Most AI training fails in the same way. People watch a tour of features, try a prompt, and go back to the old workflow on Monday. The headlines were exciting. The habits did not move.
A useful day starts from the work the team already does. Research, drafting, analysis, triage, customer replies. Pick the tasks where a model can save time without creating a compliance problem.
Everyone should leave having made something with a real tool, plus a short list of what must never be pasted into one. Accuracy, data and IP are part of the curriculum, not a footnote.
One workflow with an owner
Then write the 90 days. Which use cases, which owners, what good looks like. Training that stops at inspiration is a keynote. Training that names the next workflow is a capability.
Build the day as an AI training workshop around one workflow the team already runs. What that day should produce is in what a corporate AI training day should produce. The rules around a send sit in AI governance.
The OECD AI Principles are the short public list I use when a role page needs a boundary.