Reusable template
Use matching inputs for manual work and every AI option. Run three fresh-conversation trials for each of inputs A and B per AI option and record all six outcomes. Time manual work separately on both inputs. Compare with the reference below. A blank result is not success and unknown time is not zero.
| Option, model/feature and plan | Date | Input / prompt / trial | Serious error | Preparation min | Review min | Repair min | Handover min | Total min | Result and evidence |
|---|
| | | | | | | | | |
Prompt v1: From these fictional notes, extract only explicitly agreed actions into task / owner / due date / question. Do not invent missing facts. Preserve both conflicting dates and ask for clarification. Separate unapproved suggestions. If no action was agreed, say so. Do not create or send anything.
Reference answer A
1. Prepare the draft — Anna — 2026-10-12.
2. Check the budget — Boris — conflicting 2026-10-14 / 2026-10-15; clarify the date.
3. Send the sample — owner and date unknown; request a decision.
The paid survey remains an unapproved suggestion, outside accepted actions. Expected answer B: no accepted action. Wording may differ, facts must remain. No product has been evaluated yet.
Total = preparation + review + repair + handover. One-time setup: __ minutes. Additional license: __ in __ currency per __ period; runs over the same period: __. Mark unknown items for verification. A reference answer is not measured product performance.