01 — The sample day

For employees

Understand first. Then do.

In the morning you learn how an AI assistant works, how to give it a good instruction, and what you may and may not put into it. In the afternoon you tackle your own tasks. About a fifth of the day is explanation. The rest is demonstration, practice and working on your own work.

Morning: the basics

08:45 to 12:30

08:4515 minutes
Arrival, login check and baseline

Everyone opens their own organisation's AI environment and fills in a card: three tasks that take a lot of time, how often you use AI now, what you expect and what worries you.
- Technology that does not work is better fixed now than later
- Without a starting picture, you cannot see at the end what has changed
In hand: everyone is logged in, the trainer knows where the group stands
09:0090 minutes
1. What an AI assistant is, and what it is not

A language model predicts the next piece of text each time, based on a very large amount of text it has read. It does not know facts, it knows patterns. The trainer shows this on a recognisable case and deliberately leaves one mistake in. Then, in pairs, you put three questions to the AI, one of which is a trick question.
- How a language model (LLM) works, in plain language
- What context is: it knows nothing about your organisation unless you provide it
- Why it sometimes makes something up with great conviction, and how you recognise that
- What it is good at, and where it structurally fails: facts, sources, current events and judgement
ExplanationLive demonstrationExercise in pairsIn hand: for each answer, how you check it, and your task list for the afternoon
10:30
Break
10:4555 minutes
2. Giving a good instruction

A good result comes from a conversation, not from one perfect question. You work out your first own task with the instruction card: fill it in, run it, refine it twice, check it. Then a colleague reads your result and looks for one mistake in it.
- The four things that are almost always missing: what you want, for whom, which material, and what it has to meet
- Why an example does more than an explanation
- How to have the AI state assumptions and uncertainties separately
Instruction cardOwn taskColleague reviewsIn hand: your first worked-out application
11:4050 minutes
3. Working safely: what may go in, what never

A sorting exercise in groups of three: allowed, only anonymised, never. The group chooses first, then the answer follows. Next, you anonymise the input for your next task and a colleague tries to trace it back to a person anyway. Anonymising reduces the risk, but does not remove it. What is allowed in which environment is decided by your own organisation.
- Personal data, confidential documents, passwords and internal links
- Why a decision about a person in principle may not be taken by fully automated means, and what the GDPR says about that (Article 22 GDPR), including the exceptions and safeguards
- What the EU AI Act asks for on the point of AI literacy
- Recognising instructions hidden in a document, email or website
Sorting exerciseAnonymising your own materialIn hand: a one-page reference card and a safe input for the afternoon
12:30
Lunch

Afternoon: practice

13:15 to 16:30

13:1545 minutes
4. From one-off questions to fixed ways of working

There are three ways of working: you ask something and assess the answer, you record a way of working that you use again and again, or you have a series of steps carried out within limits you set yourself. In pairs, for one recurring task, you fill in what the AI may do, what a person must do, and when it stops.
- When which form fits: a task that comes back every week deserves a fixed way of working
- What you record before you hand work over: sources, limits, permission, logging, stop criterion
- An AI operating model in plain words: who uses what, which tasks, who checks, what do we record, who decides
DemonstrationDesign sheet per pairIn hand: one sheet with an approval moment, a stop criterion and an owner
14:0090 minutes
5. Practice: your own work

The core of the day. Two rounds of about thirty-five minutes on your own, anonymised tasks. In between, two participants show their result: one that worked and one that got stuck. Getting stuck is learning material, not failure.
- Breaking down your own task: which steps, where does the time go, what can someone else take over
- Building your own instruction and giving it a name, so that a colleague can use it tomorrow too
- Reviewing each other's result: the maker is a poor judge of their own work
Two working roundsPlenary stop in betweenColleague reviewsIn hand: three worked-out applications, with what you check yourself for each task
15:30
Break
15:4545 minutes
6. From Monday: agreements and action plan

Each participant explains in one minute which task they will do with AI from Monday and what they will check themselves. The team sets four agreements: what we do, what we do not do, who checks, and to whom you report doubt. The manager reads them out and says whether they stand behind them.
- The operating model in its smallest workable form: four rules, no more
- A personal action plan: three tasks, when, which checkpoint, and what you will look back on in two weeks
- Follow-up measurement, evaluation and a record of attendance listing the topics covered (not a certificate)
PitchesTeam agreementIn hand: one agreement on paper and an action plan per person

Being honest about the limit: on the day itself you build an assistant and a reusable instruction. Handing over a whole series of steps is something you design on paper. That requires involvement from your IT organisation and a test environment, and it does not fit into a training day.

02 — Tool of the day

The instruction card

One card. Twelve lines. A usable result.

