Tracks for every role from senior owner to front-line enquiry staff, modules on the duties public bodies carry, a practicum on real casework, sign-off on evidence that would survive an appeal, and renewal triggered by policy changes as well as the calendar. Above all, it should teach caseworkers to give reasons a citizen could read.
What tracks should it have?
| Role | What they learn | Ends with |
|---|---|---|
| Senior responsible owner | Depth, verdicts, what the agent may never decide, impact assessments | Signed depth decisions with reasons |
| Legal and policy leads | How the agent's limits encode policy; where automated decisions are restricted | Review of the agent's limits against the law and policy |
| Caseworkers and decision makers | Reviewer path: held drafts, provenance, reasons, red lines | Certification after a practicum on live cases |
| Team leaders | Calibration across offices, cover, drift | Certification and a calibration run |
| Front-line enquiry staff | Vocabulary, handoff, what the agent tells the public | A short test on their own service |
| Digital and service design staff | Builder path | An agent certified on the agency's own cases |
| Internal audit and oversight | Tracing a decision from the record | A sample traced end to end |
What public-sector modules should it include?
Duties of public bodies. Public bodies face additional transparency and fairness duties, and under the EU AI Act some public uses of AI are high-risk. Certain deployers must carry out a fundamental rights impact assessment under Article 27 before using a high-risk system. Staff should understand what those duties ask of them day to day, not only at launch.
Automated decisions and data. Data protection law restricts decisions based solely on automated processing that have legal or similarly significant effects, and often requires a data protection impact assessment before high-risk processing. Caseworkers must know that their role in an entitlement decision is a real decision, not a signature on the agent's work.
Reasons. Every approval, refusal or request for evidence should carry a reason in plain words. The module should practise writing them from the agent's draft and the record: which fact decided the outcome, and what class of evidence it was.
Evidence by class. An applicant's statement is declared; the agency's own register is a sealed record; a letter from a verified employer is signed; an unverified attachment is external. Eligibility should rest on the strongest class available.
Citizens in difficulty. Signs of vulnerability, accessibility needs and the rule that the agent holds those cases for a person.
How should practice on real decisions work?
A practicum on live casework during the agent's shadow period: caseworkers decide real held drafts, a senior caseworker checks each, and a missed red line, such as an entitlement decided on declared information alone, means repeating it. Calibration then runs across offices and regions on the same cases, because citizens should get the same outcome wherever their case is handled. Oversight staff should compare outcomes across applicant groups as part of each calibration cycle.
Who signs off?
The senior responsible owner, with the legal lead, per capability and depth. The evidence should include the agent's certification on the agency's own cases, caseworkers' practicum results, calibration records across offices, and a traced sample showing each decision could be defended on appeal.
What should trigger renewal?
Three things. The fixed cycle of certificates. Changes to the agent's sources, limits or depth. And changes to the policy or law the agent applies: when an eligibility rule changes, every caseworker on that queue needs a short update before the agent's limits change, so that both move together. Secondments and rotations should trigger a check before approval rights return.
An example
Take a hypothetical agency administering business grants. The senior owner and legal lead complete their track and the impact assessment first. Thirty caseworkers certify during a four-week shadow in which the agent drafts eligibility decisions. Team leaders run calibration across three regional offices; one office is approving more applications on declared turnover and retrains. Oversight traces 40 drafts. Front-line staff learn the handoff. When the grant criteria change six months later, caseworkers take a one-hour update before the agent's limits are amended. Planning detail is in the AI agent training plan template for government agencies, and the budget in what AI agent training costs government agencies.
Where the Academy fits
Pryme Intelligence Academy issues certificates through VerifyDoc.ai, checkable at source by anyone, and a public register lists certified people by name, credential and expiry. For bodies that must show competence openly, that register is a direct answer.
Questions people ask
What order should the tracks run in?
Senior owner and legal first, then caseworkers, as set out in how government agencies should train staff before deploying AI agents. A readiness check, as in how to run an AI agent readiness assessment in a government agency, shows where to begin.
Should elected or appointed board members be trained?
Briefed, not certified, unless they approve agent actions, which they should not.
What do agencies most often leave out?
Reasons and policy-linked renewal. More in what government agencies get wrong about AI agent training and the guide to training teams for AI agents.