Five Things Every AI Trainer Should Prepare -- and the One That Saves You
In March 2026, I ran a five-day creativity workshop for Adaptig's trainer cohort from a hotel room in Tokyo -- live sessions over Roam, patchy Wi-Fi, convenience-store coffee. The workshop teaches trainers to run AI-powered creative builds for corporate audiences: from blank brief to finished product concept using Gamma, NotebookLM, and Google AI Studio. The exercise is a product launch for a fictional energy drink called Blue Bear, competing with Red Bull -- research, positioning, packaging visuals, pitch deck, presentation, all inside two hours, all with AI tools.
By the final session, the preparation for any AI training engagement had crystallized into five specific outputs. These didn't have names until enough failures forced them into shape.
1. A prompt pattern
The reusable structure for the core prompt in your session -- not a single prompt but a pattern: role assignment, context block, output format, constraints. You adapt it per client, per audience, per industry, and test it against three scenarios before the engagement. If it breaks on one, you fix the pattern, not the scenario. Having this built means you're not improvising the most critical input of the session under pressure.
2. A demo script
The exact sequence you show, in what order, with what tool -- not a slide deck but the choreography of a live demonstration. Which tab opens first, what you type, what you say while the model is generating. Those ten seconds of processing time are dead air unless you've planned what to narrate during the wait.
3. A critique lens
The specific criteria you use to evaluate AI output in front of the audience. "Is this good?" is not a critique lens. "Does this match the brief's target audience, and would you send it to a client without editing?" is. The lens gives participants a framework for judging output instead of accepting it on instinct -- and different sessions need different lenses.
4. A participant exercise
The hands-on task participants build during the session, with clear inputs, a time limit, and a defined output. Blue Bear is ours -- yours should be relevant to the audience's industry. The exercise works when participants produce something they take back to their desk. It fails when it's a toy example they'd never replicate at work.
5. A failure-recovery move
This is the one that matters most and is hardest to prepare. A failure-recovery move is what you do when the AI generates something useless in front of thirty people and you have about four seconds before the room decides you're winging it.
The first time this happened -- live demo, corporate audience, thirty minutes of build-up, one unusable image, a silence that lasted longer than it should have -- the recovery came from instinct, not planning. Narrate what happened. Explain why the prompt probably failed. Adjust it on screen. Let the room watch the iteration. That sequence turned out to be more educational than the demo working on the first try -- but knowing that afterward and executing it with thirty people watching are different skills entirely.
Which is why it has to be practiced. In the trainer program, everyone recovers from a deliberately broken demo at least once before delivering to a client. The move is: name what happened plainly, resist the urge to apologize, keep your hands on the keyboard, and let the room see you think. Most audiences find the failure more interesting than the success, as long as you let them watch the process.
The list doesn't cover everything. It doesn't teach charisma. It doesn't fix a trainer who knows the tools but can't hold attention. But it solves the dependency problem -- the part where every session runs through one person's calendar because nobody else has the material prepared well enough to deliver independently. Five outputs, each built and tested before the engagement, the fifth one practiced until the recovery is reflex.
