Lessons from real work: production software, AI-accelerated development, and building this studio in the open.
Which framework, which database, which agent stack? Wrong question. Four rules we choose by, the boring defaults we start from, and where novelty earns its risk.
The keep-or-kill test told you where you stand. This is the road after — four stages, the question each one answers, and the checklist that gates promotion.
The prototype isn't failing — it's doing its job. The failure is promoting it. Five ways AI-era prototypes die when real users arrive, and the test that tells you what to keep.
Most first drafts of code here are AI-generated. This is the discipline that makes that safe — including the places where AI doesn't get to write, and the multiplier we refuse to publish.