Evidence of competence is the record that shows a person can perform a task to the required standard: assessment results, observed work, an assessor's sign-off, a current credential. It proves ability, not attendance. A list of courses someone completed is evidence of training; a record of their correct decisions on real cases is evidence of competence.
Why does it matter for AI agents?
Supervisors, auditors and clients ask a simple question about the people overseeing agents: how do you know they can do it? Under the EU AI Act, deployers of high-risk systems must assign human oversight to people with the necessary competence, training and authority; how that reads in practice is covered in Article 26 deployer duties and staff competence in banking and in payments. The answer has to be a record, not an assurance.
Evidence also protects the organisation internally. When a held payment is wrongly released, the first question is whether the approver was competent. A record made before the incident answers it; one assembled afterwards does not.
What counts as evidence?
Evidence is strongest when it is specific, recent, independent and checkable. In rough order of strength:
- Assessed real work: a sample of the person's actual review decisions, judged by an assessor against a standard.
- A current certificate, checkable at source, that required assessed work to earn.
- A scenario test result set in the person's own industry and role.
- A knowledge exam result with a known pass mark.
- Course completion records and attestations, which show training, not competence.
Most organisations need a mix. A good evidence pack names the person, the role, the standard, the evidence, its date and who judged it. The training evidence pack template gives a structure.
What is an example?
An auditor reviewing a lender's use of an agent to draft affordability assessments asks for evidence that the credit officers approving them are competent. The lender provides, for each officer, a certificate number checkable at source, the result of a written exam, and an assessor's sign-off on two weeks of real approvals with error counts. The auditor samples three officers, checks the numbers and moves on. A list of course completions alone would have prompted further questions, as how to prepare training evidence for an AI audit explains.
How is it different from a training record?
A training record shows what training someone received: course, date, version, result. Evidence of competence shows what they can do. The two overlap when a training record includes an assessment of real work, but a record of attendance alone proves nothing about ability. The gap between them is often measured as time to competence: how long after training a person reliably meets the standard.
Which terms are related?
- Work-based assessment: the strongest source of evidence.
- Assessor: the person whose judgement makes it independent.
- Verifiable credential: evidence a third party can check.
For the full topic, read AI literacy and certification for people who work with agents.
Where the Academy fits
Certified Pryme Operator produces evidence on both counts: a written exam of 40 questions in 60 minutes with a pass mark of 32 correct, and a two-week practicum of real review decisions signed off by a Pryme assessor. The certificate is checkable at source through VerifyDoc.ai, and the team admin view shows results and expiry dates.
Questions people ask
How old can evidence be?
Only as old as the agent and the role it describes. If the agent has gained capabilities since the person was assessed, the evidence covers less than it did. Credential expiry sets a hard limit.
Should evidence be kept per person or per team?
Per person. Teams change, and an auditor will sample individuals. Roll it up into a team view, but keep the underlying record named.