Brief a board on AI agent training with a two-page paper that answers three questions: who is trained, to what level, and what that training allows them to approve. Tie each level to the depth of the agents those people oversee, show which certificates lapse soon, and end with the open gaps and the decision you need from the board.
What does a board actually need to know?
Boards do not need to know what a module covers. They need to know whether the people approving agent actions are competent to do it, and whether the organisation can show that to a supervisor, an auditor or a court. The board's broader questions about agents are listed in what a board should ask about AI agents; training is the part that answers "who is watching, and are they able to?"
That means the paper is about authority, not learning. An agent at depth 2 drafts and a person sends; an agent at depth 4 moves money or takes actions that are hard to undo. The people who approve held actions at each depth carry the real control. The board wants to know that each of those people is trained to the level the depth demands, that their certificates are current, and that someone acts when they are not. How the whole training effort is organised is set out in training teams for AI agents.
How do you prepare the paper, step by step?
- Person (paper owner, usually the COO or chief risk officer): agree the question with the company secretary. Is this an update, or is the board being asked to approve something, such as a policy that approval rights depend on a current certificate?
- Agent, depth 1: reads the training admin data and the agent register, listing people by role, training level, certificate expiry, and the agents and depths they oversee.
- Agent, depth 2: drafts the summary table and the gaps list for the paper owner.
- Person (paper owner): checks every figure against source and writes the narrative: what is good, what is not, what is being done.
- Person (risk or internal audit): reviews whether the paper fairly represents the gaps.
- Person (paper owner): presents and records the board's decision.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Frame the question | Nothing | none | Paper owner, company secretary |
| Gather | Reads training data and agent register | 1 | Choosing what matters |
| Draft table | Drafts summary and gaps | 2 | Checking every figure |
| Narrative | Nothing | none | Paper owner writes it |
| Challenge | Nothing | none | Risk or internal audit |
What goes in the paper?
Page one: the summary. One paragraph on what the board is being asked to do. Then a table like this:
| Agent and depth | Who approves its held actions | Training level required | Current and certified | Lapsing in next quarter |
|---|---|---|---|---|
| Supplier payments, depth 4 | Payments team leads | Operator, plus Architect for the owner | All | Two |
| Claims drafting, depth 2 | Claims handlers | Operator | Most | Several |
| Board pack assistant, depth 1 | Company secretariat | Foundations | All | None |
Use the organisation's own figures. Keep the table to the agents that matter most; the full list can go in an appendix.
Page two: gaps and actions. Who is approving without a current certificate, why, and by when it will be fixed. Any agent whose owner is untrained. Any agent that dropped to depth 1 after a run that cleared no depth, and whether reviewer competence played a part. And the decision you need.
The paper should be written as plainly as an agent's brief. The same discipline that goes into writing a brief for an AI agent applies: say what it is for, what sources it uses and what it does not cover.
What controls should the paper evidence?
- Every approver is named and certified. Evidence: the admin view of certificates by person, with expiry.
- Certificates can be checked independently. Evidence: a certificate number or QR code that verifies at source.
- Lapses are acted on. Evidence: the list of lapsed certificates and what happened to those people's approval rights.
- Agent owners are trained to build and certify. Evidence: owner certificates.
What questions will the board ask?
Expect these. What happens when someone's certificate lapses: do they lose approval rights, and who checks? How do we know the training worked, beyond pass marks? What do supervisors expect? (In the EU, the AI Act's AI literacy duty requires providers and deployers to take measures to ensure a sufficient level of AI literacy among their staff.) Who is accountable if an untrained person approves a bad action? And how does this compare with the training we give for other controls, such as signing authorities?
Have one-sentence answers ready. Internal audit will ask similar questions in more depth; briefing internal audit on how AI agents are governed covers that conversation.
What should the board see over time?
A short, repeated view is more useful than a one-off deep dive. Each quarter: the share of approvers with current certificates, certificates lapsing next quarter, gaps open longer than a quarter, and any incident where reviewer competence was a factor. Trends tell a board more than a single snapshot.
Where does it usually go wrong?
The paper reports course completions, not authority. A board told that most staff completed training learns nothing about whether the people approving depth 4 actions are certified.
Figures are not checked. An agent-drafted table goes to the board with a stale export. Every figure must be checked by a person.
The paper asks for nothing, so the board nods and moves on. If gaps exist, ask for a decision: budget, a policy, a deadline.
Where the Academy fits
Pryme Intelligence Academy's Foundations track, about four hours and free, is written for owners and executives, including board members who want the vocabulary of depth, the decision chain and certification before reading such papers. Certificates for operators, builders and architects are issued through VerifyDoc.ai and can be checked at source by number or QR code, and a public register lists certified people by name, credential and expiry.
Questions people ask
Should board members themselves be trained?
At least to a foundational level, so they can challenge the paper. They do not need operator skills.
How often should the board hear about training?
Quarterly is common in organisations where agents act on customers or money, alongside other control reporting.
Can the agent write the whole board paper?
It can gather and draft. The narrative, the figures' accuracy and the recommendation belong to the paper owner.