Pull up any candidate record in your ATS right now.
If an AI agent ran the initial screen, scored the resume and ranked that candidate in the last 30 days, that fact almost certainly does not appear in the record. No agent name. No model version. No timestamp. No note that a non-human actor touched this person’s file before a recruiter ever saw it.
The same is true when an agent approves a time-off request, routes an HR service case or flags a performance concern. Work happens. The HRIS records nothing.
That’s the gap this article addresses. This piece goes deep into the systems and what CHROs can actually do about it.
Josh Bersin put the stakes plainly in his February 2026 piece in this publication: the market is moving from task-level automation to full process automation, and HR’s operating model is being rewritten. Peter Cappelli challenged the “treat agents like employees” framing in his piece the same month, arguing the real issue is that agents have to be supervised.
Both are right. The agents are already here, and they’re doing real work.
But the conversation has stayed at the conceptual level. The harder question is architectural: Most enterprise HRIS platforms were not built to store, manage or audit workflows that involve non-human workers. And almost no one is asking what that means in practice.
The assumption your HRIS was built on
Every HRIS built in the last four decades was designed around one central concept: the employment relationship.
That assumption runs deeper than a software design choice. The employment relationship is a specific, regulated construct. It has a legal start date, a compensation structure, a reporting hierarchy, protected-class obligations and an audit trail HR is legally required to maintain. Your HRIS was built to manage that relationship. Payroll, performance and succession all sit on top of it.
AI agents carry no employment relationship with your organization. They’re not employees or contractors. They carry no W-2, no 1099, no position in your job architecture, no manager of record.
And yet they’re doing work that used to be done by people who had all of those things.
This is what I’d call HR’s conceptual debt. We’ve spent decades building workforce management systems on one shared assumption: “Worker” means “person with a legal relationship to this organization.” That assumption stopped being complete. The system was never wrong. The world outran it.
3 gaps that open the moment agents go live
Once AI agents start doing real work in HR processes, three specific gaps open up inside the HRIS. I’ve seen each of them firsthand.
The visibility gap. No single place in most HRIS platforms shows which agents exist, who owns them and what HR processes they’re touching. Agents are provisioned by IT as service accounts. HR finds out they’re operating the same way it finds out about most things—after the fact.
The audit gap. When an agent screens a candidate, approves a time-off request or routes an HR service case, the system of record typically captures the outcome but not the actor. A candidate is moved to the next stage. A request is approved. A case is closed. The record shows a result with no indication that a non-human process produced it. This is the gap that creates legal exposure.
The authority gap. Traditional workforce management runs on delegation chains your HRIS captures. A CHRO delegates to a VP, who passes authority down to directors and managers. When something goes wrong, HR can trace back to who delegated what and when. AI agents break that chain entirely. When an agent screens out a candidate based on a skills matching model, it isn’t acting under a delegation HR approved. Legal accountability flows upward to HR. The actual decision traveled down a path HR wasn’t part of. I call this the reverse delegation problem, and it has no precedent in HR’s history.
A 3-step framework for CHROs
This is not a developer problem. It’s an organizational governance problem, and it belongs in the CHRO’s portfolio. Here’s where to start.
Step 1: Name the workforce. Before HR can govern AI agents, it has to know they exist. That means building or requiring an inventory that treats agents as workforce entities, not just IT service accounts. Each agent doing work in an HR process needs a record: what it does, who owns it, which HR processes it touches and what data it can access. This inventory doesn’t have to live in the HRIS on day one. It can start in a spreadsheet. The point is that HR owns it, not IT.
Step 2: Build the audit trail before the problem arrives. Every AI action in an HR process should leave a record in the system of record: which agent acted, on which record, using what inputs, with what outcome and whether a human reviewed or changed the result. This is the architectural requirement CHROs need to put in writing with their HR technology teams and their vendors. If an agent screens a candidate and that decision is later challenged, HR needs to produce that trail from the HRIS, not from a vendor dashboard or a server log.
Step 3: Draw the authority line. Every agent operating in an HR process needs a named human owner and a documented scope of authority: what decisions the agent can make independently, what decisions require human review and what triggers an escalation. This governing structure doesn’t exist in most organizations today. Building it is the CHRO’s job, not the CTO’s. It’s workforce governance, not systems engineering.
What to demand in 2026 contract renewals
Most enterprise HR technology contracts renew this year. CHROs and HRIS leaders should use that moment to put specific requirements on the table. Here’s what matters.
Agent registration capability. Can the platform register and manage AI agents as distinct workforce entities alongside human employees? Workday’s Agent System of Record, generally available in early 2026, moves in this direction. Ask every vendor where their roadmap stands.
AI action logging. When an AI agent takes an action inside an HR process, does the platform capture it in the system of record? Not in a separate analytics tool—in the record itself, where HR and legal can access it on demand.
Transparency controls. Can HR define what data an agent is permitted to access, at what level of sensitivity and in which geographies? Can that scope be changed or revoked without an IT ticket?
Audit export. Can HR pull a complete log of every AI-involved decision in a given time period, in a format legal counsel can use, without relying on the vendor’s support team to produce it?
Vendors who can answer these questions with working features belong on your renewal list. Vendors who answer with a slide deck belong on your watch list.
The org chart is a legal document
Regulators, courts and plaintiffs’ attorneys examine the org chart as a document that defines the authority structures behind challenged employment decisions. As this publication explored in its reporting on whether CHROs still hold real strategic authority, HR’s influence is only as credible as the structures it actually controls.
The ghost org chart accumulates legal exposure outside any structure HR controls right now. The problem isn’t the technology or the platform maturity. The problem is that agents are doing real employment work, and the system of record shows none of it.
The organizations that address this early will close all three gaps: visibility, audit trail and authority. They’ll hold vendors accountable to deliver the infrastructure in 2026 renewals. They’ll treat agent governance as workforce governance, owned by HR, not IT.
Credit: Source link
