevent it.

Extend the controls already in place

The good news is that many of the disciplines required by the legislation should already be familiar. They build on existing practices around privacy, security, risk management, accountability and operational governance, and much of the challenge is around applying established governance disciplines to a rapidly evolving technology landscape.

Organizations that have taken data protection regulation seriously, and already understand impact assessments and effective governance processes, are well positioned to meet many of these requirements – there is considerable overlap.

The task is to extend them for AI, not replace them. A privacy impact assessment can identify data risks, but it cannot on its own establish whether a model is dependable in a particular context, whether someone can meaningfully question its output or whether an automated process disadvantages a group.

Those questions should be built into existing workflows, particularly when AI is bought rather than built. Before a capability is embedded in software or a critical process, organizations need clarity on data treatment, service and model changes, incident response, support for transparency duties and the evidence a provider can supply. Even where the underlying model comes from a third party, organizations still need to govern how it is selected, configured and used.

The sovereignty question

Alongside the AI Act, the other crucial question impacting the way we think about governance in Europe is sovereignty. As the debate gains momentum across Europe, the practical issues raised are becoming clearer. Where is data stored? Which providers do we depend on? How resilient are our technology choices? What happens if commercial, legal or geopolitical circumstances change?

It is worth being precise about what sovereignty can realistically mean for enterprises. Full compute sovereignty, including control over chips, infrastructure and supply chains, is far beyond the reach of most organizations. The more practical goal is compute resilience: understanding where critical dependencies sit, ensuring no single vendor, cloud provider or model becomes a point of leverage, and retaining control over the layers of the AI stack that remain within the organization's influence.

Open source can support this by providing greater transparency and control over code, but it does not remove infrastructure dependency on its own. An organization may use an open model and still rely on external compute or cloud capacity to operate it.

Organizations should adopt a sovereign-by-design approach to innovation. Sovereignty should not be considered after technology decisions have been made, it needs to be embedded into decisions on data architecture, vendor selection and risk management from the outset, with resilience, not full independence, as the realistic target.

It is also important to note that not every dataset, process or workload carries the same level of risk. Organizations should take a tiered approach, applying the strongest sovereignty requirements where data sensitivity, regulatory exposure or operational importance demand it.

These are not abstract concerns. Viewing AI governance and sovereignty as interconnected challenges creates an opportunity to build a more coherent approach. This is not a case for technological isolation or for stepping back from global technology ecosystems, the aim is narrower than that. It is about ensuring organizations retain meaningful freedom of action in an increasingly complex and uncertain environment.

The real opportunity

The debate about innovation, regulation and sovereignty will continue. Europe will keep refining its approach to AI, and organizations will keep adapting to new technologies, new providers and new risks. The more important question is whether organizations use this window to build the capabilities they will need long term, rather than treating each regulatory milestone as a separate deadline.

Governance should not be viewed as a compliance exercise that slows innovation. It is an institutional capability that enables organizations to innovate sustainably, adopt new technologies with confidence, and remain resilient as markets, regulations and geopolitical conditions evolve.

The ultimate goal is innovation built on trust, and we have the ability to deliver that regardless of regulatory deadlines. The real test is not whether an organization built every part of the system it relies on, but whether it understands its dependencies well enough to change course if circumstances demand it.

Store your data in the cloud with the best cloud storage service.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

I review tech for a living, so asking AI for the “best website for tech reviews” came with a small risk of bruised pride. Would TechRadar get a mention, or would the answer send readers straight to the competition? That was one of the questions I put to HubSpot AEO.

HubSpot’s answer engine optimization tool tracks selected questions across ChatGPT, Gemini and Perplexity. It shows which brands appear, which pages the answers cite, and suggests content work based on those findings. What interested me was being able to read the answers behind the score, then check the sources for myself.

I used the free trial to set up TechRadar and track 11 of the 25 available prompts. My hands-on review covers setup and reporting, with product demonstrations and documentation filling in the workflows I didn’t try. One day of results can’t prove long-term gains, but it gave me enough to judge whether the tool helps make sense of AI search or simply adds another chart to the working day.

HubSpot AEO: Plans and pricing

The standalone price for using HubSpot’s AEO tool is $50 a month, or $45 a month billed annually, which comes to $540 a year before tax. You also get a 28-day free trial that requires no credit card; there’s no need to buy another paid HubSpot product first.

The AEO features are also included in Marketing Hub Professional and Enterprise plans. And, the HubSpot for Marketers option bundles Marketing Hub with Content Hub and include AEO too. Those prices cover the wider software package, and not an AEO surcharge.

Now, if you’re just on the AEO plan and want more credits, it costs $20 a month for another 10 prompts and 1,000 monthly answers. A standalone account tracking 45 prompts therefore costs $90 a month on monthly billing. Not to forget, tracking another brand requires a separate AEO instance.

Content generation has a separate cost too. Generating a piece from an AEO recommendation requires Marketing Hub Professional or Enterprise and uses 1,000 HubSpot Credits. Standard Professional plans include 3,000 monthly credits, while Enterprise includes 5,000. Extra credits cost $10 per 1,000, and unused monthly credits expire. CRM-based prompt suggestions also require Marketing Hub Professional or Enterprise.

There’s one pricing wrinkle, though. Accounts enrolled in HubSpot’s flexible seats-and-credits program have separate terms, including two credits per AEO answer. That is a different model from the standard prompt allowances above. So, do check which model your account uses before budgeting.

HubSpot AEO: Setup

I signed up with my email and answered a question about my role. From there, HubSpot took me through the AEO setup one step at a time.

Next, I entered TechRadar’s name and website address. That led to the competitor and audience settings, where I could narrow down what I wanted to track.

I reviewed the suggested competitors and chose TechRadar Streaming, focusing on deal-seeking shoppers in the United States. I wasn’t convinced by every competitor suggestion, though. Some were publishing companies rather than the review sites readers might compare.

HubSpot used those choices to suggest questions to track. I could edit or remove prompts before continuing. Several asked much the same thing, so I found the suggestions useful for getting started, but too narrow to cover different reader needs on their own.

I then started the checks and followed HubSpot’s progress as it worked through the prompts. From there, I could open the dashboard and start exploring the answers, brand mentions and cited sources.

HubSpot AEO: Features and specs

Prompt tracking

The panel also separates Answers, Prompt Analysis and Recommendations. I like that structure because a weak score has somewhere to lead like you can inspect the answer, compare the sources, then consider a change. These are repeatable checks of chosen questions, not a log of what every buyer asks.

Brand visibility and competitor analysis

The earlier dashboard screenshot showed 6.67% brand visibility while the account held 10 prompts. Its date range ran from October 27 to October 27, with a single point on the trend chart. That was a baseline, not enough history to judge progress.

The broader review-site question above comes from the later 11-prompt view. Comparing that single question with the earlier average would be misleading. The contrast shows how much the choice of questions matters, not that visibility suddenly improved.

Visibility measures how often the brand appears in the tracked answers. Share of voice compares its mentions with the selected competitors. Filters for engine, product, audience and buying stage help explain where a gap sits. And, Sentiment adds context about how the brand is described, rather than just whether it appears.

For TechRadar, I also need to separate being named as a recommended review site from having a guide cited in an answer. Both can be useful, but a mention score alone doesn’t capture that difference.

Citation analysis

This is the report I find most useful for planning content. A mention tells me the brand appeared. Citation analysis shows the pages and domains used as sources, including sources owned by other businesses.

My Citations screen splits the results by content type and channel. The first view concerns formats, such as blogs, comparisons, and documentation. The second concerns where the material comes from, with categories including owned, earned, peer, review-site, and user-generated sources.

Earned and review-site sources occupied large parts of my channel chart, while owned content had a thin slice. I read that as a reason to inspect outside sources before assuming another page on our own site is the answer. These are sources across the tracked prompts, though. They don’t all necessarily mention TechRadar.

The wider report lets you inspect top domains and individual URLs, and compare owned-domain citations with citations whose source pages mention your brand. That helps distinguish if you should improve your own pages or earn coverage elsewhere. Neither count proves that someone clicked through or bought anything.

Content recommendations

HubSpot uses the prompts and cited sources to suggest workarounds for owned content, social amplification, and outreach. Each recommendation has a content type, channel, and priority. Opening it adds a suggested title, summary, audience, and keywords, with the prompts and citations behind the idea.

Suppose a tracked question keeps citing third-party comparisons. Here, updating a relevant comparison page or seeking inclusion in one could be more useful than publishing another broad guide. This is exactly where the source list helps make that choice.

The recommendation can then be marked complete. I see this as a useful brief for a writer or outreach lead, rather than a finished strategy. Someone still needs to check the cited pages, decide whether the suggestion fits the business, and do the work.

Content generation and impact tracking

For eligible Marketing Hub accounts, the Generate asset feature takes an owned-content or social recommendation into Content Agent. It uses the brief alongside brand voice, audience information, and cited pages. The draft appears in the recommendation’s Content tab, ready to open in the relevant editor.

I like the direct handoff because the source research and writing brief stay connected. It still needs an editor. That’s because a draft can follow a useful outline and still remain thin on facts, examples, or a reason to choose that business.

After publication, the Impact tab accepts the page URL and tracks whether it earns citations for the target prompts. This gives the recommendation a follow-up check instead of ending at “published”. It doesn’t prove that the tool caused an increase, and it isn’t sales attribution.

Is HubSpot AEO worth it?

I recommend starting with the trial for one clear product, audience, and set of questions. My setup exposed the main risk early: a tidy dashboard can still rest on repetitive prompts and a questionable competitor list. Fix those inputs before spending time on the score.

For a small team, the standalone plan is a reasonable way to replace repeated manual checks and find useful source pages. Existing Marketing Hub Professional and Enterprise users get more value from the connected content tools. That’s why I don’t see a case for buying the wider suite solely for AEO.

The buying decision comes down to what the reports change. Find a worthwhile page to improve, a source to approach, or a buyer question you’ve missed. Without that next step, more daily data won’t earn its monthly fee.

FAQs

• YouTube announced it's hit 150 million subscribers across Premium and YouTube Music
• The platform also generated $11.2 billion in ad revenue
• Users say free tier ads are only getting worse, with some calling them 'unbearable'

YouTube announced that it has surpassed a whopping 150 million subscribers across its Premium and YouTube Music experiences — marking a huge milestone for the video platform and one of the best music streaming services.

Alphabet Inc CEO, Sundar Pichai, announced the figures in the platform’s Q3 earnings release on Wednesday (October 28), which increased by 25 million from a total of 125 million subscribers in March 2025, though individual numbers for Premium and YouTube Music weren’t disclosed. Pichai also shared that YouTube’s ad revenue rose to more than $11.2 billion compared to $10.3 billion in 2025.

Speaking about the milestone, Google’s Chief Business Officer, Philipp Schindler, said "With Premium and Premium Lite, viewers have choice. As more people opt into premium experiences, it’s a win for viewers, creators, and our business. Subscriptions help fund the creator ecosystem and drive the platform forward."

YouTube has accumulated over 25 million subscribers between now and when it last confirmed its total membership count in March 2025. When you break that down, that’s approximately 1.25 million new subscribers signing up each month, and it’s only going to increase.

While it’s not a traditional on-demand film and TV content platform, YouTube has emerged as a serious competitor among the best streaming services. In 2017, the company launched YouTube TV, and according to a recent report from Nielsen, it has surpassed Netflix in popularity for TV viewing. In July 2026, 49% of TV time was viewed via streaming platforms, with YouTube making up 14.2% of that, followed by Netflix at 7.8%.

There’s no denying YouTube Premium is rising in popularity, but as the company puts more effort into improving its paid services, a number of users have said its free version is becoming increasingly difficult to use because of the number of ads.

Earlier this year, I bit the bullet and signed up to a YouTube Premium free trial and, annoyingly, the experience was vastly more enjoyable. Not only does it grant you ad-free viewing, you get background and offline listening, as well as small but handy control tools and full access to YouTube Music. But once you go back to its free version, that’s when you truly realize how out of control YouTube’s ads have become — and I’m not the only one who’s noticed (see below).

In addition to YouTube’s ‘completely unbearable’ ads, users have also pointed out the rise in AI-generated sponsored content which one Redditor claims advertise "obvious scam products". The combination of the two is the main catalyst for users abandoning YouTube as a platform altogether, and tech-savvy users are already opting for alternative ways to bypass ads.

One of the most popular ways YouTube viewers are working their way around ads is downloading browsers such as Brave, which users on Reddit have praised. Browsers like this come with built-in ad blockers, so it can automatically skip over ads when you’re streaming YouTube for free.

While YouTube gives you the option to tailor your ad experience and block certain types of promotional content, some users claim that when they do this, YouTube only feeds these ads to them even more. One user on Reddit used Old Spice as an example, saying "it's like blocking them has a reverse effect." Similarly, another user had an experience where they reported a potential scam ad, only for YouTube to email them with a counterargument.

I’m with a lot of other users who continue to call out YouTube for how expensive Premium has become (it now costs $15.99 / £12.99 / AU$22.99 a month), especially when you factor in your existing streaming subscriptions, so many of us just put up with its free service. Despite enjoying what Premium had to offer during my free trial, it doesn’t justify adding another subscription plan to my roster — my bank account will fall out with me.

• Windows 11 has a new preview (optional) update
• This introduces the ability to fully tailor the right-click (context) menu to your liking, as well as speeding up its performance
• That's great, but you're best off steering clear of a preview update and waiting for the full release in under two weeks – but even then, this feature will be on a gradual rollout

Windows 11 now has one of the most-wanted interface capabilities for many people, namely the ability to customize the right-click menu – but it's not quite fully finalized yet (although that's imminent).

The menu that pops up when you right-click on a file (or folder) is also known as the 'context' menu as it presents a bunch of options that vary contextually based on the type of file clicked – but in Windows 11, it's currently a sprawling affair.

Those who hate the lengthy right-click menu will now be able to customize it after installing a new Windows 11 update, although this is an optional release, meaning it's still in preview. This feature will be in next month's patch, the November update, though, so you haven't got long to wait for the full release (less than two weeks, in fact).

The customization settings for the right-click menu include the ability to remove Windows commands you're not fond of – like 'Send to' for example – or add abilities that you'd like to have, and also choose the apps you want to appear in the context menu.

It's a full range of customization and you can also elect to have the top panel of the menu (cut, paste, rename and so forth) directly integrated instead, so these functions appear alongside everything else (in a Windows 10 style – and you can shift the layout around too, moving Properties to the foot of the menu for example).

Microsoft boasts that the new right-click menu is faster, too: "The updated menu opens more quickly, reduces clutter, and gives you more control over the commands and app extensions that appear when you right-click."

As you may have noticed, all this work has been in testing for a while now.

Analysis: rolling frustrations and a lengthy wait

The immediate response on the likes of Reddit seems positive, and there are Windows 11 users who have taken the plunge with this preview update who are observing that the right-click menu is indeed a lot faster.

One Redditor enthuses: "The new context menu is so good! Way faster, customization is cool."

Others acknowledge that this is undoubtedly a good move, but don't understand why it took Microsoft so long to get here: "Not bad imo. Crazy it took them 5 years though."

While the feedback is generally positive, there are complaints from those who have installed the update but haven't received the new context menu. This is because the right-click revamp has been implemented via a gradual rollout, so it's only being drip-fed through to a certain amount of PCs at a time (of those who have installed the optional update). As another Redditor puts it: "Those gradual rollouts though…"

You can force the feature to be deployed to your PC using a Windows configuration utility (ViVeTool), but I wouldn't recommend that approach. Microsoft tests the waters in a limited way for good reasons, especially with features still in preview, so you're running a larger than normal risk of some nasty bug ruining your day.

Anyway, you've waited five years for this performance and options tuning, as those on Reddit are keen to point out – so what's another couple of weeks? I'd hold on for the official November update, keeping clear of this preview, but even then, it'll still be a gradual rollout, and you may need some patience.

Still, the good news is that the context ball is rolling, and it hopefully won't be too long before the majority of Windows 11 users gain access to this important upgrade for a key piece of the UI.

Modern digital security frameworks were originally built to correct a flawed historical assumption: the belief that if a user or system was inside the company network, they could be trusted implicitly. Over the past decade, the rapid rise of cloud storage, subscription-based business software, and remote working destroyed that network boundary.

In response, our security models shifted focus. We began verifying the identity of the person trying to log in, inspecting the hygiene of their computer, requiring multiple verification steps during login, and continuously checking that they were who they claimed to be.

This model works well when a human is logging in to perform a task. But with the rise of autonomous AI agents, the real security risk no centralized at the login screen. It starts after access has been granted.

An AI agent does not behave like a standard piece of software. It reads a human instruction, pulls context from company databases, utilizes multiple sets of tools, chooses its own digital tools, connects to external software communication links, and triggers complex automated workflows.

Verifying the initial login simply tells you that the agent is officially recognized by your system. It tells you absolutely nothing about whether the next action the agent plans to take is safe, ethical, or correct, focus has shifted to context.

The evolution of the machine identity problem

IT teams have already experienced a simpler, quieter version of this challenge with non-human identities. For years, companies have used automated service accounts, security tokens, and digital certificates to allow different software programs to talk to one another.

Most security leaders are all too familiar with the risks these credentials present: accounts are created for a temporary project, granted permanent access to database files, and then completely forgotten long after the project has ended. They become silent, unmonitored backdoors into the network.

However, AI agents complicate this familiar problem by introducing the element of independent judgment.

A traditional service account follows strict, pre-written code. By contrast, an AI agent interprets a high-level goal, analyses real-time context, selects its own tools, and actively decides how to proceed.

As a result, security policies can no longer stop at the front door. We must protect every internal data or compute asset the agent can reach and monitor every single decision it makes.

Matching autonomy with executive control

To manage this risk, security leaders must separate two distinct concepts: how much independence an AI agent has, and what level of file access it holds.

The combination of these two factors dictates the level of control required. For example, a basic agent that merely summarizes long documents only needs limited reading rights and clear activity logs. However, an agent designed to update sensitive, production databases, access code repos, write new types of code, or alter business workflows requires mandatory human approval steps, strict operational boundaries, and a clear path to reverse any mistakes.

We can categorize these digital workers into 3 distinct levels of independence, applying tighter security controls as their autonomy increases:

1. First, there are Conversational AI Agents (GenAI), productivity agents usually built and managed by employees & teams as part of their productivity (e.g. managed agents, self hosted chat bots). When driving security, we must enable proper RBAC authorization, associated with the employee and team, ensuring access to non-sensitive or production-related data, enabling logging, and validating the response to ensure proper integrity.

2.Second, there are Agents Acting with Semi/Full Human Guardrails (AI Agents), such as customer support tools that draft emails or prepare database updates for review. Here, security teams must validate the parameters of the tools being used and require a human to approve any actions involving writing, deleting, spending money, or sending information externally. All above controls apply

3.Finally, there are Fully Autonomous Agents (Agentic AI), which reset passwords or trigger complex workflows on their own. These require the strictest boundaries, from visibility (agent architecture, tool usage, and more) and focus on active set of controls. The controls include strict rate limits, continuous real-time policy checks, network access restrictions, run time security controls (Public/Private Cloud) continuous identity and minimal authorization verification, and finally complete, unalterable audit trails.

The industry is leaning toward autonomous agents seeking human approval upon abnormal behavior, can that be triggered in your agent?

The need for real-time inspection

To enforce these policies, security teams need a dedicated vantage point within the network. They must be able to inspect the human prompt, the agent's response, the digital tools being used, and the movement of data while the work is actually happening.

This requires an architecture that can discover all active agent activity across the business, govern how those agents use internal tools, and maintain complete visibility over data flows and automated workflows.

The true value of this approach lies in context. When the behavior of an AI agent is viewed alongside network traffic, user activity, and cloud database logs, security teams can investigate incidents with clear evidence. They can instantly see which agent ran, who in the organization owns it, what data was moved, and which automated tool was called.

For business leaders, this means cleaner, more consistent governance across human users, software applications, and digital agents alike.

How to begin: Start with a single workflow

Securing this new landscape does not require a complete overhaul of your entire security department overnight.

