Nearly a third of the hiring managers who cut a role because of AI have already hired someone back to do it. Of the leaders who made those cuts, 55% now say the decision was wrong.
The technology did not fail them. The order of their decisions did. They automated first and redesigned never.
If you run a company of 20 to 200 people, you cannot absorb that mistake the way a large enterprise can. You can also avoid it far more easily than they can.
What the Rehiring Data Shows
Start with the size of the wave. In the first half of 2026, employers in the US attributed 101,743 job cuts to AI, roughly a quarter of all announced cuts. AI was the single most cited reason for four months running.
Now look at what came next. Robert Half surveyed nearly 2,000 hiring managers. 32% had already rehired for a role they had eliminated because of AI. Finance was worst at 44%, then human resources at 35%, then technology at 32%.
Orgvue asked the leaders themselves. Of those who made AI-linked redundancies, 55% now say they got it wrong.
That is not a handful of bad calls. It is a pattern, and it is consistent enough to plan around.
The Bet Came Before the Proof
Here is the part that matters, and it has nothing to do with the technology failing.
Harvard Business Review surveyed more than 1,000 executives. 60% said they had already cut headcount in anticipation of what AI would be able to do. Only 2% had made large cuts after the technology actually delivered in production.
Sixty percent acted on a forecast. Two percent acted on a result.
It gets sharper. Orgvue found that 23% of the companies making AI-linked redundancies based the decision on general assumptions about what AI can do. They never went role by role to work out what the people they were letting go actually handled all day.
The pattern repeats everywhere. Someone sees a demo, believes the headline, and cuts headcount before a single workflow is redrawn. They treat AI as a drop-in replacement for a person, when it is really infrastructure that only pays off if the work around it changes.
The operators who see a return do it the other way round. They redesign the process first, then put the tool into the new version of it. Skip that step and you have not transformed anything. You have bought expensive automation, and the bill arrives a few months later in the rehiring line.
What Breaks When You Skip the Redesign
Ford leaned hard on automated inspection and let experienced engineers go along the way. The systems missed defects that veteran inspectors caught in minutes. Worse, the people left before anyone could get their knowledge into the models.
Charles Poon, Ford’s vice president of vehicle hardware engineering, said it plainly: “Mistakenly, we thought that by just introducing artificial intelligence and adjusting the design requirements that we had, that that would produce a high-quality product.”
Ford brought back 350 veteran engineers to mentor younger staff, rebuild the data floor feeding its AI, and catch what the automation missed. The company then took the top spot in J.D. Power’s Initial Quality Study for the first time in nearly twenty years.
Klarna hit the same wall in a chat window. It cut roughly 700 agents’ worth of work after its AI assistant proved it could handle the volume. Then satisfaction slipped on the complicated cases. Chief executive Sebastian Siemiatkowski later told Bloomberg that “cost unfortunately seems to have been a too predominant evaluation factor”, and the company started hiring human support back for the nuanced work.
IBM’s version is narrower and more useful. Its AskHR agent settles about 94% of routine staff requests. The other 6% are ethical calls and judgement calls, and those still land with a person. IBM’s chief human resources officer, Nickle LaMoreaux, made the second-order argument: stop hiring juniors and the pipeline of people who can handle that 6% simply dries up. IBM tripled its US entry-level hiring instead.
Four companies, four industries, one wall. The work AI could not absorb was never a technical edge case. It was judgement, context, and someone willing to be accountable when it goes wrong.
And the reversal is not cheap. Careerminds puts the true cost of a workforce reduction at about $1.27 for every dollar it was meant to save once severance, lost productivity, and replacement hiring are counted.
Elevation, Not Replacement
Ford and Klarna paid to learn something IKEA worked out first.
IKEA’s Billie assistant now handles about 47% of customer service enquiries. The company did not cut the 8,500 people whose workload it absorbed. It retrained them as remote interior design advisers. That channel now brings in around 1.3 billion euros a year, and the group wants it at 10% of sales.
The work did not disappear. The people moved up into work that needs taste, judgement, and a relationship, instead of repetition.
That is the model, and it scales down cleanly.
A copywriter does not get replaced by a prompt. They move into content direction, briefing agents on strategy and brand voice while the machine handles first drafts.
A designer does not get pushed out by image generation. They move into creative direction, using their taste to brief and refine what the agents produce.
A sales rep does not get displaced by automated outreach. They become the closer who steps in once the AI has qualified the lead.
