Appreciating generative AI’s DevOps benefits

Computer code on a monitor
(Image credit: Shutterstock / DaLiu)

Generative artificial intelligence (AI) will lead to the development of a lot more code at increasingly faster rates. The challenge now is to manage the accelerated pace of development when many organizations are already struggling to manage existing DevOps workflows at scale.

Just as challenging, not all the code generated in the short term by developers using AI will necessarily be of the highest quality. General-purpose large language models (LLMs)-based platforms such as ChatGPT have been trained using code collected from across the web. Much of that code contains vulnerabilities and other flaws that are finding their way into AI-generated code–and developers don’t always have the expertise to identify and correct those mistakes.

Arguably, the most important and immediate task for any DevOps team is to identify those issues before any of the code is used in a production environment. As more code than ever starts to course through pipelines, achieving that goal will require DevOps teams to employ modern DevOps tools and platforms – themselves likely infused with AI technologies – to address this challenge.

Sacha Labourey

Co-Founder and Chief Strategy Officer at CloudBees.

Six AI areas of attention for DevOps

There are six areas where AI will make it easier for DevOps teams to cope with the onslaught of code that is already starting to move through existing pipelines. They include:

Application Code Management: Generative AI, in addition to writing code, will also be used to highlight bottlenecks and constraints that present opportunities to reduce overall toil. It will be able to define what is being produced, velocity, the types of defects encountered, assess the overall level of security, and determine the impact merge requests might have on a build.

Release Management: Generative AI will enable DevOps platforms to surface more accurate release forecasts that identify, for example, the probability that a build will pass or fail. Change failure rate metrics will substantially improve over the next two years as generative AI makes it easier to understand dependencies and overall complexity by tracking patterns that will make release management more predictable. In addition, the overall impact on the business will become more apparent as AI is infused into value stream analytics tools.

Testing: Generative AI will make testing much more effective. DevOps teams will be able to better understand not just what to test, but also reduce cycle time and processing expenses, such as IaaS/cloud costs, by defining what subset of the tests to run. Generative AI will also be able to very efficiently provide a foundation of unit tests for areas of the codebase that are currently not properly covered. Collaborative code reviews will become streamlined as generative AI becomes one of the active “peers” in the review process.

Cybersecurity: Generative AI will eventually enable developers to identify security issues as they write code and enable that code to be more thoroughly tested. Also, the kind of analysis that LLMs offer on source code can happen at a higher-level, providing analysis for more complex scenarios rather than, for example, known syntax issues that are nowadays trivial to spot. DevSecOps teams will also benefit from an enhanced ability to model threat data to create classifiers that show how a defect might be exploited, blocked, isolated, or remediated. They will also be able to use synthetic data to mirror actual data and better ensure compliance mandates are met.

Monitoring: Generative AI will make it easier to leverage metadata to identify patterns in the massive amount of logs, metrics, and traces DevOps teams collect. Those patterns can then be fed back into a DevOps platform to plan and possibly automate remediation before there is an incident that disrupts application availability.

Reliability: Mean time to recovery (MTTR) will substantially improve in the next one to two years. Today, ensuring reliability is challenging simply because there are so many tools needed to manage a DevOps workflow. Generative AI will make it simpler to aggregate the data generated by these tools in a way that will significantly improve fast identification of issues - even proactively detecting anomalies - and positively impact application uptime.

These points were also echoed during several recent DevOps World 2023 sessions.

Benefits and risks

With AI, software development costs will drastically decrease. However, there are still several issues that require more work. They include:

  • Fine-tuning LLMs to further reduce hallucinations;
  • Maintaining alignment on the meaning of words, as AI models are exposed to more prompts and data;
  • Identifying biases that exist in the AI model training data that result in suboptimal recommendations;
  • Ensuring that the data being used hasn’t been deliberately poisoned to create a deliberate hallucination that might be difficult to detect and challenging for an AI model to unlearn.

It is clear that AI has incredible potential to make software development faster and easier and ensure the resulting software product is of higher quality. But we aren't quite there yet. The AI and software development industry must foster full confidence in the recommendations generated by software development tools and platforms infused with generative AI. There simply is no substitute for LLMs trained using domain-specific data that DevOps experts have vetted.


It’s clear generative AI will soon transform DevOps workflows for the better in ways that we have only begun to appreciate. The AI genie is out of the proverbial bottle, and there is no going back. DevOps has always been about making a commitment to ruthlessly automate manual processes whenever possible. Generative AI simply takes automation to another level.

This is a challenging undertaking, but it is also an opportunity for generative AI to be used safely and sustainably in conjunction with other advances in data science and machine learning algorithms.

We've featured the best web development tools.

This article was produced as part of TechRadarPro's Expert Insights channel where we 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:

Sacha Labourey is Co-Founder and Chief Strategy Officer at CloudBees.