After my son picked up a lacrosse stick in 2025, he also picked up a cross-country habit; the same sport I ran forty years ago. I remember my old coach handing us a new article every Monday: stories of extraordinary human performance. He literally retrained our brains. I wanted to give my son the same edge.

So I built something.

The goal: an AI-powered content pipeline that generates weekly mental performance articles for young athletes; automatically, reliably, and without an eye-watering cloud bill. The constraint: no heavyweight orchestration, no Claude running cron jobs, and no architecture that punishes the client to compensate for an expensive backend.

This talk is the honest account of building that pipeline on Cloudflare Workers AI; every sharp edge, every "wait, that's not how this works" moment, and every pattern that finally clicked.

We'll dig into Worker AI, how to decide when AI is the right tool and when a ten-line script does the job better, and what it really means to handle AI output defensively, treating it as untrusted input at every step. Along the way we'll cover the hidden complexity inside Browser Rendering, how Service Bindings handle credential passing between Workers, and how to manage model configuration in an ecosystem where the models themselves change daily.

Real code. Real trade-offs. Real lessons from a project that had to stay lean. If you're building AI-powered applications on Cloudflare (or wondering whether you should), this talk will show you what the tooling can actually do, and how to do it without breaking the bank.