“You can't just take an LLM and throw it at a problem” - why Salesforce is pushing for a smarter way for everyone to do AI

AI is now a common presence in enterprises across the world, helping transform workflows for employees and customers alike, bringing in new levels of productivity and efficiency.

But, in order to unlock these benefits, your business has to still make sure it is utilizing and enforcing AI correctly, not just doing it because everyone else is.

At the recent Dreamforce 2025 event, Salesforce was understandably keen to show off its dominance and expertise when it comes to all things AI - and it seems a lot of this confidence comes from within.

The Agentic Enterprise

"We are expanding into the enterprise in a big way,” Muralidhar Krishnaprasad, President & CTO of Engineering, Salesforce, told me at Dreamforce.

“Previously we were always just relegated to saying you are only doing sales, or service marketing - but now we are going beyond that, we're managing customers, we're managing employees, and we're going to be managing operations...and most excitingly, we're managing agents as well! All managed by our unified platform.”

The theme of Dreamforce 2025 was arguably that of the "agentic enterprise" idea - with Salesforce CEO Marc Benioff declaring simply deploying enterprise AI models is no longer enough for businesses if these solutions aren't integrated and grounded in proper governance.

Krishnaprasad agrees, noting, “for a SaaS company like Salesforce, we're making AI useful, just like we did making the dotcom useful for the enterprise, and we believe this is the best way to get AI working for the enterprise.”

A large part of that comes from being customer zero for these new agents and AI tools, as Benioff had highlighted how the company “is on the journey” with its customers to the age of the agentic enterprise.

Krishnaprasad highlights the work his team did around the initial launch of Agentforce in 2024, building and launching agents on its help site in just a few weeks - agents which have, in a year, handled more than 1.8 million conversations, freeing up human support workers for potentially more crucial cases.

There have been well-publicized concerns around the effectiveness of AI tools in some areas, with Krishnaprasad noting an MIT study showing 95% of AI projects fail.

“You can't just take an LLM and throw it at a problem,” he notes, “(but) this is a constant fear we all have as humans - when the dotcom came we all feared for our jobs, when Tesla introduced auto driving, we all wondered what would happen - but the reality is - humans are the best at adapting - over millennia, we have adapted to so many different things."

“We used to look back after electricity came in and say, imagine the pre-electricity days - it changed our whole race, right? AI is going to do the same, and we shouldn’t be fearing it, because it’s just going to make us better - because guess what? We have so many drugs left to discover, so many processes we can make better.”

Krishnaprasad does paint an optimistic view of the future where AI agents and systems work alongside humans to maximize the potential of each other. He highlights findings which claims 40% of a developer’s life is maintaining code, meaning these workers are often stuck with maintaining old code rather than creating something new.

“This is where, if AI can really help solve a whole bunch of issues there, imagine what new things we can create in just 20 years,” he says.

“This acceleration is happening because of technology, and with AI, it’s just going to go even faster…we will be able to leave out the drudgeries of maintenance, instead focusing on innovation, new ways of interacting, new ways of helping our human race together.”