A company rolls out AI training.
Employees attend the webinar. They see examples. They learn a few prompts. Someone shows them how to summarize a document, draft an email, brainstorm ideas, and speed up basic tasks.
People leave curious.
A week later, most of them are back to the same habits.
Some use AI for low-risk writing. Some avoid it because they are unsure what is allowed. Some try it, get a weak answer, and decide it is overhyped. A few power users move ahead on their own. Managers do not know what to coach. Teams do not agree on where AI belongs in the work.
The training happened.
The work did not change.
That is the problem.
Most AI training fails because it teaches the tool and skips the transfer. It gives people exposure without giving them a clear path into daily work.
The wrong goal: tool familiarity
A lot of AI training is built around the question:
How do we teach people to use this tool?
That sounds practical. It is too small.
Tool familiarity is easy to teach. Employees can learn where to type, how to ask a better question, how to upload a file, how to rewrite an answer, and how to try different prompts.
Those are useful skills. They are not enough to change performance.
The better question is:
How should this work change because AI is now available?
That question is harder. It requires looking at the tasks people actually do, the judgment those tasks require, and the points where AI could help or create risk.
If training never reaches that level, employees leave with tips instead of new work habits.
AI training has a transfer problem
Learning and development has studied this issue for decades. Training transfer refers to whether people use what they learned back on the job. Baldwin and Ford’s classic review made the basic point clear: transfer depends on more than the training event. It is shaped by the learner, the design of the training, and the work environment.
AI training often fails on all three.
Employees may be interested but uncertain. The training may be too generic. The work environment may send mixed signals. Managers may not know what good use looks like. Policies may be unclear. Workflows may stay untouched.
So people do what people usually do under pressure.
They return to the familiar way of working.
That is not resistance. It is gravity.
The training is too far from the work
Generic AI examples are useful for awareness. They are weak for adoption.
A prompt that works in a demo does not automatically help a claims analyst make a better recommendation. A clever use case from marketing may not help a compliance specialist decide what needs human review. A productivity example may not help a manager coach an employee through sensitive performance feedback.
People need to see AI inside their actual work.
That means training should be built around real tasks:
reviewing a customer issue preparing for a difficult conversation comparing policy language drafting a project update analyzing feedback creating a first version of a job aid identifying risks in a proposed change summarizing meeting notes into decisions and owners
The task matters because the task carries the judgment.
Without the task, AI training becomes a tour of features.
Prompt training is usually overvalued
Prompting matters. But prompt training is often treated as if it were the whole discipline.
It is not.
A well-written prompt cannot fix a poorly understood task. It cannot decide what quality means. It cannot tell an employee when the output is incomplete. It cannot replace the context people need to judge whether an answer is useful.
Better AI training teaches employees to ask:
What am I trying to produce? What context does the tool need? What would a good answer include? What risks should I check? What should I verify somewhere else? What decision still belongs to me?
That is a different kind of learning. It is closer to judgment training than software training.
The goal is not to make everyone a prompt engineer.
The goal is to help people use AI without outsourcing their thinking.
Experts are missing from the design
The people building AI training often involve technical teams, learning teams, and compliance teams.
They should also involve the people who understand the work.
Every function has experts who know where the real judgment lives. They know the common mistakes. They know which shortcuts are safe. They know where the official process leaves room for interpretation. They know when a good-looking answer is wrong.
If those experts are absent, AI training stays generic.
Employees learn how the tool works in theory, but not how the work should be done with the tool in practice.
This is where expertise translation matters.
Before AI training can stick, expert judgment has to be made visible. That means translating what experienced people know into shared language, criteria, examples, review habits, and decision rules.
Training can then teach those translated practices.
Without that step, the organization is asking people to adopt AI without showing them what good use looks like.
The manager layer gets skipped
Managers are often expected to support AI adoption after a short briefing.
That is not enough.
Managers need to know what to reinforce. They need examples of acceptable use. They need to understand where AI can help their team and where it creates risk. They need language for coaching employees who are experimenting, avoiding, overusing, or misusing the tool.
If managers cannot answer basic questions, adoption becomes uneven.
One team experiments carefully. Another team avoids AI entirely. Another quietly uses it for work that should have review. Another treats AI use as a personal productivity trick instead of a team capability.
Manager enablement is part of AI training.
If managers are not prepared to coach the behavior, the training event has a short shelf life.
Policy is being confused with learning
Employees do need rules.
They need to know what data they can enter, what tools are approved, what uses are restricted, what must be disclosed, and what requires human review.
But policy alone does not create skill.
A policy can tell people what not to do. It rarely teaches them how to think through a real task. It does not show them how to improve an output, evaluate quality, or decide when AI is the wrong tool.
Some organizations overcorrect here. They lead with restrictions because risk feels urgent. Then employees either avoid the tool or use it quietly without much guidance.
Good AI training should connect policy to practice.
For example:
Show employees what a risky input looks like. Compare a weak AI output with a better one. Give examples of tasks that are safe for experimentation. Show where human review is required. Let people practice deciding whether AI belongs in a task.
Policy tells people the boundary.
Learning helps them work inside it.
A better goal: changed work
AI training should be judged by what changes after the session.
Did employees use AI on the job?
Did they use it in the right tasks?
Did the quality of work improve?
Did managers reinforce the expected behavior?
Did the workflow change?
Did the team agree on review standards?
