How I run AI agents on production code
For six months a fleet of AI coding agents has done most of my production work. The agents were never the hard part. This is the machinery that makes it safe.
Lucas Firmo · Senior iOS engineer
Practical notes from 13 years of Swift and from running a fleet of AI coding agents on production code. Short, tested, and written for working engineers.
For six months a fleet of AI coding agents has done most of my production work. The agents were never the hard part. This is the machinery that makes it safe.
What I learned moving a website from Gatsby to Next.js one route at a time, and why the status code was the least useful check.
Each agent takes one page. I decide the order, approve anything that touches production, and fix the rules when something slips. Here is how it works and what still went wrong.
My agents kept saying "all tests pass". So now they cannot open a PR without proof inside it.
My agents could see every problem. None of them could fix one, so every problem waited for me.
No big-bang rewrite. Small PRs, a warnings baseline on every PR and tests on the old code before touching it.
SwiftUI, Swift 6 concurrency and the patterns that hold up in a large production app.
How to run coding agents on a production codebase with review, proof and guardrails.
Small reviewable steps, tests that fail first, and measuring before and after.
Remote from São Paulo, working US hours. Contractor or EOR, no visa needed.
Newsletter
One email when something new is out. iOS, Swift and AI agents on real codebases.