Start by picking a single AI workflow that is currently in use or close to being launched. Map out its digital identity, its level of independence, the files it can access, the tools it uses, the points where a human must approve an action, and how you would undo a mistake.

This simple mapping exercise will quickly highlight where your security gaps lie. If you cannot see the agent's actions in real-time, approve its decisions, block its access, or explain its behavior after the fact, your security controls are simply too far away.

From there, build a baseline on how your secure your coding agents, AI harness (or AI orchestrator) up to utilizing of tools/skills/MCP and moving forward to data assets access and utilization of identity that will allow you to set the correct foundations and guard rails.

The core principle of a "never trust, always verify" security model must follow the agent into its daily work. The most meaningful decision is no longer who is logging in, but whether their next action should be allowed to happen, under these exact conditions, right now.

We've featured the best endpoint protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Pixelmator Pro gets a new AI-powered Generate Fill tool
• Keynote, Pages, and Numbers add AI assistants and OpenAI model support
• Final Cut Pro gains thousands of MotionVFX creative assets in November

Apple's Pro and Tell week is marching on, and alongside the reveal of the M6-powered iMac and 14-inch MacBook Pro, the Cupertino-based giant is also shining a spotlight on software. Specifically, Apple Creator Studio is getting a wide-ranging update that brings a fresh batch of features and, ultimately, packs in more value.

As it stands, Apple Creator Studio is a software subscription priced at $12.99 / £12.99 / AU$19.99 a month or $129 / £129 / AU$199 per year, giving you access to Pages, Numbers, Keynote, Pixelmator Pro, Final Cut Pro, and Logic Pro across Mac, iPad, and iPhone. All that stays the same, including the especially excellent education discount — $2.99 / £2.99 / AU$4.99 per month or $29.99 / £29.99 / AU$49.99 per year — but there are some pretty big updates here, with AI playing a starring role.

Leading the charge is Pixelmator Pro, which is getting a new Generate Fill tool. This lets you select an area of an image, describe what you want to add or change, and let AI do the heavy lifting. It's the kind of feature that could come in handy when you're editing a photo and don't want to jump between apps to get the result you're after. It'll also help level the playing field between Pixelmator Pro and Adobe Photoshop by filling a feature gap.

Keynote, Pages, and Numbers are also getting AI assistants designed to help you create presentations, write documents, and work with spreadsheets. Most interestingly, you'll be able to use OpenAI models alongside Apple's own, opening up more possibilities for working with AI in these apps. The apps also support compatible external AI tools, potentially making them a more useful part of your broader workflow.

With Final Cut Pro, Apple is bringing MotionVFX content directly into its pro video editing suite, making thousands of creative elements, including transitions, templates, effects, and titles, available for use in your projects. It's another welcome addition that should let you get more done directly inside Final Cut Pro, rather than having to rely on other tools.

All of this is coming in November as a free update for existing subscribers, though Apple isn't sharing specific timing just yet. These updates should also be right at home on the new Macs announced this week, but I'm curious to see if Apple introduces any more surprises, like the much-rumored touchscreen MacBook Pro, during its Pro and Tell event tomorrow.

• Windows 11's internet speed test currently redirects to a Bing search to facilitate the test
• This is clunky to say the least — and an annoying piece of Bing promotion — but now Microsoft is changing things
• In a new preview build the speed test is fully integrated into Windows 11 — as it should have been in the first place

Microsoft has thought better of one of the methods it was using to push Bing in Windows 11 — via an internet speed test — and we can but hope that maybe it's seeing sense from a broader perspective about such promotional trickery.

Windows Central reports that, in a recent preview release for Windows 11 (build 26340.9767), Microsoft has applied some changes to the Quick Settings panel (accessed via the taskbar).

In the release notes, Microsoft says: "We're making it faster and easier to check the performance of your internet connection in Windows. Windows Insiders can now run an internet speed test and see the results directly in Quick Settings, without being redirected to a browser."

As it stands, when using this feature it simply opens an online speed test in your default browser — but it's in the Bing search page (giving Microsoft some web traffic, of course). With this change it's actually a native speed test that works in Windows 11, as it should always have been in the first place.

You can start the new speed test from Quick Settings, or access it by right-clicking on the network icon in the system tray on the right of the taskbar.

Note that this is just in testing for now, and it's still at an early stage (the Experimental channel), but it should eventually come to all Windows 11 users.

Analysis: fixing Windows 11 includes getting rid of this kind of nonsense

It's useful to have a built-in speed test right in the OS for those occasions when you're curious to learn what kind of online speeds you're getting — maybe you're streaming a video and it's buffering or glitching, and you're wondering if that's the fault of the service, or whether your internet connection has hit a rocky patch.

However, it's hardly convenient to be redirected to a web browser to have that test conducted.

Microsoft says of this change: "This updated experience builds on the internet speed test we previously introduced for Windows Insiders, bringing the test and its results directly into Windows."

Which makes it sound like the current implementation was always just a stopgap or a temporary stepping stone, but let's be honest about this: we all know it was another way that Microsoft was trying to push traffic to Bing.

It's far from the only time that Microsoft has tried this kind of trick, and most recently, we've witnessed the Bing Wallpaper app once again crowbarring in Bing searches for people who merely click on a blank part of their desktop (see the further explanation of that incident here).

No, the Bing Wallpaper client isn't part of Windows 11 by default, but it's nonetheless a piece of Microsoft software, and it demonstrates that the company's attitude is still leaning towards exploring annoying ways of promoting its own services — this is another great example — at the cost of the user.

As one Redditor observed of the current Bing-laden internet speed test in Windows 11: "This is the kind of stuff that pushed me to Linux. Overall it wasn't too bad, but Linux (Fedora in my case) feels like a breath of fresh air."

All of these kinds of practices need to end, and at least this fresh move in testing is a hopeful sign that Microsoft is finally realizing this, given that it's battling to fix Windows 11 and change public sentiment around its desktop operating system this year — needless Bing promotion certainly won't help that cause.

• A new Windows 11 feature is in testing, but it doesn't work yet
• It's an option to enable Intel's 'Advanced Performance Extensions' or APX to boost the performance of some apps
• This will benefit Intel's next-gen CPUs, and AMD's processors will be in line for this speed boost eventually, too

There's a new Windows 11 feature to boost the performance of apps that's been spotted in testing, but only PCs with Intel's next-gen desktop CPUs (codenamed Nova Lake) will benefit from this initially.

Windows Latest noticed that a regular Windows 11 watcher, XenoPanther on X, posted a screenshot of an option for 'Advanced Performance Extensions' (APX) which is present in a recent preview release of Microsoft's OS (build 29695).

Intel explains that APX "expands the entire x86 instruction set with access to more registers and adds new features that improve general-purpose performance."

This means that app developers can take advantage of this feature to speed up the performance of their software. How much of a boost any given app might get is, as ever, going to be variable, but some applications might end up with a nice performance bump (games probably won't get much benefit, in case you were wondering).

There are a couple of catches here, however, including that apps will have to be recompiled to take advantage of APX, although Intel observes that actual changes to the application's code aren't expected to be necessary.

It's worth noting that Linux already added support for APX some time back.

Analysis: there are barriers to entry – but AMD chips will get this

Right now, this APX toggle is present in Windows 11 test builds, but it doesn't do anything at all, which is hardly surprising given that there aren't any Intel CPUs which support the feature yet. What this shows is that Microsoft is preparing the ground for this capability, which will be supported with Intel's next generation of desktop silicon. These Intel Core Ultra Series 4 processors are expected to debut early in 2027, so they are still some way off.

Existing generations of silicon from the Intel Core Series 3 families and earlier won't ever get the benefit of APX, as this is something that needs to be baked into the chip at production. So, existing PCs will not get the speed boosts from this feature, which is a shame.

Theoretically, this could be an additional reason to upgrade your CPU, but there's another catch here in that Intel's Core Series 4 CPUs require a new processor socket, and therefore a new motherboard for your PC. So that means an extensive upgrade for a desktop PC – or an impossible one for laptop owners – meaning that both Intel and Microsoft could be hoping that APX is another way to shift some new laptops in difficult conditions when the market is dipping.

How about AMD processors? This is an Intel technology, as noted, so you might assume you can forget about seeing APX with a Ryzen CPU, but actually as Windows Latest points out, Team Red is collaborating with Team Blue to get the benefit of this feature.

With AMD involved, it's more likely that developers will use the feature, and that'll help Intel's case for its usefulness. However, again, the support needs to be baked into the chip, so the earliest this might happen on the Ryzen side would be the Zen 6 family – or possibly the following generation (rumors seem to favor the latter). At any rate, support is coming for AMD chips as well.

Meanwhile, some of the online commenters remain skeptical. One Redditor observes of this potential speed boost: "I'll believe it when I see it!"

Others aren't happy about the support here, and the need for an upgrade, complaining: "Imagine locking a performance boost behind a platform..."

But as noted, this is a feature that requires new hardware in the chip itself, it's not something that can be added to finished silicon.

As far as Windows 11 itself is concerned more broadly, Microsoft is busy working to boost performance in all kinds of ways, including applying a couple of new tricks to speed up the launch times of apps and menus. There's also a major goal to ensure that Window 11 runs better on 8GB of RAM, too.

As the AI buildout continues and models improve from generation to generation, fears persist over how useful this technology will actually be in practice and whether AI will reach a ceiling before it makes a meaningful impact.

Although roles and responsibilities, and jobs, are being replaced in the name of AI, does that mean it's changing things for the better?

Energy scarcity

Microsoft CEO Satya Nadella was speaking at the World Economic Forum in Davos in 2026 about the impact of AI on the global economy and the business world.

In his remarks, Nadella acknowledged the tensions at the heart of the AI expansionism that has captured not only big tech but the vast majority of the corporate world.

For AI to continue growing at the pace it has been – not just in terms of the compute required to train the models but also for inference as usage grows – we'll need plenty of data centers to churn through tokens. And for these tokens to be created, the AI pioneers will need plenty of energy.

Encountering bottlenecks

Burning as much electricity as AI companies plan on doing so for AI has caused alarm among the public – particularly as much of this has demanded infrastructure built close to populated areas in towns and cities.

Nadella acknowledged there has been a productivity gap between the hype around AI and the meaningful impact on people's lives – warning that without something to point to, unrest will bubble up and boil over into something more obstructive.

As time passed, we're seeing more tensions rise as people – and even legislators – object to new data center projects due to concerns over the impact on the energy grid, the environmental damage, and local communities.

• Windows 11 has an alternative take on Recall in testing
• It's an app called the 'Context Vault' offering a similar supercharged search experience
• In this case, though, you have to manually add content to the vault – there's no automatic regular screenshotting of the activity on your PC

Hate the idea of Recall in Windows 11? You wouldn't be alone, so Microsoft is trying a new spin on the idea of an AI-powered supercharged search which is a lot more limited than Recall – although the initial reaction has still been rather negative.

Windows Latest discovered that Microsoft is working on Project Pantheon, the current name of which is the 'Context Vault', and rather than being built into Windows 11 like Recall, this has to be installed from the Microsoft Store.

This isn't just an optional feature, then, it's an app you have to actively seek out, and it works quite differently to Recall in that it doesn't automatically take regular screenshots (snapshots) of the activity on your PC.

Instead, you must manually add the files or content you want to use with the Context Vault, so its AI powers can then make use of them.

To what end? Similar to Recall, you can ask the AI questions about the contents of the vault, or get it to organize material, planning the likes of a summer trip (the example Windows Latest gives) from documents, images, web pages, or even clipboard text saved to the vault.

Note that the app isn't yet officially available, and not listed in the Microsoft Store as such, but Windows Latest managed to find it nonetheless.

Context Vault doesn't necessarily need a Copilot+ PC – or a powerful local NPU, in other words, as it can leverage cloud-based AI – and you can give third-party AI like Claude access to the content in your vault, too, we're told. There's even a panel that tracks your AI token usage in the current build of the vault.

Bear in mind that this is still an early version of the feature, and it sounds pretty wonky right now. It's not clear if this app is something Microsoft will definitely carry forward to a release version for Windows 11, as ever with functionality in testing.

Analysis: is a retake on Recall going to fly?

On the one hand, the full user control over what content Context Vault can access compared to Recall could be a big selling point. On the other hand, using the cloud for running the AI processes is obviously more of a privacy concern (although you don't have to go that route to be fair).

Windows Latest has been led to believe that Microsoft is considering other ways to make the Context Vault more secure, including encryption and per-app access controls, as well as requiring a Windows Hello login for access (as Recall does) to protect the vault from possible local attackers jumping on your device.

As noted at the outset, the reaction to this idea in testing is generally skeptical, and many folks simply see it as 'yet more AI' that Microsoft is cramming into Windows 11 (although it isn't actually confirmed this is coming to the OS, of course).

As one Redditor puts it, this is a classic case of "solving [a] problem [that] doesn't exist."

Another agrees: "Yeah, Microsoft can't even be creative in thinking up something even marginally useful in an extremely far fetched scenario. Literally never in my life felt the need or wanted a tool like Recall or this equally s****y version of it. This problem was solved circa 2000 with bookmarks and Word/Excel documents. Not whatever convoluted trash this is."

In fairness, others are more positive, and there are Recall users out there. One commenter on the same thread says: "I use Recall all the time, it has been a gamechanger for me, I just wish it synced between my devices so I could look up something on my desktop I saw on my tablet."

The one thing that seems clear enough is that the anti-Recall folks aren't going to be swayed by a different take on the same concept, even one that sports a pretty radical change.

The kind of people this might persuade are those who most likely won't be drawn to comment on this idea in testing one way or another. I think it's certainly an alternative that's worth trying for Microsoft, provided it's implemented in a secure enough manner, of course.

• Windows 11 is cutting some Copilot features on the right-click menu
• At the same time, a Microsoft exec said that "AI has already helped our design team accelerate significantly"
• The exec added that "we're even seeing designers contribute directly to code fixes" and this has caused quite a stir online with anti-AI pushback

Windows 11 is getting tweaked to cut back on Copilot functionality, while at the same time, a Microsoft exec has admitted that the operating system's design team are using AI to help them – which has provoked something of a reaction, as you might expect.

First up, the move to cut back on AI, which as Windows Latest reports consists of stripping out some Copilot options from Windows 11's context menu, the one which pops up when you right-click on a file or folder.

In a post on the Microsoft 365 Message Center, the company explained: "We're retiring the OneDrive Copilot Summarize, Create FAQ, Compare Files, Ask a question, Explain image, Explain images, Compare images, Extract text actions from the File Explorer context menu."

These are Copilot actions which are tied to OneDrive, then, that folks who don't use Microsoft's cloud storage wouldn't have seen anyway.

You can still reach these options using the 'Ask Copilot' icon in the right-click menu, so overall, this is a streamlining change for the better (albeit one that some users won't even realize has happened).

It's not being implemented right away, but from mid-November, and by the end of the year, these options will have disappeared from Windows 11's context menu.

Now for the more controversial development that Windows Latest also noticed, namely a comment from a Microsoft executive on X. Diego Baca, who is VP of Design for Windows, posted in reply to a comment pushing for "Windows design unification" (better consistency) and using AI to help to that end.

Baca observed: "On AI agents, 100% agree. I'm not the best person to speak to the engineering side, but AI has already helped our design team accelerate significantly. We're even seeing designers contribute directly to code fixes that improve craft, consistency, and quality. That's exciting progress, and we're only getting started."

The key part of this that's getting picked up on social media is the admission that AI is being used by the Windows 11 design team to ensure tasks are completed more swiftly, and that it's helping designers to contribute code fixes. As you might guess, there are some unhappy people out there due to this admission.

Analysis: a chorus of noticing

As noted, the reaction on Reddit has been predictably full of barbs and withering fire at Microsoft. This upvoted post complains: "We noticed [the use of AI in coding], the rapid decline in quality, every patch bricking at least some devices, major releases bricking entire fleets, and the introduction of major security vulnerabilities that are easily exploitable makes it clear the codebase is being vibecoded with insufficient code review and testing."

In fact, there's quite a chorusing of "we noticed" or "we know" regarding Microsoft turning to AI in coding, as well as the inevitable "slop" accusations, including the likes of: "And this is how Linux goes mainstream, not because it becomes better, but because Microslop becomes unusable thanks to AI integrations." (I should note that as you may have seen, Linux has its own controversies with AI, but that's a whole other story).

At any rate, I think we need to dig down a bit into the meaning of Baca's statement here. When the exec says, "AI has already helped our design team accelerate significantly," it's a pretty broad statement – perhaps referring to assistance on the design brainstorming front more than coding.

Yes, designers may "contribute directly to code fixes" now with the help of AI, but how often is that really happening? I'm guessing not that much, and that this work is more about simple interface tweaks and polish than anything, and this code will be passed to software engineers to be fully checked and reviewed – the main hope being it saves a bit of time. There's a lot of the Windows interface to fix, after all.

I think the hostility here is a bit unfair given that Baca is just trying to be honest about what Microsoft is doing, and I don't think the reality is that Windows 11's interface is suddenly being vibe-coded (by AI) to any great extent. What's happening is that Windows 11 users are jumping on this and venting because they're fed up of too many bugs, and that has nothing to do with the design team.

For me, this is about Microsoft's broader failure with Windows 11 bug testing and flaky updates, which isn't something you can lay at the door of AI as such. Rather, this is about the overall QA (quality assurance) processes which aren't picking up these problems and sorting them out before release (whether any given bug was introduced by AI, or not).

This is what's really the root cause of the outbursts on social media here, and the comments about patches going wrong and general unreliability. And this is what Microsoft really needs to fix, and fast – it remains the biggest problem with Windows 11 that needs addressing.

• The Bing Wallpaper app has seen an annoying feature reintroduced
• This means those clicking on a blank space of the Windows 11 desktop will get a Bing search fired up in their default browser
• Unlike the last time Microsoft tried this stunt, there's no way to disable the ability this time around

Microsoft is up to its old tricks again, except even worse this time — with a scheme whereby Windows 11 users with the Bing Wallpaper app installed are sent to its search engine with a mere stray click on the desktop.

Windows Latest reports that an old, potentially off-putting shortcut has been reintroduced to the Bing Wallpaper app, whereby if you click anywhere on the Windows 11 desktop, a search in the (separate) Bing engine is fired up (in your default web browser).

If you recall, this actually happened a year ago, but Microsoft reversed course, although now appears to be trying to reintroduce this feature. The difference is that this time around, Windows Latest notes that there's no option to turn off the functionality (which was present last year, in the app's settings).

However, there are some points to note here: firstly that you must choose to install the Bing Wallpaper app (which cycles Bing images daily for your desktop background). It's not a default Windows 11 app, so this isn't something Microsoft is piping directly to Windows users.

Secondly, the Bing search doesn't trigger every time you click on a blank piece of desktop (a space without a file, folder or icon). It does this once, and then the Bing search won't be triggered for another 24 hours. You'll only experience this once per day, in other words.

Analysis: Microsoft must do better

Just because this app isn't part of Windows 11 by default doesn't give Microsoft the excuse to push functionality like this. It's just going to annoy or even confuse people if they accidentally click the desktop and end up with a Bing search being loaded.

The search relates to the wallpaper image, in case you're curious, and fair enough – maybe some folks might be inquisitive and actually want to know more about the picture by clicking on it.

However, this should be an ability that's optional and can be turned on, not something that's enabled by default. What's worse still is that in this new incarnation of the click-to-Bing farce, you can't even turn it off if you want to do so.

Why is Microsoft engaging with this kind of nonsense to promote its own services? I really can't begin to guess given that 2026 is the year of fixing Windows 11 and curing all the ills around the OS, and while this app may not be part of the operating system, it's still a piece of official Microsoft software – and there shouldn't be any place for this cynical Bing traffic grab or any other remotely anti-consumer practices.

Whether it's a part of Windows by default, or not – or whether you have to be 'foolish' to install the Bing Wallpaper app, as some online commenters argue – is besides the point. Microsoft is supposed to be better than this now, but another episode with a recent Harry Potter advert in this Bing app – one that provoked a considerable rebellion – suggests that it very much needs to try harder.

