A hospital programme should train people by the decisions they will sign, not by job title. It needs four parts: a short common grounding for everyone who touches agent work, deeper tracks for those who review it daily, practice on real held decisions from the hospital's own wards and offices, and a sign-off that expires and must be renewed.
Which tracks does a hospital need, and who goes on each?
Hospitals put agents into administrative and operational work first: discharge letters, clinic correspondence, referral triage, bed board updates, coding suggestions and stock reorders. Clinical judgement stays with clinicians. So the tracks follow who approves what, not seniority.
| Group | What they do with agents | Track |
|---|---|---|
| Board, executives, clinical directors | Set red lines and accept the risk | Common grounding |
| Patient flow coordinators, medical records supervisors, clinical coding leads | Approve or correct held drafts every shift | Daily reviewer track with a practicum |
| Digital and informatics staff | Design, test and certify agents | Builder track |
| Clinical information and information governance leads | Lead deployments that span wards and sites | Architect track |
| Clinicians who receive drafts | Read, correct and sign | Common grounding plus a ward briefing |
The reviewer track is where most hospitals underinvest. A coding lead who approves forty agent suggestions a shift is making forty decisions that end up in the patient record and the hospital's activity data.
What may the agent do, and what stays with staff?
Training should make this split second nature, because reviewers act on it hundreds of times a week.
- Depth 1, read: the agent reads bed boards, theatre lists, referral letters and the notes it is given.
- Depth 2, draft: it drafts discharge summaries, clinic letters and coding suggestions; a named person signs or sends them.
- Depth 3, routine within bounds: it books follow-up slots in set clinic templates or reorders stock below a fixed level.
- Stays with a person: every clinical decision, every change to a medication or allergy list, every safeguarding concern, and every message that tells a patient something about their diagnosis.
Depth 4, moving money or taking hard-to-undo actions, has little place in hospital administration. A good programme says so plainly, so nobody asks for it by accident.
Why must practice use the hospital's own cases?
Generic exercises teach generic judgement. A coding lead needs to see what the agent gets wrong in this hospital's case mix, and a discharge reviewer needs to see the shortcuts this ward's notes take. The practicum should replay recent held drafts from the reviewer's own service, with each decision recorded as approve, correct or reject, and a reason.
Practice should also cover refusals. When the agent declines to draft a letter because a source is missing, that is the system working. Reviewers who treat refusals as faults start filling gaps by hand without recording why.
An example: discharge letters on a surgical ward
Take a surgical ward where an agent drafts discharge letters from the inpatient notes. The medication list comes from the pharmacy system, which is sealed provenance. A letter from the patient's family doctor, scanned in at admission, is external. In one illustrative case the agent copies a drug from that scanned letter that was stopped during the stay.
A trained reviewer checks where each line came from before signing, and catches it. An untrained one reads the letter for sense, finds it fluent and signs. The training difference is one habit: read the provenance before reading the prose.
How should sign-off and renewal work?
Sign-off should come from someone outside the reviewer's own team, against a fixed standard: enough correct decisions over a set period, and no missed red line. It should lapse after a year, because the agents change. Agent certificates lapse after 90 days, and a run that clears no depth drops an agent to depth 1 the same day, so every reviewer must also practise the fallback: what the ward does by hand while the agent can only read.
A hospital can make approval rights depend on a current certificate, depth by depth. The AI agent training plan template for hospitals sets out the schedule, and what hospitals get wrong about AI agent training covers the traps. To scale beyond the first wards, a train-the-trainer programme helps, and the wider method is in training teams for AI agents.
Where the Academy fits
Pryme Intelligence Academy runs the four tracks above as Foundations, Operator, Builder and Architect. Operator ends with a 40-question exam passed at 32 of 40, and a two-week practicum of real review decisions, 95 in every 100 correct with no missed red line, signed off by a Pryme assessor. Lessons can be listened to as audio, which suits shift rotas.
Questions people ask
Should consultants and senior nurses take the full reviewer track?
Only if they approve agent work every day. Most clinicians receive drafts and sign them, which needs the common grounding and a short ward briefing on what to check.
Does health data change what reviewers must learn?
Yes. Health data is special-category data under GDPR, and in the US it falls under HIPAA. Reviewers should learn to see only what a task needs and to report any agent output that exposes more than that.
How long before a hospital team is ready?
The common grounding takes a morning. The reviewer track takes about two working days of study plus two weeks of supervised review, which can run alongside normal duties on the ward where the agent goes live.