Checks That Run Without You
The Checklist Learns to Run Itself
Vibecoding taught you to write a checklist and run it after every change — the whole list, every time, not just the part you touched. That habit was right, and this lesson does not replace it; it promotes it. The next step is checks a machine runs for you. In plain language: a list of promises about your product — a taken slot never shows as free, a double-tap makes one booking, the groomer's email goes out within a minute — that gets tried automatically every time anything changes. When every promise holds, the change goes through. When one breaks, the change is refused before it reaches anyone, and you are told which promise failed. That is the whole idea. Engineers have elaborate names for the machinery involved, but the machinery is not the point. The point is that your checklist stops depending on you remembering to run it.
- A machine-run check is a promise, tried automatically on every change
- When a promise breaks, the change is refused before it reaches customers
- Your Vibecoding checklist is the raw material — the promises come from it
- The machinery has technical names; the idea needs none of them
Why Hand-Run Lists Decay
Why bother, when running the list by hand worked perfectly well last month? Because of what you are actually signing up for. A real product gets changed for years — small fixes, the second groomer, a price change, closing for a week in August — and every change can quietly break something far from the thing you touched. Nobody re-runs a twenty-item checklist by hand every Tuesday for years. You will skip it once, because the change was tiny, and you will be right. Then you will skip it again, and be right again. Hand-run lists decay exactly as fast as attention does, and attention decays fastest when nothing has gone wrong for a while. Machine-run checks have no attention to lose. They run identically on the tiny change and the frightening one, at two in the afternoon and at eleven at night, on the day you are on holiday.
- Products get changed for years, and any change can break something distant
- You will skip the hand-run list on a tiny change, and be right — at first
- Hand-run lists decay at the speed of attention; machines have none to lose
- The same checks run on the tiny change and the frightening one alike
What This Honestly Costs
The good news: writing the checks is cheap now. Ask the AI to build each promise-check alongside the thing itself — "add the cancellation flow, and add a check that a cancelled slot shows as free again" — and it will, in the same sitting, for barely any extra effort. The honest cost is not the writing; it is the habit. Every new promise needs its check. Every check that fails needs to be believed rather than deleted. And the discipline of never shipping past a failing check is yours alone to keep, because in a one-person product nobody will catch you cheating. One boundary, so you know what this lesson has not covered: if your product has AI inside it, testing whether the AI's answers are any good is a different discipline, and 'Does Your AI Actually Work?' is the course that teaches it. For everything else, promises and checks are enough.
- Ask for the check alongside the feature — one sitting, little extra effort
- The habit is the expensive part: every promise gets a check, always
- A failing check is information to believe, never an obstacle to delete
- Testing AI answer quality belongs to 'Does Your AI Actually Work?'
Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.