Essential
Required for the site to work. Always on.
Enablement and change
You don't adopt AI by buying a licence
Tools are bought in an afternoon. The judgment to use them is built, and it is built per role: the person who signs doesn't need what the person who specifies needs, and neither needs what the person answering an auditor does.
- We train
- By role, not by template
- Where
- Inside real work
- What remains
- Evidence, not attendance
The gap isn't knowledge. It's judgment
Most teams have already used an assistant. What's missing is rarely how to write a prompt: it's knowing when the output is good enough, when to stop it, what must never be sent to a model, and who answers if it goes wrong. That doesn't transfer in a two-hour session, and an attendance certificate doesn't prove it.
A responsible-use policy sitting in a shared document isn't training. It's the intention to provide it.
Three ways, and almost always all three
Standalone training is forgotten and training without context is never applied. So the usual answer isn't picking one, but combining them around a case the team actually has in front of them.
-
Training programme
The catalogue itinerary, for teams that need common ground before touching anything. It is the closest thing to a course, and it is what delivers least on its own.
- Common vocabulary and shared judgment
- Adapted to the sector and the client's systems
- A way in, not a destination
-
Inside the rollout
The part that actually moves the needle. While something is being installed, the client's team works alongside ours on their own cases, their own data and their own constraints.
- Learning on the real case, not on an example
- Documentation and handover as part of the deliverable
- The explicit goal is that you can carry on without us
-
Sessions and workshops
Short format to unblock one specific group: a committee that has to decide, a legal team facing a new obligation, an area that doesn't know where to start.
- Half a day or a full day
- About a concrete decision, not about «AI» in general
- Also open format, at our own events
What each one has to be able to do
This table is the skeleton of the plan. It isn't a syllabus: it is what each role must be able to do by the end, and what proves it. The concrete content comes from your systems and your cases.
| Role | Has to be able to | Proven by |
|---|---|---|
| Executive committee | Ask what needs asking: which AI systems we use, which risk category they fall into, who can stop them and on what evidence. | Minutes where those questions are on the record, with the name of who answers each one. |
| Legal and compliance | Turn a regulatory obligation into a requirement an engineering team can implement, and recognise when a technical answer doesn't cover the obligation. | The system inventory mapped to its risk category, and vendor contracts reviewed against it. |
| People | Design and evidence AI training for staff who use or oversee it, proportional to their role. | A record with date, content and attendees. Not a policy: the record that it was delivered. |
| Product and business | Specify a case precisely enough to be built and evaluated, and decide when an output is good enough. | Specifications with acceptance criteria and a metric agreed before starting. |
| Engineering | Work with coding agents without lowering the bar: specify, review someone else's output, and catch the subtle failure the model misses. | Reviews where real defects get caught, and tests shipped alongside what was generated. |
| Operations and support | Live with a system that is right almost always: recognise the rare case, escalate it, and refuse a plausible answer as good enough. | Escalation routes actually used, and cases returned with a reason. |
What we use is what we publish
There is no briefcase of sample material. What we put in front of a team is the same material that is published and signed, and that you can read right now without talking to anyone.
- Briefings and whitepapers
Ten published documents
Sector reports and executive briefings, each with its author and its sources. They are the pre-reading behind almost any session.
- Self-assessment
AI readiness check
Six questions and a radar by area. Used to open a conversation with a committee: it shows where the imbalance is before anyone argues about it.
- Compliance
AI Act checklist
A step guide with the official sources, to work from the binding text rather than third-party summaries.
- Vocabulary
Glossary
So a room stops arguing about words. More useful than it sounds in the first sessions.
- Method
Enterprise AI-SDLC
How software is built and maintained with agents without losing control. The reference material for engineering training.
- Direct query
MCP server
Our public knowledge base connected to Claude, ChatGPT or Cursor, no API key, read-only. A team can ask it instead of searching the web.
All of the above is public and free. If any of it solves your problem without hiring us, better.
Four steps, and the first one is looking
A training plan written before knowing where the imbalance sits is a course catalogue. That's why the first step isn't training.
-
Where you are
What is used today — including what is used without permission —, who decides, and which obligations already apply to you. The output is a map, not a note.
-
What's missing, per role
On that map, what each role must be able to do and the real distance to get there. This is where you decide what doesn't need training at all.
-
Train on real work
Sessions are built on a live case of yours. By the end there is something that works, not an exercise.
-
Leave the evidence
A record of what was delivered, to whom and when, and what each team can now do. It is what you will be asked for, and the first thing missing when they ask.
Attendance is not a metric
Counting people trained measures what you spent, not what changed. What gets looked at is whether the work is done differently, and whether there is something to show when an outsider asks.
- Work that wasn't done, or was done by hand, and now isn't
- Decisions stopped in time because someone recognised the rare case
- A training record that survives an auditor's question
- Teams that stop depending on us for what they already know
Let's start with where you are
Tell us which teams will use it and what you already have running. The map comes out of that, and the plan out of the map.
Talk to the team