Thinking · Growth Library

AI-Assisted Premortem

Practical protocol

Ask an AI to generate failure modes, then verify them yourself

A person comparing notes and marking one useful insight
Protocol summary · practical

Try this

Describe the plan without sensitive information, ask an AI for plausible ways it could fail, group the suggestions, verify the important ones yourself, and choose mitigations only for risks that survive the check.

Check-in: Which risks were genuinely relevant after verification, and which model suggestions were generic, impossible, or based on missing context?

Evidence: Practical guidance; no evidence-like claim is detected in the current source record.

Review note: Public content is replaced by a bounded AI-assisted premortem. Model output is treated only as candidate failure modes, sensitive information must not be shared, and important risks require verification against the real plan before mitigation.

A premortem asks what could make a plan fail before the failure happens. An AI can widen the list quickly, but it does not know your full context and should not decide which risks deserve action.

1. Sanitize the context

Describe the goal, constraints, and broad plan without sharing protected or confidential information.

2. Generate possible failures

Ask for concrete failure modes across people, process, dependencies, timing, data, cost, and assumptions.

3. Verify important items

Check high-impact suggestions against the real plan. Delete generic, impossible, or context-free items.

4. Choose bounded mitigations

Add a prevention, detection, fallback, or owner only for risks that remain plausible after verification.

Questions to consider

What should I avoid sending to the AI?

Do not paste secrets, personal data, confidential client material, credentials, or information you are not allowed to share.

Should I treat every generated risk as real?

No. Model output is a list of possibilities. Verify high-impact items against the actual system, people, constraints, and available facts before acting.

Maintenance record

Review history

Current status: Reviewed. An editorial or evidence review is recorded and no later event changes that conclusion.

  1. 2026-08-20 · reviewed · Brali editorial agent Source: protocol-content-overrides-zone-starters.json

    Replace inherited AI and risk-reduction claims with a bounded premortem workflow where AI generates possibilities and the user verifies them before acting.

Browse the review ledger · Challenge or update this hack

Implementation observations can trigger a review, but they cannot change an evidence conclusion by themselves. Commercial relationships do not control review status or outcomes.

Continue from here

Choose the next useful path.

Explore zoneThinkingSee the wider Growth Zone that contains this protocol.Browse life areaLearning & ThinkingStep back to a broader life-area view before choosing another protocol.Explore topicError PreventionFollow the maintained knowledge topic behind this protocol.

Related protocols

Continue with another reviewed or low-risk practical entry that is close to this topic or Life Area.

Article versions

Brali keeps substantive article history visible. Later reviews may refresh wording, sources, examples, or boundaries without silently replacing the record.

Canonical source record · Evidence state: practical.