Connect the dots between having AI write tests, testing things yourself, and your Definition of Done.
Let's think about it this way for a sec
In vibe coding, don't measure 'done' by how you feel about it. Use a spec checkbox, at least one automated test, and clicking through it yourself in the browser. AI can write tests — but they're only useful if they actually assert the behavior you care about. A generated feature with no tests is an easy thing to accidentally break the next day.
Connecting it to everyday life
For an 'add note' feature, a test might be 'empty titles shouldn't be allowed.' Tell AI to 'write a test that fails if empty titles are accepted.' If that test is still red, the feature isn't done. Even if you can't do full E2E testing, click through the checklist yourself. Done = spec checkbox + review + at least one verification.
Let's try it together
it("does not add a blank note", () => {
const notes = addNote([], " ");
expect(notes).toEqual([]);
});
it("adds a trimmed title at the front", () => {
const notes = addNote(["Old"], " New ");
expect(notes).toEqual(["New", "Old"]);
});You'll be able to show two pieces of evidence that prove a feature is actually done.5-Minute Try-It
For your mini app's first feature, write two tests and three manual click-through steps.
A Quick Word of Caution
Never treat AI output as automatically correct. Have a human read through the code and test it — and double-check secrets and user data — before you rely on it.