Tests as the Spec: TDD when the AI writes the code

AI made writing code cheap. It made verifying code expensive. Almost nobody has adjusted.

You can generate a feature in ninety seconds now. Reading it properly still takes an hour, and when the model touches forty files in one pass, nobody reads it properly. They skim, they run the app, they say "works, I checked", and they ship.

That sentence is where the bugs live.

Write the test first, so the model has a target it cannot argue with,
Watch it fail before any code exists, because a test that passes early is testing nothing,
Forbid the model from editing test files, because it will change the assertion to go green,
Test money, permissions, parsing and every bug you already fixed,
Skip layout, copy and anything you would delete without regret 🔥

I wrote the whole thing up. Copy-paste prompts, a real worked example with the failing test and the instruction that produced it, and an honest section on over-testing.

Free. No email wall.

Comment "TESTS" 👇 and I'll send it to your DMs.