Rating
- size:
5 - complexity:
5
Purpose
Practice disciplined AI use in a high-ambiguity scenario where the model can help organize signals but cannot safely replace investigation judgment.
Scenario
Customers report duplicate orders.
Available information:
- API logs show some duplicate request IDs and some unique request IDs
- worker logs show retries after timeout from a downstream payment provider
- one engineer suspects UI double-submit
- another suspects missing idempotency around the queue handoff
- AI produces a polished root-cause story blaming retries
Additional constraints:
- customer impact is ongoing
- rollback would disrupt unrelated checkout improvements
- support needs a short update within
30minutes
Task
Use AI to help organize the problem, but do not let it declare the answer.
Produce:
- a working hypothesis map
- immediate containment options
- a verification plan
- a short, uncertainty-aware stakeholder update
Expected output
- clearly labeled hypotheses
- near-term action list
- evidence still needed
- concise update that does not overstate the root cause
Good AI uses
- organizing timelines and hypotheses
- drafting a structured stakeholder update
- comparing containment options
Verification focus
- separate symptom narrative from confirmed mechanism
- identify which evidence would confirm retry-based duplication versus UI duplication versus queue idempotency gaps
- keep stakeholder communication accurate under uncertainty
Anti-patterns to watch
- treating a coherent AI story as proof
- skipping confidence labeling in the stakeholder update
- optimizing for narrative neatness over operational truth
Debrief prompts
- what part of the AI output was most tempting to overtrust
- what did you deliberately keep uncertain in the stakeholder message
- what would count as enough evidence to narrow the hypothesis set