How do AI agents verify work history?

On their own, they mostly cannot. An AI agent can find a CV, a LinkedIn profile, a portfolio and a personal site, and it can summarise what they say — but it has no way to tell which claims are true, which are out of date, and which have ever been checked by anyone. Verification has to come from somewhere outside the agent: evidence attached to the claim, confirmation from people who were there, or a credential issued by an institution.

Why can’t the model just check?

Because a language model evaluates plausibility, not truth. A well-written claim about leading a project reads exactly like a true one, and the model has no independent record to compare it against. What it can do is notice inconsistency — two sources disagreeing about a date — which is a signal about the sources, not about the fact.

Searching the web does not solve it either. Most professional claims leave no public trace: internal work, private projects, contract engagements and anything under an NDA are invisible to a crawler no matter how capable it is.

What actually counts as verification?

Evidence tied to the specific claim — the work itself, or a document produced by it — rather than a general reference. A claim about shipping a design system means more when the system is there to look at.

Confirmation from people who were present. Someone who worked on the thing can confirm it happened, which is different in kind from a recommendation that says you are a pleasure to work with.

Credentials from an issuer. A degree, a certification or a licence is verifiable because an institution stands behind it and can be asked.

What does this change for the person being assessed?

It moves the advantage from the people who write well to the people who can show their work. When a claim carries evidence, it is worth more than a better-phrased claim that carries none — which is the opposite of how a CV pile currently sorts.