DevOps didn't fix your culture, and AI won't either.

There is a pattern that repeats itself in engineering organizations, and most leaders only see it when it is too late.

When DevOps arrived, companies rushed to adopt it. New tools, new pipelines, new team structures on paper, but many still failed, not because the tools were wrong, or the method was flawed. They used DevOps to solve a problem it was never designed to solve: They tried to fix culture with a toolchain.

I have been in enough of those rooms to find out how it goes:

• Someone presents a new tool.
• The team gets excited.
• Leadership approves the budget.
• Six months later, nothing really changed. The tool is there and the problems are there too.

Now the same story repeats with AI, only much faster.

The tell is always the same

You can spot it early if you know what to look for.

A team with no real ownership asks which AI tool will make them faster, a codebase full of undocumented decisions gets a documentation agent, a broken deployment process gets wrapped in automation and renamed a pipeline.

The symptoms get treated, but the causes are still getting ignored.

The teams that struggled most with DevOps were not the ones with the worst tools. They were the ones where nobody agreed on who owned what. Where shared responsibility meant, in practice, that nobody was responsible.

Those same teams are now the most enthusiastic adopters of AI agents.

Tools do not change behavior. They amplify it.

Leaders underestimate this part, and it is the single thing I repeat most in conversations with engineering teams right now.

If your team lacks accountability, AI will scale out that lack of accountability. If your processes are unclear, AI will run faster and with more chaos. If nobody reads your documentation, a documentation agent will only produce more content than nobody reads or trusts.

The problem was never the absence of tools, it was the absence of structure.

The technology was almost secondary. DevOps only worked in organizations that agreed to change how their teams operate, guaranteeing:

• Clear ownership.
• Responsibility that was shared.
• New ways to define and measure success.

AI asks for the same level of change, probably even more.

What you really delegate

Before any conversation about AI adoption, I used to ask engineering leaders what the job descriptions said or what the org chart shows.

Although important, now I realize these are not the most critical questions.

The real value is in asking the engineering leaders:

• What does your team own?
• What do people feel accountable for?
• And how do you know?

Most of the time, the answer is vague. That vagueness is exactly what makes AI risky in those environments.

When you delegate to an agent, you do more than automate a task. You decide what that task is, what good output looks like, and who answers when the output is wrong. If those decisions are not clear in your organization, you do not accelerate your team, and you accelerate your ambiguity.

A code review agent will only expose the lack of ownership, faster.

The harder question

Most conversations about AI in engineering focus on capability:

• What can it do?
• How fast?
• At what cost?

Fair questions. Also, the easy ones.

The harder question is behavioral:

• What does your team need to stop, change, or take more seriously before AI can help?
• Which assumptions about how work flows through your organization deserve a challenge first?

DevOps did not fail because companies lacked CI/CD tools. It failed because those tools exposed gaps that organizations were not ready to face.

AI will do the same and worst: faster, at a larger scale, and with less patience for everything we left unresolved for years.

The leaders who handle this well will not be the ones who adopted first, but the ones brave enough to ask the harder questions first.

The hard part of delegating agents is to know what should be delegated, and why.

So, before you decide which agents to build, ask for something more difficult.

What behaviors in your team need to change?

If your organization is navigating AI adoption and the numbers in your business case don't match production reality, we've been through this transition across multiple projects and teams. We'd like to help.

Reach out at contact@xgeeks.com to talk about what your engineering data is actually telling you.