What AI-DD actually changes: a field map
Five places delivery shifts when you rewire around AI: code structure, CI/CD, ticketing, rituals, docs. Before-and-after patterns for each.
Replace this body with the real essay before public launch. Stub for reading-time, OG generation, and homepage cards.
Moving an engineering org to AI-DD isn’t one change. It’s five changes that have to land together — skip one and the whole thing plateaus at felt-20%.
Code structure. Folder layouts written for humans (“this is where auth lives”) are illegible to agents without a map. The shift is toward flat-by-domain directories, explicit README files at every level, and types that name the thing rather than describe it.
CI/CD. Pipelines designed for humans-reading-failures don’t serve agents-acting-on-failures. Signal-to-noise matters 10× more. The shift is toward structured-failure reporting, exit-codes that classify, and build logs an agent can read without retries.
Ticketing. Tickets that were fine for a human lead (“refactor the payment flow”) are hostile to agents. The shift is toward ticket shapes that encode acceptance criteria, the file set to touch, and the test to write.
Rituals. Standups, planning, retrospectives — all designed around human-in-the-loop assumptions. The shift isn’t to cancel them. It’s to refactor what they’re for.
Docs. The classic “we’ll write docs when we have time” is now a throughput bottleneck. AI-DD teams write docs because the agent reads them. That shifts the economics.