Organizations can be under huge pressure to make changes to their IT estates. In particular, there are two drivers of this change.

Firstly, there’s the pressure to innovate. The rapid advancement of technology and AI coupled with shifting business priorities can put organizations under pressure to adopt the latest tools, with vendor messaging advocating for the speed of adoption. Simultaneously, ongoing inflation and slower economic growth can mean there is a need to find more cost-effective solutions.

Secondly, there’s the change that is forced. Operating systems might be reaching their end of life and need upgrading, or there are outages and cyber threats to defend against. There have also been several recent examples of software vendor prices rising significantly due to ownership changes, like with VMware, and companies therefore needing to find new solutions for their budget and reduce vendor dependency.

However, in dealing with these pressures, the trap companies can fall into is trying to deliver all of this change at once and within a small timeframe. Yet in doing so, this change becomes harder to manage and the risk of outages and system failures increases. So, how can they deliver any change that is needed safely?

The dangers of too much change at once

The main challenge in changing too much too quickly is visibility. Companies have to oversee complex IT estates – on-prem systems, the cloud, servers, end-user computing, networks – where change is constant and security updates are largely automated. As this change happens, there are more interactions between older and newer systems, different operating systems, different products, and this means the risk of failure can increase.

Configuration drift, for example, occurs when changes, which haven’t been recorded or authorized, cause a system to drift away from its intended baseline (of how it should perform). Without the right monitoring software, this drift can remain invisible, creating added risk to every subsequent change – and that’s when an outage or cyber incident happens.

In May, for example, Google Cloud suspended Railway’s production account without warning because Railway couldn’t see that the platform had a technical dependency on GCP that caused it to fail, leading to an eight hour outage.

What’s more, as we’ve seen with the adoption of AI tools, change can happen faster than training and processes can keep up with, further increasing the risk of disjointed workflows, security vulnerabilities and data breaches. But even when teams can see the changes that need to take place, they can also be faced with the challenge of deciding how to modernize their existing applications.

Finding alternatives to the buy-or-rebuild dilemma

Of course, trying to slow down change isn’t always possible. A company might have been priced out of its VM software or needs to upgrade its applications for security purposes. In this position, traditionally, IT teams have had to either buy new software or rebuild existing apps, which can both be expensive and time-consuming.

This can lead to vendor lock-in, as IT doesn’t see a viable alternative and doesn’t have the engineering capacity to rewrite applications tied to their original operating systems.

To overcome this dependency, as agentic coding matures, organizations may increasingly opt for a third option: using AI agents to build their own replacements for SaaS tools. But while this can appear to remove vendor lock-in and license costs, it also transfers responsibility for maintenance, security, configuration and ongoing support in-house, making it even harder to oversee change without additional resources or capacity.

However, there is another route. Packaging and migrating existing applications can allow teams to modernize without taking on the burden of building from scratch. By packaging an app’s or virtual machine’s existing operating system dependencies, teams can move them into containers and modern desktop infrastructure without needing to rewrite the app.

The app maintains its existing behavior, but can now work on the latest operating systems and in workspaces like AWS.

Delivering change at scale

With a modernization strategy formed, how do teams gain ongoing visibility across their entire estate and know when apps need to be packaged?

With so much change taking place, manual monitoring and reconciliation processes will struggle to keep up with the demands of infrastructure modernization. Instead, teams should consider using configuration monitoring software to gain continuous visibility into change across their entire estate.

Drift detection tools, for example, can automatically measure change across an estate against established configuration baselines and alert IT teams whenever a change triggers a system to move away from its baseline. This means that an automated patch by an external provider or a manual intervention by a user gets detected and flagged to IT, changes that might otherwise go unnoticed and trigger an outage further down the line.

Imperatively, this estate-wide visibility means teams can see and understand the impact of any unauthorized changes taking place and intervene before they lead to failure.

It also gives them the information they need to understand how a transformation project might interact with the systems currently running in their estate, helping to inform the feasibility and implementation of the proposed changes and deliver more targeted, impactful and safer change.

Alongside configuration visibility, it’s also possible for teams to automate the packaging and delivery of apps, saving them a large amount of time and cost in manually performing a typically laborious task. This subsequently creates a repeatable and scalable process that enables apps to be regularly deployed and made available to users.

Escaping the trap

With so much pressure to change, above all, escaping the transformation trap requires a mindset shift. It involves, one, seeing change as an ongoing problem to manage, and two, not attempting to carry out too many innovation projects at once.

Small, incremental changes can deliver more meaningful impact and ensure systems are properly deployed or integrated, while this targeted approach also provides more capacity for IT teams to deal with any forced change that might surface.

Yet by prioritizing change visibility, IT teams can mitigate forced change by addressing risks earlier and seeing when upgrades are needed in good time. And by automating processes like packaging apps, they gain more capacity to prioritize innovation projects as routine changes take place faster and more securely.

When these processes are joined together, instead of grappling with an overwhelming and uncontrolled level of change that increases risk, teams can ensure change remains a driver of security, resilience and innovation.

We've featured the best IT automation software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Enterprises are being forced into software upgrades that they can't really get out of, report finds
• 91% have upgraded hardware in the past three years just to maintain software
• Enterprise protection legislation and software support obligations could be the answer

Legacy tech has become something of a hot topic in recent years with AI, but while hardware and data foundations have been core to the discussion around artificial intelligence, a new Origina report has revealed that around one-third of enterprise IT budgets are being earmarked for software upgrades.

The report found that 28% of large enterprise IT budgets are being spent on vendor-driven changes that they would not have otherwise chosen in a trend the company's calling a 'software tax'.

But more than just consuming budgets, software changes are also impacting worker time and leading to potentially lost productivity as a result.

Software upgrades cost more than you might think

"That's not a one-off cost, it's a recurring drag on capacity, and every year it repeats is a year those people aren't working on the initiatives that actually differentiate the business," said James McGlennon, former EVP and Global CIO of Liberty Mutual Insurance.

More than four in five (84%) report that senior engineering staff always or often spend time dealing with vendor-driven software changes, taking them away from more meaningful work that they're employed to do. A similar number (80%) also reported that leaving a major software vendor could lead to widespread disruption within their business, ultimately forcing them to stay with that software vendor and enforcing vendor lock-in – with exit costs and migration technicalities to blame.

More than three-quarters now agree their software vendor dictates the timing and terms of changes, not them (78%), and that they have very little negotiating power (76%).

In many instances, enterprise customers don't even feel that an upgrade was totally necessary. Nearly all (91%) had retired fully functional hardware within the past three years just to comply with software vendor support deadlines. While Origina doesn't specifically state it, Windows 10's end of support is likely a large factor in this statistic.

Looking ahead, most (91%) support legislation to provider greater protection for enterprise customers, and four in five (82%) support software support obligations similar to what the EU has introduced for hardware repairability.

There has been an explosion of AI-powered products released over the past four years, with most built on top of leading foundation models. Startups, enterprise software vendors, and even large incumbents route their core intelligence through APIs from the leading AI labs, and a growing but still narrow set of open-weight alternatives.

This arrangement has most certainly enabled a genuine burst of AI products, but it has also created a structural fragility which can be remedied through the democratization of model creation and training.

Defining a wrapper

In its simplest form, an AI wrapper is a product whose core function is an interface sitting on top of a third-party foundation model's API, with little to no proprietary technology beneath that layer.

That approach can have a short lifespan and Jasper can illustrate that. Once the standout of the AI writing boom with a $1.5 billion peak valuation, Jasper saw its core market vaporized when ChatGPT introduced free, native writing features. Its revenue fell from approximately $120 million in 2023 to approximately $55 million in 2024. Jasper did survive, by redirecting its efforts toward enterprise workflows and brand-specific training.

Generally speaking, analysis shows that, among AI startups that have wound down in the US in 2025, 81% were "wrappers and apps". Copilots and assistants layered on foundation models, AI productivity tools, content generators, and vertical SaaS products whose core differentiation was an LLM front-end rather than proprietary data or infrastructure.

In Jasper’s case, and in many others, what pressured the company wasn't a competitor out-executing them. It was the platform they built on top of simply absorbing the feature they'd built their company around.

Not all wrappers are equal

It would be a gross generalization to say that all AI wrappers are the same. A thin prompt-and-display wrapper is different from a product that does heavy retrieval, tool orchestration, agentic workflows, or domain-specific fine-tuning. The latter can have real defensibility even while sitting on top of a third-party base model. The model call is one component, but certainly not the whole product.

Some companies also use multiple foundation models or fine-tune/distill their own smaller models for specific tasks after prototyping on a frontier API. Open-weight models add variety too as not all wrappers call a closed-source frontier model.

The wrapper problem

Regardless of where an AI product sits on the wrapper complexity scale, they all share the same problem. When a company's core value proposition depends entirely on an API call to a third party, that company does not fully control its own product.

It cannot control pricing changes, rate limits, deprecations, shifts in model behavior, or the possibility that the platform provider decides to compete directly in the same vertical.

This is not a theoretical risk, though how much it affects any given company likely depends on how much of its value sits above the model layer.

The economics of the wrapper

So what is the defensibility of an AI wrapper in practice?

The short answer is that it comes down to intellectual property.

Where a company benefits from an existing distribution advantage, established customer relationships, regulatory licensing, network effects, or years of accumulated brand trust, model-layer differentiation may not matter as much. But if what makes a wrapper distinctive is observable from outside the organization and replicable by a competent competitor within a limited period, it seems reasonable to expect it will eventually be replicated.

That is the unfortunate economic truth beneath the wrapper model. The cost of replication is near zero for the platform that controls the underlying model, and near total for the startup that doesn't. A wrapper's moat, if it has one, has to come from something other than the model call itself. Absent those, a wrapper is unlikely to have a lasting competitive advantage.

The barrier to the solution

At this point, you might be inclined to agree that model ownership is one solution to this problem. But that's easier said than done.

The reason so few companies own their models is not a lack of desire. It’s, more plausibly, financial. Training a frontier-competitive model requires enormous compute budgets, specialized talent, proprietary or expensively licensed data pipelines, and the infrastructure to run reinforcement learning at scale. These costs put model creation out of reach for all but the best-funded companies.

For companies to build on their own models as a norm, this barrier has to fall meaningfully. Several forces are already pushing in this direction, open-weight models being one of them. But even then, hosting and fine-tuning an open-weight model is no small task. It still demands technical talent and compute infrastructure most startups don't have access to.

True democratization of model ownership

Democratization, in this context, doesn't mean every company training a frontier-scale model from scratch. That goal is neither realistic nor necessary. It means democratization reaching the training layer the same way access to AI has been democratized for end users.

Democratizing model training and ownership could be what gives founders a lasting chance at turning an idea into a business that doesn't eventually get absorbed by a model provider as a simple feature. And if that happens, the stakes may go beyond individual business durability.

As things stand, access to models is cheap. Access to training is not. This needs to change. The ability to train an AI model should be fully democratized so that the value, the control, and the responsibility can be distributed rather than centralized.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Windows 11 has a new preview build with a change to boost performance
• It's just for Edge to begin with, and it means that if the browser is already running, new instances of the app will launch faster
• Microsoft's plan is to introduce this to a wider range of Windows 11 apps

Microsoft is making another move to tune up Windows 11 so it launches some apps more swiftly, although this is just in testing for now.

Windows Latest noticed that the latest build in the Release Preview channel (the final stage of testing) applies a change to bring in "accelerated app launches" for Microsoft Edge in a certain scenario. Namely that when Edge is already running, further instances of the web browser being opened will appear at a faster speed.

Microsoft tells us that: "When you select a link or open Edge from the taskbar or Start menu while Edge is already running in the background, Windows now routes your request directly to the running instance instead of first starting an extra, temporary process."

This shortened launch path for when Edge is 'warm' (already running) means a considerably faster launch time, as Windows Latest verified.

The tech site also points out that only the first 'cold' start of the browser will be slower (with the longer launch process), as even if you open Edge, then close it, if you run the app again, it's still 'warm' as the relevant processes have spun up in the background.

Analysis: broader benefits, but why did all this take so long?

If you don't use Edge (and that's doubtless most of you), then you might be thinking: what's the big deal anyway? Well, this is a first step in testing towards a grander goal, namely a "broader Windows capability that any app can adopt to speed up its own launches" as Microsoft explains in the preview build's release notes.

Eventually, we could see multiple apps benefiting from what's an apparently considerable performance boost when launching, assuming that they're already running (and warmed up). It could be some time before Microsoft opens this up to third-party developers, mind, and then they have to get around to implementing the functionality.

At any rate, it's a useful new feature, and one that comes on top of Microsoft's existing work in enabling preloading for certain Windows 11 apps, notably Edge itself and File Explorer. We've also witnessed the introduction of the Low Latency Profile feature, which represents a nice CPU speed boost for Windows 11's menus and also apps.

The main question is: why did it take Microsoft so long to get around to these performance improvements? File Explorer being slow was a complaint from the start with Windows 11, and yet we've had to suffer this – and various other instances of sluggishness – for years before they've been fixed. It was only this year, when Microsoft properly turned its attention to fixing Windows 11, that we've seen solutions flooding forth.

And don't get me wrong, that's great, but it also shows that Microsoft didn't take these performance problems seriously until the Windows 11 userbase was more or less in open rebellion over AI features being pushed while all these issues continued to frustrate people. It's clear enough that the optimization of Windows 11 could have been done in the past, if Microsoft was willing to graft at the task, it was just that the company didn't see it as enough of a priority – bafflingly.

I concede that even when Microsoft is doing a good thing here, I'm moaning, and that in this respect the company can't win. Fair enough, and all's well that ends well as they say, but what I'm really hoping is that Microsoft can learn from this in the future and act earlier to troubleshoot its desktop OS when it clearly has problems.

Whatever next-gen Windows turns out to be, it needs to take this new attitude and go forward with it, until Microsoft has an operating system which is running smoothly for daily operations in every important respect, from performance to usability – not to mention reliability and stable updates which is still a big issue that's left to resolve.

When a competitor appears in an AI answer and your brand doesn't, the obvious reaction is to publish more content. But that skips the useful questions like:

Which answer?

What was the person asking?

And did the assistant reference your competitor's website, an independent review, or neither?

Those questions shaped this review. To test the platform, I compared Saily eSIM with two of its close competitors, Airalo and Holafly. The aim was to find something worth investigating and see whether the Brand Radar data holds up to the hype.

My hands-on work focused on its existing AI Visibility Index, moving from combined-platform reports to ChatGPT-specific checks using recent data. Here’s how it went.

Ahrefs Brand Radar: Plans and pricing

First, separate the two purchases. The AI Visibility Index lets you search data Ahrefs has already collected, and the Custom Prompts run the questions you choose from the point you set up tracking. Now, Brand Radar is sold separately, so you don't need to buy the wider Ahrefs SEO suite first.

Ahrefs pricing breaks the options down as follows. These are monthly US-dollar prices, accurate as of October 15, 2026.

Lite, Standard, and Advanced subscribers already receive 150, 300, and 600 custom checks per month, respectively. That gives existing customers a small starting allowance without buying index access.

The catch is that checks aren't interchangeable with prompts. Each platform, location, and scheduled run consumes allowance. At the standard one-check rate, 20 prompts checked daily across three platforms and two countries use 3,600 checks over 30 days. By the way, Claude consumes eight checks per execution, so its allowance disappears faster.

For an agency researching different brands every week, the index has a clear purpose. But for a small, growing business following ten buying questions, paying for the entire archive before using those smaller allowances feels premature to me.

Ahrefs Brand Radar: Features and specs

AI Visibility Index

I entered Saily, an eSIM provider, along with its competitors (Airalo and Holafly), and opened the overview.

Mentions, citations, estimated impressions, and AI Share of Voice sit together, with platform comparisons underneath. The index already includes historical data to explore, which is what makes it attractive. That means you can investigate a competitor before deciding which questions deserve ongoing tracking.

Saily's smaller mention footprint gave the investigation a direction. Scrolling down brought its own performance history and associated topics into view, making it easier to move from a broad comparison to a specific question.

You still need to understand the sample here. Ahrefs' methodology combines search-informed questions, including Google's People Also Ask, with semantic expansion. These aren't private conversations gathered from ChatGPT users. Its newer AI-adjusted volume applies platform-specific usage estimates to a parent keyword's Google search volume.

I prefer that qualification to treating Google demand as AI demand, but it remains a model. A rising line isn't proof that more customers saw your brand, and dataset changes can affect comparisons. AI Share of Voice also depends on the brands and filters you selected.

Competitor gap analysis

The comparison became more useful when I isolated ChatGPT answers mentioning competitors but excluding Saily. This turns an intimidating report into a reading list. You can investigate the questions behind the gap instead of guessing what another brand is doing right.

Opening “What's the best eSIM in the UK?” revealed a stored response dated October 9. Brand Radar's tracked-brand panel listed Airalo and Holafly, with no Saily. The question also mixed domestic mobile service with travel data, a useful reminder that apparently relevant prompts can cover different needs.

That response exposed some friction too. Parts of the answer displayed raw markup, which made it harder to read. The sidebar reported no citations while listing pages under “Found but not cited.” Keeping those categories separate was helpful, even if the presentation needed tidying.

This is where your judgment matters. A missing brand is worth investigating when the question fits its product. It isn't automatically an instruction to write an article, and a brand mention doesn't necessarily amount to a recommendation.

AI citation analysis

The Cited pages report was the natural next stop. Its page-type view groups results into formats such as articles, landing pages, and reviews. That gives you a starting hypothesis about the material appearing around your category before you inspect individual URLs.

Switching to individual pages made the distinction between discovery and citation much clearer. “Found in” includes pages retrieved while producing an answer. “Cited in” records answers that actually cite the page. A source can appear frequently in the first column without receiving much visible credit in the second.

TechRadar's own best eSIMs guide appeared alongside review website listings and provider-owned pages. More interestingly, the guide's “Mentions on page” indicators included Saily even though this investigation had excluded Saily from the AI answers. A roundup can mention your product without an assistant carrying that mention into its response.

That is a better research lead than a score alone. You can inspect the source's coverage and ask what information an answer used or omitted. However, “Other” wasn't a ready-made outreach list because Holafly-owned URLs were present too. So, always check actual domains and your brand setup before treating a result as independent coverage. Not to forget, getting included on a cited page also doesn't guarantee a future AI recommendation.

Brand sentiment analysis

Ahrefs initially described sentiment analysis as a work in progress, but a native Sentiment report was available in my review account. In simple terms, this report adds a useful second question to the comparison, i.e., when AI mentions a brand, how does it describe it?

Saily's lower mention volume sat alongside a higher positive-response share than the two competitors. I found that contrast useful because, in my opinion, being discussed less often and being described less favorably are different problems. The former might prompt discovery research, while the latter calls for examining what the answers say.

I didn't independently audit the classifications, so this isn't a verdict on sentiment accuracy or customer satisfaction. So, use the report to identify answers worth reading, then check their wording.

Custom prompt tracking

The existing index is convenient until the question you care about isn't there. Custom Prompts let you add it, choose platforms and locations, organize questions with tags, and schedule daily, weekly, or monthly checks. Its history starts with collection, not arriving pre-populated.

For ChatGPT and Perplexity, Brand Radar also exposes Fan-out queries (extra searches an assistant generates while working through a prompt). They suggest useful supporting topics, although Ahrefs only shows the queries returned with the response, not necessarily every search the model ran.

The public indexes cover ChatGPT, Perplexity, Gemini, Copilot, Google AI Overviews, and AI Mode. However, platform coverage needs a closer look than a row of logos. Plus, Claude is available for Custom Prompts rather than in the public index. Grok collection is paused, with historical data retained. Such distinctions are pretty vital in case either platform is central to your audience.

AI research tools

Brand Radar also offers search-demand and web-visibility reports. YouTube research covers titles, descriptions, and transcripts, while Reddit visibility focuses on threads appearing in Google results. TikTok has its own beta report.

These surrounding reports have separate Ahrefs-plan and beta conditions, so don't assume an index purchase includes every sidebar item. I see these as ways to investigate surrounding coverage, not proof that every mention influences an AI answer.

