Every AI literacy programme needs the same core: what an agent can do, how far it may act, where its facts come from, how its decisions are recorded, and when to stop it. Role and risk change the depth. A buyer needs the concepts; a reviewer applies them; a builder and an owner set them.
What is the core everyone should learn?
Seven topics belong in every version of the programme, from the board to the front line:
- What an agent is. Software that takes actions, not just a chat tool that answers. The difference changes what can go wrong.
- Depth. How far an agent may act on each capability: reading, drafting for a person to send, executing routine work within bounds, moving money or taking hard-to-undo actions, and coordinating other agents.
- The decision chain. Every action meets several bounds at once, such as the agent's role scope, its certification, its sources, the business's own limits and the platform's guardrails, and the lowest one wins.
- Verdicts. Allowed, held for a named person, refused as outside the agent's bounds, or withheld at the platform floor for every agent.
- Provenance. Where each fact came from, and why an emailed claim cannot carry the same weight as an entry in the system of record.
- The record. Every decision is filed so it can be reconstructed, and people's approvals are part of it.
- Escalation and incidents. How to stop or report an agent that is behaving oddly, including suspected manipulation through its inputs.
Prompt literacy, knowing how to instruct a model well, is useful but secondary. Most staff never write an agent's brief; they review what it does.
How should the programme change by role?
| Role group | Depth of coverage | What they must be able to do at the end |
|---|---|---|
| Board and executives | Concepts, risk and accountability | Ask the right questions about depth, certification and incidents |
| Everyone working near agents | Core seven topics | Recognise verdicts, escalate, explain to a customer that a person decides |
| Reviewers and approvers | Core plus applied review | Read the record, find the deciding bound, approve, edit or decline with a reason |
| Builders | Core plus design and certification | Write a brief, choose sources, set bounds, get an agent certified |
| Owners and architects | Core plus governance | Set limits, sign off changes, answer for outcomes across agents |
How should the programme change by risk?
Risk sets the assessment, not just the content. Where agents only read and draft, a knowledge test may be enough. Where agents execute or move money, reviewers need a practical assessment on real decisions and a credential that expires. Risk literacy is itself a topic: staff should understand which actions are reversible and which are not, because that is what decides whether something is held.
Regulated sectors add a layer on the rules that apply to them. In the EU, Article 4 of the AI Act asks providers and deployers to ensure a sufficient level of AI literacy among staff and others dealing with AI systems on their behalf, taking into account their knowledge, experience and the context of use. How to set up an AI literacy programme under the EU AI Act covers that route.
An example: one company, three tiers
Take a mid-sized insurer introducing agents in claims. All 400 staff take a short foundation course on the core topics. The 40 claims handlers who will approve held settlements take a reviewer course with a practical assessment on claims cases. The six people who will build and own the agents take a builder course ending in a certified agent. The policy records which tier each role needs, using the fields in the AI literacy policy template.
What each certificate covers is set out in what an AI agent certificate covers, and the overall structure in the guide to AI literacy and certification for people who work with agents.
Where the Academy fits
Pryme Intelligence Academy maps onto these tiers. Foundations, about four hours, covers the core: why agents must be governed, the decision chain, depth and certification, the workspace, cost, ownership and data. Operator is for reviewers, Builder for those who design and certify agents, Architect for people leading regulated, multi-agent deployments. Lessons, labs and tests are free.
Questions people ask
How long should foundation literacy take?
A few hours is enough for the core concepts if the material is focused on agents that act. Longer courses often drift into general AI history that does not change behaviour at work.
Should the programme include hands-on practice?
Yes, for anyone who will review or build. Reading a real record and deciding is the skill; practice in a safe workspace is the fastest way to learn it.
How do we keep the programme current?
Tie updates to product and agent changes, version the material, and require short update modules when renewing credentials.