A university needs a programme that works across a devolved institution. It should include a shared grounding for staff and, where useful, students; certified reviewer tracks for registry, admissions and student services; practice on the university's own student cases; clear lines around academic judgement and student welfare; and sign-off that lapses and is renewed.
Why is a university hard to train?
Authority is spread out. Faculties and departments run their own processes, professional services run others, and academics guard their judgement closely. An agent that drafts an admissions decision in one faculty may run under different rules from the same agent in another. Training has to work at two levels: one institutional core, and a short local layer for each service or faculty that runs an agent.
Universities also hold many students' futures in their records. A wrong mark, a missed deadline extension or a lost welfare message can change someone's degree. The programme should be built around those decisions.
Who takes which track?
| Group | What they do with agents | Training |
|---|---|---|
| Senior leadership and governing body | Set red lines, accept risk | Common grounding |
| Registry and exams officers | Approve agent drafts of records, transcripts and progression letters | Reviewer track with practicum |
| Admissions teams | Review agent checks of applications and draft offers | Reviewer track with practicum |
| Student services and wellbeing teams | Review triaged messages | Reviewer track with heavy drill on escalation |
| Academic staff | Receive summaries and drafts, keep all academic judgement | Common grounding |
| Digital services and data protection | Build agents, answer for the record | Builder track; architect track for whoever leads cross-faculty deployment |
Students can take the common grounding too. It helps them understand how decisions about them are made and makes them better informed when they graduate into workplaces that run agents.
What may the agent do, and what stays with staff?
- Depth 1 to 2: read applications and records; draft offer letters, progression letters and replies to routine student queries.
- Depth 3: book appointments, chase missing documents, update records for routine changes within set rules.
- Depth 4: refunds of fees and any change to a confirmed mark or award; held for a named person.
- Stays with a person: every academic judgement, every admissions decision that is not purely mechanical, every welfare concern, and every disciplinary step.
An example: a mitigating circumstances claim
Take a university where an agent reads mitigating circumstances claims and drafts a summary for the panel. A student submits a claim with a letter from a doctor and a short personal statement. The doctor's letter is external provenance; the statement is declared. Buried in the statement is a sentence suggesting the student is at risk of harm.
The agent's job was to summarise for the panel. Because the university set any risk-to-self language as a red line, the item is held for student services first. A trained reviewer sees why and refers the student the same day. An untrained one, focused on clearing the panel queue, forwards the summary. Practice built from the university's own anonymised claims teaches reviewers to read for more than the task.
How should sign-off and renewal work?
Sign-off should come from someone outside the reviewer's own office, against a fixed standard with no missed red line, and lapse yearly. Exam boards and admissions cycles are natural points to schedule renewal. Agent certificates lapse after 90 days, and a run that clears no depth drops an agent to depth 1 the same day, so registry teams should rehearse handling results release by hand.
Pryme Intelligence Academy's Foundations track is open to students as well as staff, and its badge is free with no expiry.
Use the AI agent training plan template for universities to schedule, and read what universities get wrong about AI agent training first. If the university has no programme at all, start with how to build an AI agent training programme from scratch. Teaching hospitals linked to a university should also see the programme for hospitals. The general method is in training teams for AI agents.
Questions people ask
Should academics be certified as reviewers?
Only those who approve agent work, such as programme directors releasing progression letters. Most academics need the grounding so they know what the agent did before a draft reached them.
Can the agent mark student work?
It can draft feedback against a rubric at depth 2, but the mark is academic judgement and stays with the marker.
How do we handle data protection for student records?
Treat the agent as another processor of personal data under GDPR or its equivalent, and train reviewers to report any output that exposes more than the task needs.