There is more assistance than the old “research only” description suggests. “Ask Ahrefs” is built into the interface, and the agent previously called Agent A now lives under Letaido, which offers additional analysis and automation.

Is Ahrefs Brand Radar worth it?

The strongest reason to buy Brand Radar is the ability to move between a competitive overview, an individual answer, and its source material. My eSIM investigation showed why that matters. Having lower visibility, positive sentiment, source-page mentions, and actual AI recommendations can tell different stories.

For agencies and in-house teams investigating several competitors, that depth can justify paying for research access. Entity matching, which distinguishes a company from unrelated uses of its name, is another useful safeguard when your brand isn't as distinctive as Saily.

For a small business with a short list of questions, the smaller Custom Prompts allowance or a limited manual check is a more proportionate start. The expensive part shouldn't arrive before you have a plan for using the findings.

I came away with useful leads and a clearer understanding of their limits. Brand Radar makes the evidence easier to inspect. Whether that evidence becomes better content, corrected information, or relevant coverage still depends on what you do next.

FAQs

There’s never been so much change happening across an IT estate. And it’s constant. On a day-to-day basis, IT teams might be creating and monitoring user permissions, deploying any scheduled operating system (OS) updates or integrating new third-party software tools. This is not to mention all the change carried out by external providers, outsourced teams and employees.

Change itself is not a problem. Indeed, it’s essential for ongoing security, resilience and modernization. Change that isn’t actively managed, however, is a different story. This could be change that goes unnoticed, like an external patch that automatically updates a system without IT oversight, or change that is forced, like rushing an OS upgrade to meet an impending end-of-life deadline.

The trouble is many organizations have to monitor incredibly large and complex IT estates that can contain hundreds or thousands of business-critical endpoints. And so much change is now automated or carried out by external parties. Consequently, without the right software and processes, this means that IT teams can have limited visibility into many changes taking place.

These changes might seem small or innocuous on their own. But left unmanaged, they mean systems are drifting away from their intended baseline, which adds risk to each subsequent change – and that’s when systems can fail and damaging outages occur.

Change that is not managed effectively also increases vulnerability to cyber attacks. Misconfigurations or gaps in security coverage materially enabled more than 90% of cyber incidents investigated in recent research, showing how seemingly minor changes or lapses in oversight can create openings for attackers.

So, what are some of the ways IT teams can effectively control change across their estate?

Here are three key steps:

1. Understand

The first step is to gain greater visibility into change across the IT estate, so teams can identify potential risks and understand their impact before they cause disruption. Achieving this requires the use of automation tools that enable continuous drift detection. With this in place, IT teams can monitor their entire estate – servers, end-user computing, networks, on-prem and cloud – for configuration change in real time.

To do this, the technology needs to scan every layer in the infrastructure to establish a known-good configuration baseline to use. The platform can then flag any time a change causes a device to drift from this intended baseline, while also allowing teams to gain visibility on the gap between the change that was expected and approved and what is actually running in the estate. This means a manual fix or external patch gets flagged, rather than flying under the radar until it triggers an outage or failure.

2. Respond swiftly

One of the key challenges IT teams face is reconciling and approving change. Inevitably, with so much change taking place, trying to manually perform this process leaves room for errors and can delay approvals. So, to reduce manual effort, the next step is to automate change approval processes, using software to automatically reconcile every detected change against approved baselines and validate what steps were taken against those planned in the change request.

With this, teams will then be able to rapidly respond to unapproved changes and use their time to take action on issues before they escalate and impact operations. More broadly, this also speeds up the delivery of new upgrades or projects and gives IT teams greater freedom to innovate safely.

3. Protect

There’s always a risk of a cyber incident taking place. But when teams understand what’s happening in their estate and accelerate their approval processes, they can also detect and remediate issues faster and reduce their mean time to resolution. This can help to mitigate a failure escalating into an outage, or minimize the cost of an incident, while also strengthening overall resilience.

Protecting systems also means ensuring they are compliant and that teams can export defensible evidence on demand whenever a regulator requests to see it. This also highlights why continuous compliance that generates clear and auditable trails is so important. Otherwise, audits are often accompanied by a time-consuming, manual search through spreadsheets and screenshots for evidence, and this also means information can be out-of-date by the time it’s collected.

Controlled change unlocks innovation

IT change is constant, and with the rapid development of AI and cloud tools, it’s also only going to increase. This provides more opportunities for companies to modernize, but it also creates more risk when it happens without control.

To maintain control and prevent damaging incidents from taking place, IT teams need to understand when configuration change happens across their estate, accelerate the rate at which change can be approved and reconciled, and protect their systems through faster remediation and establishing a real-time audit trail.

While it can be easy to put off resilience projects in favor of innovation, the threats facing IT are becoming stronger and more varied, so controlling change has to be a key priority. If anything, it provides the foundation to innovate successfully and safely.

We've featured the best firewall software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

For eight years, Surfer had one job, and that was to help you rank on Google.

You wrote your content, Surfer told you what to fix, and your page climbed. Then the whole AI wave showed up. People started asking ChatGPT and Perplexity for recommendations instead of scrolling through blue links, and ranking first stopped meaning what it used to.

So Surfer built its own AI Tracker. It's a paid add-on that watches whether AI engines mention your brand when people ask about your industry, and shows you who gets named when you don't.

That's a big promise, so I put it to the test. I tracked a well-renowned brand across all five engines Surfer covers. I checked its numbers against what I saw by asking ChatGPT myself. I dug into where the data holds up, where it wobbles, and where the tool quietly stops being useful.

Some of it genuinely impressed me. Some of it didn't. Here's everything I found.

What is Surfer AI Tracker?

Surfer AI Tracker is a paid add-on inside Surfer that shows you how often AI engines mention your brand when people ask about your industry, and which sources they cite instead of you.

Surfer itself is an SEO tool that has helped writers rank on Google since 2017. The AI Tracker, launched in 2025, is its answer to a new problem, i.e., buyers now ask ChatGPT and Perplexity for recommendations and not just Google.

Here is what it tracks and how.

What it tracks:

• Whether your brand appears in AI answers, and how often (Mention Rate)
• How prominently you show up when you do appear (Average Position)
• An overall Visibility Score across engines
• Which websites AI pulls from to build its answers (Top Sources)
• The sentiment of each mention, positive or negative
• The same data for your competitors, side by side

How it tracks it:

• You give it real prompts your buyers type, like "best CRM for small teams"
• It runs those prompts across ChatGPT, Perplexity, Gemini, Google's AI Overviews, and Google AI Mode
• It scrapes the actual answer a real user sees, not a stripped-down API response
• It re-runs them on a schedule (daily on Pro) so you can watch the trend

How I tested Surfer AI Tracker

Before I trusted a single chart, I wanted to know whether the numbers on my screen matched what a real person would actually see if they asked ChatGPT the same thing.

That's the lens I judged it through. These are the things I cared about, and why.

• Is the data real? Surfer's big claim is that it scrapes the actual answer a user sees, citations and all, not a stripped-down version pulled from an API. That's the entire foundation. If it's off, nothing else matters. So this got the hardest look.
• Does it hold still? AI answers shift. Ask twice, get two replies. I wanted to know if the tool's readings wobble so much that any single number is meaningless.
• Do you need every engine? The cheaper plan tracks ChatGPT only. The pricier one adds Gemini, Perplexity, Google's AI Overviews, and Google AI Mode. I wanted to see if those engines actually tell different stories, or if you're paying extra to hear the same thing four times.
• Does it tell you what to do? A gap you can't act on is just anxiety with a dashboard. Surfer's mention-gap and fan-out reports promise a to-do list. I wanted to know if that's real guidance or a prettier version of the problem.

Then I actually used it. Here's what that looked like.

• I picked a brand I know inside out (I already knew its real rivals and roughly how often it gets talked about), and let Surfer build the competitor set from the brand name, so it had something to measure against.
• I loaded 40 prompts real buyers type into AI, like "best [category] tool for [use case]". Not vanity searches for the brand name. The questions that actually decide who gets picked.
• I let it run a full pass across all five engines, then ran the same prompts by hand myself and compared what I saw to what Surfer captured.
• I watched how much the numbers moved day to day, checked how the story changed engine to engine, and dug into the Sources view, clicking different gaps to see if the advice ever actually changed.

Setting up Surfer AI Tracker

The AI Tracker sits in its own AI Visibility workspace, apart from Surfer's SEO side. You're tracking how a brand shows up in AI answers, not auditing your own pages, so there's nothing to plug in.

Even better, I barely had to type anything. I entered the brand domain name of "Zoho," and Surfer built the rest itself.

Business type, industry, a paragraph on what Zoho sells, its customer base, a list of competitors, and the topics worth tracking. It knew Zoho was a B2B SaaS suite for small and mid-sized teams; it knew privacy and all-in-one bundling were part of the pitch, and it named real rivals like Salesforce and HubSpot.

I find this smarter than most trackers bother to be. The usual tool asks for a brand name and stops there. Surfer uses that name to understand who you actually are, then uses that understanding to judge whether a mention is really about you later. The catch is that you should still read what it generates. It's a starting point, not gospel, and a wrong detail here quietly skews every chart down the line.

Then come the prompts, and this is where the tool genuinely impressed me.

Surfer generated 40 prompts on its own, sorted into eight topics, five prompts each, from accounting and CRM to collaboration and payroll. And they weren't lazy.

These are the exact questions a real buyer types into AI: "Which accounting software is the best for small businesses in the USA and Canada," "Best project management software for a one-man team," "What is the best collaboration software with built-in VOIP for a small business?" You get 50 prompt slots on Pro, and Surfer filled 40 of them for me before I'd done anything.

You can still edit, delete, or add your own, and you can pull your top queries from Google Search Console if you'd rather. But the fact that a single brand name produced 40 usable prompts is a real time-saver, especially if you don't already know what your buyers ask AI.

The last piece is competitors, and here's how it pulls up.

The competitors Surfer tracked weren't only the ones it listed at setup. When the data came back, the table showed brands I never entered, like QuickBooks, Xero, Pipedrive, and Gusto. Surfer surfaces whoever actually shows up in the AI answers next to Zoho. That's the honest way to do it, because your real competition in AI search is whoever the model names, not whoever you think you compete with.

It's also the first hint of a theme you'll see throughout this review, i.e., the tool is very good at showing you what's happening, and Zoho landing second here, behind HubSpot, is exactly the kind of thing you'd want to know.

Then you let it run. Surfer sends every prompt to the engines, records what each one says, and builds your baseline. Load your prompts early. The sooner they're in, the sooner your trends are worth reading.

Surfer AI Tracker dashboard

The dashboard runs on four numbers– Visibility Score, Mention Rate, Average Position, and Top Sources. Once you know what each is telling you, the tool clicks into place.

Start with the Visibility Score. Zoho landed at 51. On its own, that means nothing. It's a blended score out of 100 that mixes how often you're mentioned with how high up you appear. It only starts to mean something when you see it against your rivals, like in the chart below.

It tells you if you're winning, never why. For why, you need the other three.

Mention Rate is the honest one. Zoho showed up in 33.4% of answers. Plain and concrete, and the number I'd track week to week, because when it moves, something real moved.

Average Position was 4, and it comes with a trap. It only counts the times you were actually mentioned, so a brand can post a great position and still be nearly invisible if it shows up once and vanishes everywhere else. Read it alongside Mention Rate, never alone.

Adding to this, you can also go through the Topic breakdown.

Surfer splits your score by topic, and that's where the real story hides. Zoho scored 86 on SMB CRM selection; it's home turf. But on Collaboration it scored 1%, and on Payroll and HR just 10%, both markets Zoho actually sells into.

AI is barely naming it there. That gap is worth more than any headline number, because it shows you exactly where you're missing from conversations you should be winning.

So which numbers are signal and which are vanity? Mention Rate and the topic breakdown are signal. Visibility Score is a summary, useful at a glance, shallow up close. Average Position only means something when read beside Mention Rate. Look past the big number on top, and the dashboard gets a lot more honest.

Surfer AI Tracker accuracy

I checked Surfer's captured answers against what I got by asking ChatGPT the same questions myself, and on both prompts the tool matched reality.

I picked two prompts, opened ChatGPT, ran them, and lined my answers up against what Surfer had captured for the same engine.

I chose the two on purpose i.e. one where Surfer said Zoho shows up, and one where Surfer said Zoho was invisible. If the tool is accurate, it has to get both right.

Start with the accounting prompt, "Which accounting software is the best for small businesses in the USA and Canada?" Surfer scored this one high for Zoho, a 71.4% mention rate over the week.

My own run backed it up. ChatGPT crowned QuickBooks Online as the overall winner, then ran through a table of alternatives, and there was Zoho Books, sitting in the "best value" row. Not the top pick, but clearly named, exactly as Surfer reported.

Next, I tested the project management prompt, "Best project management software for a one-man team." Surfer scored this one a flat zero. According to the tool, Zoho is never mentioned at all.

I ran it myself, half-expecting to catch the tool missing a mention. I didn't. ChatGPT named Todoist, Trello, Notion, and ClickUp.

So does the data match? On both prompts, yes. But there's one honest wrinkle you should carry with you, and it's visible right inside Surfer. Look at the trend line on that accounting prompt: over a single week, it swung from near the top down to almost nothing and back up again.

That's not Surfer being unreliable. That's AI answers being inconsistent. The models reshuffle their picks day to day, so any single reading is a snapshot of a moving target.

Sources, gaps, and recommendations

Here's where Surfer is brilliant and frustrating in the same breath. It shows you exactly where you're losing, down to the specific web pages, and then mostly leaves you to figure out what to do about it.

Let me walk you through it, because the diagnosis really is good.

The Source section tells you which websites AI is actually reading to build these answers. For Zoho, it had pulled from 367 domains, 797 URLs, and over 6,000 references, and that pile keeps growing the longer you track. Each source shows whether your brand was mentioned, how often it was shown, and a confidence rating.

The pages themselves are exactly what you'd expect AI to lean on, like Forbes, TechRadar, and big software roundups. These listicles are what the models quote, which is the whole reason visibility in them matters.

Another interesting metric here is the Mention Gap analysis.

Each card pits you against one competitor and splits the sources three ways.

Shared are the pages that mention both of you. Gap is the pages that mention them and not you. Against HubSpot, Zoho had 101 gap sources. Against QuickBooks, 93. It's a literal list of the exact pages where your rival gets named, and you don't.

There's also a Fanout Queries report, which shows the follow-up questions AI asks itself while researching your topic.

So a prompt like "best CRM" quietly branches into "best CRMs integrating with Google Workspace and QuickBooks," which alone spun off ten separate fan-out queries. Those branches are basically a content to-do list the AI handed you for free.

Pricing and plans

If you only want AI tracking and none of the writing tools, there's a tracker-only plan, AI Search Analytics, at $95/month. It covers 50 prompts daily across every engine.

The cheapest plan with content tools, Standard, is $119/month. But, it only tracks ChatGPT, and refreshes weekly. If ChatGPT is all you care about, fine. If you want the full picture of where buyers actually search, you're paying to see a quarter of it.

Pro, at $219/month, is where the tracker gets serious. 50 prompts, refreshed daily, across all five engines. This is the plan I tested, and the one I'd point most people toward.

Above that sits Peace of Mind at $359/month, adding unlimited workspaces, API access for agencies running many brands, and 100 daily prompts.

I’d suggest skipping the Standard tier unless ChatGPT-only genuinely covers you. Pro is the real tool, and everything in this review runs on it.

Who should buy it, and who should skip it

FAQ

AI coding assistants are no longer experimental. In organizations that have adopted them, AI is responsible for roughly half of committed code. The question is, therefore, no longer whether developers will use AI, but whether security programs can govern what those systems produce. AI-generated code is remarkably good at producing syntactically correct code, with a 99.9% pass rate. In contrast, security performance tells a very different story.

As it stands, the average security pass rate is only 56%, meaning roughly 44% of tested code-generation tasks introduce a known vulnerability. As the volume of AI-generated code increases while security performance remains largely unchanged, organizations risk introducing more vulnerable code into their development environments. As a result, software vulnerabilities are now the top breach entry point at 31%, overtaking stolen credentials.

Closing that gap means rethinking how security fits into AI-assisted development. Organizations need to make security part of the development process from the start, choose models with risk in mind and verify code in context before it reaches production. The goal is not to slow developers down; it’s to make sure security can operate at the same speed and scale as the software being created.

Model choice is a security decision

AI models learn from enormous volumes of historical and publicly available code, including examples from open-source projects and developer communities. That training data contains secure implementations, but it also contains outdated libraries, weak validation and insecure coding patterns. While models are designed to produce useful, functional code, without the right controls they can reproduce those insecure patterns at scale.

Organizations cannot control every source used to train a model, but they can control which models enter development. The strongest model tested achieved a 68% security pass rate, while six of the 11 models clustered between 50% and 53%. Model selection is, therefore, not simply a productivity decision but a risk consideration.

Choosing a stronger model can reduce exposure, but it does not make the output trustworthy by default. AI-generated code should still be tested throughout the software development lifecycle using capabilities like static analysis and enforceable security policies. Security must be part of the delivery process itself.

Security needs application context

AI-generated code can appear to work as intended, compiling successfully and passing functional tests while still containing security vulnerabilities. Under deadline pressure, that can create misplaced confidence, particularly when developers accept output without fully understanding how it behaves inside the wider application.

The difference in security performance across vulnerability types shows why context matters. While models achieved pass rates of 83% for SQL injection and 87% for insecure cryptographic algorithms, performance fell to 15% for cross-site scripting and 12% for log injection. Some weaknesses can be identified through a recognizable coding pattern, but others often depend on data flows, authorization, trust boundaries, and how different parts of an application interact.

That’s why verification needs to happen in the workflow and in the context of the application. AI-native analysis can add contextual risk detection alongside established static analysis, surfacing issues such as logic flaws and broken authentication directly in pull requests. The key is to combine contextual analysis with independent testing before code is merged and risk has the potential to spread.

Prioritize risk, not just findings

As AI increases the volume and speed of software development, it also increases the amount of risk security teams have to assess, prioritize and remediate. That adds additional pressure to organizations who are already carrying significant backlogs of unresolved vulnerabilities.

That backlog enables security debt to build as flaws are identified but remain unfixed, meaning known risk sits in applications for longer. The more that debt accumulates, the harder and more expensive it becomes to reduce. Today, security debt affects 82% of organizations, while critical security debt affects 60%.

Those figures point to a growing capacity problem. Organizations are generating and discovering risk faster than they can reduce it. Treating every finding as equally urgent will only make that problem worse. Teams, therefore, need to focus on exploitability, application criticality and business impact so they can direct limited remediation capacity towards the vulnerabilities most likely to cause real harm.

AI-powered remediation can help close the gap between discovery and action by proposing fixes inside the developer workflow and reducing repetitive manual work. But remediation still needs to be governed. A generated fix must be tested, validated and checked for regressions. The objective should be to reduce meaningful exposure, not simply increase the number of findings or fixes moving through the system.

What leaders need to do now

AI-assisted code has become embedded into software development. As a result, leaders should treat its security implications as a change to the software operating model.

That means knowing where AI-generated code enters the organization, evaluating models against relevant security requirements, building verification into developer workflows, and focusing remediation on exploitable business risk. It also means retaining evidence of what was tested, what was fixed, and what risk was knowingly accepted.

The organizations that succeed will not be those that slow AI adoption, but those that evolve their development practices so security can keep pace with software creation. AI is making it easier to generate working code, but the real test is whether businesses can prove it is secure before it reaches production.

We've featured the best endpoint protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

There is no shortage of published thought leadership on the potential of artificial intelligence. But what often gets overlooked is the gap between theoretical potential and the reality of deploying it successfully inside a global enterprise. Gallup’s 2026 State of the Workplace report reveals a stark disconnect: while 65% of employees say AI has boosted their individual productivity, only 12% strongly agree the technology has actually transformed how work gets done in their organization.

