#claude-code
Agentic Software Factory, In Action
In this post I show you my agentic setup consisting of three repos, one of which is a completely self-contained software factory. It defines agents that complete specs, write plans, break them down into frontend and backend work, then pass it onto the agents specializing in that type of development, who are working concurrently and communicating between each other through a 'mailbox' mechanism. Once they are done, pull requests are submitted and another adversarial agent performs a PR review. Only after addressing the feedback, going green on CI, do the PRs become ready for merge and for human evaluation. I did this deliberately, but sooner or later the majority of PRs will auto-merge and auto-deploy.
How I built three dry-cli extensions, then the forum's AI bot thought I was a spammer
Three Ruby CLI annoyances became three gems. Then my announcement got caught by a spam filter, and a conversation I didn't expect followed. If you write CLI utilities, especially in Ruby, and in particular using dry-cli framework, you'll want to read this. Pinky promise.
Condensing Twenty Years of Wisdom in One Markdown
I've been shipping at a genuinely ridiculous pace lately, and somewhere in there I stopped being an engineer and became the bottleneck in a twenty-person team of AI agents. Don't worry, I am working on unblocking this too. But in the meantime, this is the story of how I started dumping two decades of PostgreSQL scar tissue into a single markdown file that my agents load on demand - the entire file, verbatim, for you to steal. Includes a bonus rant on transaction ID wraparound, the horror almost nobody outside the PG core team and your friendly neighborhood LLM actually understands.