We hated ClickOps, so we wrote Infrastructure as Code. Then we built portal UIs on top - ClickOps with extra steps. We fought for reproducibility with Docker, Kubernetes, and Terraform. Now we’re plugging non-deterministic GenAI agents into deterministic pipelines and calling it innovation.

That’s a moonwalk - it looks like forward motion, but you’re sliding backward. Each new layer is a confession that the previous one wasn’t finished.

If we’re treating our platform as a product, where’s the customer feedback that said “add a large language model”? It doesn’t exist. What the feedback actually said was: “make it work.”

This talk names (and possibly shames) the moonwalk pattern. But fear not, it also proposes a framework for where AI actually belongs in platform engineering (and where it absolutely doesn’t), and argues that the most radical thing your platform team can do this year has nothing to do with AI.

Target Audience: Platform engineers, SREs, DevOps leads, and engineering managers evaluating AI agent adoption for their infrastructure and developer platforms.