Having led tech transformations from both sides — as an executive driving innovation at Canadian National Rail and as a partner accelerating AI outcomes for customers with Tenarai — I’ve learned that successful digital transformation thrives on people. In the agentic era, the unlock for enterprise AI outcomes is empowering employees to unlearn the tools and habits they have relied on for years, helping them embrace new capabilities with confidence.

As the industry shifts from passive AI assistants to autonomous AI agents, traditional software delivery models fall short because they overlook the human-AI connection. Bridging this gap requires elevating change management into a foundational pillar of enterprise AI transformation, aligning tech with the context and culture of the enterprise.

From assistants to active agents

To understand why legacy engineering structures fall short on scaling AI outcomes, let’s look at how technology changes human workflows. Generative AI serves primarily as a task-based drafting tool, keeping humans comfortably in the driver’s seat to review and approve.

Agentic AI, in contrast, is goal-driven, context-aware, and collaborative; acting more like an autonomous digital co-worker. This shift is already reshaping complex enterprise operations, but its success hinges on people’s ability to adapt.

A recent data migration we led for a global logistics leader illustrates this challenge. While moving its data pipeline from a legacy SAS architecture to the cloud presented significant engineering complexity, the greater challenge was actually guiding its long-tenured business analysts in unlearning many years of established workflows.

By pairing technical migration with hands-on change management initiatives, we helped these analysts understand the clear value and benefits of change. Ultimately, winning over their belief system did far more to accelerate our deployment timeline than the underlying technical migration.

This tension reveals the fundamental flaw in traditional AI adoption. When employees transition from controlling task-based tools to supervising autonomous agents, a vulnerability gap is exposed. Legacy IT structures, often operating in silos, are inherently built to fail in this environment, leaving frontline staff stranded between old habits and new autonomous capabilities. Bridging this divide requires installing a modern delivery model engineered around human-centered collaboration.

Architects of trust: FDEs and agentic design leads

To close this adoption gap and scale agentic AI from proof-of-concept experiments to real business outcomes, organizations must deploy a dual-role framework. This model embeds change management, domain expertise, and operational empathy directly into the technical pipeline through two key roles: the Forward Deployed Engineer and the Agentic Design Lead.

Traditional software teams frequently write code in silos based on slow-moving specifications. The Forward Deployed Engineer (FDE) model eliminates this bottleneck. FDEs operate as adaptable technical architects embedded directly at the intersection of business processes, corporate context, and live production code.

Working within cross-functional units, FDEs collaborate closely with specialized AI agents to compress build cycles from months into weeks. By working alongside business users, FDEs rapidly translate operational goals into scalable, production-ready systems while resolving friction in real time.

While the FDE drives technical execution, the Agentic Design Lead focuses entirely on the human-AI connection and organizational alignment. This specialized role reshapes business workflows around agentic capabilities before development begins, securing active adoption at every stage. By analyzing user behavior, the Agentic Design Lead remodels the design process with the actual users in mind, ensuring a seamless human-AI connection is pulled all the way through the final solution.

Managing the risk of unlearning

Guardrails and governance are essential because unlearning decades of defined software habits can be highly uncomfortable for an organization. To scale agentic workflows safely, leaders must establish a clear map of where unlearning is encouraged and where it is not, dividing enterprise operations into two distinct risk categories: deterministic and probabilistic.

Deterministic controls govern mission-critical environments where errors risk physical operations or human safety; such as warehouse operations, heavy rail dispatching, air traffic management, or industrial manufacturing. These high-consequence domains demand rigid, non-autonomous safeguards.

Conversely, probabilistic frameworks support domains where autonomous agents excel, including demand forecasting, targeted marketing, and automated document analysis. For these adaptive environments, leaders should establish secure sandboxes where cross-functional teams can safely experiment, evaluate performance logs, and scale proven use cases incrementally.

Change management as an internal marketing campaign

Even the most advanced AI architecture fails if the workforce resists adoption. Enterprise transformation cannot succeed through executive mandates or static dashboards. Leaders must execute it like an internal marketing campaign, earning employee buy-in across the organization and purposely guiding teams through grassroots engagement.

Organizations can build momentum through interactive workshops, innovation days, tailored training, and immersive boot camps that demystify new tools through hands-on exposure. Meanwhile, low-pressure forums such as team connection meetings or informal brown-bag sessions provide safe spaces to address concerns, establish trust, and cultivate creative confidence across teams. We’ve done all of these activities and though some may seem simple, they work.

Winning in the agentic era isn't about outspending competitors on models, it's about understanding people and navigating them through change. Operational success belongs to the organizations that view change not as a technical hurdle, but as their most important product. By guiding their employees in the art of unlearning, leaders transform workforce friction into adoption, ensuring that as agents take on the heavy lifting, human potential remains the true engine of innovation.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

AI has changed many aspects of the professional world, especially the practice of software development and engineering. A swathe of tools and platforms now enable programmers to use AI assistants and copilots – as well as swarms of agents – to work on building systems autonomously, based on instructions. And for one key figure in the AI space, this was something of an inevitability.

Going viral

In a viral post on the social media platform X, one of the key figures behind the conception of OpenAI, Andrej Karpathy, remarked on the future of programming, provoking a reaction from many in the industry at the time.

ChatGPT had launched a few months prior to that, and developers were already familiar with the basic level of coding that the chatbot could perform, especially in popular languages such as Python and HTML. Much of this was also based on simple text-based instructions – meaning that the AI models could understand and process natural language and churn out complex programming in a way that hadn't been seen before.

Karpathy, who in 2024 co-founded Eureka Labs, an AI education platform, perfectly encapsulated the feeling, with the tech industry in the process of understanding the full implications of this technology.

Good vibes only

Karpathy added in a follow-up post that many examples out there in scientific literature showed that natural language was capable of wielding AI to engage in complex programming. These findings served as a collective precursor to many enterprise features that we see in widespread use today, such as GitHub Copilot, Claude Code, and others.

However, the rise of vibe coding and instructions-based programming also means that the nature of the work is changing for many software developers. Many now simply review code instead of writing it themselves, engaging in validation and bug-squashing.

Thus, the role of software development is becoming something that requires fewer and fewer juniors. This, however, may lead to other unintended consequences – including the fact that the traditional career pipeline is permanently changed.

• A new vibe-coded app offers a simple, browser-based office suite
• Each file is 'saved' in the code for the URL
• The apps come with a variety of options for sharing links/files

AI vibe coding is now at the stage where these apps and agents can clone big-name software packages just by observing how they work. Now there's a new web app, apparently written with AI's help, that puts a simple office suite right in your browser, with no logins or cloud access required.

It's called Tiny Little Office, and it comes from veteran developer and 'technology thinker' Paul Ford. There are 10 different components here: word processor, outliner, presentation tool, spreadsheet, calculator, drawing tool, diagram maker, animator, calendar, and trip planner. You can just click around and start working.

There's one main principle the office suite is built around: the link is the document. Any changes you make to a file are saved as codes in the URL itself, and sharing that URL shares an exact copy of the file. Someone else can take that URL and make their own edits (though they'll stay local and won't sync back with yours).

When you share links, they can be protected by passwords, configured as read only, and set to expire after a certain time period (from an hour to a month). For your own use, multiple files can be saved locally in the browser, or you can choose to wipe everything each time you close down the page.

Six years in the making

Introducing Tiny Little Office tinylittleoffice.com?c=xdgpfud#/v...

— @tinylittleoffice.bsky.social (@tinylittleoffice.bsky.social.bsky.social) 2026-10-16T13:08:46.635Z

The story of the project starts six years ago, when Ford's colleague Rich Ziade built TinySheet, a small ten-line spreadsheet that ran in the browser like Tiny Little Office does now. That was followed by TinyCalendar, but the projects were eventually abandoned, as were the URLs associated with them.

Now the idea is back as a full 10-app suite of products, because "it really is easier to build small than ever before". Ford says he uses the tool to leave himself notes in Slack, and that "if Signal was an operating system, this could, one day, aspire to be its productivity suite" due to its private, simple approach.

"There are lots of tradeoffs," adds Ford. "This is not, in any way, how web apps are expected to work. The links can get unwieldy pasted in chat, and you can't upload video or images, full stop. It's not collaborative like Google Docs. It's collaborative like handing out zines at a street fair."

The Tiny Little Office code is open source and can be viewed and used by anyone, there's support for LLM agents, and you can also find Tiny Little Office on Bluesky. Quite how much of it was vibe coded isn't clear, but according to the docs posted by Ford, Claude Code was doing at least some of the work.

For years, quantum computing lived in a strange place in the cybersecurity conversation: important enough to worry about, but sufficiently distant to leave for another day. When we asked CISO what's keeping them up at night, quantum computing rarely made the top five. It still feels distant, theoretical, someone else's problem for 2030-something.

That calculation is starting to look wrong.

Gartner predicts that by 2030, advances in quantum computing will make conventional asymmetric cryptography unsafe to use. Four years away sounds like breathing room. It isn't. Replacing the cryptography embedded across a modern technology estate isn't a patch cycle. It's a multi-year transformation touching applications, infrastructure, identity systems, suppliers and developers all at once.

The clock is already running. Most organizations just haven't noticed it started or don’t appreciate how little time they really have.

This was never really a technology question

Quantum innovation is moving at pace with IBM, Amazon and Google publishing roadmaps or breakthroughs on error correction and qubit stability. Taken together, they tell a consistent story: the timeline to a cryptographically relevant quantum computer is shrinking, not stretching.

But for the average enterprise, that's not the most useful question to be asking right now. The more useful question is far more uncomfortable: how much cryptography do we actually have, do we know where it lives, and could we change it if we had to?

Cryptography isn't a product category you can point to. It's woven into everything, buried in vendor software, embedded in Transport Layer Security (TLS) handshakes, quietly doing the work inside identity and access management systems nobody thinks about. Gartner's research puts a number on the blind spot: 61% of organizations lack full visibility into their own cryptographic systems.

CIOs and IT leaders cannot migrate away from a risk they cannot see. That single statistic is arguably the real concern than the quantum breakthrough itself.

The attack has already started

Here's the unnerving part. You don't need a working quantum computer today for the threat to be live today. Adversaries, particularly well-resourced state actors are already engaged in what the industry calls "harvest now, decrypt later": hoovering up encrypted data now, sitting on it, and waiting for the day decryption becomes trivial.

For anything with a long shelf life such as government records, medical files, IP, legal contracts, identity documents, a five-year-away breakthrough is still a today problem, because the theft happens now.

Encrypted securely today doesn't mean secure for the data's whole useful life. That's the sentence every chief information security leader should be testing their current strategy against.

Crypto-agility is the actual prize

There's a temptation for CIOs and IT leaders to treat post-quantum cryptography as a swap: rip out an old algorithm, drop in a quantum-resistant one, done. It doesn't work that way. New algorithms aren't drop-in replacements.

Larger key sizes strain memory on older systems, signing operations run at different speeds, and performance assumptions baked into legacy applications quietly break. Moreover, additional standards are still being developed meaning today’s replacement might be tomorrow’s cryptographic debt.

Which is why the capability that actually matters isn't any specific algorithm. It's crypto-agility - the ability to change cryptographic algorithms, keys and protocols quickly, without re-engineering the business around them. And organizations are starting to focus on building it, alongside prototyping or evaluating post-quantum algorithms outright.

Build crypto-agility properly and you're not just preparing for quantum. You're fixing something older: cryptographic technical debt that would have caused pain at the next major algorithm weakness regardless of what triggered it. Quantum is the deadline. Technical debt is the disease.

Additionally, hybrid cryptography which blends classical and post-quantum mechanisms during the transition buys some breathing room. Treat that as scaffolding, not a destination. Classical methods have a shelf life measured in single-digit years.

Why now, when nobody can predict the exact date

There's an understandable instinct to wait for certainty, to see exactly when a cryptographically relevant quantum computer arrives before committing budget. That instinct is the wrong strategy for a transformation measured in years, not sprints.

Organizations don't need to know the date. They need to start building visibility and flexibility now, so they can act when required. The unglamorous first step: build an inventory. Scan the estate, catalogue every cryptographic component, feed the findings into whatever technical-debt dashboard has executive attention. It surfaces expired algorithms, outdated TLS versions and vendor dependencies — and gives the migration a factual basis instead of a guess.

Then prioritize. A marketing database that's stale in eighteen months carries a very different risk profile to a decade-long government contract or an identity credential meant to last a generation. Ask which information still needs protecting after 2030. That list moves to the top. Then ask which use cases could disable the business and tackle those next.

This cannot be handed to the security team

A mistake is treating post-quantum migration as a security project with a security owner. It isn't. Developers need to understand how algorithm changes hit application performance. Cloud and network teams inherit protocol changes. Identity teams own the authentication dependencies.

Procurement needs to interrogate vendors on their own timelines, because a huge share of enterprise cryptography arrives bundled inside third-party software nobody directly controls.

Equally, pushing this down to individual system owners without central coordination is a guaranteed way to fail, leading to inconsistent decisions, no organization-wide picture of progress, and gaps precisely where attackers look first.

A cross-functional Cryptographic Centre of Excellence, pulling IT, development and data security under one roof, isn't governance theatre. It's what keeps the response from fragmenting into a patchwork.

The real question isn't about quantum at all

Strip away the acronyms and this stops being a story about qubits. It becomes a test of how resilient an organization's architecture actually is, whether it knows where its cryptography sits, understands which data needs long-term protection, and can change cryptographic mechanisms without rebuilding from scratch.

The organizations most exposed won't be the ones with the weakest algorithms. They'll be the ones still treating quantum as a distant technology story rather than a resilience problem sitting on today's books. By the time the threat is obvious to everyone, the hard part—untangling years of cryptographic dependencies—will already be behind schedule.

The quantum era is still approaching. The migration effort isn't.

We've featured the best endpoint protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Windows 11 is now 'up to 20% more responsive,' Microsoft tells us, based on benchmarking typical interface workloads
• Apparently, this performance honing 'can make an especially meaningful difference on PCs with 8GB of memory'
• A new Photos app is also out, which is a good deal speedier, alongside changes for Windows 11's Gallery and photo viewer

Microsoft has underlined how it's making Windows 11 faster and ensuring the OS uses system RAM more optimally, while also introducing a new and more performant Photos app.

In a blog post, Microsoft explained how it's been tinkering with Windows 11 to make it run faster, and that: "The result is a Windows experience that's up to 20% more responsive, with improvements you can feel as soon as you boot into your PC."

That 20% figure is quite vague, but Microsoft says it's based on measured speed increases in Windows 11 search, the right-click menu, Settings, Start menu, as well as File Explorer (the folders on the desktop) and the Snipping Tool.

These are performance boosts seen in the most recent October update for Windows 11 (well, they were tested in the September preview update, but it's the same thing) compared to the OS as it stood at the beginning of 2026. Naturally, performance will vary by device and configuration, but the gist is clear: there's a sizeable speed boost.

Microsoft observes that a big part of this nippier performance lies in better use of system resources and RAM, with improvements in memory management and efficiency.

The software giant notes: "That work can make an especially meaningful difference on PCs with 8GB of memory. Windows can now keep up to 2GB more of your active data in memory instead of moving it to slower storage, helping your PC stay responsive as you open more apps or browser tabs."

We're also told that Windows 11's background RAM use is being trimmed back, so on PCs with only 8GB of memory (or less), widgets load only when opened, reducing background workload by 200 MB.

In this blog post, Microsoft also reminds us of all the work it's done to fix Windows 11. From key interface changes for the likes of the Start menu and taskbar, through to improving Windows updates, we've also witnessed recent changes that make search faster, more accurate, and useful (and more efficient with RAM usage).

There's an entirely fresh revelation here, though, in the form of the new Photos app as mentioned at the outset.

Microsoft says it rebuilt the native Photo editor around WinUI 3 (the modern interface framework for Windows 11), and it now launches 25% faster than before. It sports a reorganized set of menus and features, including repair (unwanted object removal) and relight (to swiftly adjust the lighting of a photo).

The Photo viewer also launches faster (by about 15% on average, Microsoft claims), opens photos up to 30% faster, and has a cleaner, more streamlined design, so you see more of the picture.

It also adds a new Gallery that brings together photos on your PC and in the cloud (OneDrive and Apple iCloud). Again, expect faster launch times, more responsive scrolling, and natural-language search, so you can ask for 'pictures of my dog on the beach,' and Windows will find them (hopefully).

Analysis: progress in the right direction

Microsoft previously promised to work hard on memory optimization for Windows 11, and it's presenting clear evidence of that here, which is good to see. As I've said before, though, it's a puzzle why Microsoft hasn't done this before now, and you've got to wonder why, for example, widgets were ever allowed to eat up memory if they're not being used.

At any rate, this is useful progress, and there's a very handy bunch of enhancements for the Photos app, too, along with the associated Viewer and Gallery, with impressive performance improvements therein.

Microsoft is crystal clear about there being significant speed-ups present in Windows 11 all-round with the latest update for the OS, although we're yet to see any commentary from users who are noticing this. Admittedly, the October update has only been out for a couple of days, so it'll be interesting to watch the ongoing reaction.

Overall, this seems like continued progress in the right direction, and we can expect more performance tuning before 2026 is out.

Love it or hate it, Microsoft's Windows operating system has been a big part of many of our lives. If you've ever used a PC during the past four decades, then it's very likely you've used Windows in some form (unless you've stuck to Macs, or are a Linux fan).

Despite each iteration of Windows sharing basic similarities with those that have come before it, each version also has its own identity and quirks.

Some Windows releases are more iconic than others, but I am sure we've all, at some point, wondered what version of Windows best matches our personalities. So, I've put together this highly scientific quiz that will answer the question once and for all.

Please don't sue us if you end up matching with Windows 8, though. But maybe think about your life choices.

Happy with your result? Angry? Confused? Share your results below and let us know!

For technology teams, connecting telemetry has traditionally been seen as a route to greater visibility. The assumption is straightforward: the more information available about what is happening across an environment, the easier it should be to understand problems and respond to them.

But as modern technology environments have grown in scale and complexity, that equation no longer looks quite so simple and teams have found themselves dealing with fragmented telemetry and unclear signals. It now takes longer to understand what’s gone wrong, and even longer to fix it. And to add to the challenge further, the cost of storing and processing all this telemetry and data continues to be on the rise.

This is prompting organizations to reconsider where the problem should be addressed. Rather than collecting everything and attempting to make sense of it later, attention is shifting towards improving telemetry earlier in its lifecycle. Filtering, enriching and directing it from the outset can make the information reaching teams more relevant and ultimately easier to act on.

More data doesn’t always mean better insight

The volume of telemetry available isn’t the fundamental issue. Its quality is. As AI-driven capabilities become deeply embedded in software delivery and operations, the quality of observability telemetry becomes even more important. Collecting telemetry without filtering out what isn’t needed can leave employees spending more time managing information than extracting value from it.

Fragmentation makes this even more challenging. With data sitting across several tools and agents, each of which generates varied forms of signal and alert. When an outage occurs, teams may then need to piece together inconsistent information from across those tools before they can understand the root-cause of the problem. Consuming valuable time while the impact on the business continues to grow.

There is also a significant financial cost attached to this approach. Processing and storing telemetry and data at scale requires considerable resources, particularly when organizations are retaining information that contributes more noise than insight. In the traditional sense, this isn’t a technology problem. It’s a telemetry and data quality concern, and the longer it’s ignored, the worse it usually gets.

Rethinking where telemetry is processed

One way organizations are responding is by moving intelligence further upstream in the technology stack. In practice, that means making decisions about telemetry earlier – filtering out what isn’t useful, enriching important information and routing meaningful telemetry before it reaches downstream systems.

Teams are beginning to collect, process and route relevant observability telemetry in real-time before it is ingested into downstream systems. As a result, there is a less volume to handle, reduced expenses and cleaner telemetry to deal with since only the relevant signals are sent downstream.

Open standards such as OpenTelemtry are making it increasingly possible to take this approach across complex environments. They allow teams to gather telemetry consistently without being dependent on a particular vendor or set of tools. That flexibility is particularly valuable when working with intricate, distributed systems that span several clouds, security and storage platforms.

Furthermore, as environments continue to evolve, it makes adapting to them that much easier.

