The session goes well. People lean in, ask good questions, and try a few things on the spot. Someone stays behind to say it was the most useful hour they have spent on AI all year.
A month later, very little has changed.
A few people use the tools every day. Most tried something once, hit a snag, and went back to the old way. One or two have started pasting customer emails into whichever tool was open. Nobody could tell you which group is largest.
AI training usually fades because it runs as a one-off event. It sticks when the organization treats the session as the first step in changing how people work, and puts four things around it: colleagues nearby who can answer questions, one place to find the prompts and tools that already work, data rules everyone knows, and a regular check on which habits are catching on. Good training still matters. It just cannot do that work on its own.
Why does AI training fade so quickly?
A workshop changes what people know. It does nothing to their calendar.
On Monday morning they return to the same inbox, the same deadlines, and the same way of doing each task. Trying a new approach costs time the day does not have, and when it goes wrong, there is nobody nearby to ask. It looks like resistance. Most of the time it is just a busy week winning.
Two other things happen quietly at the same time.
The first is that good discoveries stay with the person who made them. Someone works out a reliable way to draft a weekly report or summarize a long thread, and it lives in their head or their chat history. The colleague at the next desk solves the same problem from scratch, or never does. When that person changes roles, the method leaves with them. It is the same dependency I wrote about in One resignation away.
The second is that risky habits stay hidden too. Without clear rules, people make their own judgments about what is safe to paste into a tool. Most judge well. Some do not, and nobody finds out until it matters.
What counts as AI adoption?
AI adoption is the point where a new way of doing a task becomes the default, without anyone reminding people to use it.
That definition describes behavior, not attendance. It is easy to count how many people went to a workshop. It is much harder to count how many people changed the way they work. So organizations measure the first, plan for the first, and are surprised when the second does not follow.
Training teaches people what a tool can do. Adoption is what they keep doing after the trainer leaves.
What makes AI training stick?
Four things around the training make the difference: champions people can reach, a shared toolkit kept in one place, rules people can remember, and a regular look at what people use.
Champions people can reach
An AI champion is a person inside a team who has deeper training, some protected time to help colleagues, and a direct line to whoever sets the rules.
Champions turn a question that would have gone unanswered into one that gets answered quickly, by someone who knows the team's work. They also notice what is working. When a colleague finds a better approach, the champion is the person who hears about it and passes it on.
The role only works if the time is real. Without protected hours, helping colleagues is the first thing to go when the week gets busy.
A shared toolkit, kept in one place
A shared AI toolkit is a single place to keep the prompts, custom assistants, and short guides that have held up on real tasks. Its job is to make one person's good method everyone's starting point, so the report template someone worked out in week two does not stay in their chat history. One named person keeps it current and prunes it when a tool changes or an entry stops working.
Ideally, the best entries come from the people doing the work, not the trainer. When that starts happening, the program is taking hold.
Rules people can remember
Clear rules can encourage experimentation rather than discourage it. People are more willing to try things when they know where the edges are.
Keep the rules short enough to remember. They should answer a few plain questions: which tools are approved, what never goes into a general-purpose or unapproved tool, and who checks AI output before it reaches a customer or a decision. The list of what stays out should be concrete, for example customer records, personnel files, and anything covered by a confidentiality agreement.
A regular look at what people use
Keep it light: a short survey before and after, a quick check on who is using what, and a rough sense of where any time saved is going. Use it to decide which habits to spread and which to drop, and to find the person who has quietly worked out something better. Nobody is being graded.
Should you train everyone at once or start with a pilot group?
For most organizations, start with a pilot group and widen from there.
Train broadly when the tools are already licensed across the organization, the rules are agreed, and the main problem is that people do not know the basics. A wide rollout gets everyone to the same starting line quickly.
Start with a pilot group when the tools or rules are still being decided, or when the work varies a lot from team to team. A small group of champions can test approaches on real work, and what they learn shapes the wider rollout. Mapping a workflow before an AI pilot covers how to choose the work to start with.
In practice, that usually means a short leadership briefing so decision makers understand what they are funding, then a pilot group, then a wider rollout built on what the pilot found.
How do you know if AI training is working?
Look at behavior, and look early. A check about two weeks after training, and another at six to eight weeks, gives you an early read and a settled one.
Look for: questions that shift from "how do I start" to "how do I do this better," a toolkit that grows from contributions rather than only from the trainer, and champions who are busy.
Be cautious when: usage stays concentrated in the same few people, nothing new has been added to the toolkit, or the rules get quoted but not followed.
Where should you start?
Start small. One leadership briefing of about an hour, or one workshop of 60 to 90 minutes built on the tasks your people already handle each week, is enough to begin. Name a champion or two on each team before the session, not after. Write the rules down, even if they fit on one page. Then check in at two weeks. If nobody has changed how they work, fix that before you book more sessions.
If you are planning AI training for your team, here is how we approach it at Joust. I lead the sessions myself, and we build them around your team's own work.
The workshop is the easy part. Plan for the month after.
Ron Davis
Founder
Three decades building enterprise platforms. Started Joust to close the gap between strategy decks and the work they're supposed to change.