Most AI training is a lecture: someone shows impressive demos for ninety minutes, everyone nods, and within a week the team is back to exactly how they worked before. The information was fine. The design was wrong. After years in the classroom, here's what we've learned actually moves the needle.
Use their work, not toy examples
The fastest way to lose a room is a generic prompt about writing a poem. The fastest way to win it is to have people bring a real task from their actual job and improve it live. Relevance is the difference between "neat" and "I'm using this tomorrow."
Teach judgment, not tricks
Prompt tricks age out in months. What lasts is judgment: knowing when AI is the right tool, how to verify its output, and where it tends to be confidently wrong. We spend as much time on checking AI as on using it, because a team that can't catch a bad answer is more dangerous with AI, not less.
Build in the awkward part
Real skill comes from doing the thing and getting it slightly wrong with help nearby. Good workshops are mostly hands-on-keyboard, with the facilitator circulating, not presenting. If everyone's just watching slides, no skill is forming.
End with a commitment, not a recap
The last ten minutes matter most. Instead of summarizing, we have each person name one workflow they'll change this week and write down how. That single act of specificity is what separates a workshop people enjoyed from one that actually changed how they work.
None of this requires a bigger budget, just a different design. Make it their work, teach judgment over tricks, get hands on keyboards, and close with a commitment. Do that, and the workshop outlives the room.