Turn "it works" into acceptance criteria

I am about to have AI build: {{feature}}
Who uses it: {{users}}

Before any code, write the acceptance criteria:
1. 5-8 concrete conditions that decide whether this feature is done, each verifiable by actually using it.
2. Phrase each as "when ... then it should ...".
3. Mark which are must-pass (not done without them) and which are nice-to-have.
4. Add the edge cases: slow or dead network, empty input, very long content, double-clicking. What should happen in each?

Rules:
- Specific enough that someone can click through them. "Feels smooth" is not a criterion.
- Do not skip the edge cases - that is exactly where AI-generated code falls over.

Replace the {{}} parts with your own content

Why it works

1. The "when ... then ..." form forces actionable steps and blocks unverifiable filler like "smooth UX"
2. Listing edge cases before coding plugs the branches AI most often forgets
3. Splitting must vs nice-to-have tells AI where to spend effort, and tells you when to stop
4. Shared criteria make later testing and rework factual instead of a matter of taste

What it looks like

For a login modal, criteria reach the level of "on wrong password, show the specific reason without clearing the username field" and "double-clicking login fires exactly one request".

Where it bites

• !Covering only the happy path, so the code breaks the moment input is empty or oversized
• !Writing criteria and never checking against them at delivery, which makes them pointless

Browse categories