The Copilot

The Copilot confronts participants with a workweek they cannot handle alone. Across two sprints, they choose which missions to delegate to a generative AI tool, then face the consequences of their choices. The game brings to life the trade-offs between time saved, quality, confidentiality, accuracy, and human relationships. It is designed for teams that want to build concrete, safe day-to-day uses of AI.

Walkthrough

1. 1

   Check-in

   7 min

   In pairs, each person shares a task they have already entrusted to a generative AI tool, or one they would never dare entrust to it. Bring back 3 examples in the full group, without technical debate. “Think of a real task: either you have already entrusted it to AI, or you definitely would not want to. What guides your choice?”

   Tip — Welcome very different practices without judgment: the game will put the criteria up for discussion.

2. 2

   Game rules

   8 min

   Present the mission, mode, incident, and safeguard cards. Explain the 12 h budget, mode costs, trust starting at 10, and the final score. Show the sprint board. “Your challenge is not to use AI everywhere. Your challenge is to deliver value without losing trust.”

   Tip — Give a cost example: a 5 h mission in Copilot = 3 h, in Autopilot = 1 h.

3. 3

   Sprint 1

   11 min

   Each table draws 6 missions. They place them in the Human only, Copilot, or Autopilot columns, calculate hours, and check that the total does not exceed 12 h. No safeguard is available. “You have too much work and no shared rules yet. Decide quickly, then take responsibility for your trade-offs.”

   Tip — If a table gets stuck, ask: “Which risk do you consciously accept, and which one do you refuse?”

4. 4

   Sprint 1 incidents

   7 min

   Draw 4 incident cards for all tables. Read them one by one. Each table checks whether the condition is met on at least one sprint mission and applies the effect on trust. “Incidents do not mean AI is bad. They reveal what happens when context, verification, or a framework is missing.”

   Tip — Apply an incident only once per table, even if several missions match.

5. 5

   Mini-debrief

   5 min

   Ask tables what cost them the most trust and what protection they wish they had. Write on the board: data, accuracy, relationship, creativity, responsibility. “What should have stayed human? What could have been accelerated, but not abandoned?”

   Tip — Do not try to build a full charter yet: simply prepare for sprint 2.

6. 6

   Sprint 2 with safeguards

   14 min

   Each table draws 6 new missions. Make the 6 safeguards available. Teams may buy them, 2 h each, within the 12 h. They plan missions, modes, safeguards, and total time. “This time, you can buy safety. But each safeguard consumes time: choose the ones that truly protect your decisions.”

   Tip — Remind participants that a safeguard is active for the whole sprint, not just for one mission.

7. 7

   Incidents and score

   8 min

   Draw 4 new incidents. Teams apply the effects, unless a listed safeguard is active. Then they add up the values of missions delivered across the 2 sprints and final trust. “Verification costs time, but it can save trust. Let’s now see what your choices produce overall.”

   Tip — If the score is low, value the learning: an error in the game prevents an error at work.

8. 8

   Group debrief

   8 min

   Facilitate the debrief questions in three phases: Feel, Understand, Transfer. Invite several tables to speak, especially about trade-offs that changed between the two sprints. “The score is a pretext. What matters is the criteria you take away to make decisions tomorrow.”

   Tip — Reframe responses as simple criteria rather than general opinions about AI.

9. 9

   Charter in 5 rules

   7 min

   Hand out the charter canvas. Each table formulates 5 rules: what to do, why, with an example. Ask for one rule to share in plenary and consolidate convergences. “A good rule must help you decide on Monday morning: concrete, observable, and short enough to remember.”

   Tip — Push teams to write actionable rules: “anonymize before entering,” not “be careful.”

Variants

• Remote : Use a collaborative board with the cards as thumbnails and one board per subgroup. The facilitator draws incidents in plenary. Teams announce their budget verbally.
• Large group : Up to 24 people: keep 4 tables of 6. Draw the same incidents for everyone to compare strategies and speed up the debrief.
• Sensitive sector : Add a constraint: any 🔒 mission cannot be in Autopilot. This variant strengthens the compliance and responsibility discussion.