AI needs reliable information to make reliable decisions

The implications become particularly important as organizations introduce more AI-powered capabilities into their operations. AI systems depend on accurate, contextualized telemetry and data to produce insights and automate decisions.

Teams are not only slowed down by incomplete or cluttered telemetry, but it’s this simple: poor quality telemetry leads to poor quality insights. When those insights are applied to large-scale automated decision-making, the margin for error drastically lowers.

The same principle applies to trust. Teams responsible for automated systems need to understand what has happened and explain why decisions about performance or security have been made. Doing so becomes much more difficult when those decisions are based on inconsistent information.

And this requirement for confidence matters far beyond just the technical teams. As senior management and board members are being increasingly asked to support AI-driven decision making, they need to trust the information those decisions are built upon. Credible, consistent telemetry and data can make that conversation much simpler.

The operational advantages are also evident. Teams can work quicker if they aren’t continuously battling erratic telemetry and data. They spend less time on manual research, detect issues early, and automate with greater confidence.

A strong telemetry and data foundation and a unified, context-driven platform not only enhance daily operations but also increases the successes of the AI-powered capabilities the organization invests in over time.

Quality has to be built in from the beginning

Ultimately, faster innovation and better customer outcomes depend on organizations having useful telemetry and signals available when they are needed. The good news is that it doesn't require starting from scratch, it's about making choices early in the telemetry lifecycle.

By introducing intelligence at that stage, teams can improve the quality of what ultimately reaches their observability platforms and ensure it can scale as their environments grow. The advantage will not come from simply collecting more. It will come from ensuring the telemetry an organization does collect is relevant, trustworthy and ready to support action.

We've featured the best business intelligence platform.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• A new version of Tiny11 Builder is out that's 'completely redesigned'
• The developer has revamped the interface and made it more modern
• This means it's easier to use — and faster — when creating your own slimline take on Windows 11, for those keen on de-bloating the OS

If you feel Windows 11 is too bloated, one way of tackling that is to make your own light version of the OS – and it's possible to do that with a utility that's been around for a long time now.

Tiny11 Builder lets you create a custom version of Windows 11 without all the bloat you don't want — and the good news is that a new release of this tool makes it a lot more user-friendly.

The Register spotted that NTDEV, the creator of Tiny11, has released a "completely redesigned" Tiny11 Builder, with an all-new fancier interface that makes decluttering Windows 11 a lot easier.

Previously, the Builder tool relied on a cruder setup process leveraging PowerShell scripts, but with this new version you're presented with a WinUI 3 desktop app. In other words, a fully modern Windows 11 app, as you can see from the screenshots, which is a far nicer — and easier — way of working than before.

Builder is also faster now, too, with the developer explaining: "The new Builder is massively more reliable, with features like auto-resume, custom configurations (!) and leveraging the full power of modern PCs in order to speed up the process."

Analysis: a slicker way of cutting down Windows 11

In short, making your own installation image is less intimidating for those who aren't so technically confident, whether you're concocting a custom take on Windows 11, or just using the standard Tiny11 build. The latter removes a whole bunch of Microsoft apps and general bloat, but it keeps the central pillars of the OS, including Windows Update and Microsoft Defender (the built-in antivirus).

That said, there are caveats for those wishing to avail themselves of a super-cut-down version of Windows 11. As a custom installation of the OS, the end result isn't officially supported by Microsoft, and if you go for the default Tiny11 configuration, bear in mind that it's quite aggressive in what it strips out (so you may find features that you use suddenly aren't present).

Definitely steer clear of the 'Core' version of Tiny11, as that removes all sorts of important facets such as Windows 11 feature updates (though not security patches) and Microsoft's recovery environment (for a super-light install). Even the developer recommends that the Core install is for testing in virtual machines only.

Note that a Tiny11 install still requires a valid Windows license. While the app can be used to install Windows 11 on an older PC which doesn't meet the official system requirements, the developer doesn't guarantee that this will work flawlessly, noting that "local-account and hardware-check settings are best effort, not guaranteed bypasses on new builds". As stated, you can also use the tool to install Windows 11 with a local account (something Microsoft is still working to prevent), at least in theory, but again the bypass for this isn't guaranteed.

It's also worth bearing in mind that as with any third-party software, you use Tiny11 at your own risk.

Eric Steven Raymond, widely known as 'ESR' has been a large presence in the software development space for many decades, making key contributions to the school of development, programming languages, and the open source movement. Given the rise of AI, ESR has signalled the end of the closed-source model of software development.

Reverse engineering

ESR delivered this verdict on X recently, adding to his recent history of commenting on the rise of large language models (LLMs) and the impact on software.

In his post, the software pioneer remarked on how a fun experiment involving the use of AI to decompile an old game and port it into a new format alerted him to the potential for this practice to happen on a much grander scale.

"The thing is, I thought of this as a fun stunt but didn't have any confidence that the technique would scale up to large, real programs," he wrote, adding: "I have since learned that people are now doing this sort of thing with entire AAA games, which are among the most complex software artifacts ever to be shipped as a binary. If they can be decompiled, anything can be."

A paradigm shift

The implication of this practice, ESR said, is that there is no longer a secrecy moat around almost any software at all – and that the only way for the industry to maintain this "moat" is by hiding their executables behind SaaS shields. In that sense, cloud-enabled software will be much harder to crack open and reverse engineer.

Rather than closed source being gradually driven toward extinction by cost gradients, as he foresaw, he now believes that closed-source software may be "wiped out in a technoapocalypse" – albeit with the caveat that some laws will prevent the mass decompilation of binaries.

His thoughts mirror the views of Nvidia's CEO Jensen Huang, who put a twist on Marc Andreessen's famous comment that "software is eating the world" by suggesting in 2017 that "AI is going to eat software". This is something that we are arguably seeing play out as AI improves and becomes more prevalent across the enterprise.

• Ray-Ban Meta users have noticed unwanted voice recordings
• Meta says the issue is resolved but hasn't said what went wrong
• Turning off voice activation for the AI seems to fix the problem

If you're already facing a barrage of criticism over perceived privacy intrusions, then you really don't want more allegations of spying loaded on top — but that's what Meta is dealing with when it comes to its Ray-Ban Meta smart glasses.

Over on Reddit (via Android Authority), one Ray-Ban Meta user says they've found audio conversations and text transcripts saved by the glasses that were never requested or associated with a "hey Meta" command to invoke the on-board AI.

According to the individual, it's happened on three different pairs of specs (including 2nd and 3rd generation models), and others have posted in the conversation to say they've seen the same behaviour on their own Ray-Ban Metas.

Meta has responded to Android Authority, without providing much detail, and says the issue "has been resolved". We've reached out to Meta ourselves to see if there's any further comment, and will update this article if we hear back.

Check your voice logs

If you own a pair of Meta smart specs, you can check voice recording activity by opening the Meta AI app on your phone, then selecting Glasses > Device settings. Tap on the name of your glasses, then Glasses privacy > Voice activity log.

You can delete voice recordings from here, but you can't stop them being captured in the first place. In theory, they should only be recorded when you say "hey Meta", or activate the AI with a button push.

The associated privacy notice does mention voice services being "inadvertently" activated, which may be what's happened here. It also says some recordings may be reviewed by Meta staff to "help improve Meta's voice services".

Some Redditors have resorted to turning voice activation off for the integrated AI, which you can do from the same device settings screen in the Meta AI app. Clearly though, that's an inconvenience users shouldn't have to put up with to stop this from happening.

AI Agents are unlike any software development witnessed before. Traditionally, trustworthy software systems have followed a structured approach to being built and developed. One that plans, builds, tests, and maintains software to ensure its reliability and scalability. This created the Software Development LifeCycle (SDLC) (1) framework.

But AI agents are different: they are a team of autonomous, probabilistic actors that work most effectively in a real-time, event-driven environment, enabling the agents to coordinate with real-time data to develop, learn, and work.

To scale an enterprise workforce of AI agents, organizations need to combine the principles of employee management with software development rigor. These agents are not static software that can be coded and developed, due to probabilistic language models. They are built, evaluated, and governed differently.

This requires organizations to move from traditional SDLC to an Agent Development LifeCycle (ADLC) (2), which utilizes the structure and repeatable process of the SDLC framework and combines it with the development and management of AI agents.

A helping hand from HR

Since AI agents perform roles within an enterprise, the ADLC should draw on lessons from the human employee lifecycle.

When an organization brings on a new team member, there is a deliberate process: their role is defined; they’re given access to the systems they need; they are trained; and they are integrated into a team. Once they are empowered and put to work, a manager oversees their activity, monitoring and developing their performance over time.

As autonomous actors operating within a system of governance, AI agents deserve and require that same treatment.

AI agents need a similar structured approach to onboard and integrate them into a business and deliver the efficiency benefits they’re designed for – that means they need real-time contextual data to do their jobs!

Treat AI agents like humans at the hiring stage

Picture an organization where every employee is a generalist, there’s no hierarchy or structure, and all communication happens via synchronous one–to–one phone calls. No email. No Slack. No shared systems.

The organization can’t scale. Information doesn’t flow. People drown in data from every direction, with no real understanding of what’s going on or what to do. There are no accountability structures, no access controls, no coordination mechanisms beyond individual conversations.

You wouldn’t drop a new human hire into this chaos. Effective organizations look different, counting on people who are specialized and organized hierarchically, so coordination happens in parallel and communication is premised on asynchronous channels that let information flow without requiring everyone to be online at the same time. The same goes for onboarding AI agents.

AI agents and employees are no different in the eye of an agent mesh

Most early agentic AI deployments work as complex monolithic agents tasked with doing everything, accessing all data, communicating synchronously, and trying to maintain very large context or working synchronously with sub-agents. The result? A system that’s brittle, expensive, inconsistent, and completely unable to handle real enterprise complexity.

This is where an agent mesh can act as a development and runtime platform, helping build AI agents that actually work for a real-time enterprise, specialized for particular functions, organized hierarchically, and orchestrated by a development and runtime platform that specializes in delegating tasks to appropriate agents.

Communications between agents are asynchronous and event–driven, and access to data and systems is governed by role-based access controls, giving agents exactly the permissions they need and nothing more.

So, let’s look at how agents can be “onboarded” through an ADLC much like a new employee joining the organization:

1. Treat agent setup like drafting a job description

Before a human employee starts, you write the job description. You define what role they’ll fill, what they’re responsible for, what behaviors you expect, and what boundaries they operate within.

An agent mesh’s agent builder can use an internal AI agent to provide a guided interface for defining an agent’s purpose, scope, and configuration. You set up the instructions and system prompts that shape the agent’s persona and behavioral parameters. And you configure guardrails – hard constraints on what the agent can and cannot do – that prevent it from going off–script or taking unsafe actions.

2. Get the agent up to speed with tool and system access

A new employee’s first weeks are spent getting access to tools, systems, and data. Agent onboarding is the same.

An agent mesh should include a number of data connectors – pre-built integrations to enterprise databases, data warehouses, data lakes, enterprise application APIs, and MCP servers. This enables agents to access real-time data from day one.

Access is provisioned through role-based access controls (RBAC), enforcing least-privilege principles so agents only see what they need to see. It can also define skills that go with internal tools so agents know best how to use them.

3. Passing the probation period – stress-test agents before deployment

After getting access to systems, human employees undergo training. This is where general ability becomes company and job-specific skills. You wouldn’t put a new hire in front of customers without first verifying they know what they are doing. The same logic applies here.

An agent mesh can provide an Eval function that uses AI to suggest tests for agents, allows you to add more and then run them against your agents. This empowers the organization to validate agent competence against defined success criteria during development. This supports both initial testing and regression tests as you make changes.

4. Keep humans in control – validate AI actions in real-time

Even the most capable employee gets close oversight when they are new to a role. Supervision isn’t micromanagement; it’s the safety net that ensures quality, builds trust, and catches errors before they compound.

An agent mesh with human-in-the-loop architecture would support routing of specific agent actions or decisions to human reviewers for approval before execution. This is important in agentic systems because LLMs are non-deterministic, they make mistakes so when the impact of agents being wrong is too risky, humans can validate their actions, responses or conclusions.

5. Teamwork makes the AI agent dream work

The most transformative phase of the employee lifecycle is when individuals become part of high-performing teams, where collective capability exceeds the sum of its parts. For AI agents, this is where things get really interesting.

An agent mesh would support workflows that define repeatable processes, such as the steps involved in approving a loan, providing an insurance quote or opening an account. Dynamic orchestration is where the mesh comes alive: orchestrator agents route work to the right specialists, in sequence or in parallel, adapting in real time when the input is fluid and the path forward must be reasoned through.

The result is a multi–agent mesh topology – hierarchical agent organizations coordinating specialist agents across various functions that mirror the structure of an effective human organization.

6. Continuous AI improvement – catch performance drift early

Effective organizations don’t deploy employees and forget about them. They monitor performance, provide feedback, and create mechanisms for continuous improvement.

An agent mesh’s visualizer would give a real-time graphical interface for tracing agent interactions, tool and LLM calls, and decision pathways, so you can see exactly what your agents are doing and why. Ongoing evaluations detect performance drift or emerging failure modes via online evals that run in the background, monitoring production execution.

Further, Open Telemetry instrumentation can surface performance trends over time, giving you the data to make informed decisions about when and how to tune your agents.

Turning AI agents into mission-critical assets requires a new deployment approach

To turn AI agents from simple experiments into high-performance operational tools requires a whole new mindset for building and developing them. Traditional SDLCs work for predictable code, but AI agents act on their own! As a result, they require clearly defined roles, secure tool access, strict testing, human oversight, and constant monitoring.

To successfully deploy, manage, and scale a workforce of AI agents with confidence, organizations must deploy an ADLC on a real-time and event-driven platform.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The hardest part of retail AI may soon be the part nobody puts in the demo.

A model can recommend a markdown, predict a stockout, adjust a picking route or identify a pattern in store traffic in milliseconds. But a retailer only gets value when that decision becomes an accurate action in the physical store. Whether it’s a price change or a sensor reading, it has to arrive promptly for it to be useful.

There is already a striking gap between AI ambition and execution. Deloitte's 2026 survey of 200 retail and consumer products executives found that 75% consider AI a top strategic priority, yet only 16.5% can quantify the return they are getting from it. Clearly, retailers believe in AI. The harder problem is turning isolated capabilities into something that works consistently across the business.

For physical retailers, the challenge does not end with models, data or software. AI-driven decisions have to pass through stores containing thousands of devices, existing networks and years of installed technology. In complex environments, the building itself can affect connectivity. As the number of connected endpoints grows, so does the importance of the wireless foundation carrying information across the store.

For CIOs and CTOs, that means the next constraint on AI may be something that has been underappreciated: whether the store estate underneath it can execute reliably at scale.

A store is not a cloud workload

Physical retail introduces conditions that software teams cannot abstract away. Stores have walls, metal shelving, refrigeration, back rooms, loading areas and legacy systems installed. They also have trading hours. Infrastructure changes that look simple on an architecture diagram can become expensive or disruptive when repeated across hundreds of sites.

Take inventory, for example. A forecasting model can improve a replenishment decision, but store-level execution may depend on shelf sensors, handheld devices, picking tools and accurate product information all exchanging data.

Computer vision, digital displays and environmental monitoring add further endpoints and a wider mix of data, from routine device updates and sensor readings to visual content and real-time operational data. These workloads place very different demands on the network.

Some require high bandwidth or low latency. Others involve large populations of battery-powered devices sending regular updates across difficult environments. Individually, none is necessarily a network problem. Collectively, they can become one.

If a pricing decision cannot reach every store quickly, a replenishment signal arrives too late or each new application requires another costly infrastructure layer, the retailer loses part of the value the technology was supposed to create.

One response is to add another gateway, controller or application layer for the next project. That can work for a pilot. Repeated often enough, though, every new device gains a separate integration path, security owner, management console and failure mode.

In difficult store environments, adding project-specific infrastructure to overcome coverage gaps can also change deployment economics. The technology works, but operating it becomes progressively harder.

The scaling problem appears after the pilot works

Retail technology teams are used to asking whether a new application meets its business case. AI adds another question: what happens to the infrastructure if the application succeeds?

A pilot may involve one store and a limited number of endpoints. Scale changes the economics. Coverage has to work throughout the building, updates must remain reliable during normal trading and security teams need a consistent way to onboard and manage devices. Operations teams also need clear ownership when something fails.

Connectivity should therefore be treated as an architectural choice, not a specification buried inside each procurement. The objective is not to replace every existing network or create one for every application, but to match workloads to the wireless layer best suited to their requirements while keeping the estate manageable.

Current designs may be adequate in some stores. In large or difficult formats, or where suitable retail IoT endpoints are growing quickly, an additional purpose-built wireless layer may be justified. The decision should be deliberate rather than inherited one project at a time.

Different workloads need different wireless foundations

The future-ready store is unlikely to depend on one network doing everything. It will more likely use a combination of connectivity layers, each supporting the workloads it suits best. Now, retailers should consider whether enterprise Wi-Fi should remain the de facto transport layer for every new category of connected device.

As electronic shelf label estates expand and retailers introduce more sensors and retail IoT applications, suitable workloads may be better served by a purpose-built wireless layer alongside the enterprise network.

This should not mean an isolated network for each use case, but a scalable, reusable connectivity layer able to support large device populations over time. Security belongs in the same conversation. Every endpoint needs consistent management, from trusted onboarding and credentials through to decommissioning.

Open standards remain important because they provide a more consistent foundation for future devices, applications and platforms. But openness alone is not the architecture. The decision is which workloads belong on which connectivity layer; standards help those layers remain manageable and adaptable.

The most successful retail AI programs will be judged by more than the quality of their predictions. As connected estates grow the test will be whether organizations can turn those predictions into dependable action across real stores.

Before approving the next AI pilot, retailers should ask what will their estate need if it succeeds, which workloads will grow, and should they all continue to be carried in the same way?

If success means more devices, richer data, more frequent updates or deployment across a much larger estate, the wireless architecture supporting those workloads has to be a central part of the AI strategy.

Retailers are not investing in AI tools or connectivity for their own sake. They want to protect margin, improve availability and make stores easier to operate. AI capability and AI scalability are different problems.

The objective should be an architecture in which new workloads can use the wireless layer best suited to them without forcing the estate to be rebuilt each time. The retailers that design for both will be in a much stronger position to turn ambition into reliable, repeatable execution.

We've featured the best AI chatbot for business.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Clicking 'Learn more' would skip Microsoft account demands when setting up Windows 11
• It now shows user data policies, so you'll have to sign in with your account (or create one)
• Having an account can make certain things easier, but it's one less choice for consumers

Microsoft has finally fixed one of the easiest ways for attackers to bypass its account requirements when installing Windows 11, but it's bad news for anybody who didn't want to sign in with – or create – a Microsoft account during setup.

Previously, users only had to click the 'Learn more' hyperlink on the sign-in screen to be taken to a local account creation screen.

Ultimately, it marks another step in the company's ongoing efforts to make Microsoft accounts mandatory during the initial setup of Windows 11 on consumer PCs.

Microsoft wants you to use an account to activate Windows 11

Before the fix, users could click the 'Learn more' link and be taken to a screen that allowed them to enter a local username, create a password, configure security questions and complete Windows 11 setup without ever entering their Microsoft username or password.

Instead, the link now opens a page that explains how the company managers user data.

The change "has been recently quietly patched server side," Bob Pony said in a Bluesky post after revealing that "a new method for skipping the mandatory Microsoft Account login in Windows 11 Home Edition has been discovered" just one month ago.

Local accounts remain supported to a certain degree, but creating them either requires technical knowledge to use a third-party tool or command-line methods, or enterprise versions of Windows rather than consumer-grade operating systems.

There are some benefits to signing in with a Microsoft account from the get-go, like speedier setup of system integrations with OneDrive, the Copilot app and other Microsoft products, however it eliminates user choice for those who prefer not to tie their PCs to their online account.

Discussions about passkeys replacing passwords as the go-to sign-in method have rapidly accelerated. The National Cyber Security Centre (NCSC) has acted as the catalyst for this transition, describing the shift as "overhauling decades of security practice". Finally, the industry has reached a collective point of recognition: traditional, static passwords are a weak defense against modern, automated threats.

