Skip to content
A working meeting in a bright office in front of a switched-on screen, where one participant has paused mid-task and turned towards the colleague standing at the screen.

AI agents

AI agent training: the person who orders one does not need to be able to build it

We train people in what decides whether a tool that carries out steps on someone's behalf can be trusted at work: what it may touch, what it may not touch, when it has to stop and how it is halted. As an open course with dates, or as a session of your own.

What is usually unclear

The word agent is currently attached to five different things

For a public authority or a company where someone has already started talking about AI agents, training in how to order and halt them is the right thing, because the risk sits in what the tool is allowed to do on its own and not in how it was built.

Vendors now call almost everything an agent. A chat function with a search box is called an agent. A script that moves rows between two systems is called an agent. And then there are tools that genuinely choose their own steps and carry them out. The difference is not visible in the name.

That difference is not academic. A tool that answers can be wrong. A tool that acts can be wrong and then do something about it, in your case management system or in your inbox.

Which of them you have in front of you is something we find out before we arrive. You do not need to have the answer ready when you get in touch.

What participants can do afterwards

Four things, and none of them requires you to be able to program

The session is built around what an employee or a manager actually needs to be able to do once a tool starts carrying out steps on someone else's behalf.

Tell an answer from an action

Participants see whether the tool merely answers or whether it carries out steps for someone, and recognise when something is called an agent without being one. That is the difference that matters when things go wrong, and it rarely shows in the product name.

Write an order

What the tool may touch, what it may not touch and when it has to stop. Participants write such an order with their own task as the starting point, and have it read back to them.

See what it did

Where the record sits, how it is read afterwards and what is missing from it. Without that part, the question of what actually happened cannot be answered.

Halt it mid-run

How the work is stopped while it is running and who is allowed to do it. That part sits on the full day on site, because it needs to be tried rather than described.

None of the points is about building an agent. Building is another profession. Ordering, bounding and halting is supervision, and that is where most organisations stand right now.

What usually surprises people

A tool that is wrong is wrong faster than anyone can look

An employee who has misunderstood an instruction does one thing wrong. A tool that has misunderstood the same instruction does it forty times before anyone reads the email.

That is not an argument against agents. It is the reason the order and the stop condition are written before the tool is let loose on anything that matters, and the reason someone is appointed to own it.

We go through it with your own examples. An invented case teaches no one what is at stake in your organisation.

Whether an agent should be put into operation, who answers for what it does and what the EU AI Act requires of your organisation are legal and organisational assessments. We do not make them. Those questions belong with Ampliro.

Where the session sits today

It already sits in four courses, and looks different in each

We do not sell a separate agent course. The session sits in the courses where it belongs, because an agent is rarely a subject of its own but something that turns up in the middle of the work.

AI in the workplace

Module 7 covers ordering, bounding and halting an agent. On the full day on site, in twenty minutes.

See dates and price

AI for managers

Twenty minutes on the online course and an hour on the full day on site. The full day also covers where you see what the agent did and how it is halted mid-run.

See dates and price

AI for HR

On the full day on site, in twenty minutes. You write an order for a tool that carries out steps on your behalf.

See dates and price

AI Strategy & Implementation

Module 5. The weight here is on what can be seen afterwards and who answers for what the agent does.

See dates and price

If a whole staff group needs the same picture, a session inside an open course is rarely enough. Then we build the occasion around the tools you actually have.

Two ways to take it

An open course with dates, or a session of your own

Which one suits you depends on how many you are and whether the exercises should be built on your own tools.

Open course

For one or a few individuals. The agent session is part of the courses above, held on site in Stockholm and Gothenburg and online. Dates and prices are on each course page.

A session of your own

For a whole department or the whole staff. We come to you, and the exercises are built on the tools you already have rather than on a generic example.

Before you order

What you need to know, and what you are spared

Participants need neither install anything nor create accounts. The session can be run without letting any agent into your systems, and the first time round that is usually how it should be.

If you already have a tool that someone calls an agent, say which one when you get in touch. The exercise is then built against that tool rather than against an example.

Every participant who has completed the course, or worked on their own during an assignment, gets a record of completion, issued digitally and possible to verify afterwards. For an assignment you also get a delivery log saying what was done, who took part and when.

For an HR department that has to document what the staff have been through, a record only the recipient can show is not enough. It needs to be verifiable by someone outside the organisation.

Common questions about AI agent training

AI agent training teaches participants to handle tools that carry out steps on their behalf rather than simply answering questions. The content covers how to tell an answer from an action, how to write an order saying what the tool may touch, what it may not touch and when it has to stop, where the record of what was done sits, and how the work is halted mid-run. At AIUC the session sits in four courses and can also be run as a session of your own.

No. No part of the session is about building an agent. It is about ordering one, bounding what it may do, reading the record afterwards and being able to stop it. That is supervision and it requires no technical background.

Possibly. The question that decides it is whether the tool merely answers or whether it carries out steps for someone in your systems. Many products sold as agents do the first. The difference rarely shows in the name, and the key is what the tool has permission to do without anyone approving each step. Tell us which tool it is when you get in touch and we will say what we see.

No, and that is a deliberate choice. AIUC trains staff to use and govern AI at work. If something is to be built, put into operation and maintained at your end, that is a different kind of assignment, and the question belongs with Ampliro.

Yes, and that is where the question comes up most often. A public authority that has to be able to say what happened in a case needs to know what the tool did, when and on whose instruction. The exercises are taken from the administration's own working week rather than from a generic office example. We have trained staff at government agencies, in regions and in municipalities.

As a session inside an open course it is twenty minutes, and an hour on the full day on site in AI for managers. As a session of your own the length is shaped by what you want to achieve. An hour is enough for a group to share the same picture of what an agent tool is and is not. Half a day is needed for participants to write an order of their own and have it read back.

The open courses have a fixed price per seat, shown on each course page, and the agent session is included in it. A session of your own is priced per assignment, and what governs it is the number of participants, the format chosen, the length, whether we travel to you and how much preparation is included. Tell us what it looks like with you and you will get a figure in the conversation.

No, and nobody can issue one today. A certification is issued by an independent body against an established standard, and no harmonised standards under the EU AI Act have been published yet. Every participant who has completed the course, or worked on their own during an assignment, gets a record showing what they went through and when, which can be verified afterwards.

Both. The session exists in an online format, and a session of your own can be run online when staff sit in several locations. We also run the training in English for international groups. On site you gain the exercise where the work is halted mid-run, which needs to be tried for it to stick.

Yes. AIUC submits tenders in public procurements and delivers on call-offs. Tell us how you buy and we will say what is required from our side.

Start with a conversation

Tell us who is to be trained, what they work with and whether anything in your organisation is already called an agent. We will say what we think is reasonable to do, and what it costs.