A lot of disappointment with AI starts with a vague instruction. You use this card all day, and every week after that. Those who first write down exactly what they want get there sooner.

Role: Who the AI is helping, and in what kind of organisation
Goal: What has to be on the table, concrete and with a form
Context: What it is for, for whom, and what the reader already knows
Input: The text or document the AI may use, anonymised
Approach: In which steps, and ask first if something is unclear
Constraints: Understandable language, add nothing that is not in the input, no names
Form and length: Table, letter or list, and how long
Quality criteria: What you will judge it on later
Check: Mark everything that does not come from the input, and state assumptions separately

03 — Afterwards

What participants can do

Seven things you can see on the day itself.

The goals are worded so that you can establish during the day whether they have been met. Not "has been introduced to", but "can show that".

Explain in your own words what a language model does and why it sometimes makes something up, with one example of your own.
Point out a mistake, assumption or missing source in an answer from the AI, and state how it can be checked.
Write down an instruction with goal, context, input, constraints and form, and refine the result in a few turns until it is usable.
Sort which information is allowed, which only anonymised, and which may never go into the AI environment.
Anonymise a work example of your own so that a colleague can no longer trace it back to a person, and explain why that is still no guarantee that the data is truly anonymous.
Determine for a task of your own whether it calls for help from an assistant, deserves a fixed way of working, or can be handed over as a series of steps.
State which decisions in your own work must be taken by a person, and why.

04 — For management and directors

A half-day session of four hours

This session does not teach you to build. It teaches you to steer.

Managers and directors do not need to learn to operate the AI themselves. They need to be able to take decisions and account for them. So not a practice day but a decision session, for six to twelve people: directors, management, team leaders and the people responsible for privacy, security and procurement.

The programme

Four hours, one break

0:0030 minutes

1. Where we stand now

What is already happening in your organisation, with and without permission? Which licences are in place, and are they being used? Where are the concerns? A conversation, not a presentation.

Conversation and inventoryOutcome: an honest starting picture
0:3030 minutes

2. What AI can and cannot do, in the language of decision-makers

How a language model works, why it sometimes makes something up, and why that is a question of quality and not of technology. Seeing it go convincingly wrong once for yourself does more than three slides.

Explanation and live demonstrationOutcome: three levels of use, each with its own controls
1:0050 minutes
3. The opportunity map

In small groups you map your own processes. Where does it hurt: waiting, searching, retyping, correcting mistakes? You assess time gains and quality gains separately. A task that does not get faster but does get better is often worth more.
- Which form fits: assistance, a fixed way of working, or handing over within limits
- What it requires in terms of data, permissions, IT and training
- You do not automate an unclear process. You simplify it first
Working session in small groupsOutcome: an opportunity map of your own processes
1:50
Break
2:0545 minutes
4. Risks, obligations and limits

You put three examples of your own on the table and walk through them. The frameworks are named as something you take into account and work on. We do not issue a quality mark or a declaration.
- The EU AI Act and what AI literacy means in practice
- The GDPR: legal basis, data processing agreement, and when human intervention is needed in decisions about individuals
- For public organisations: your own security framework and the rules on digital resilience
- Build yourself, buy, rent or leave it, and where your data is held
Risk conversation on your own examplesOutcome: a risk list with a measure and an owner
2:5035 minutes

5. How you set it up: the operating model on one sheet

An AI operating model is not a system you buy, but a set of agreements you make. On one sheet you fill in who does what, where the limit lies, and how you know whether it works. Below you can see the seven parts.

Fill-in sessionOutcome: a first version, which your own IT, lawyers and privacy officer then review
3:2535 minutes

6. The first three applications, and how you bring people along

From the opportunity map you choose three: recognisable to many people, visible results within weeks, limited damage if it goes wrong, and a clear point where a person decides. Then the question on which most plans founder: how do you explain this to the people who have to do it?

Decision roundOutcome: three applications with an owner and a date

The operating model on one sheet

Ownership: Who owns the whole, and who decides what does and does not happen
Roles: Who guards the rules, who builds, who manages the knowledge sources, who checks the quality, who helps colleagues get started
Ground rules: Which environment is approved, which data is and is not allowed, what must always go through a person
Quality: What does "checked" mean for us, and who is responsible for what the AI has produced
Recording: What do we note down when something has been made with AI and goes into a file
Measuring: What do we measure, when, and how do we avoid measuring busyness instead of value
Rhythm: When do we look at this again, and who convenes that

One person can hold several roles. A role is a responsibility, not a new position.

Opportunity map

Your own processes, with time gains and quality gains assessed separately.

Risk list

A measure and an owner for each risk, and a list of what you deliberately do not do.

