TDD was never really about the tests. It was how we designed a feature while we built it, one failing test at a time. AI spins up whole features in minutes and backfills tests written to pass, which quietly turns the practice inside out. Here's why that's a regression, and how we taught our agents to write the test red first.

I build knowledge-base engines for a living, so I pointed one at my own team. It mined a thousand PR discussions into the judgment behind our code reviews, and now it reviews our PRs and learns from whether we take its advice. Here's how the Table22 Oracle works, and how I built it in an afternoon.

When AI agents do the building, the slow part stops being code and becomes coherent intent. Here's the in-repo project management system I built to keep a fast-moving solo project from drifting, and the ideas behind it that transfer to any team.