Agents are not employees, but they change every job they touch. HR needs to redesign roles around review and exceptions, set training that proves people can oversee agents, make clear that accountability stays with named people, and rewrite job descriptions before go-live. The agent’s authority is earned and expires; so should the authority of the people who approve its work.
How do roles change when agents arrive?
Preparation moves to the agent; judgement stays with people. A claims handler stops keying data and starts deciding held claims. A payroll administrator stops chasing timesheets and starts guarding bank detail changes. A support adviser stops writing routine replies and starts reviewing difficult ones.
Each agent works at one certified depth, and each action needs a depth: 1 read, 2 draft (a person sends), 3 execute routine work within bounds, 4 move money or take hard-to-undo actions, 5 coordinate other agents. The higher the depth, the more the human role becomes about exceptions. HR should know which depth each agent runs at in each team, because that determines what the job now is.
New roles also appear: reviewers and approvers inside teams, agent owners accountable for each agent’s brief, builders who design and certify agents, an AI operations manager, and a governance lead.
Who is accountable when an agent acts?
People, always. An agent acts within bounds set by the business and filed in an append-only record. When an action is held, a named person decides it, and the record shows who. When an action is allowed at depth 3, the person who set the limit answers for it.
HR’s part is making that accountability real in contracts and job descriptions. If a role now includes approving held refunds up to a limit, write it down. If a manager owns an agent’s limits, put it in their objectives. Where a regime such as the UK Senior Managers and Certification Regime applies, accountability maps must include agents.
What training do people need, and how do you prove it?
Three layers, matched to the role.
| Who | What they need | Proof |
|---|---|---|
| Everyone working near agents | Why agents are governed, depth, the four verdicts, what to escalate | A foundation test |
| Reviewers and approvers | How the agent decides, how to review, red lines, reading the record | An exam and a period of observed real review |
| Builders, owners, governance | Designing, certifying and governing agents | Exams and practical work |
Under the EU AI Act, deployers must take measures to ensure staff dealing with AI systems have sufficient AI literacy, and must give human oversight to people with the competence, training and authority to exercise it. HR is usually the function that can evidence this.
A practice some businesses adopt: tie approval rights to a current certificate. A reviewer whose certificate has lapsed loses the right to approve held items until they renew. This mirrors how agents work: certificates lapse after 90 days; a lapse, or a run that clears no depth, drops the agent to depth 1 the same day.
How should job descriptions change?
Before go-live, not after. A job description written for a world without agents describes tasks the person no longer does and omits the ones they now carry. Add review and approval duties, with limits. Add escalation duties for red lines. Add responsibility for correcting agent drafts and recording why. Remove preparation tasks the agent now does.
Grading may change too. A role that now decides exceptions all day may carry more judgement than before, not less.
Which HR processes touch agents directly?
Two that deserve care. First, any agent used in HR work, such as drafting letters or screening queries, reads personal data about employees; scope its access tightly and agree it with the data protection officer. Second, decisions with significant effects on employees, such as dismissals, grievances or pay disputes, stay with people.
HR’s own team needs approver rules and a reviewer rota like any other: see how to set an approver rule for the HR team and how to build a reviewer rota for HR. A competency matrix for the HR team maps who is ready.
Who should HR work with?
Team leads, who turn new role designs into rotas and coaching; see what a team lead does differently when agents join. Compliance, who will test whether reviewers are competent; see the compliance team’s view. The role-by-role overview is in AI agent skills by role.
Where the Academy fits
Pryme Intelligence Academy gives HR a ready structure: Foundations for anyone, Operator for reviewers and team leads, Builder for agent designers, Architect for leaders of regulated deployments. The admin view shows progress, results, certificates and expiry dates, which is the evidence HR needs. Certificates can be checked at source through VerifyDoc.ai.
Questions people ask
Should agents appear on the organisation chart?
Not as employees. Show them in process maps with a named owner, so it is clear who answers for each one.
Do people need consent to work alongside agents?
Consultation duties vary by country and contract. Involve employee representatives early where they exist.
Will reviewers need higher pay?
That depends on the job evaluation. Review and approval roles often carry more accountability, which grading should reflect.