The Human Element

The HR work AI should never touch: terminations, compensation decisions, final hiring calls, and sensitive employee relations. That's not just our position — it's the consensus that emerged across more than two dozen conversations with HR executives.

We asked people leaders where they draw the line on AI, and while they disagreed about many things — tech stacks, rollout speed, even whether to hire new grads — on this question, they kept landing in the same place. As Q Hamirani, Chief People Officer at HighLevel, put it, "Our job is not to implement AI. It's to figure out where the humans sit in the AI construct."

Knowing AI's limits isn't AI skepticism. Gartner says the market is crowded with "agent washing" — vendors rebranding ordinary chatbots as autonomous agents — so being competent in today's tech landscape includes precision about what AI shouldn't do. Something interesting we found is that the leaders most enthusiastic about AI are often also the most exact about where its boundaries should be. Here's where they drew them, and why they say the lines make adoption better, not worse.

What HR Decisions Should Never Be Automated?

Across our conversations, four categories came up again and again. Each one shares a trait: The decision changes a person's life, and they deserve to know a human was behind it.

1. Hiring, Firing, and Compensation

Joshua Horenstein serves as both CHRO and Chief Legal Officer at Innophos, a global food ingredient manufacturer. His dual vantage gives him a unique perspective, and he was firm about this: "I would never let it make a termination decision. I would never let it make a promotion decision. I would never let it make a salary or compensation decision."

His reasoning, though, is less about legal exposure and more about what the technology actually is.

"AI stands for almost intelligent," he said. "It doesn't stand for totally intelligent."

Public sentiment backs him up. A Pew research report showed 71% of Americans oppose AI making final hiring decisions, while only 7% are in favor. A majority also oppose AI-driven analysis being used to decide whether someone gets fired.

2. Judgments on Performance and People

Liz McSavaney, Chief People Officer at Zurich Canada, runs an HR function inside one of the most regulated industries on Earth. Still, she's an unapologetic advocate for AI in insurance work. 

Her boundaries are just as unapologetic: Judgments about humans (performance reviews, how an employee relations case resolves) stay with humans. AI can assemble the file, but it can't make the final call.

Heather Oxley, CHRO at Perficient, echoed the same idea, but she called it, "Moments that matter." AI can automate answering policy questions, ticketing and scheduling all day. But moments that matter between people require people.

3. Succession Planning

Azuree Montoute-Lewis, Global Chief People Officer at Burson, leads a people function that has built around 9,000 internal AI agents — hardly the work of a technophobe. But she draws a line when it comes to planning the organization's big picture future: succession planning.

Her reasoning: Succession decisions depend on nuance that's "only visible in interacting with leaders" — the behaviors, judgment, and presence you observe over years. No dataset captures it, which means any model that claims to is guessing.

4. Sensitive Human Situations

Preet Hansra Michelson, Chief People Officer at TMS and Morgan Street Holdings, used automation to achieve performance review completion rates above 95%. Still, she routes a whole category of work away from AI: complicated FMLA cases, immigration outcomes, and family crises.

"A human touch not only demonstrates HR prowess, but also loyalty," she said. Hard conversations aren't a burden to automate away; they're where HR proves what it's for.

Larry Ott, CHRO at automotive supplier Cooper Standard, put it a little more bluntly: "AI is not going to help me address employee relations issues." His 200-person HR team uses AI for job descriptions and communications, but nobody consults it about a tough conversation.

Is There a Simple Test for What Stays Human?

In a word, no. Across our conversations, we heard about different frameworks people leaders use, but there's no one simple rule.

Here are three of those frameworks that stood out:

Ani Nazaryan, Chief People Officer at the Siegel Group, keeps a human in the loop if the work falls into any of these three buckets:

1. If it impacts someone's livelihood

2. If it impacts the organization's culture

3. If it requires judgment

Something might fall into multiple buckets; for example, a termination falls into all three at once.

Angela Briggs-Paige of Acelero Learning, who leads HR for a frontline early childhood development workforce, compressed the division of labor into just four words: "AI recommends, humans decide." For example, her team uses AI to screen credentials and licensing requirements at volume, but only humans make the actual hires. 

Q Hamirani, Chief People Officer at HighLevel, said it all comes down to accountability: "You cannot blame the AI agent." That means humans own all outcomes, whether they originate with people or AI agents. 

What Should AI Handle In HR?

The same executives who drew hard lines about what AI shouldn't do are automating aggressively on the other side of them.

The routine parts of HR are high-volume, repetitive, and backed up by documentation:

• Answering policy questions
• Explaining benefits
• Scheduling
• Answering status requests
• Writing first drafts of nearly everything

Christy Harris, CHRO at CCC Intelligent Solutions, chose to automate expense reports early in her company's AI adoption journey — precisely because it was routine work that everyone hated doing.

AI excels at HR operations work, but you need domain-specific AI that's equipped to handle the nuance and compliance needs of the HR world. Harper is the world's first AI HR generalist, always on, fully contextualized, and embedded in your HR operations to handle cases from policy compliance to performance management with speed, expertise and care, just like your best team member would. 

Harper is trained on Wisq's proprietary HRLM, the first and only language model built specifically for HR Operations — plus your company’s content, knowledge, culture, and policies. It resolves 40% of cases in just seconds and cuts 80% of effort on the rest — handling intake, triage, and full case resolution end-to-end.

Where Do You Draw the Line?

In conversations with 15 HR executives, a pattern emerged: The leaders furthest along with AI are most precise about where it should stop. As Q Hamirani put it, "AI is never going to be able to read the room." In other words, your job isn't just to implement AI — it's to decide what stays human — and then to defend it.

Frequently Asked Questions

Can AI handle complex HR policy questions?

Partially. AI handles routine policy questions well when it's grounded in your organization's policy documents: PTO accrual, benefits eligibility, leave procedures. Complex questions that require weighing competing policies, jurisdiction-specific employment law, or individual circumstances should be triaged to a human. A well-designed HR agent escalates them automatically rather than guessing.

Can AI interpret HR policy?

AI can read, summarize, and apply written policy to straightforward situations. Interpretation that sets precedent, resolves ambiguity, or affects an individual employment outcome is judgment work, which should stay with HR professionals.

Should AI make firing decisions?

No. Every executive we interviewed drew this line. AI can organize documentation and surface relevant policy, but a decision that ends someone's employment requires human judgment and accountability.

Can AI make hiring decisions?

AI can screen credentials, schedule interviews, and surface qualified candidates at volume — but the final hiring decision should be made by a person.

Can AI handle employee relations?

Routine intake and documentation, yes. The cases themselves, no. Sensitive employee relations work — like conflicts, investigations, accommodations, or anything involving a person's wellbeing or standing — requires the discretion, empathy, and accountability of a human professional.

What is the human-in-the-loop rule in HR?

Human-in-the-loop means a person reviews, approves, or owns the outcome of an AI system's work rather than letting it act with full autonomy.

The leaders in this piece described an AI that knows where to stop. That's how we built Harper: she handles policy questions, intake, and triage end to end, and escalates the cases that need a person.

Learn more about Harper. Book a demo today.