Before You Cut Jobs Because of AI: The Rehire Math and the California Rules for a Small Employer

Map the tasks before you cut the people. Gartner expects at least 30% of positions eliminated because of AI to be refilled by 2029, at a higher cost than the original role. For a small California employer, that means counting the hours a role really carries, pricing the rehire, documenting how you picked who goes, and meeting the final pay, notice and release rules if you still cut.

I am Mac Malbrough. I hold a J.D. and an MS-HRM, I am not an attorney, and I write for owners who run lean. If a board, an investor or a competitor’s press release is pushing you to announce an AI-driven reduction, read this first. Every dollar figure below is a hypothetical I built to show the arithmetic, not a client result.

What does the research say about rehiring after AI layoffs?

On September 9, 2026, Gartner announced that it expects 30% of positions eliminated by AI-related layoffs to be refilled by 2029. Gartner analyst Tori Paulman said the figure for large enterprises will likely be closer to 40%. The same coverage says companies will face steep increases in recruiting, training and onboarding costs, and that executives are attributing cuts to AI that have other causes, which Gartner calls AI washing. The coverage gives no sample size or method, so treat the percentages as a forecast, not a measurement.

A forecast is still useful. It tells you where the mistake usually happens: the company cuts the title, discovers that the title carried work nobody wrote down, and hires someone back, often at a higher rate because the market now knows the job exists. A small employer feels that more than a large one. A bank with 40,000 people can absorb a bad cut. A company with 60 people loses the one person who knew why a customer’s invoice is always paid on the 20th.

Why do AI-justified cuts backfire?

Three reasons come up again and again.

• Titles hide tasks. A “coordinator” role often includes quality checks, customer apologies and workarounds that never appeared in a job description. The AI tool automates the written duties and the unwritten ones vanish.
• The tool is less reliable than the pilot suggested. A tool that performs well on a clean sample may stumble on your messy customers and exceptions, and the people who handled the exceptions are the ones you let go.
• Remaining staff absorb the gap. The people left behind pick up the leftover work. Some of them will leave, which makes the rehire list longer than the layoff list. For a team under that strain, my guide to quiet cutting shows how it feels from the employee side.

How do you map the work before you cut?

Do a task inventory, one role at a time, before any decision about headcount. You need a spreadsheet and two hours with the person who does the job and their manager.

1. List every recurring task and the hours per week it takes, including the exceptions and the fixes.
2. Mark the cost of an error for each task: a typo, a late invoice, a compliance fine, a lost customer.
3. Sort each task into three groups: a tool can do it, a tool can draft it and a person must check it, or a person must do it.
4. Add up the hours in each group. The share left in the second and third groups is your human share of the role.

If the role is 70% in the “check or keep” groups, the cut is not an AI savings. It is a workload transfer. Large companies have started calling this a human-to-agent ratio. For you it is just the share of hours a person still must spend, and it is a better planning number than a headcount target.

What does the rehire math look like?

Here is a hypothetical. You cut one role paid $70,000 and save the salary. The tool covers 40% of the work. Six months later you rehire. Use your own numbers, but include every line:

• $70,000: annual salary you saved by cutting.
• $35,000: six months of the saved salary you actually kept, if the rehire starts at the same pay.
• The recruiting cost: advertising, recruiter fees or the hours of your own time.
• The ramp-up: months in which the new hire is paid in full but produces less than the person you cut.
• The tool: its license and the hours someone spends supervising it.
• The people who left in between, and the cost of replacing them.

If the rehire needs a raise, the savings shrink further. Gartner’s coverage says the rehired roles cost more, and the 2027 raise data shows why a market that knows the job is back will not sell it cheap. The point of the exercise is not to prove a cut is wrong. It is to put a number on the risk before you commit.

What does California require if you still cut?

If the math still says cut, the legal steps do not change because the reason is AI. These are the ones I see skipped most often in small companies:

• Document the selection criteria. Write down why each position was chosen before anyone is told. If an AI tool ranked employees or flagged roles, the Civil Rights Council’s regulations on automated-decision systems apply to employers with five or more employees, and you must keep the related records for four years. See my guide to AI hiring tools and the Civil Rights Council rules.
• Check for a pattern. Compare who is cut against age, sex, race, disability and leave status. A pattern is evidence for a claim even if no one meant it.
• Know the notice thresholds. Cal-WARN applies at covered establishments with 75 or more employees and is triggered by larger cuts, closings or relocations. My Cal-WARN guide has the details, including the 2026 notice content.
• Pay everything on the last day. A discharged employee’s wages and accrued vacation are due immediately. The penalty for late final pay is in my final paycheck guide.
• Handle releases carefully. A severance agreement that asks employees age 40 or older to release age claims in a group layoff has extra federal requirements under the Older Workers Benefit Protection Act, including at least 45 days to consider the agreement, seven days to revoke it, and written disclosure of the job titles and ages of those selected and not selected. My termination documentation and severance guide covers the release basics.

I would also keep the written task inventory in the file. If an employee later says the reason was a pretext, a dated record showing how the work was divided is better evidence than a manager’s memory.

Should you say AI is the reason?

Only if it is true, and only if you can show it. Gartner’s own coverage says executives are blaming AI for cuts that have other causes. For an employer, a stated reason is a statement you will have to defend later. If you tell a laid-off employee, an unemployment office and your remaining staff that AI replaced the role, and then post the same job two months later, the sequence reads badly to a plaintiff’s lawyer and to every employee watching.

Give the same reason everywhere it appears: the notice to the employee, the separation agreement and your answers to the unemployment agency. If the real reason is a mix of cost, a slow quarter and a tool you are trying, say that. A mixed truth is easier to keep consistent than a clean story that did not happen.

Are there gentler options than a cut?

Often, yes, and they cost less to reverse. A hiring freeze lets attrition do the work. Shifting a role’s tasks to include the tool supervision you will need anyway turns a layoff into a redeployment. A pilot, where the tool runs next to the person for a quarter and you compare errors, gives you your own data instead of a vendor’s. Reduced hours or unpaid furloughs have wage and benefit consequences in California, so run them past someone before you announce them. Each option buys time to find out whether the tool does what the demo promised.

What should you do this week?

Pick the single role under the most pressure, run the four-step task inventory, and write down the human share of hours and the cost of the errors. Then price a rehire with the six lines above. If you decide to cut anyway, have the selection criteria, the notice analysis and the final pay plan in writing before the first conversation.

My California employee handbook template includes the separation and final pay language, and the HR compliance check-up is where I would review your layoff plan before it goes to anyone.

The notice side of an AI-driven layoff is covered in what a technology displacement layoff notice must say under SB 951, and the discipline side is covered in what SB 947 lets an employee request when software drives discipline.

The people side of the same shift is covered in what happens to HR’s job when AI coordinates more of the work.

Sources

• CIO, Gartner on AI layoffs and rehiring, September 2026, read October 31, 2026
• California Civil Rights Council, automated-decision system regulations, June 30, 2025
• 29 U.S.C. 626(f), waiver requirements for age claims

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 October 2026. Sources are linked where each figure appears.