Posts tagged “ai”
~/blog ❯ ls -lt tag/ai/
total 9
How Claude spawned 100 agents to audit my backend
Three audits of Horsemate in two weeks: five agents, then ten, then a Claude Code workflow with 100. What each layer of agents added, the one serious hole it found and fixed before I'd read the report, 29 PRs in 26 hours, and what it all would have cost on the API.
How to train your (AI)pony
Horsemate got a chatbot named after my old Gotland pony. The same evening it went live, I built a 61-question test battery, let Claude Opus grade every answer, and found that most of what was wrong was in how I measured, not in the pony.
Ten Years and Three Minutes
I spent ten years looking for a lost mixtape track from 2013. Every website, every Wayback capture, unanswered emails and DMs. Tonight Claude found it in under half an hour, sitting on Bandcamp's servers behind an address I had looked at again and again without seeing.
The best thing I've built with Claude Code isn't software
1.2 GB of Apple Health XML, 2.6 million records, and the series the whole project is about is 106 of them. What building a health project in Claude Code taught me about using LLMs on data too big to read.
GPT-5.3 Codex writes cleaner code and I still hated it
I spent 20 days using Codex as my only coding tool. The code quality was fine. The experience was not. Here is the honest field report.
I built an AI project manager that tells me no
My tech-lead workflow evolved into a full project manager skill that challenges assumptions, orchestrates specialist agents, and refuses to build the wrong thing. Here's how it works and why the best AI collaboration starts with pushback.
AI doesn't write perfect code (and that's fine, if you're paying attention)
Real examples from my codebase where AI-generated code looked correct but wasn't. The bugs were subtle, the lessons were obvious in hindsight.
How I stopped letting AI write my code (and started collaborating instead)
Claude Code out-of-the-box felt like hiring a developer who never asks questions. Here's how I built a workflow that actually works.
I gave Claude Code root access to my Linux system (and it was amazing)
What happens when you stop treating AI as a coding assistant and start treating it as a systems administrator? Spoiler: your .bashrc has never been this clean.
~/blog ❯ cd .. # all posts