In October 2025, at Raiffeisen Engineering Day, I framed AI in DevOps as a choice between a teammate and an executor.

At that time, my position reflected our operating context. In a nine-engineer startup running more than one hundred services, AI should extend human judgment rather than replace it. It could review pull requests, summarize logs, explain infrastructure errors, and draft SQL queries, but responsibility remained fully with engineers. Autonomy beyond that boundary introduced more risk than leverage.

That model was appropriate for our stage. We were optimizing for speed while preserving accountability. The engineering function included coding, deploying, monitoring, and fixing. AI augmented that loop without redefining it.

Since then, the context has evolved.

Agents are now integrated directly into the development environment. They can reason across files, refactor flows, generate tests, and execute scoped tasks end-to-end. Implementation is no longer the primary bottleneck. The constraint increasingly shifts toward problem definition, clarity of intent, and quality of constraints.

This introduces a change in emphasis. Engineers write less code directly and spend more time shaping outcomes. The center of gravity moves from syntax to product thinking. The value lies in defining problems precisely, designing boundaries clearly, and evaluating trade-offs deliberately.

The next step is not isolated autonomous agents operating independently. It is coordinated teams of agents with distinct roles, operating under human direction. Planning, implementation, and review can be decomposed and delegated structurally. The engineer remains accountable for system design, validation, and final judgment.

This reframes responsibility rather than removing it. The engineer becomes a designer of systems that include both human and artificial contributors. The unit of work becomes intent. Quality depends on orchestration and clarity rather than keystroke volume.

The original question – teammate or executor – remains useful, but it is no longer sufficient. AI is gradually becoming part of the infrastructure of software development. As that integration deepens, the defining skill shifts toward reasoning, constraint design, and product judgment.

The tooling has changed. The responsibility has not.

References