Build it in tiers, not in one course. Everyone needs a short grounding in what agents may do. People who use agents need training on the ones they use. People who approve agent actions need tested competence and a certificate. People who design or own agents need the deepest training. Only the top two tiers need certification.
Why one course for everyone fails
A single organisation-wide AI course is easy to roll out and easy to report. It is also too much for most people and too little for the few who matter most. The finance assistant who occasionally reads an agent's summary does not need to know how certification gates work. The payments approver who releases agent-prepared transfers needs far more than an awareness module.
Under the EU AI Act, Article 4 asks providers and deployers to ensure a sufficient level of AI literacy, taking account of each person's role and the context of use. That is a tiered requirement in all but name.
What are the four tiers?
| Tier | Who | What they need | How it is evidenced |
|---|---|---|---|
| 1. Everyone | All staff, including contractors | What agents are, how they are governed, what to do if something looks wrong | Completion of a short module |
| 2. Users | People who work alongside agents or read their output | The specific agents they use: what they do, their depth, what they refuse | Role training records |
| 3. Approvers | People who approve held actions | Reading the decision chain, provenance, held versus refused versus withheld, practised judgement | A certificate with a practicum, checked at source |
| 4. Owners | People who design, train, certify or are accountable for agents | How agents are built, certified, monitored and retired | An advanced certificate |
Each tier has a different kind of literacy at its core. Tier 1 is close to governance literacy: knowing that agents act within bounds and who sets them. Tier 3 depends heavily on decision literacy: understanding how a decision was reached and whether to accept it.
How do you roll it out?
- Map roles to tiers. Start with who approves agent actions today and who will soon. Those are tier 3. Everyone else follows from there. A skills matrix for an AI agent team helps.
- Start tier 1 for everyone at once. It is short and creates a shared vocabulary: depth, held, refused, withheld, the record.
- Certify approvers before agents act at depth 3 or above. Their competence is the control.
- Train owners alongside the first deployment. The people building and certifying agents need the deepest grounding first.
- Tie rights to tiers. A business can make approval rights depend on a current certificate, depth by depth.
- Keep records. The AI literacy training record template captures who reached which tier and when.
What does this look like in a mid-sized firm?
Take an insurer with 1,500 staff deploying agents in claims and customer service. All 1,500 complete a short tier 1 module. About 400 in claims and service complete tier 2 training on the agents they work with. Around 80 claims handlers and team leads who approve agent-prepared settlements are certified at tier 3, each after supervised practice. A dozen people in claims operations, risk and the technology team who design and certify agents complete tier 4.
The board and executive team take tier 1 plus a briefing on how the organisation governs agents. When the regulator asks how staff are trained, the insurer can show each tier, who is in it and why.
What keeps literacy current?
- Refresh on change. When an agent's depth, sources or limits change, brief the tier 2 and 3 people affected.
- Renew certificates before they lapse.
- Use the record. Overrides and near misses show where training is not landing.
- Revisit the tier map when new agents or roles arrive.
The guide to AI literacy and certification explains how literacy and certification differ and connect. For the wider governance view, see how to govern AI agents across a whole company.
Where the Academy fits
Pryme Intelligence Academy's four tracks follow a similar shape. Foundations, about four hours and free, is for anyone. Operator is for the people who run a live workspace every day. Builder is for those who design, train and certify agents. Architect is for those leading regulated, multi-agent deployments. Each track is the key to the next.
Questions people ask
Should executives be certified?
Most need tier 1 and a governance briefing. Executives who approve agent actions or own agent deployments should go further.
How long does tier 1 take?
Keep it short enough that everyone completes it, typically a few hours at most. Its job is shared vocabulary, not expertise.
Do contractors need to be included?
Yes, if they work with agents on your behalf. Article 4 covers others dealing with AI systems for the organisation, not only employees.