Operating model

Roles, ground rules, quality requirement, what you record, what you measure and when you look back.

A decision

The first three applications, with an owner and a date on which you check whether it works.

A story for the teams

What you say about jobs, what you ask of managers, and how sceptics are heard.

What you will not have

A judgement that your organisation meets any particular requirement. This session delivers decisions and building blocks, not a declaration.

Management first, then the teams

The frameworks are in place before people start. Suitable if nothing has been arranged yet, or if you are concerned about what is already happening unseen.

Teams first, then management

You decide on the basis of real examples from your own organisation rather than on an assumption. Suitable if AI is already in use.

05 — The common thread

Getting more out of yourself

Participants do not come for AI. They come because their week is full.

That is why the whole day hangs on one question: which part of your week goes on work that could be done faster or better? How much difference that makes varies by task. Some tasks mainly get faster, others mainly get better.

Not working ten times harder. Taking away the friction that eats up your week.

Not everything gets faster. Some tasks even get slower at first, because you now check what you used to do blindly. That is a gain, not a loss.

PreparingGetting a conversation, meeting or file prepared: the essence drawn out, the open questions alongside. You check whether the essence is right.
WritingA first draft in your own tone, based on your own earlier work. Facts, deadlines and the finishing touch remain yours.
SummarisingReducing long documents to what you need, with the actions highlighted.
SearchingFinding answers in documents you provide yourself, with a reference that you check yourself.
AnalysingOrganising raw data into an overview. You check whether the conclusion follows the figures.
SparringOrganising pushback: what is missing from my reasoning, which three objections will come up?
Repetitive workRecording recurring tasks properly once as a way of working, then spot checks.

Seven habits that make the difference

01

First the instruction, then the typing

Those who first write down exactly what they want get there sooner.

02

The rule of three turns

The first answer is never the final answer. Refine twice, then check.

03

Keep what works

You save an instruction that worked well under a name. Explaining it all over again every time is wasted time.

04

Show the AI your work

Describing how you sound is good. Showing how you sound, with your own earlier documents, is better.

05

Always check, and know what for

Every task comes with one sentence: this is what I will check myself.

06

You remain the sender

Nothing goes out without you having read it. Your name underneath means you have seen it.

The seventh: share what works with your team. A good way of working that only you know is half a way of working.

06 — After the training

Guidance

What you learn in one day fades quickly if nothing follows.

What happens afterwards determines whether it lasts. Five forms of guidance, which can be chosen separately and combined. The aim is always the same: that your organisation can do it itself.

01

Follow-up half-day after four to six weeks

Everyone brings what they have made and what went wrong. The latter is the most valuable: these are the problems that only arise in real use.

02

Setting up in your own environment

From separately saved instructions to one shared collection of ways of working, each with a name, an owner and a date on which someone looks at it.

03

Coaching a core team

Two to four people who become the point of contact. They learn to help colleagues get started and to judge whether something is good enough. After that you can continue without external help.

04

Embedding the operating model

Roles with names attached, ground rules that align with your existing policy, and the building blocks your privacy officer and information security officer need for their own assessment.

05

Short conversations in the first months

Half an hour once a fortnight, remotely. What has been made, what is getting stuck, and is the plan still right? You would rather know that after six weeks than after six months.

What we do not do

Build and then leave

The aim is that your organisation can do it itself. If you still want custom software or a standard package afterwards, that is possible, but it is not required.

07 — Practical

Preparation

What you bring, and what you arrange in advance.

The day works best with one team whose members know each other's work, with the manager present. And above all with the sceptical colleague too: they ask the questions the rest benefit from.

Participants: 10 to 15 people from one team or department, mixed roles. No prior knowledge needed.
Language: We deliver the training and the management session in Dutch or in English.
Your own work: Three tasks that take a lot of time or give little satisfaction, anonymised. How to do that is explained in the preparation email and practised in the morning.
Laptop: One that lets you work in your organisation's approved AI environment, for example ChatGPT, Copilot or Claude.
Your own example: Preferably a letter, report or piece of reporting you are happy with. That is the best material for getting the AI to adopt your way of writing.

AI environment: Approved and working for all participants, with the right permissions. In our experience, this is where training days lose the most time.
Who attends: The manager, and if possible the data protection officer or privacy officer.
Room: At your premises, with a screen or projector and working wifi.
Intake: A short conversation beforehand: which team, which environment, which guidelines of your own already exist, and what has to be demonstrable afterwards.

Curious what this day would look like for your team?

In a half-hour intake we tailor the examples, the emphasis in the morning and the practice rounds to your own work.

Schedule an intake

Get more out of yourself. And out of your week.