Did fewer people rely on the same overloaded expert for routine questions?
If the answer is no, the training probably created awareness rather than transfer.
Awareness has a place. It is not adoption.
What to do instead
Better AI training starts with the work, not the tool.
Here is a practical sequence.
- Choose one workflow
Do not start with every possible use of AI.
Pick one workflow where better support would matter. It might be onboarding, customer escalation, meeting follow-up, performance support, policy interpretation, proposal drafting, project planning, or knowledge capture.
A narrow workflow gives the training something concrete to change.
- Map the judgment points
Look at where people make decisions in that workflow.
Ask:
Where do employees slow down? Where do newer people make mistakes? Where do people ask an expert for help? Where does quality depend on context? Where would AI save time? Where would AI create risk?
This keeps the training grounded in actual performance.
- Translate expert practice
Bring in experienced people and ask them to walk through real examples.
Do not ask only for tips. Ask them how they think.
What do they notice first? What makes them pause? What does a good answer include? What would they never let AI decide? What would they want a newer employee to check before moving forward?
Turn those answers into usable training assets:
examples review criteria decision guides practice scenarios approved prompts checklists escalation rules before-and-after samples
Now the training has substance.
- Teach with real tasks
Let employees practice on work that looks like their work.
A good practice activity should include enough context to require judgment. It should also include review. Employees need to compare outputs, revise them, explain what they changed, and decide what still needs human attention.
That is where learning happens.
The value is not in watching AI produce an answer.
The value is in learning how to judge the answer.
- Build manager follow-up
Managers need a simple way to keep the learning alive.
Give them questions they can use in team meetings:
Where did AI help this week? Where did it waste time? What output needed the most revision? What did we decide should stay human-led? What example should we add to our team guide?
This does not require a large program. It requires a routine.
AI adoption becomes more durable when teams keep refining how they use it.
- Put the learning into the workflow
Training should leave behind something people use.
That might be a job aid, a team agreement, a prompt guide, a quality checklist, a review rubric, or an example library.
If the learning artifact sits in a folder, it will fade.
Put it where work happens: inside the project template, onboarding path, team meeting rhythm, knowledge base, manager checklist, or workflow tool.
Training sticks when the environment supports the new behavior.
What good AI training looks like
Good AI training is practical, specific, and tied to performance.
It does not promise transformation in one session.
It helps people make better decisions inside real work.
A useful AI training session might sound like this:
Today we are going to use AI to improve customer escalation summaries. We will look at three real examples, identify what makes a summary useful, use AI to create a first draft, revise the draft against expert criteria, and decide which cases still need human review.
That is better than:
Today we are going to learn ten ways to use AI at work.
The first version has a task, a standard, a workflow, and a review habit.
The second version has curiosity.
Curiosity starts adoption. It does not sustain it.
The role of L&D
Learning teams should not position themselves as tool trainers.
That is too narrow for the work ahead.
L&D is well suited to help organizations answer the adoption questions other groups may miss:
What performance problem are we solving? What does good use look like? What expert judgment needs to be translated? What practice will help employees build skill? What support will managers need? What should change in the workflow? How will we know whether the training transferred?
Those are learning questions. They are also change questions.
AI adoption needs both.
The mistake to avoid
The common mistake is treating AI training as a launch activity.
The tool launches, so training is scheduled.
That sequence makes sense administratively. It often fails behaviorally.
People do not change how they work because a tool is available. They change when the new behavior is clear, useful, supported, and easier to repeat than the old one.
That takes design.
It takes translated expertise.
It takes managers who can coach.
It takes practice that resembles the job.
It takes follow-up after the novelty fades.
Closing thought
AI training fails when it stays at the level of tool exposure.
Employees need more than a demo. They need to know where AI fits, what good use looks like, what risks to watch, and which parts of the work still require human judgment.
The better question is not:
Did people attend AI training?
The better question is:
Did the training change how the work gets done?
That is the standard worth designing for.
References
Baldwin, T. T., & Ford, J. K. (1988). Transfer of training: A review and directions for future research. Personnel Psychology, 41(1), 63–105.
Blume, B. D., Ford, J. K., Baldwin, T. T., & Huang, J. L. (2010). Transfer of training: A meta-analytic review. Journal of Management, 36(4), 1065–1105.
Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition and the culture of learning. Educational Researcher, 18(1), 32–42.
Burke, L. A., & Hutchins, H. M. (2007). Training transfer: An integrative literature review. Human Resource Development Review, 6(3), 263–296.
Eraut, M. (2000). Non-formal learning and tacit knowledge in professional work. British Journal of Educational Psychology, 70(1), 113–136.
Grossman, R., & Salas, E. (2011). The transfer of training: What really matters. International Journal of Training and Development, 15(2), 103–120.
Jarrahi, M. H. (2018). Artificial intelligence and the future of work: Human-AI symbiosis in organizational decision making. Business Horizons, 61(4), 577–586.
Salas, E., Tannenbaum, S. I., Kraiger, K., & Smith-Jentsch, K. A. (2012). The science of training and development in organizations: What matters in practice. Psychological Science in the Public Interest, 13(2), 74–101.
Tofel-Grehl, C., & Feldon, D. F. (2013). Cognitive task analysis–based training: A meta-analysis of studies. Journal of Cognitive Engineering and Decision Making, 7(3), 293–304.