One AI setup for every repo: how I work with coding agents, and why it became a factory

Wed, 08 Jul 2026 00:00:00 +0000

I write software with AI agents in three places: a terminal session while I work, cloud sessions I fire off and forget, and agents that live in my repos’ CI. All three follow the same committed rules, read the same committed memory, and hand everything to me as a pull request that I merge. This post is a tour of that setup: what each piece is for, what actually changed in how I work, and why keeping it consistent across repositories eventually forced me to build a small platform for it (ai-factory).

Workflow on Onur Celep

One AI setup for every repo: how I work with coding agents, and why it became a factory