Proof, not promises

There's a sentence I've stopped trusting: "It works now." Not because people lie — but because AI has made it hopefully easy to BELIEVE something works without ever having watched it.

Generation got cheap. Verification didn't

A couple of years ago, the bottleneck was WRITING the code. Today an AI agent writes a function in seconds. The bottleneck moved — it now sits in proving that what got written actually does what it was supposed to. Does the code compile? Fine, that's a precondition, not proof. Does the API return 200? Fine, that proves an HTTP status, not a user experience.

I've seen it happen too often: a feature "works" according to the terminal, but in a real browser the button is invisible, the form doesn't submit, or the redirect points to the wrong place. The terminal wasn't lying — it was just never asked the question that actually mattered.

The rule we work by

A task doesn't change status to "done" until it's been clicked through in a real browser — filled in, submitted, and watched doing what it was supposed to. Not an assumption. A screenshot, a completed action, actual proof.

It sounds boring. That's the point. The boring discipline is exactly what separates "I think it works" from "I watched it work."

Why it's worth sharing

Because AI has made it cheap to PRODUCE and expensive to SKIP verification. The more an AI agent can build on its own, the more important the question becomes: who — or what — checked that it actually worked? If the answer is "no one, but the code looks right," you haven't built something. You've built a hypothesis.

Proof beats promises. Every time.