Knowing about AI means you can explain how it works. Being trusted with it means someone has watched you make real decisions about its output, to a standard, and would withdraw that trust if you stopped meeting it. Most organisations train for the first and grant the second anyway. That gap is where agent deployments go wrong.
What does "knowing about AI" actually cover?
Knowledge is the layer most training programmes deliver. People learn what a model is, why it can produce confident nonsense, what an agent does differently from a chat assistant, and why governance matters. They learn the vocabulary: depth, verdicts, provenance, the decision chain. Good programmes test this, and people who pass can talk sensibly about AI in meetings.
That is valuable. Everyone in a business that uses agents should have it, and in the EU, Article 4 of the AI Act expects deployers to take measures for a sufficient level of AI literacy among staff dealing with AI systems. The same expectation runs through sectors that might not think of themselves as AI-heavy, as the piece on AI literacy for agents in hospitality shows.
But knowledge is a property of a person's head. It tells you what they understand when asked. It does not tell you what they do when a held refund is waiting, the queue is long and the customer is angry.
What does trust require that knowledge does not?
Trust, in the sense that matters for approving agent work, is a decision someone else makes about you, based on evidence. That evidence has four parts.
Observed judgement on real work. Not quiz answers, but decisions made in a live workspace, on held items, and scored. Did the person read which bound decided the hold? Did they check the source of the key fact? Did they catch the one item that should have been declined?
A standard. A defined bar, with red lines that cannot be missed at all. Approving a payment to bank details sourced from an email is not a minor error to be averaged out.
Currency. The evidence is recent and on the system as it now runs. Trust earned on an older version of the workspace is weaker.
Revocability. The trust can be withdrawn if the person stops meeting the standard, and everyone can see when it has been.
Notice that this mirrors how governed agents themselves earn authority. An agent's depth is earned on evidence from its own cases, enforced at the moment of action, and it expires. People who approve agents' work should be held to a parallel standard.
Why do organisations blur the two?
Because knowledge is cheaper to deliver and easier to measure. A course has a completion rate. A practicum needs assessors, real cases and time. When a deployment deadline is close, it is tempting to run everyone through a module, mark them complete, and give the reviewers approval rights.
The confusion is built into the language too. "AI training" can mean a general awareness session or role-specific preparation for reviewing a particular agent's work, and the two are very different, as the difference between AI training and AI agent training explains. Similarly, a badge that records knowledge and a certificate that records trusted competence look alike on a profile. The article on badges versus certificates in AI agent training draws the line.
What is the trade-off?
Separating knowledge from trust slows things down. If only people with observed, current certificates can approve held actions, you need enough of them before go-live, and that takes weeks. Some capable people will wait for approval rights they could probably handle.
It can also create a two-tier feel in a team: those who are trusted and those who are not yet. Handled badly, that breeds resentment.
But the alternative is worse. Granting trust on the strength of knowledge means approval rights go to people nobody has watched make the decisions. When something goes wrong, the record shows who approved it, and the organisation has nothing to show that it had reason to trust them. That is unfair to the individual and indefensible for the business.
The two-tier problem has a simple fix: make the path to trust visible, quick and open to everyone who needs it. People resent hidden criteria, not clear ones.
What does this look like in a support operation?
Take a contact centre bringing in an agent to handle refunds. Every adviser takes a short course on how the agent works: knowledge. A smaller group, those who will approve held refunds over the limit, complete an observed practicum: trust. Team leads, who judge other people's approvals, complete a practicum that includes coaching and sampling.
The case for certifying support team leads and the case for certifying contact centre managers both rest on this distinction. Knowledge spreads widely. Trust is granted deliberately, to the people whose decisions shape what the agent does.
What should you do differently?
Write the distinction into policy. State which roles need knowledge only, and which need a current certificate earned on observed work before they approve anything. Tie approval rights to the certificate, not to course completion. Make the route to certification open and well signposted. And when you audit, ask not "who completed the training?" but "who was trusted to approve, and on what evidence?" The guide to AI literacy and certification for people who work with agents sets out how to map the two.
Where the Academy fits
Pryme Intelligence Academy draws the same line. The free Foundations badge records knowledge: why agents must be governed, the decision chain, depth and certification. Certified Pryme Operator records trust: a written exam plus a two-week practicum of real review decisions, signed off by a Pryme assessor, valid for 12 months and checkable at source.
Questions people ask
Can someone be trusted with AI without formal certification?
They can be trusted informally, by a manager who has watched their work. But informal trust leaves no evidence, does not expire and cannot be checked by anyone else, which is a weakness when an auditor or regulator asks.
Is AI literacy training a waste if people still need certification?
No. Literacy is the base everyone needs, and it makes certification faster for those who go on to it. It just should not be mistaken for the authority to approve.
Who decides when someone is trusted?
The business does, using evidence. A certificate from an independent assessor is strong evidence; the business still decides which approval rights it grants on the strength of it.