Build it

1. 5 min

   Boards out. Read your own brief back to yourself before touching anything.

2. 10 min

   First card, on paper, before any prompt.

   What does a good version look like? Two things, specific.

3. 20 min

   Prompt, read what came back against the card, circle one of four verbs. Repeat.

   How would you know if that came back wrong?

4. 10 min

   Run the drill on the one thing you built.

5. 5 min

   Count the cards out loud.

• The person who got a good result on the first prompt and stopped. It looks like the room's success story and it is the least useful outcome in it.
• Fix it circled where Ask differently is the honest answer. Fix it puts the problem in the output; Ask differently puts it in the request, which puts the problem somewhere less comfortable.
• Anyone typing before their card is finished. The card takes ninety seconds and skipping it is the whole failure mode in miniature.
• Results that could belong to anyone in the room. That came from a request that could have come from anyone, and the fix is on the board, not in the prompt.
• Three cards all circled Keep it. Worth asking what the second-best version would have looked like.
• The fourth line of every card is the gate. Read three cards over shoulders mid-build. A fourth line reading "I'll look at it" tells you where to spend your next five minutes.

• Send stuck people to their board, not to a better prompt. It is almost always the right diagnosis and it is the most useful thing you will do today.
• You are watching the cards, not the screens. A room where everyone is typing looks like a room that is working.
• Let the count at the end be public. The numbers do the work without anyone being named.
• Running late, the drill drops to two questions and the count can be skipped. The card cannot, since a prompt written without one is the thing this session exists to prevent.

The print pack: this plan and its sheets

• card on its own
• drill on its own