What Happens to HR’s Job When AI Coordinates More of the Work?

HR’s job shifts from keeping tasks moving to deciding what should happen. AI can chase paperwork, track status, draft options and flag exceptions. A person still has to assess fit, talk with employees, approve changes and answer for the result. In California, some decisions about discipline and layoffs now require human review by law.

I am Mac Malbrough. I hold a J.D. and an MS-HRM, I am not an attorney, and this is general information, not legal advice. This guide is written for HR professionals, and for owners of 10 to 150 person organizations who run HR themselves, and it covers what software can take over in an HR week and what it cannot.

What does it mean when AI coordinates the work?

Two things are happening at once. The first is assistants: tools that help a person prepare a decision, such as drafting a role description, summarizing a file or listing candidates for an internal move. The second is agents: tools that carry out a task that has been authorized, such as sending a reminder, updating a record or scheduling a meeting, and that flag something when it needs a person. Vendors have started calling the combination “autonomous HR.” I treat that as a marketing phrase, not a standard. What matters is which step in a process is handled by software and which step is handled by someone who can be held to account.

Take a common example, moving an employee into a new role. Someone has to check qualifications, talk with the employee about the opportunity, approve the change, arrange training, update payroll and systems, and confirm the person is ready. In many small companies, one HR person spends days just keeping those steps moving. That is the part software is good at. My guide on what an AI agent can do in HR for a small California employer goes through the task-level details.

Which HR tasks move to software first?

The first tasks to move are the ones that are repetitive, rule-based and easy to check:

• Chasing missing forms, signatures and training completions.
• Tracking where each case, hire or change request sits and who is holding it up.
• Drafting a first version of a job description, offer letter or policy update for a person to edit.
• Answering common questions about benefits, holidays and procedures from approved documents.
• Flagging a pay figure, date or record that does not match the rest of the file.

None of these require judgment about a person. All of them eat hours. If you run HR alone, getting even part of that time back is worth a careful pilot.

Which parts of HR stay with a person?

The parts that stay are the ones where someone has to weigh a situation, speak to a human being or be accountable for the outcome:

• Deciding whether a person is ready for a role when the evidence is mixed.
• Having the conversation when a change is welcome and, more often, when it is not.
• Approving a pay change, a discipline decision or a termination.
• Handling a complaint, an investigation or a leave request with sensitive facts.
• Deciding what to do when the rules do not fit the case.

A useful test: if the person affected would want to hear the reason from a human being, a human being should deliver it. My piece on the cost of AI layoffs and rehiring shows what happens when a company hands that conversation to a process.

What does California already require a person to do?

Two 2026 laws put a human in the loop for some of the highest-stakes decisions. SB 947 takes effect July 1, 2027. It bars an employer from relying solely on an automated decision system to discipline or fire someone, requires a person to corroborate the output when the system is the primary basis, and requires a separate plain-language notice to the employee. I walk through it in my guide to SB 947 and what employees can request. SB 951 adds content to a Cal-WARN layoff notice when AI or automation causes a layoff, and it also takes effect January 1, 2027, as I explain in the SB 951 notice guide.

Neither law says that a role change is discipline. But a role change that follows from a score or a flag, such as a demotion or a removal from a position, can slide toward discipline in practice. If a tool feeds that decision, have an employment attorney read the definitions before you rely on it. The safer habit is the same one the law points to: a person reviews the data and the other evidence, writes down the reasoning and owns the decision.

What new skills does an HR person need?

I would build five skills, in this order:

1. Process design: knowing which steps in a workflow can be automated and where a person must approve.
2. Checking output: reading what a tool produces, spotting what is wrong or missing and keeping a record of corrections.
3. Data literacy: asking where a number came from, how it was counted and what it leaves out.
4. Hard conversations: explaining a decision, listening to a reaction and staying steady. My hard conversation scripts are built for this.
5. Employment law basics: knowing when a situation needs an employment attorney.

The skills at the top of the list are about the system. The skills at the bottom are about people, and those get more valuable as the routine work becomes cheaper.

How should a small employer plan for it?

Start with a task map, not a vendor demo. List what HR does in a month, mark each task as repetitive, judgment-heavy or both, and look at the hours. Pick one repetitive workflow with low risk, such as onboarding paperwork reminders, and run it in read-only mode for a month before you let a tool change a record. Write down the approval points, meaning the steps where a named person says yes. Keep a simple log of what the tool did, what a person changed and what went wrong. The core HR and payroll software guide covers the tools most small employers already own, which often have these features switched off.

Be honest with employees about it. A short note that says which tasks software now handles, who approves decisions and how to reach a person will head off the fear that a machine is deciding their future. The same note helps when an employee asks whether software was involved in a decision about them.

How can an HR professional stay valuable?

Move your time toward the work that needs judgment, and make it visible. Keep a private record of cases where your reading of a situation changed the outcome: the complaint you caught early, the manager you coached before a bad decision or the change you slowed down because the facts were thin. Describe those in resumes and reviews without client names or invented numbers, only what you did and what you saw. If you are job hunting, my career blueprint guide helps you decide which kind of HR role to aim for as the work changes.

What can go wrong when software coordinates people work?

Three failures show up again and again. The first is silent drift: a tool keeps doing a task after the rule behind it has changed, such as a leave or pay rule, and no one notices because the output looks normal. The second is false confidence: a summary or score reads smoothly, so a busy approver signs off without checking the underlying file. The third is lost context: a person used to catch the odd case in conversation, and a workflow that removes the conversation also removes the catch.

The defense is dull and effective. Review a sample of the tool’s work every month against the source records, set a date to re-check every rule it uses, and require that approvers write one sentence on what they checked. If an approver cannot say what they checked, they did not review it.

How does this change who you hire for HR?

If you are deciding between a first HR hire, an interim manager and an outside service, weigh judgment above administration. An administrator can be supported by software. A person who can sit with a manager, read a situation and say “do not do that” cannot be replaced by it. For a team of 10 to 150 people, that often points toward a senior part-time or fractional resource who owns the judgment calls, with software handling the routine. My comparison of fractional HR, a PEO and a first HR hire sets out the tradeoffs.

How does the UHR onboarding kit fit an AI-assisted HR team?

The problem for an owner is that software can now coordinate the steps of people work, but it can only coordinate a process that exists. If your first 90 days live in someone’s memory, an agent has nothing reliable to follow. The 30-60-90 onboarding program kit ($99) gives you the process in writing: manager checklists, a week-one schedule, a buddy guide and the check-in agendas. Once the steps are written, you can decide which ones a tool may schedule and which ones a person must run.

What should you do this week?

Block one hour and write your task map: every recurring HR task from the last month, the hours it took and whether it was routine or needed judgment. Circle the one routine task that costs the most time and ask your current software vendor whether it can automate it. Then write the approval point: who must say yes before anything changes. If you want help sorting the list, the HR compliance check-up includes a review of your current HR tools and processes.

Two practical follow-ups: a safe prompt library for HR tasks and the four stages of AI use on an HR team.

Related reading: ten AI pilots for a small HR team, ranked by risk; five reusable AI skills with refusal rules.

For a related angle, see HR strategy versus HR operations.

Sources

• SB 947 (McNerney), Chapter 859, Statutes of 2026, approved September 30, 2026, read November 4, 2026
• SB 951 (Reyes), Chapter 860, Statutes of 2026, approved September 30, 2026, read November 4, 2026

McKinley holds a J.D. but is not a licensed attorney. Articles here are general information, not legal advice. For your specific situation, talk with an employment attorney.

Data current as of November 2026. Sources are linked where each figure appears.