Diagnose a weak AI output by stating the observable failure and success criterion before editing the prompt.
Maya has a polished but unusable launch brief and five minutes before the VP review. What is the strongest first move in the prompt-debugging loop? Write the observable failure and the passing standard before touching the prompt. Right. This creates the eval target. You can then patch the prompt against a specific gap instead of chasing vibes. Ask the model to try again with a more professional tone. Tone may be one symptom, but this skips diagnosis. It risks making the same incomplete answer sound more polished. Switch models immediately because the first output was weak. Model choice can matter,…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in