AI is reshaping software delivery, and it’s not going away. The question for platform engineers isn’t whether to use AI in the CI/CD pipeline—but where and how. This session shares lessons learned from integrating AI into a continuous delivery tool, including what worked, what didn’t, and where guardrails are essential.

Because AI is inherently non-deterministic, it should not decide the actual steps executed in a pipeline—doing so introduces risk and compliance concerns. Instead, AI can add value by generating deterministic pipeline definitions, triggering workflows, analyzing failures, and recommending remediation steps.

Platform engineers must define the boundaries. The key takeaway: people are non-deterministic too. Treat AI like you would a human contributor—constrain what it can do based on trust. For some organizations, that may mean “intern-level” access limited to dev and test. For others, it may act like a senior engineer—able to propose process changes, but never deploy directly to production.