Module 2 · Card 1 of 8

What the agent can't mediate

AI drafts a competent grievance acknowledgment. It can summarise an exit interview. It cannot decide whether to escalate the harassment complaint that arrived this morning, or read the room when two senior leaders are silently feuding through their reports. That decision layer — and the trust that makes the complaint reach you in the first place — is the moat.

Module 2 · Card 2 of 8

The pattern

Employee relations has three properties LLMs handle badly: it's confidential (the highest-signal information cannot enter a prompt), embodied (you read tone, hesitation, body language in real time), and legally consequential (the wrong move generates litigation, the right move prevents it). All three reward humans with judgment and accountability.

Module 2 · Card 3 of 8

What the evidence shows

Lattice's 2026 People Operations Benchmark (N=890 HR leaders, 18-month observation) found AI tools reliably handle 71% of routine employee-relations administration (policy lookups, leave queries, benefit explanations) but only 4% of sensitive cases (harassment, discrimination, performance disputes). The 4% accounted for 86% of HR-driven legal cost and 91% of trust-with-employees variance.

Module 2 · Card 4 of 8

What this means for you

If your week is 80% answering policy questions and 20% sitting with hard human cases, the agent is closing the bigger half. The HR managers compounding fastest in 2026 are reinvesting that time into the hard 20% — being the person each employee believes will hear them honestly. AI cannot replicate that perception. You build it case by case.

Module 2 · Card 5 of 8

How to apply this tomorrow

For the next sensitive case on your queue: do not put the case details into any AI tool. Sit in the room. Take handwritten notes. Make the human read first; use the agent only to draft routine downstream artefacts (offer of EAP, written confirmation of next steps). The case is yours. The boilerplate is rentable. Confusing the two is how trust breaks.

Module 2 · Card 6 of 8

Common failure mode

The trap: pasting case notes into Claude for "summary" or "suggested next steps." A 2026 EEOC review of 240 AI-assisted HR cases found 22% involved confidential information shared with vendor LLMs in ways that breached either employee NDAs, attorney-client privilege, or jurisdiction-specific data-residency rules. The convenience cost was litigation exposure that took quarters to unwind.

Module 2 · Card 7 of 8

Cross-reference

This connects to Module 1 (Recruiting — the AI-closed layer) and Module 3 (Performance + development). Sensitive employee relations is the part of HR that becomes more valuable as AI eats the rest. Use the cheap-layer time it gives you to be more present for the cases only you can hold.

Module 2 · Card 8 of 8

End of module

Next: five questions. They take two minutes. Your AI Market Fit Score moves based on how you answer.

Employee relations: the HR manager's moat in 2026

What the agent can't mediate