Training teams for AI without the awareness deck
The hour-long AI awareness session changes nothing by design. What works: role-shaped practice on real work, and the colleagues next desk over.
The standard corporate response to AI is a training module: an hour of slides, a definition of a token, a quiz. Everyone passes, nothing changes, and the box is ticked. Meanwhile the actual training is happening anyway, unsupervised: the surveys show people adopting these tools chiefly because they can see how to work faster, policy or no policy. Your staff are already in class. The question is whether the company shows up to teach.
Train by role, on the role's real work. A claims processor, a marketer and an engineer need to practice on their own tasks, not on generic prompting theory. The format that works is embarrassingly simple: ninety minutes, the team's actual documents and tickets, a facilitator who does the job rather than merely the AI. Each session ends with two or three workflows the team will run with the tools this week, and the paste-this-not-that rules for their specific data (the one-page policy becomes concrete exactly here).
Judgment is the curriculum. Using the tools is easy; knowing when not to trust them is the skill. Good training shows failure on purpose: the confident wrong answer, the citation that leads nowhere, the subtle error in generated code. People who have watched a model be persuasively wrong calibrate faster than any slide about hallucinations can manage.
Then fund the neighbors. The durable teacher is the colleague one desk over who figured something out. Give those people time, visibility, and a channel where wins and horror stories circulate freely. A workshop opens the door; the practice community keeps people walking through it. The awareness deck, for the record, was neither.