The method
Why practice beats a course.
Most AI training for employees is a video course. People watch it, nod, and go back to using AI the way they did before, because watching somebody else be competent has never made anybody competent. The thing that transfers is doing the work and being corrected while you do it. That is the whole method, and everything below is how it is built.
Why AI training for employees usually fails
Nearly every company has now given its staff an AI tool. Far fewer have made their staff good at using one, and those are completely different problems. A licence is a purchase. A skill is a practice. Handing someone a chatbot and a policy document does not produce the second one, which is why so much AI spend shows up on an invoice and nowhere else.
The gap is measurable, and it is not a gap in enthusiasm.
What the research says
These are findings about AI and AI training in general, from primary sources, each linked. None of them is a measurement of Trainwork learners: we have not run that study, and we will not present somebody else's result as if we had.
- 64%of employees say the AI training they get is not enough.
BCG, AI at Work 2025 - 2.2 hrsgiven back every week to each employee who uses AI, 5.4% of their hours.
St. Louis Fed, 2025 - 79%of employees given more than five hours of training become regular AI users, against 67% who get less.
BCG, AI at Work 2025 - 15%more work done per hour by people who know how to work with AI. The least experienced gain the most.
Brynjolfsson, Li & Raymond, QJE 2025
One caveat worth stating, because it is the kind usually left out: the Quarterly Journal of Economics study varied whether workers had access to an AI assistant, not whether they were trained. It is honest to say the least experienced gained the most, since the assistant worked by surfacing expert practice to everyone. It would not be honest to claim it measured trained against untrained workers.
AI course or coached practice: which one teaches the skill
Not a criticism of video as a format. It is a good format for explaining a thing. It is a poor one for building a skill, and AI at work is entirely a skill.
| A recorded AI course | A coached conversation | |
|---|---|---|
| Who does the work | You watch someone else do it | You do it, and get corrected while you do |
| Fit to your job | One fixed path, whatever your job is | The conversation adapts to your role and your industry |
| The examples | Examples from a company that is not yours | Your own work is the example, every time |
| When you find out it did not land | Weeks later, at your desk | In the session, while it can still be fixed |
| What you leave with | Notes, and a completion tick | The thing you came in needing |
| Shelf life | Out of date the quarter after it is recorded | Coached against how the tools behave this month |
Why every skill is 3 conversations, not one
A single session teaches a trick. Three sessions in the same skill, shaped differently, teach the skill. The shape is the same everywhere in the curriculum.
Foundation
Do it once, with the reasoning made explicit
The first conversation does not tell you a rule and move on. It makes you build the thing, names why each part works while you are building it, and refuses to hand you the finished answer. Being told the four parts of a good prompt is worth very little. Assembling one for a message you actually have to send is worth a great deal.
Advanced
Do it again when it goes wrong
The second conversation starts from the failure case, because that is where the skill really lives. A prompt that works is easy. Knowing what to change when a good prompt still returns something useless is the part that separates someone who uses AI from someone who gets value out of it.
Mastery
Do it a third time, and keep the artifact
The third conversation ends with something reusable that goes into your notes: a prompt library, a template, a checklist you wrote. Practice that leaves nothing behind decays. Practice that leaves a tool behind gets used again next week, which is the only retention mechanism that has ever reliably worked.
15 skills at 3 conversations each is 45 reps, and finishing them earns the Fluency Certificate, which carries a public verification page anyone can check. See the full curriculum.
Last reviewed . Every figure on this site links to its source.