What universities get wrong about AI agent training

Universities usually file AI training under teaching and assessment, so the effort goes to academics and student conduct policy. The professional services staff who actually run agents in admissions, finance, research administration and student support get a committee-designed module, and nobody practises what to do when a student's message to an agent reveals they are in distress.

Why is AI training aimed at the wrong staff?

When universities talk about AI, the conversation is mostly about students using it in coursework and how to assess fairly. That is a real issue, and it absorbs the training budget and the attention of senior academics.

Agents that act on the university's behalf are a different matter. They read applications, draft replies to enquiries, prepare refunds and bursary payments, check research grant expenditure and route student support requests. The people overseeing them are admissions officers, finance staff, research administrators and student services advisers. They need to read a verdict, judge the strength of a fact and refuse with a reason. A session on academic integrity teaches none of that. Split the two topics and give professional services their own programme.

What goes wrong when a committee designs the course?

University governance runs through committees, and training often does too. The result tends to be a broad module that satisfies every stakeholder and prepares nobody in particular: a little on ethics, a little on data protection, a little on the platform, the same for every department.

Reviewers learn by deciding real cases from their own queue. An admissions officer needs held applications; a research finance officer needs held expenditure claims. Let the committee set the standard and the assessment, and let each service own its case material. Agent certificates lapse after 90 days, and a run that clears no depth drops an agent to depth 1 the same day, so each service also needs its own short guidance on what that means for its queue, not a university-wide notice.

Why does research administration get missed?

Admissions and student finance are visible; research administration is not. Yet agents there handle money and compliance. An expenditure agent checks grant claims against the award terms at depth 1 and drafts queries at depth 2. A procurement agent prepares supplier payments from research budgets, held at depth 4.

The evidence varies in strength. The award terms in the university's own grants system are sealed. A supplier invoice emailed by a researcher is external and cannot authorise a payment on its own. A researcher's note that equipment was "essential to the project" is declared. Research finance staff need to know which is which, because funders audit spending and a payment released on a weak basis can cost the university more than the item.

What happens when a student discloses distress to an agent?

Student-facing agents answer questions about deadlines, fees, accommodation and services. Sooner or later a student will reply with something else: they cannot cope, they are unsafe, they are thinking of leaving. The agent should be bounded so that any such message is routed straight to a trained person rather than answered.

The failure is on the human side. The adviser who picks up the routed message may be a temporary member of staff on the enquiries rota, unsure whether it is their job. Train everyone on the rota to recognise a welfare disclosure, to respond promptly through the university's support process, and to record what they did. Rehearse it with anonymised examples. Students should also be told when they are interacting with an AI system; in the EU, Article 50 of the AI Act requires it. This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

Why are fixed-term and peak staff a risk?

Universities lean on fixed-term contracts and seasonal staff, and sometimes student ambassadors, to handle admissions and enrolment peaks. If they are given approval rights over held actions, the least-prepared people make the most consequential decisions in the busiest weeks.

Keep approval rights with certified permanent staff and let temporary staff draft and route. A university can adopt the practice of granting approval rights only to staff whose certificate is current, depth by depth. Lenders face the same issue with seasonal collections staff; what lenders get wrong about AI agent training covers it from their side.

What should a university do instead?

1. Separate agent oversight training from academic integrity, and fund both.
2. Let committees set standards; let services own their case material.
3. Include research administration and procurement from the start.
4. Train everyone on student-facing rotas to handle welfare disclosures.
5. Keep approval rights with certified permanent staff.

The general order is in the guide to training teams for AI agents. For a university's version, see what an AI agent training programme for universities should include, the AI agent training plan template for universities, and how much AI agent training costs for universities.

Where the Academy fits

Pryme Intelligence Academy's Foundations track works for students and academics who need the vocabulary, and it is free. Professional services staff who approve held actions take the Operator track, with module tests set in their own industry and role and a two-week practicum of real review decisions. Seats come with paid Pryme Intelligence plans, and the admin view shows progress, results and expiry dates by department.