Teaching one colleague, in twenty minutes
Training courses do not work for this. What works is one person showing one other person a task they both already do.
Companies that buy training courses for AI tools generally find that attendance is good, satisfaction is fine, and usage six weeks later is unchanged. What produces lasting use is smaller and much cheaper.
Why courses fail here
Because they teach the tool rather than the task. An hour of features produces someone who knows what is possible and has no idea what to do on Tuesday morning with the email in front of them.
People adopt a tool for one specific job and expand outwards from there. Starting with the general case inverts that and mostly produces nothing.
The twenty-minute version that works
One person who uses it sits with one person who does not, and they do a real task together. Not a demonstration — the colleague’s own work, from their own inbox, on their own screen.
Three rules make the difference. Use their task, not a prepared example. Let them type, even when it is slower. And go until the output is actually good enough to use, including the second and third attempt, because watching a first attempt fail and then get fixed is the entire lesson.
That last point is the one people skip, and it is why the 14 August problem persists: staff conclude the tool is poor after one bad answer, having never seen anyone recover from one.
Who should do the teaching
Not the most technical person. Whoever does the same job as the person learning, which usually means someone from the January conversation about what people tried over Christmas.
How it spreads
Each person who has been through this can do it for one more. Five pairs over two months covers a department, costs nothing, and produces use rather than awareness.
Write down what was taught — one paragraph per task, the 10 February format. Otherwise you are rebuilding the same knowledge every time somebody leaves, which is the 17 November problem in a different coat.
The measure
Not whether they liked it. Whether, three weeks later, they did it again without being asked. That is the only signal worth counting, and it takes one question to collect.