Using AI

Start with trusted protocols

These entries currently meet Brali's reviewed/practical retrieval bar. “Trusted” is a publication state, not a promise that every protocol has equal scientific support.

• Ask AI for Failure Cases Before You ShipDescribe the intended behavior, constraints, known happy path, and expensive failures. Ask for boundary cases, invalid inputs, state transitions, concurrency or timing issues where relevant, and user misunderstandings. Remove duplicates, verify each candidate against the real system, and write tests for the cases worth keeping.practical
• Attempt First, Use AI for Feedback, Retest Without ItName the skill you are actually trying to acquire. Attempt a problem, explanation, translation, derivation, or answer unaided. Ask AI to identify gaps, give a contrasting example, or explain one sticking point. Check the feedback against trusted material when accuracy matters, then perform a fresh attempt without AI.reviewed
• Benchmark AI on the Task Before You Trust the WorkflowChoose a small sample containing easy, ordinary, and awkward cases. Define what counts as acceptable before running the model. Compare AI output with the source or a trusted human result, record failure patterns, and decide which cases can be assisted, which need review, and which should stay human-first.reviewed
• Draft Professional Writing with AI, Then Take Back the EditProvide the audience, goal, source facts, must-include points, and constraints. Ask for one draft. Compare it with the source material, remove invented details, restore your own judgment and voice, and send only the version you can defend without the model in the room.reviewed
• Extract a Table with AI, Then Sample-Check the RowsDefine the columns and allowed values before extraction. Ask the model to return null rather than guess, preserve a source snippet or locator for each row, and flag ambiguous cases. Check a random sample plus all high-impact rows against the original text before using the result downstream.practical
• Give AI a Rubric and Ask for Specific CritiqueWrite three to six criteria that describe a good result. Give the model the draft and rubric, ask for evidence from the draft for each criticism, and request no rewrite on the first pass. Triage the feedback, then make the changes you agree with.practical
• Let AI Draft Routine Replies, Keep the Human on the ConversationProvide only information you are allowed to share, plus the customer question and approved policy or knowledge source. Ask for a concise draft. Check that it answers the real issue, does not invent policy or status, and gives the correct next action before sending.reviewed
• Rehearse a Difficult Conversation with AI Before the Real OneState the goal, your opening, and relevant context without unnecessary private detail. Ask AI to play a neutral, skeptical, and confused counterpart. Practise responding with facts, boundaries, and questions. Write one shorter opening after the rehearsal and use that in the real conversation if it still fits.practical
• Summarize with AI, Then Reconcile Against the SourceProvide only material you are permitted to share. Ask for a summary with sections, explicit uncertainties, and source locations when available. Mark the claims that affect your decision, then open the source and verify those claims directly before reusing them.reviewed
• Turn Meeting Notes into Actions, Then Reconcile ThemUse only notes or transcripts you are allowed to process. Ask for five buckets: decisions, actions, owners, dates, and unresolved questions. Compare each populated field with the source, remove inferred commitments, and publish the checked version quickly while context is fresh.practical

4 more trusted protocols are listed in the full archive below.