PrymeIntelligence

AI literacy and certification·Answer

What records should a company keep of AI agent training?

Keep records of who was trained on AI agents, what they passed, when it lapses and what it lets them approve, joined to the decisions they actually made.

Keep four facts for every person: their identity and role, what they completed and passed, when each credential lapses, and what it allows them to approve. Then join that record to the decision record, so you can show any approval of an agent's held action was made by someone qualified at that moment.

What belongs in the record for each person?

Field Example Why it matters
Name and staff ID A. Mensah, 40217 Ties the record to the decision log
Role and team Payments reviewer, treasury operations Sets which training level is required
Training completed Foundation course, reviewer course Shows coverage
Assessment passed Written exam, practical assessment Separates completion from competence
Credential and number Reviewer certificate, number on file Allows checking at source
Issue and expiry date Issued in spring, expires 12 months later Drives renewals and pauses
Approval rights granted Release held payments at depth 4 up to the customer's limit Links training to authority
Date rights granted and by whom Granted by the head of operations Shows a named person decided
Status checks Date each time the certificate was checked at source Proves the business looked
Changes Rights paused at lapse, restored at renewal Shows the control worked

The last three rows are the ones most companies miss. A spreadsheet of course completions shows that training happened. It does not show that the business used training to decide who may do what, or that it noticed when a credential lapsed.

Why join training records to decision records?

Because that is the question an auditor will ask. Take any held action an agent raised, such as an update to a customer's bank details. The decision record shows the agent's draft, the bound that held it, the approver and the reason. The training record should then show that the approver held a current credential covering that kind of approval on that date.

If you cannot answer that for a random sample of decisions, your training records are a file, not a control. The argument is made in training records are evidence; treat them that way.

An example: a CRM update agent

Take a company running an agent that corrects customer records in its CRM. Routine corrections, such as fixing a postcode format, run at depth 3. Changes to bank details or legal names are held for a person. The process is described in how to let an AI agent update customer records in your CRM.

Six months in, a customer disputes a bank detail change. The investigator pulls the decision record: held by the customer's own limits, approved by a named reviewer with the reason "confirmed by call-back to number on file". The training record shows that reviewer held a current certificate on that date, had approval rights for bank detail changes granted by the operations lead, and their status was last checked two weeks earlier. The investigation moves to the call-back, not to whether the reviewer was allowed to decide.

What about boards and senior managers?

Directors who oversee agent deployments should have their training recorded the same way, with less detail on approvals and more on governance. Boards in logistics and payments have different needs; see what AI agent training a logistics company board needs and what AI agent training a payment company board needs. Their record should show what they covered, when, and how it relates to the risks the board signs off.

How long should you keep it, and where?

Keep training records at least as long as the decision records they support, since a decision may be examined years later and the qualification of its approver will be part of that. Store them where they cannot be quietly edited: changes should be added, not overwritten. Hold the certificate number rather than relying on a stored PDF, and check status at the issuer.

Where people are certified through Pryme Intelligence Academy, the team admin view shows progress, results, certificates and expiry dates, which covers most of the training side. The approval rights and the join to decisions remain your record to keep. The levels these records track are set out in the guide to AI literacy and certification for people who work with agents.

Questions people ask

Do we need records for people who only read agent output?

Yes, at a lighter level. Record that they completed foundation training, because they still act on what the agent shows them and may escalate.

Should failed assessments be recorded?

Record attempts and outcomes for your own management, but keep the formal record focused on credentials held. A failed attempt is not a credential and should not grant anything.

Is a learning platform report enough?

It covers completion and results. It rarely records approval rights or the status checks you made, so you need those alongside it.

Train the people who run your agents.

Pryme Intelligence Academy: free courses on the real product, audio lessons, labs in a free workspace, and certificates anyone can check on VerifyDoc.ai. Teams get seats, an admin view and a free readiness assessment.