New post is up!

Twenty-one services to move from ECS to EKS -- the final wave of an eight-quarter migration. Each one is wired differently: different CI, different logging framework, different infrastructure setup. The layers from the previous post defined what to change across all of them. The how varied per repo -- one service keeps its logging in logback.xml, another in a custom NodeJS module, a third in a shared library. You can't script that discovery. And handing it to an AI agent just produced a different approach every run.

The fix wasn't a better prompt. It was everything around the prompt.

The post is about the harness we built to close out the migration -- a pipeline that mixes prompts, scripts, AI discovery, and validation in an ordered sequence. I walk through the six constraints that made AI predictable enough to trust across 21 differently-wired repos, and the parts we honestly got wrong.

I welcome you to read this new article on my blog!

Just click the button below to open it in your browser: