I Taught an AI to Write My Docs (So I Don't Have To)
Documentation is always the last thing to ship and the first thing customers notice when it's missing. As a sole technical writer supporting five engineering squads, I knew that keeping up wasn't a process problem. It was a scale problem. So I built an agent to help. In this talk, I'll walk through how I designed and iterated on an AI-powered documentation generation system using the Claude API, GitHub Actions, and a Python agentic loop, running across multiple repositories and integrated into our existing CI/CD pipeline. The agent monitors code changes, generates documentation drafts, and flags what needs a human decision. My job shifts from writing everything to reviewing, testing, and publishing. I'll cover what works, what doesn't, and the unexpected lessons from handing a core part of my job to an AI: how to prompt for structure and consistency, how to handle the cases the agent gets wrong, and what it means for the technical writer role when the boring parts start automating themselves.
I've been a technical writer for 17+ years — and I've spent the last few trying to automate myself out of the boring parts. Now working at the intersection of documentation and engineering, I build AI-powered systems that generate, update, and ship docs at the speed of the product teams I support.