HR is one of the highest-leverage places for AI agents and one of the easiest to get wrong. Use them on the high-volume work that takes HR away from real people problems; keep humans on anything that affects employee outcomes.
Start with agents for recruiting (resume screening with humans-in-the-loop) and employee FAQs. Add onboarding and people-analytics agents next. Stack: an HR-aware agent layer plus Claude or ChatGPT for ad-hoc work. Budget $300-$2,500/mo. Time saved goes back into the work HR actually wants to do.
HR work splits into transactional (questions, paperwork, recruiting volume) and judgmental (performance, conflict, policy). Agents handle the first cleanly; humans must own the second.
The HR-agent pattern: agents take the transactional volume; humans own anything that touches an individual's outcome. Skip the human review step on something sensitive and trust erodes fast.
These are real implementations, not theory. Each shows what the agent handles, what the human decides, and roughly how much time it saves.
The most common mistake in HR agent deployments is giving the agent too much autonomy on sensitive decisions. This table shows where agents add value without creating risk.
| HR task | Agent role | Human role | Risk if agent goes solo |
|---|---|---|---|
| Resume screening | Score and rank applicants | Final selection and advance/reject call | Disparate impact claims under EEOC |
| Employee FAQ | Answer policy questions with citations | Escalation for edge cases | Low. Well-contained. |
| Onboarding tasks | Sequence and send docs and reminders | Day-1 connection and manager coaching | Low. Process only. |
| Leave approval | Provide policy info and forms | Approve or deny the leave request | FMLA/ADA liability if agent misclassifies |
| Performance review | Draft and organize language | Review, revise, and own the final rating | Discrimination claims if agent rates without oversight |
| Compensation benchmarking | Pull data and flag gaps | Compensation decisions and offer approvals | Pay equity claims if agent sets comp autonomously |
| Employee relations cases | None. Do not involve agents. | All of it. | Severe. Legal exposure and trust destruction. |
HR-agent stacks for mid-market typically run $300 to $2,500 per month.
Two measurements: HR ticket deflection (% of inbound HR questions resolved by agent) and recruiter time per hire. Most teams hit 40-60 percent ticket deflection within 90 days and cut recruiter screen-time by half on high-volume roles.
The $1,500 AI Audit produces a written agent rollout plan in 5 business days: workflow selection, vendor choice, and the human-in-the-loop design that keeps quality high.