Teach a new HR adviser the policy library before the agent. In week one they learn which policies the agent reads, which questions it answers, which letters it drafts and which topics it hands to a person. In week two they review and send its drafts with a senior adviser checking each. Sensitive cases stay with experienced advisers.
What does the HR agent do, and what is left?
In an HR team that runs agents, the agent typically answers employee questions on leave, pay dates, benefits and procedures from the policy library (depth 3 for routine factual answers within the organisation's limits), drafts letters for absence reviews, flexible working requests and probation outcomes for an adviser to send (depth 2), and prepares case summaries. It must hand over grievances, disciplinary matters, health disclosures, harassment and anything outside its bounds.
A new adviser therefore spends less time answering "how many days of leave do I have?" and more on the drafts and hand-overs, which are the cases where judgement and care matter. How the role shifts is set out in what changes in an HR adviser's day when AI agents arrive. The HR manager owns onboarding; a certified senior adviser coaches.
What does each week involve?
- Day 1. Person (HR manager): introduces the policy library the agent reads: which policies, which versions, and who owns each.
- Day 1 to 2. Person (new adviser): learns the agent's bounds: what it answers alone, what it drafts, and the topics that always go to a person.
- Day 2 to 3. Agent, depth 2: drafts letters and replies, which the coach reviews aloud while the new adviser follows.
- Day 4. Person (new adviser): practises on past queries with the actual replies hidden, then compares.
- Day 5. Person (new adviser): reads site and local agreements that vary from head office policy, because these cause many agent errors.
- Week 2. Person (new adviser): reviews and sends live drafts, each checked by the coach before it goes.
- End of week 2. Person (HR manager): signs off routine drafts. Sensitive cases stay with experienced advisers for longer.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Policy questions | Answers routine ones from library | 3 | Anything outside routine |
| Letters | Drafts absence, flexible working, probation | 2 | Sending and the decision |
| Case summaries | Drafts from case notes | 2 | Judgement |
| Sensitive topics | Hands over | none | Grievance, discipline, health |
| Corrections | Nothing; reason filed | 1 | Reason, without names |
What does a new adviser need to learn first?
The library is the agent's whole world. If a policy is not in the library, or an old version is still there, the agent will be wrong with confidence. New advisers should know how to check which version the agent cited, and who to tell when it is out of date.
Local agreements change the answer. A head office overtime or leave rule may not apply at a site with its own agreement. Advisers should check which agreement covers the employee before sending.
Sensitive topics are not drafts. A health disclosure in the middle of a leave question changes the conversation. If the agent drafted a routine reply, the adviser takes it over and records the miss.
Tone is part of accuracy. An absence review letter can be correct and still wrong for someone grieving a close relative.
Confidentiality between cases. An adviser may know about one employee's situation from another case. Neither the agent nor the adviser should carry facts from one case into another employee's reply. Advisers check that drafts use only the facts of the case in hand.
How should corrections be written?
With a reason and without personal data. "Applied head office leave rule to a site with a local agreement" is useful. The employee's name, health or pay is not, and should never be in a correction reason. Good reasons let the agent owner change the agent's sources or brief. The team's reviewer onboarding checklist for the HR team includes a check on this, and the AI agent training plan template for HR advisers has a module on it.
Which controls and evidence should onboarding leave?
- Checked sends in week two. Evidence: both advisers named on each letter.
- Practice comparison. Evidence: past queries reviewed and differences noted.
- Correction reasons free of personal data. Evidence: a sample checked by the HR manager.
- Staged sign-off. Evidence: dated rights for routine work, later for sensitive cases.
What should the HR manager measure?
Drafts changed by the coach and why. Sensitive cases the new adviser spotted that the agent missed. Letters sent with an outdated policy citation. Reasons that needed redaction. After sign-off: complaints or grievances arising from letters the adviser sent.
Where does it usually go wrong?
Treating the agent as the policy expert. New advisers trust its citations and never open the policy itself, so they cannot tell when the agent is wrong.
Personal data in reasons. Advisers used to detailed case notes write the same detail into correction reasons.
Sensitive cases too early. Two weeks is enough for routine work, not for a harassment complaint.
Advisers in other fields follow similar plans; see those for a new wealth adviser and a new student services adviser. The wider approach is in training teams for AI agents.
Where the Academy fits
Pryme Intelligence Academy's Operator track teaches how a governed agent decides, human review and keeping knowledge current, which suits advisers whose agent depends on a policy library. Module tests are set in the learner's own industry and role, so an HR adviser practises on HR cases, and lessons can be read or listened to as audio.
Questions people ask
Can the agent decide a flexible working request?
No. It can draft the letter and gather the facts. The decision belongs to the manager and HR, under the organisation's policy and the law.
What if an employee asks whether they are talking to an AI?
Tell them. Employees should know when they interact with an AI system.
Who updates the policy library?
The policy owner, with a named link to the agent owner. New advisers should report outdated policies rather than fix them.