The AI was perfect. For the first 46 files.

02
Then File 47

Three localization keys missing.

No EN. No DE. The AI got the pattern right, then drifted. And you'd never check file 47 manually.

Specs describe. Tests verify.

04
How It Works

A test runs after every file.

Like a linter—not just for code. For behavior. For content. Catches what manual review can't.

Not 100% coverage. 100% of what matters.

06
The Pragmatic Way

Try ideas. See what sticks.

Then pin the important stuff with tests. Before you scale—not before you code. Some tests stay forever. Some get deleted after the job.

The ultimate test

Nagelprobe

/ˈnaːɡl̩ˌproːbə/

After drinking a toast, you turned the cup on your thumbnail. Not a drop could remain. That's how thorough your verification needs to be.

08
The Carbonara Fix

What's implicit becomes probabilistic.

Tests make it explicit. Not more documentation—executable constraints that the AI can't ignore.

The best 'done' signal? Test green.

Stop trusting. Start verifying.