Task → do → verify
Build-along · ~10 min at your own keyboard
The single biggest difference between people who get magic from AI and people who get slop: the loop. Give it a task, let it do the task, give it a way to verify its own result. No verification means you're the QA department for a very fast, occasionally wrong intern. And one good run proves less than it seems. The model is non-deterministic: the same prompt on the same files can come back different next time. So the check has to run every time, not once when you first set it up.
The long version: The verify loop ↗
Do this
0/4Checkpoints
- ▸You should now see your definition of done written into the prompt.
- ▸You should now see a verification table or list from Claude after the task.
- ▸You should now see the planted mistake flagged on the re-run.
The three ways this goes wrong
- Claude says done without checking. — Put the check in the prompt: 'when finished, list each item against the rule and flag misses.'
- The verification passes but you spot an error. — Your check was too loose. Tighten it to something you can count or match exactly.
- The planted failure is not caught. — Ask: 'why did your check miss that file?' Then patch the check and rerun.
Where these ideas come from: Non-determinism (AI Hero dictionary). Paraphrased here.
Share what happened
Paste your definition of done, the verification output, and proof the broken case got caught.
Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].
🔒 Don't paste confidential, personal-health, or payment info.