AI has turned cybercrime into a highly efficient, industrial-scale enterprise. With researchers uncovering agentic ransomware attacks, we are fast approaching a reality that outpaces traditional human-centric defenses. Cybercriminals routinely deploy industrial-scale AI tooling to orchestrate and deploy vast credential-stuffing attacks and hyper-realistic phishing campaigns at machine speed and scale.

For organizations, the adoption of passkeys and other passwordless verification methods is dependent on the maturity of existing identity and access management (IAM) systems. Of the businesses with less mature, or no IAM strategy in place, only 30% are embracing password alternatives. Without passwordless systems, threat actors will continue to undermine digital trust.

User friction and weak protection

For years, identity security was treated as an isolated concern for the operations department. However, as customer experience dictates corporate revenue, delivering seamless IAM has become a commercial imperative.

Passwords force a broken trade-off: making users create unique, complex strings of characters results in severe password fatigue, while cutting corners by reusing credentials and using easily guessable phrases leaves gateways vulnerable to threat actors. When security kills usability, the business model breaks.

Another barrier passwords create is that, although most of the British public is genuinely worried about identity theft and fraud, only 17% fully trust organizations managing their identity data. Passkeys directly resolve this trust deficit through cryptographic authentication.

Because private keys stay securely on the user's local device and are never shared with a central server or application, there is no central database for hackers to breach. Passkeys are also phishing-resistant, removing the risk of users falling prey to social engineering attempts. Unlike passwords, they are also cryptographically bound to a specific, verified domain name, adding another layer of technological defense against attackers.

Securing autonomous identities

The shift towards agentic AI – fully autonomous systems operating with minimal human intervention – shatters traditional authentication models. Passwords were fundamentally built for human brains to remember.

An autonomous AI agent, such as autonomous digital proxies, personal shoppers and synthetic workflows, cannot "remember" a password without creating severe vulnerabilities, such as hardcoded credentials or shared session tokens exposing permanent risks.

If an agent inherits over-privileged human credentials, a compromised or rogue AI can exfiltrate sensitive data at machine speed, vastly outpacing the response time of traditional security methods.

To secure the growing autonomous workforce, AI agents must be given their own distinct, first-class identities with task-specific, time-bound permissions. Passkeys and decentralized, cryptographic credentials secure these machine identities and allow enterprises to securely verify and audit agentic workflows and interactions, without exposing underlying human credentials.

From passwords to passkeys

The technology industry is driving the momentum behind passkey adoption. Tech giants like Google and Amazon are seeing the need to improve customer convenience and account security, positioning passkeys as their default authentication standard. Businesses that delay moving towards passwordless frameworks are left defending a perimeter built for an entirely different era.

Although most organizations now see the importance of adopting passwordless options, transitioning away from traditional passwords is hindered by real-world complexities. Many core IT systems and legacy infrastructure are reliant on passwords, which remain deeply embedded within business processes.

IT leaders therefore fear that updating these systems will demand massive code rewrites and extensive specialist development resources, meaning replacing them requires careful orchestration.

To address this challenge, modern, low-code or no-code identity platforms allow organizations to visually map, test and deploy sophisticated authentication journeys seamlessly. This enables businesses to introduce modern passwordless options like passkeys or biometric prompts progressively, without disrupting the active user workflow or having to replace entire underlying legacy systems.

By allowing teams to modernize security processes without affecting user experience or uptime, advanced identity orchestration becomes a critical business enabler.

The future of trust starts with identity security

The threat landscape continues to be redefined by automated threats, deepfakes and autonomous software. Against these sophisticated attacks, the concept of static, point-in- time logins is obsolete. The NCSC’s call to action has been the final push many organizations need to drop passwords and rethink how to safeguard digital trust.

Adopting an identity-first approach centered on passkeys and robust identity orchestration allows enterprises to reduce user friction, strengthen the security of both human and machine identities and establish the continuous trust needed to grow confidently in an AI-driven environment.

We've featured the best antivirus software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Microsoft CEO Satya Nadella says Copilot Code and local AI make every Windows PC an infinite software factory with no cloud token spend
• Rather than killing app stores, Microsoft is moving app hosting into Copilot, while IT leaders stay wary of citizen-built tools
• Microsoft's own engineers say local inference doesn't make sessions offline, and the routing ships later this month

Microsoft CEO Satya Nadella has a new pitch for the Windows PC, and it targets one of the most lucrative ideas in consumer tech.

In a post on X summarizing Microsoft's recent Windows and Surface event, Nadella told followers they no longer have to settle for whatever an app store happens to stock, because Copilot can now build the software they need on the desktop without spending any cloud tokens.

He wrote this idea without mincing any words, saying, "Your PC becomes an infinite software factory."

From Gates' factory to your desk?

The narrative is a striking line from the head of a company that sells more software than almost anyone.

But read Microsoft's own engineering documentation for the same launch, and the infinite factory starts to look rather finite: it runs best on a $5,899 laptop, it still reaches into the cloud, and the businesses Microsoft most needs to convince have spent the past year telling researchers they are worried about exactly this kind of thing.

More importantly, the factory metaphor is not new for Nadella: In a 2025 staff memo, he recalled that Bill Gates had envisioned "not just a software company, but a software factory," and argued the AI era demanded a rethink.

At a recent customer gathering, the CEO said Microsoft had progressed from code assistance through agentic coding "to software factory all within one year," and he now wants customers to open software factories of their own, spending more with Microsoft along the way.

The October 7 2026 launch builds on this ambition: a Windows Experience Blog post by Windows and Devices chief Pavan Davuluri describes hybrid intelligence, in which Copilot reads local files with permission, acts across Windows, and hands work to on-device models.

Copilot's Code experience builds an app, dashboard, or workflow from a plain-language description, and on Windows, Microsoft says it can produce native apps from one prompt. Jared Spataro, Microsoft's chief marketing officer for AI at Work, argued that building such tools is becoming "as basic a skill as writing a memo or modeling a budget."

The engine underneath is MAI-Code-1.1 Flash, Microsoft's in-house 137-billion-parameter coding model, which Nadella's post said is now optimized to run on a PC. There is a catch, however: despite Microsoft painting it as a situation where consumers are finding their AI needs "outpacing what their cloud budgets can support."

Microsoft's own engineers are more grounded about what the process entails, saying GitHub Copilot's automatic mode decides task by task whether work runs locally or in the cloud, and one diagram caption is blunt: "local inference does not make the session offline."

That routing doesn't ship until the end of October 2026; the hybrid Copilot features reach Copilot+ PCs over the coming months, and Code is rolling out first through Microsoft's Frontier early-access program. This is interesting, but somewhat predictable, because Nadella's approach, as stated, would cannibalize Microsoft's token-based business model.

An approach that routes harder tasks back to the Azure cloud makes far more sense by conserving compute where possible while still putting Microsoft's cloud compute at the center of things.

The hardware is another factor that one might be overlooking here: according to the Command Line post, the on-device version of MAI-Code-1.1 Flash is quantized down to 53GB, roughly an 80% reduction from the cloud model, and peaked at 75.5GB of memory at full context on Microsoft's test machine, a Surface Laptop Ultra built on Nvidia's RTX Spark.

That matters because eight Surface Laptop Ultra configurations are in play; only one has more than 64GB of memory, per TweakTown: the 128GB model at $5,899, which costs $3300 more than the base model and isn't available to purchase yet, meaning that potential users can't really test out how well it does compared to cloud-based solutions at a time when all performance-related figures come from Microsoft.

The death of modern enterprise software? Not yet

Despite the hardware limitations in play, it might indicate a death sentence in the making for most basic enterprise-level software solutions, even as more ambitious ports have already been in play: Developer Brandon Thomas, for one, has just released seven open source Adobe Creative Cloud alternatives that he says he built with Anthropic's Claude Opus 5.5, declaring that "software is over."

His own roadmap, however, puts the Photoshop stand-in at roughly a quarter to a third of what a professional needs for daily work. Building software has become cheap. Building software someone can depend on has not.

The bigger issue will be enterprise IT, which will, I suspect, keep the gates closed for now; enterprises don't just buy features; they buy someone to call when it breaks, a patch schedule, an audit trail, and a vendor that carries liability.

Getting an open-source or AI-assisted app someone designed with zero liability in tow that could break things, create security risks, and may not come with future software updates that resolve either issue.

The survey data points the same way: a June 2026 Retool study of 307 CTOs, CIOs, and CISOs found 93% concerned about vibe-coded tools reaching production, though Retool sells an internal-tools platform pitched as the governed alternative.

A Harris Poll survey for Dataiku of 600 CIOs found 81% worried that citizen-built AI could expose sensitive data, and over half had already discovered unsanctioned AI in their organizations.

Gartner, meanwhile, predicts that over 40% of agentic AI projects will be scrapped by the end of 2027, with analyst Anushree Verma describing many as "mostly driven by hype and are often misapplied."

Microsoft's own launch materials quietly concede the same, even taking measures to ensure that agents do not run rogue while building software or solutions on-device.

The Command Line post admits Copilot's built-in file tools are policy-checked inside the app rather than isolated by the operating system, and GitHub requires Business and Enterprise administrators to enable a policy before staff can use MAI-Code-1.1 Flash at all, while Microsoft Execution Containers, which are now generally available, allow IT to define what agents can read, modify, or execute.

The real test for Nadella's factory isn't individuals with a $5,899 laptop and a tolerance for alpha software, but businesses willing to take the risk and trust Microsoft's safeguards. Until then, one could assume the AI factory narrative will continue to run well below its advertised potential.

• Vibe-coded clones of Word, Excel, and PowerPoint have appeared
• They're from the same developer who cloned the Adobe apps
• The developer says other open source software is 'broken'

The developer who used AI tools to clone the Adobe Creative Cloud suite and release the copied versions for free has now done the same with Microsoft Office.

As Tom's Hardware reports, there are now open-source versions of Word, Excel, and PowerPoint available, all looking very similar to the Microsoft apps and all with rebranded names: WordCraft, GridCraft, and DeckCraft, respectively.

These programs don't appear to be as far along in development as the Adobe clones, but the intention is there — and it's going to be a serious concern for large software companies wanting to protect their products.

The clones are the work of software engineer Brandon Thomas, who has been taking to social media to point out that he has years of experience in coding, even if he has been using the Anthropic AI Claude to do the bulk of the work on these recent projects.

'GIMP sucks'

There has also been criticism aimed at Thomas for choosing AI 'vibe coding' rather than supporting current open-source software like LibreOffice, GIMP, and Thunderbird. However, the developer doesn't seem to be a fan of these projects.

"GIMP sucks," says Thomas. "It has always sucked. There is no reason to save broken software." He's also recently related the story of a bad experience with LibreOffice that "killed a 40-page essay I wrote the week it was due, during finals".

The reaction on Reddit is split between those who see this as a "game-changing" positive for the software industry, and those who say "no job is safe" from the relentless march of AI and the increasingly powerful capabilities of these bots and agents.

Other commenters point out that free programs like LibreOffice and Google Docs have existed for years without damaging Microsoft Office sales too much, and question whether this lone developer will be able to keep up with bug fixes and upgrades.

• Microsoft is rolling out a new font for Microsoft 365 apps
• The 'Michelangelus' font is based on Michelangelo's handwriting
• However, it's a loose digital interpretation, and some feel that the font and the real handwriting are 'wildly different'

Think fonts are boring? Well, that's fair, they certainly aren't the most interesting topic on the planet, but every now and then one comes along that does make you sit up and take notice.

A case in point (12-point, perhaps) is that as Microsoft announced in a blog post, Michelangelo's handwriting is being rolled out to Microsoft 365 users – or a modern approximation of it at least.

The 'Michelangelus' font has been created as part of the anniversary of the consecration of St. Peter's Basilica, which took place 400 years ago (on November 18).

The font was made to be used for digital and printed materials relating to the Basilica, as well as signage inside the building, but the decision was made to share it with the world via Microsoft Office.

An Italian design agency, studiogusto, was commissioned by the Fabbrica di San Pietro to turn Michelangelo's characters into digital versions, noting that the idea was not a direct imitation of his handwriting, but to "adapt it through design into a modern tool."

Mimmo Berterame, the CEO of studiogusto, noted that: "For him [Michelangelo], [writing] was about clarity and respect for the reader. So, beauty was important, but it also needed order. Our typeface needs character and personality, but it also needs to be clear and easy to read."

In other words, the font is a modern approximation of his penmanship based on the 'rules' and feel of the writing style, so it doesn't look too over-the-top, but it still maintains some of Michelangelo's flourishes (like dotting the top of the 'h').

Analysis: fun with fonts

Some of the reaction to the font has veered towards the less-than-impressed in terms of the resemblance of the digital letters to the real handwriting. As one Redditor says: "That is certainly a very loose interpretation of the phrase 'based on.'"

A reply points out that the official press release from the Basilica di San Pietro explains that: "The typeface draws inspiration from Michelangelo Buonarroti's bold, expressive handwriting. However, it goes beyond a mere reproduction of the artist's graphic style and an imitation of his calligraphic techniques.

"The ambition is broader: to identify within his handwriting a visual logic, a system of proportions, rhythm and harmony measured by the artist's eye, and to translate this into a contemporary typographic system."

However, the original poster remains largely unmoved, noting that: "That's a noble goal and an ambitious approach, but I really don't feel their execution lives up to their intent."

Somebody else observes: "Among many other characters which don't even come close, the capital letters A and M in this interpretation are wildly different. I don't get the feeling of handwriting at all from it."

I can see where the criticism is coming from, and it stems from the fact that the design is clearly titled much more towards clarity and a more contemporary, readable look, than it is the actual handwriting. Still, regardless of how you gauge the interpretation here, the end result looks very smart.

If your curiosity is piqued and you are heading off to Word to try the new font, hold up a minute, because it hasn't been rolled out just yet.

Si Daniels, who is Principal Product Designer, Office Design, explains that the new font will be rolled out to Microsoft 365 apps at the end of the year. However, if you can't wait, you can manually download and install it from here.

• Microsoft has an initiative to improve the quality of drivers
• Intel's new Wi-Fi driver for Windows 11 contains the first glimpse of this
• More should come soon from Intel and others, and the initiative should help not just with stability, but also the performance of Windows 11

Remember Microsoft's plan to improve the quality of drivers in Windows 11? Work is well underway on this scheme, and Intel appears to be ahead of the rest of the pack in implementing this.

It's broadly known as Microsoft's Driver Quality Initiative (DQI), and Windows Latest reports that Intel has already begun to follow these new rules with its Wi-Fi and Bluetooth drivers for Windows 11.

The tech site has dug into Intel's wireless drivers and notes that they "follow early DQI practices", although the full scope of this initiative has yet to be implemented.

Nonetheless, it's good to see that the cogs are whirring on this effort, and in Intel's release notes for a recent Wi-Fi driver for Windows 11 (v24.50.0), the company said that it was applying "enhancements aligned with Microsoft's Windows ecosystem quality initiative."

With any luck, a couple of months down the line we'll see this work come to fruition, and Intel's GPU drivers should be due for the same treatment.

Windows Latest observes that Intel is set to work on its Arc drivers and knock them into shape for the DQI, too — and Team Blue has already added support for Windows 11 26H2 with its October graphics driver update.

PC hardware makers are already testing their drivers and working on DQI compliance actively. However, so far Intel is the only company to actually release an update to the computing public running Windows 11 that has benefited from this polishing.

Analysis: better quality goes beyond stability

What does the Driver Quality Initiative really mean? It should usher in a new era of more stable drivers for Windows 11, and therefore fewer crashes with the OS. Microsoft is focusing on upping the bar for reliability with Windows drivers that are provided by third-party hardware makers like Intel, but DQI goes beyond that.

This isn't just about preventing crashes, or wonky Windows 11 updates, but also ensuring tighter security, and that there's a higher quality of drivers coming through which don't drag down performance levels (or cause overheating issues with laptops, meaning combating excessive battery drain, too). Outdated drivers, or those that don't meet the new quality bar, will be unceremoniously ditched from Microsoft's OS.

It sounds like a big step forward for stability with Windows 11 in general, and as I've long argued now, this is an area where Microsoft has a serious problem in terms of the public perception of its desktop OS. I still think Microsoft must address issues around policing its own updates for Windows 11, though, and internal QA and testing processes.

• Microsoft has changed the way cloud storage works with the Microsoft 365 Family and Premium plans
• Instead of getting a possible total of 6TB — with 1TB for each of six family members — these plans now share 2TB between all users
• Microsoft argues this gives you greater flexibility, and that's true, but it works out badly for families where a lot of people are using cloud storage

Microsoft has changed the way the included storage works with its Microsoft 365 Family and Premium plans — and it's left quite a few people unhappy.

Neowin spotted the move, which Microsoft explained in a new support document, whereby what was previously a ceiling of 6TB of storage on OneDrive has been reduced to 2TB.

Microsoft explains that: "Microsoft 365 Family and Premium subscriptions include 2TB of shared cloud storage … The primary subscriber and up to five additional members draw from the same shared storage pool."

How it worked before was that each family member was allocated 1TB of cloud storage, so in the case where a full complement of six people were using the family subscription, that meant 6TB of storage in total.

However, as it now stands in the scenario of six family members using the plan, each of them is only getting a slice of the total of 2TB, so around 330GB apiece. In other words, a third of what they were previously receiving.

This new package is in effect for new Microsoft 365 Family and Premium subscribers from now, but for existing customers, it comes into play when their next plan renewal happens after (or on) May 2, 2027. (Quite what happens with multi-year deals isn't clear).

If you want to still have 5TB of total storage, the Microsoft 365 Pro plan gives you that, but it costs a hefty $99 (£95 / AU$165) per month, so it's a huge leap over the Family subscription (over 7x pricier).

Analysis: weasel-speak and greed?

Microsoft argues that this works out better for some family plan subscribers, and that "this approach provides more flexibility by making storage available to the people who need it most."

That's true to a point. Certainly if only the primary subscriber is using the cloud storage to any meaningful extent, they can now avail themselves of basically the whole chunk of 2TB rather than just 1TB. And obviously two main users get 1TB each, so that's the same as it was, but if there are more than a few active cloud storage users, this is a much poorer deal.

That's not the only problem, though, in that this change could hit existing subscribers in quite a nasty way. What if all (or most) of those multiple family users have already loaded up their OneDrive cloud lockers with 800GB or so, for example – they'll now have to cut that right down, or pay for more cloud storage.

So, if you're affected, you've potentially only got seven months (until May next year) to rejig your family's cloud storage and shift files elsewhere.

There's been quite a lot of negative reaction as you might guess, and observations that this amounts to a kind of bait-and-switch tactic. Someone on Reddit complains: "So disappointing seeing everyone getting sucked into this cycle. Something is cheap/liberal with features to gain customers, then jack up prices and cut features."

Others aren't happy about the 'weasel-speak' here (as in 'look, we're giving you more flexibility'), or they feel this is just another example of Microsoft's greed.

One Redditor complains: "Bad decision, they will find many customers cancel the auto renewal now."

Another simply states: "Greedy company. They never stop do they?"

In fairness to Microsoft, what was a great deal has become an okay deal cloud storage-wise – it's still not a bad offering, and 2TB is a decent amount of space for the money. But as noted, it does leave some folks very much in the lurch, and a bad taste in the mouth as we can clearly see from the online commentary.

The other worry is that now this has happened, Microsoft might look to trim the allowance of the Microsoft 365 Personal plan (for one individual) down from 1TB. I hope not, as this is a very different situation to the family subscription – there can be no arguments about 'flexibility' improvements if the company drops that storage limit.

• Windows 11 search is getting a major overhaul
• It'll be made faster, more efficient with RAM usage, and be tweaked to better understand what you're searching for
• Microsoft is also adding 'actions' to allow users to directly adjust Windows 11 options in the search box, a nifty and convenient touch

Microsoft has announced that it's forging onwards with making Windows 11's search function a faster and more streamlined experience.

In a blog post, Microsoft said it's rolling out this redesigned search experience to Windows 11 testers in the Experimental channel now (likely in a very limited fashion to begin with).

