Handoffs drop, between people and between people and AI
Start here
Closed-loop callouts
Echo a handoff back out loud, so every transfer of work is acknowledged and owned.
Try this first
On important handoffs, the receiver says back what they have taken on.
Borrowed from professional kitchens, and the closed-loop communication taught in health care by the TeamSTEPPS program. Read the original source.
Other moves for this problem (10 of 10)
Kind of change
- The Control Map · Map who can see and stop each stepFor when: AI has taken over steps in a workflow and nobody in the room can say who is accountable for a bad output, or who would catch one.
- say yes and deliver, or say no clearly · Commit clearly or decline clearlyFor when: Juniors defer to seniors and to the AI, and half-commitments abound.
- Challenge-and-response callouts + "question the unusual, listen, don't anticipate." · Say it out loud, confirm it backFor when: Critical handoffs are assumed rather than confirmed, or the team pushes past its own stop-points.
- CAPCOM — one voice to the crew · One voice talks to the team under pressureFor when: A team under pressure is getting conflicting inputs from many directions.
- The maintainer and the RFC · A named owner, and written proposals for big changesFor when: Ownership is unclear and big changes surprise people.
- Jidoka — "automation with a human touch." · Machines stop, people handle the exceptionsFor when: Designing or auditing a human-AI workflow, or when the AI is quietly handling exceptions it gets wrong.
- The Mixed-Initiative SeamFor when: People are either checking every AI output by hand or checking none of them.
- How Did You Know? · Walk an expert's decision backwardFor when: You are about to hand a step to a model, or already have, and the description of that step is thinner than the judgment it takes to do it.
- Redundancy as safetyFor when: Single points of failure exist and one person holds all the context.
- Provenance by DefaultFor when: People are signing off on AI-assisted work without knowing which parts the AI wrote or how sure it was.