AI training for employees: what to teach, and how to make it stick
AI training for employees works when people practise on their own tasks with a real AI agent, not when they watch one. This guide gives the skills to teach by role, the formats compared, what a request really costs, and a rollout in four phases.
- No setup
- No AI licences to buy
- Runs in the browser
What is AI training for employees?
Teaching people who do not write code to get real work done with AI: understand what the agent does, ask for what they want, build with it, then hand it whole tasks. Four levels, learned in this order.
Understand
AI Essentials
- They learn
- What an AI agent reads, counts and costs: tokens, context, the right model.
- Afterwards they can
- Tell why a request went wrong, or cost more than it should.
Ask
Prompting with AI
- They learn
- A request in four parts: context, task, guidelines, constraints.
- Afterwards they can
- Get the result in one precise request instead of three vague ones.
Build
Building with AI
- They learn
- A web page, a chart or a tool, built by talking to the agent.
- Afterwards they can
- Turn their own data into something colleagues use, without a developer.
Delegate
Build an Agent
- They learn
- An assistant with a role, rules, tools and skills, tested on real messages.
- Afterwards they can
- Hand a repeated task to an agent they have tested.
What to teach, role by role
Everyone needs the first two levels. Past them, what a team needs depends on the tasks it does every week.
| Role | Understand | Ask | Build | Delegate |
|---|---|---|---|---|
| Everyone | Core | Core | Useful | Later |
| ManagersDecide which tasks to hand over | Core | Core | Useful | Useful |
| Sales and customer serviceAssistants that answer customers | Core | Core | Useful | Core |
| HR and recruitingRecruitment is a high-risk use in the AI Act | Core | Core | Useful | Useful |
| Finance and operationsCharts and tools from their own data | Core | Core | Core | Useful |
| L&D and learning designersBuild the sessions for everyone else | Core | Core | Core | Core |
Video courses, workshops or hands-on AI training?
The three formats companies buy, asked the same five questions.
| Video course | Live workshop | Hands-on practice | |
|---|---|---|---|
| Each person does the task themselves | NoThey watch someone else do it | PartlyA few exercises, in a group | YesEvery step, with a real agent |
| Feedback on what they made | NoA quiz on definitions | PartlyFrom the trainer, when time allows | YesMilestones checked on the result |
| Reaches every employee | YesAny number of viewers | NoOne trainer, one room | YesNo trainer needed to move on |
| Up to date with today's AI | NoRecorded once | YesIf the trainer keeps up | YesThe real agents, as they are today |
| Leaves a record per person | PartlyCompletion and a quiz score | NoAn attendance sheet | YesMilestones, attempts and tokens |
Hands-on AI training, one session at a time
About 45 minutes. The learner reads a short brief, asks the agent for what they want in plain words, and sees the result appear next to the conversation.
- Milestones checked on the result. The page, the files, the agent's replies: nothing to grade by hand.
- A coach that follows. Hints and feedback; it never does the work for the learner.
- No code on screen. Learners describe what they want, the agent does the rest.
- 1Brief
- 2Learner's request
- 3The agent builds
- 4Result in the preview
- 5Milestone validated
Play a session
A recorded session of the platform, with its brief, its coach and its milestones. Send the request and watch the agent work.
The skill most AI training forgets: what a request costs
An AI agent is paid by the token, and how a person asks decides how many it spends. We measured it.
Each square is 10,000 tokens
Measured on the platform, in the session “Ask Once, Ask Clearly”.
- Its own instructions, read again each timeabout 20,700 tokens
- The conversation so fargrows with each message
- What it looks at and writes nowfiles, the page, its answer
So one step costs at least about 21,000 tokens, and a request with two to five actions 45,000 to 117,000.
Count the tokens of your own textPractise precise requests with Prompt Golf
An AI training program for employees, in four phases
Start from the tasks, not the tool, and widen once the first team has done it.
- Weeks 1 and 2
Pick the tasks
Three to five tasks per team where AI would help, and the people who do them.
- Weeks 3 and 4
Pilot with one team
AI Essentials and Prompting with AI, with a token budget per session. Read the reports together.
- Month 2
Give learning paths
Programs assigned to units, groups or people, sign-in through your SSO (Entra ID, Okta or Google).
- Month 3
Measure, then widen
Milestones reached and tokens spent across attempts; then Building with AI and Build an Agent where the tasks call for them.
How to measure AI training
Count what people did with the agent, not how long they watched.
Shows a skill
- A milestone validated on what the person built
- The same task done again with fewer tokens
- One precise request where there used to be three
- An agent tested on real customer messages
Only shows attendance
- A completion rate
- A quiz on definitions
- Hours of video watched
- A satisfaction score
The platform keeps the first column for every learner: milestones, attempts, and the tokens each one cost.
AI training and Article 4 of the AI Act
In the EU, employers that use AI systems must take measures for their staff's AI literacy.
Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf […]
- Article 4 applies
- It is enforced
No training makes a company compliant on its own. Practice on real agents, with a record of each learner's progress, is one such measure.
Check your measures with the AI literacy checklistAI literacy training
Questions about AI training for employees
What should AI training for employees include?
Four levels, in order: understand what an AI agent reads and costs, ask in four parts (context, task, guidelines, constraints), build with it, then delegate tasks to an agent. Each one practised on a real task with a real agent, not described in a video.
How long does AI training for employees take?
A session lasts about 45 minutes. Sessions are grouped into four programs, taken in order: AI Essentials, Prompting with AI, Building with AI and Build an Agent, which has 13 short sessions, two prompt games and a final assessment.
Is AI training mandatory for employees in the EU?
Article 4 of the AI Act asks providers and deployers of AI systems to take measures to support their staff's AI literacy. It applies since 2 February 2025 and is enforced since 2 August 2026. It does not prescribe a course: training with a record per learner is one such measure.
How do you measure AI training?
On what people did: milestones validated on the result they built, the tokens each attempt cost, and how those numbers change from one attempt to the next. A completion rate or a quiz score shows attendance, not a skill.
Do employees need to know how to code?
No. They describe what they want in plain words, and never see code on their screen.
Which AI tools do employees train on?
The real agents: Claude Code, Codex (OpenAI) or GitHub Copilot, as they are today, run by the platform. Your organization chooses its agent; there is nothing to install and no licence to buy per learner.
What does the AI cost during training?
Every request is counted in tokens, step by step, and some sessions have a token budget, so learners see what each request cost and learn to spend fewer. The report at the end shows the cost in money at your model's real prices. For the platform itself, book a demo: there is no public price.
Can we train employees on our own business cases?
Yes. Your learning designers make sessions of their own in the Studio by describing them, with your files and business cases, and try them as a learner before publishing.
See it with your own teams.
A 30-minute call: we get to know your project, then show you the platform.
Book a demo