This new search is built on WinUI 3 (the contemporary interface framework for Windows 11), and as a result it'll be faster and will have less impact on your system resources. Microsoft boasts that: "Early testing shows substantial improvements in performance and memory usage, so [search] feels faster and more efficient from the moment you open it."

The search panel itself has been pared back to a single list, as opposed to the current implementation with the list on the left and the preview pane on the right. The interface therefore looks less cluttered, and it's easier to see what you're, well, searching for.

Microsoft also promises it'll work on ensuring that search more readily grasps what you're trying to find, with better handling of typos and guessing what you might mean if you're, say, attempting to find a system option and have entered the name incorrectly. This is work that the company already started in Windows 11, and the post notes that Microsoft will continue to fine-tune all this over the coming weeks.

This is what's happening initially; then, later this year, Microsoft is going to bring more features into testing. That includes inline previews and details that give you all kinds of extra info on the file you're searching (like a thumbnail for an image along with its resolution and size).

Search queries like "weather in London" will bring up an in-line card (powered by Bing) with the relevant at-a-glance details, including the likes of temperature highs and lows for the day.

The other big step is to introduce 'actions' to Windows 11 search, meaning you can simply type a command to change settings in the OS directly from the search box. Microsoft shows an example of typing 'change to dark theme' which pops up the option, then you simply click apply and you're done.

You can type 'mute' to turn off sound, or 'dim my screen' to adjust the brightness level automatically, and issue a similar range of instructions via search.

Analysis: some polarized reactions

These are clearly useful changes, although as with much of the work to fix Windows 11 this year, one of the main questions is: why wasn't this done before? It's long been known that Windows 11 search is clunky and sluggish, with too much clutter in the panel — so why has it taken Microsoft so long to address these points?

A typical barb on Reddit runs like this: "It only took decades for Microsoft to sort this abysmal search out." Someone adds that: "The only time search was decent in Windows was in Windows 98 SE."

Okay, so this kind of flak isn't fair, as other Redditors point out, but you can certainly say that it's not acceptable that Windows 11's search has been in bad shape since the launch of the OS half a decade ago.

It's good to see meaningful movement towards a slicker search is happening, and I'll definitely take this. The actions are a nifty idea, too, for quickly changing options there and then (rather than simply locating the option, and having to use extra clicks to go into Settings and adjust what you want).

As others on Reddit note, this is sort of drafting some ideas over from PowerToys (the Run tool), and something I've previously argued is that Microsoft should port over some of the functionality in that suite of utilities so it's in Windows 11 by default. Maybe that's starting to happen.

Even so, there are folks on Reddit who don't want search to be anything but an ultra-streamlined search (rather than offering the likes of actions). This Windows 11 user complains: "Why isn't the search bar just for searching? Why does it do a million other things on top of it?"

Of course, everyone's entitled to their own take, but I don't agree, and there is a lot of streamlining work going on here which Microsoft should be commended for, with the additions on top being useful in my opinion. Still, it just goes to show that the level of Microsoft hate out there is still palpable, and whatever the software giant does, it'll have its critics.

One of the most important moves with search that's just gone live in the most recent Windows 11 update is the introduction of the ability to turn off web results clutter. Again, that was a long time coming, but at least Microsoft is getting there… finally.

Despite living in an age of seamless connectivity, our workplaces have never been more disconnected.

Not literally, of course. We are surrounded by relentless barrage of message pings, overflowing email inboxes, and the ability to speak to anyone, anywhere, instantly. But beneath this surface-level digital noise is a crisis: winning and retaining people’s attention.

According to a recent Gallup poll, global employee engagement has cratered to just 20%, its lowest level since 2020. To some this may seem like a soft HR metric, but it is in fact a hard financial reality with a tangible economic impact. In 2025 alone, this engagement deficit is estimated to have erased $10 trillion from the global economy in lost productivity.

These figures tell us that traditional corporate communication is failing. If you can’t command the attention of your employees, you can’t execute strategy, align culture, or drive transformation. The employee engagement deficit is a challenge that leaders must address to compete in the modern enterprise ecosystem.

To tackle this problem head on, businesses need to stop thinking like traditional corporate communicators and take inspiration from an industry that has been captivating audience attention for almost a century: broadcast media.

Creative, captivating corporate communication

Professional broadcasters have always operated under the foundational truth that if you want an audience to retain information and connect with a message, the delivery must be as visually compelling and dynamic as possible. Broadcasters spend millions on high-production set designs, real-time data graphics, and seamless visual transitions because they know that the moment a screen becomes static, viewer attention starts to wane.

Yet, in 2026, many of the world's largest enterprises still communicate their most critical messages using flat, text-heavy slide decks and passive, one-way video presentations.

This goes further than just being uninspiring, and in the modern hybrid working environment actually constitutes a business risk. In a remote and hybrid work environment, employees are not sitting in an auditorium with their eyes locked on the stage. They are at their desks, surrounded by competing digital stimuli. A static slide deck is an invitation for an employee to open another tab, check their phone, or clear their inbox.

To win the battle for attention, enterprise communications must evolve from basic information delivery into immersive visual storytelling.

The economics of engagement

The economics of engagement refers to the relationship between how much content costs to produce, how effectively it connects with the audience, and what value that connection actually generates. For the pragmatic technology buyer, the first question that comes to mind is “does high-quality visual storytelling deliver a tangible return on investment, or is it merely an expensive aesthetic upgrade?”

Today, businesses can look at ROI through engagement and business outcomes. Firstly, technology has allowed businesses to gather more accurate data on metrics such as active viewing duration and number of Q&A interactions. This means that organizations can have a clearer picture of whether their communication is actually reaching and engaging employees.

Keeping employees invested in corporate communications is also about more than just making sure they are paying attention during a townhall presentation. These meetings are usually where businesses unveil a new strategy or direction for the overall organization, and these can often be complex.

If ideas aren’t communicated clearly, employees can leave with different interpretations of what they have heard, which can lead to inconsistent action. Creating more engaging presentation and making sure the message lands properly is essential for allowing business leaders to execute their overall strategy.

Another key ROI point lies in integrating broadcast capabilities directly into existing enterprise workflows. Rather than requiring employees to adopt new, proprietary viewing platforms, modern systems inject high-production-value graphics, dynamic lower-thirds, and multi-source video feeds directly into the collaboration tools employees are already using, such as Zoom.

By elevating the tools already in place, businesses can expect higher adoption and lower friction, maximizing the value of their existing software licensing.

AI-enable scalability, without specialization

Until recently, the barrier to achieving broadcast-quality communication was very high for most businesses. It required a dedicated physical studio, expensive hardware, and a fleet of broadcast engineers with specialized knowledge. However, AI has broken these barriers down and democratized access to broadcast-level production.

Today, intelligent software can now take on many of the technically complex elements of production, from managing layouts and multi-camera switching to generating real-time closed captions and incorporating real-time graphics. This means that much smaller comms teams can produce a live broadcast that matches the production value of a cable news network, for example.

A business shouldn’t need a production crew every time they want to deliver a high-quality update. By removing the need for specialized engineering and creative talent, enterprises can scale custom, high-impact communications across the entire business.

The technology should sit in the background and allow the global division lead can host a polished quarterly update, or a sales director can deliver an interactive pitch, all while maintaining a consistent brand identity.

A creative solution to the engagement problem

The numbers don’t lie. The $10 trillion productivity drain caused by disengaged workforces is a clear signal that the old ways of corporate communication are no longer fit for purpose.

As we look further into 2026, the competitive divide between organizations won’t just be defined by their products or strategies, but by their ability to align and inspire the people behind them. The businesses that continue to rely on flat, uninspiring mediums will continue to lose the war for attention.

Conversely, the enterprises that embrace the tools of modern broadcasting, democratized by AI, secured by robust IT governance, and integrated into daily workflows, will transform their corporate communications from a passive exercise into a driver of strategic alignment, culture, and measurable business growth.

We've featured the best employee experience tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Industrial employers sitting on unfilled technical roles and a generation locked out of work are two halves of one fixable equation.

The latest figures from the Office of National Statistics reveal that around 13% of UK 16-24 years olds are not in employment, education or training. The government's own interim review has warned this could mark the makings of a "lost generation."

At the same time, close to a fifth of the engineering and manufacturing workforce is expected to retire by the end of the year, taking with it an estimated half a million skilled engineers, technicians and operators.

One side of the ledger is a generation that wants to work and cannot get in. The other is an industrial base that needs people and is about to lose its most experienced. Treated separately, both look intractable. Treated together, they start to look like a single problem with a single answer. The solution: training AI agents to become autonomous and fill this gap.

A knowledge problem, not a hiring problem

The first mistake is to see the industrial shortage as a hiring problem. Recruiters and trade bodies have been clear that external hiring alone cannot close the gap: the issue is not just empty seats but the decades of judgment walking out the door with every retirement. When a maintenance lead or a supplier audit manager leaves, the loss is rarely the task itself. It is the accumulated reasoning behind it, and the why that never made it into a manual.

That is precisely why hiring and knowledge transfer demand completely different solutions. A vacancy can be filled in months. Thirty years of tacit knowledge cannot be replicated by recruitment, and the so-called "hollowed-out middle" - the shortage of workers with five to fifteen years' experience - means there is often no intermediate layer left to pass that knowledge down. The conveyor belt that once carried expertise from veteran to apprentice has lost its middle section.

Where AI changes the arithmetic

This is where AI changes the arithmetic, and not in the way the headlines assume. The breakthrough is not a chatbot that answers questions. It is the ability to capture the relationship between an organization's structured records, the purchase orders, the part numbers, and the unstructured reasoning that surrounds them: the emails, the annotations, the informal decisions made in the margins.

Map how the two influence each other, embed that context into the daily workflow, and decades of institutional memory become an operational asset rather than a folder no one can find.

Done well, this compresses the experience curve. Instead of acquiring judgment slowly, over years, a newer engineer inherits a working version of it from their first day on the job. The relevant history, the prior decisions, and the reasoning surfaced from the very beginning. The point is not that experience stops mattering. It is that the ramp becomes far shorter.

The window to exchange knowledge is narrow, and it is closing now. Context can only be captured while the people who hold it are still in the building. Once an expert retires, the structured data remains, but the reasoning that made it usable leaves with them. Every quarter of delay is a quarter of irreplaceable judgment lost. The retirement wave is not a distant problem to be scheduled around; it is already underway.

Where AI is used the wrong way

There is an honest objection to confront. AI is widely blamed for hollowing out entry-level work, and the data gives the charge weight: the same interim review found that entry-level roles have become less plentiful and more demanding, recruitment more remote, automated and less human, and apprenticeship starts for young people are down by more than 40%.

But AI is rarely the real culprit; the mindset behind it is. Too many organizations are still buying the technology the way they bought software a decade ago - as a line item that removes people rather than a capability that makes them more valuable. Pointed at entry-level work, AI will cut it and call the result “efficiency”.

Pointed at the opposite goal, making inexperienced people productive faster, the same technology widens the door rather than narrows it. The real question is not whether AI has to close off the old way in, but whether companies choose to use it to open a new one.

A different way in

This only works if the argument is honest about who these young people are. The cohort behind the NEET numbers is not, for the most part, trained engineers waiting for a vacancy. Six in ten have never held a job at all; many are contending with poor health, disrupted schooling or years of disengagement. Employers are willing to take them on yet hesitate because an unproven candidate feels like a risk a smaller firm cannot afford.

AI lowers the cost of that bet. When a system surfaces the right context and lets a newcomer direct complex processes in plain language, the absence of a track record stops being disqualifier. For a 23-year-old walking onto a factory floor for the first time, that is the difference between being screened out and being let in.

That is the opportunity hiding inside the two statistics. The retirement cliff and the NEET crisis are usually filed under different departments: workforce planning on one side, social policy on the other. They belong on the same page. Capture the expertise that is about to leave, put it within reach of the people who are desperate to start, and the industrial sector does not merely plug a gap. It builds the on-ramp for a generation that has been told, too often, that there was no way in.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• The Norwegian government plans to issue a ban on the use of smart-glasses in public spaces
• These include parks, beaches, museums, gyms
• While the ban is temporary, Norway's government wants to introduce permanent legislation in the future

Norway has become the first country to propose a temporary ban on camera-enabled and AI-powered smart glasses in public spaces, amid growing concerns around privacy.

Ars Technica reports that the Norwegian Government is proposing a blanket ban in selected public spaces where it believes people are vulnerable to being recorded without their consent. It follows the news this week that Dutch eyewear retailer Hans Anders has pulled Meta Ray-Ban glasses from over 400 stores.

Venues where the use of smart glasses would be prohibited include parks, beaches, museums, schools, kindergartens, shopping centers, healthcare facilities, gyms, and public events. Private use of smart glasses, for example in wearers' homes, would still be permitted.

The ban would initially be temporary, but it seems that Norway’s government is set on rolling out some form of permanent restrictions in the near future, and will enlist the help of an expert group to draft legislation.

Explaining the reasons for the proposals, Norway’s Minister of Digitalization, Torgeir Micaelsen, said "I am concerned about the introduction of powerful new technology where people risk being photographed, filmed, or recorded without their knowledge."

He added that a temporary ban "will give [us] time to conduct a thorough assessment and hold an informed debate that can lay the foundation for permanent regulation".

In order for a ban to be passed the governing Labour party will need the support of opposition parties, as it doesn't have an overall majority in the Norway's parliament, but it says it’s determined to introduce the bill "as soon as possible".

Smart glasses have prompted a patchwork of bans by authorities and companies in various countries, and most agree that they're an unsatisfactory approach. Norway’s more sweeping approach would be a first, and could serve as a model for other countries who want to introduce bans or other restrictions.

The Australian government recently announced that the use of smart glasses will be prohibited in federal workspaces, service centers, and local council buildings, and is considering a ban on their use in public areas. The City of Sydney joined forces with four other local councils, vowing to ban smart glasses from public pools, gyms, childcare centers, libraries and other spaces.

Localized bans have also been introduced in other countries. In the UK, pub chain Wetherspoons and some restaurants have individual restrictions on the use of smart glasses, and UK cinema chains could soon follow (although in that case the concerns are around piracy rather than privacy).

One of the biggest concerns with smart glasses is that, while they have a light to indicate when video footage is being recorded, wearers have been able to disable or cover these. In an effort to counter this, Meta rolled out an update to stop privacy hackers, introducing a tool that stops smart glasses from recording video when the LED is blocked.

• Eyewear retailer Hans Anders has stopped sales of Meta Ray-Ban smart glasses
• The decision comes amid growing concerns over the public use of the tech
• Dutch consumer rights group Consumentenbond has said it will take Meta to court if it fails to comply with the law

The clampdown on privacy-threatening ‘pervert glasses’ continues, with Dutch eyewear retailer Hans Anders pulling the Meta Ray-Ban smart glasses from its online and high street stores in Belgium and the Netherlands indefinitely.

Hans Anders is part of the nexeye retail group, which operates over 400 stores across those two countries alone. When combined with its stores across Germany, Austria, and Sweden, it has 776 stores in total, although the Meta Ray-Ban smart glasses aren’t sold in the latter three regions.

The use of smart glasses has sparked a heated debate about whether these devices belong in public spaces, with concerns raised about general safety and privacy. Speaking with Reuters, a spokesperson for Hans Anders detailed the company’s decision to make the bold move, sharing the following:

"This decision follows the ongoing public and political debate surrounding smart glasses and the questions being raised about the conditions under which this technology can be used”.

One of the biggest areas of concern is women's privacy. According to a report from the NL Times, Meta's smart glasses have been used to capture footage of 'hundreds, maybe thousands' of Dutch women without their consent over the past few months. The outlet says the footage has been shared on messaging apps such as Telegram, as well as other harmful online sites.

The move follows a recent warning issued by the Dutch Data Protection Authority (AP), which said that photos and video footage of identifiable individuals taken using smart glasses can’t be shared without the individual's permission — though this is something that's clearly difficult to control in public spaces.

That’s not the only organization that’s decided to take an active stance against Meta. Earlier this week Consumentenbond, a consumer rights group, declared it will take Meta to court if the tech giant doesn’t amend its smart glasses to comply with privacy laws. That said, Meta says it’s always developing its smart glasses:

"Every pair with a camera ​has a bright light on the front that blinks when you take a ​photo or video — it can't be turned off, and the camera is disabled if someone covers ‌or ⁠tampers with it," a Meta spokesperson said. Despite this indicator, there’s nothing in place to stop users from covering the light to stop others from knowing if they’re being filmed.

As far as retailers go, Hans Anders is one of the first to enact a public protest against smart glasses, joining the fight to protect the privacy and wellbeing of consumers. The Netherlands’ combat against these devices is only just beginning — and other nations have expressed similar concerns.

The UK is also taking similar measures to ensure public spaces are free from the looming threat of smart glasses. Popular pub chain Wetherspoons has banned the use of smart glasses in its pubs across the country, while the UK Cinema Association says movie theaters could also introduce similar smart glasses restrictions to prevent piracy, and even stop users from livestreaming movies to social media platforms .

• Microsoft has admitted there's a bug in the latest Windows 11 update
• The September preview update causes some games and apps to crash
• According to some reports, this patch is also leading to sluggish performance, or even boot failures, so tread very carefully

Windows 11's latest update has a bug that's causing havoc with some PC games and apps, or worse still, according to reports that have found the upgrade breaking their entire system.

Windows Latest highlighted a post on the Windows health dashboard where Microsoft admits there's a problem with September's optional update, which was released almost two weeks ago now.

Bear in mind that this patch (KB5124010) is a preview update, then, but there certainly seem to be quite a few issues for gamers here based on some of the feedback on social media. These problems affect Windows 11 26H2 and 25H2 users, as well as those still on 24H2.

Microsoft tells us that: "Some applications which require the decoding of AC-3 (Dolby Digital) audio might close unexpectedly."

The company adds: "Affected applications may vary and can include games, media players, and certain productivity applications."

The bug may cause the game or app to completely fail to launch, or it might crash, and in some circumstances things could be fine until certain features are used, notably playing music.

At least Microsoft assures us that "many modern applications do not use Windows for this type of encoding and therefore will not encounter this issue," but clearly there are problems here.

There are reports on Reddit from gamers expressing their displeasure at witnessing crashing with, for example, Fallout: New Vegas, Elder Scrolls IV: Oblivion and Wardogs. In all cases, players found that the problems started after installing the September preview patch, and ceased when the update was removed.

What's more worrying is that some of the affected gamers are reporting more serious issues that include the system ending up broken with boot failures, or very slow performance for booting or shutting down the PC. Other complaints include general sluggishness on the desktop.

As one Redditor observes: "KB5124010 was a disaster on one of my Win11 25H2 PCs. It caused boot issues, very slow shutdown/search, and after uninstalling it the machine got stuck at the Asus logo and stopped booting. Clean reinstall in the end."

Another agrees: "I thought I was going insane with slow boot up. File Explorer issues. Just big grey boxes taking forever."

Analysis: this needs a quick fix

Microsoft says: "We are working on releasing a resolution for this issue in a future Windows update. We will provide an update when more information is available."

Well, this is a nasty one by all accounts, so Microsoft better muster all its bug squashing powers pretty swiftly. After all, while this is a preview update now, it becomes the full October update for Windows 11 a week tomorrow – so that's not far away at all.

Also, there are some folks who may have wanted to grab this optional update straight away because it packs a fix for a previous bug in the September (non-preview) update which broke a backup feature in Windows 11. So, they could have downloaded KB5124010 and been happy to receive this particular fix, only to find that their games or apps are suddenly in hot water (or worse). Sadly this kind of situation is all too common with Windows 11 these days.

Furthermore, this is a warning to be careful around preview updates, because seemingly the only cure for the bugs here is to uninstall the patch – there's no workaround. Also be aware that if you have selected the option to 'Get the latest updates as soon as they're available' in Windows 11 then you will get all these preview updates automatically, and possibly be hit by bugs like these.

That's why I'd be cautious about exercising this right, even if you are very keen to receive new Windows 11 features as soon as possible.