Three Things to Get Right Before an AI Agent Talks to Your Candidates

I started Reframing HR to help people leaders lead the AI shift in our profession rather than have it handed to them — practitioner-led education and advisory that treats HR as the strategist driving the technology, not the end user receiving it. One of the ways I do that is by testing HR AI agents for companies, from the inside. I go through the real candidate experience these agents create — applying for the roles, sitting through the video and text interviews, exactly the way an applicant would — and then I tell the company what I found.

If you're rolling out agents in your hiring process, here are the three things I'd get right first:

1. Keep a human in the loop on candidate-facing decisions that can't be undone — rejections, offers, withdrawals.

2. Give every agent a "stuck" rule. When it can't finish a task, it hands off to a named person — it doesn't silently drop the candidate or go quiet.

3. Build in visibility. Someone who can see what the agent did, both while it's running and after launch.

Here's why those three are important, and what happens when they're missing.

The pattern I keep seeing

Across this testing, two issues show up again and again — different companies, different tools, the same gap.

The first one is quiet. An agent hits an ordinary snag — a scheduling conflict, a calendar that won't line up — and instead of flagging it, it "resolves" the problem by ending the candidacy. The applicant comes off the requisition, and the recruiter who owns that requisition never knows. They find out, if they ever do, from the candidate.

The second is its cousin. An agent is asked to do something — call the candidate, book the screen — and at some point it can't. Because no one defined what it should do in that moment, it does nothing. The candidate is left waiting for a call that never comes.

Neither of these is the technology failing. The tools are genuinely good and getting better fast. The agents are doing exactly what they were deployed to do — there just wasn't a plan for what happens when reality doesn't cooperate. No human in the loop, and no rules for the edges. That's a deployment gap, not a product flaw. And it's most costly at the exact moment an agent touches someone outside the company, where your employer brand and candidate experience are on the line.

It's not just my testing

This is showing up in the data and in the courts.

According to a 2025 survey of 1,399 U.S. workers, three in four companies said they allow AI to reject candidates without human review — and only 26% require human oversight on every rejection.

And in Mobley v. Workday (No. 3:23-cv-00770, N.D. Cal.), a federal court granted preliminary certification of a nationwide collective action alleging that AI-driven screening systematically rejected applicants over 40 — often within minutes, without a human ever reviewing the application. The court has even required Workday to disclose employers who used certain screening features so affected applicants can be notified. The point for HR leaders is straightforward: candidate-facing automation without oversight is now a legal exposure, not only a brand one.

It also lines up with the most credible research on how agents behave in the real world. Anthropic's 2026 analysis of millions of live human-agent interactions found that effective oversight isn't about approving every action — it's about a human "being in a position to intervene when it matters." When an agent ends a candidacy and no one knows it happened, that's exactly the test being failed.

The takeaway

Governance here isn't a brake on AI — it's what lets you say yes to it with confidence. And it belongs to HR, because HR owns the candidate experience these agents now shape. You don't need a heavy framework to start. You need three things:

• A human in the loop on anything that can't be undone.

• A clear hand-off when an agent gets stuck.

• Visibility into what the agent actually did.

Get those right, and you can experiment boldly — which is exactly what you should be doing.

If you're deploying agents in your hiring process and want to know what your candidates are actually experiencing, that's one of the things what we do. At Reframing HR, we test your agents the way a real applicant would, and show you where the gaps are before they become a story. Reach out at https://www.reframinghr.com/contact .