This is also the difference between a chatbot and an agent. A chatbot waits to be asked. An agent owns a job, makes calls inside clear limits, and flags you when something is off. Think of it as the difference between a hire who runs a function and one who waits to be asked.
This is the whole idea behind AI enablement as we practise it. Let the robots do the robot work: pulling data, sorting leads, drafting first versions. That frees your people for the human work, which is strategy, creative direction, and real relationships. You elevate them from doing the legwork to supervising it.
For that to hold, the expertise has to live in the system, not in someone’s head. Our Catalysts do that job. GTM Catalyst, Sales Catalyst, and Design Catalyst put your brand, your market knowledge, and your go-to-market playbook into the workflow as infrastructure, so an agent has something to be right about and a person has something to direct.
The question worth asking is not which roles AI eliminates. It is which roles your best people should move into once the routine work is handled. That one distinction separates a team that fears AI from a team that uses it to move up.
Why It Costs You More Than It Costs Them
A large enterprise can quietly reverse a workforce decision over two quarters and bury the cost in a restructuring line. The cost is real for them too, but it is spread across thousands of people and absorbed by a budget built to carry mistakes. Nobody outside finance ever sees it.
You cannot do that. One bad call destabilises a team you spent years building, and everyone in the building watches it happen. The cost is not a line on a report. It is trust, and it is slow to rebuild.
But the same size that makes the mistake expensive makes it avoidable. We have written before about why your size is the AI advantage; this is where it pays.
A 50-person company can redesign a workflow in weeks. A bank with 50,000 people needs a steering committee, a change programme, and two years. You need a clear outcome, a clear picture of how the work should flow, and the willingness to test it before you commit headcount.
That is the advantage. Not cheaper labour, not a bigger budget. A shorter distance between deciding and knowing.
The Sequence That Actually Works
Redesign first. Automate second. Elevate people third. That is the only order that compounds.
It is the same rule we hold ourselves to on every engagement: Diagnose, Redesign, Install. Diagnose before you buy. Redesign the workflow before you deploy. The tool goes in last, into the new version. And sometimes the honest answer is that AI is not your binding constraint yet.
Here is what it looks like in practice.
Say you want to enter a new market and you think you need ten sales reps. The old model says hire ten, onboard them, absorb them into the culture, and hope half of them work out.
The redesign model says hire one strong sales manager and two or three elite closers. Let agents handle the background work: research, prep, and qualification. Your manager mentors a tight team with specific briefs. You see the output directly and coach in real time.
The team stays small, aligned, and easy to direct, because the work was drawn around what people do best and what AI handles well. That is redesign, not redundancy. It produces a compounding operational advantage instead of a one-off saving.
Notice you never cut anyone to get there. You changed the shape of the hire.
The part most people miss is that the new shape has to stay legible. When people and agents share an outcome, everyone needs to see how that outcome feeds the company’s bigger goals. That is what we built Keel for, and it is why we organise work around value pods (“keels”) rather than departments. When the line from daily work to the mission is visible, handing the routine part to an agent stops being a threat.
Redesign, Not Redundancy
The companies in the rehiring data are not victims of a technology that underdelivered. They are the ones who made a workforce decision before they had done the work that would have told them what to decide.
Test any AI decision against one thing: does it change the unit economics of a core process, or does it just remove a cost line? The first is foundational. The second is a shiny object with a payroll attached.
So the question is not whether AI can do part of someone’s job. It nearly always can. The question is what you want that person doing once it does.
Want to know where you actually stand before you touch headcount? Take the AI-First Readiness Assessment, a short readiness checklist for growing companies. If you would rather start with the work itself, the Work Redesign Audit is a free workbook for exactly this. Or book a discovery call and we will map your first AI work redesign together.
FAQ
Frequently asked questions
Are AI layoffs backfiring?
For a large share of companies, yes. Robert Half found 32% of hiring managers had already rehired for a role they cut because of AI, and Orgvue found 55% of leaders who made AI-linked redundancies now say the decision was wrong. The common cause is sequence: the cut came before the workflow was redesigned and before the technology had proved itself in production.
How many companies are rehiring after AI layoffs?
Robert Half's survey of nearly 2,000 hiring managers put it at 32%, with finance highest at 44%, human resources at 35%, and technology at 32%. Careerminds puts the true cost of a workforce reduction at roughly $1.27 for every dollar it was meant to save.
Why do companies regret AI layoffs?
Because most cuts were made on a forecast rather than a result. Harvard Business Review found 60% of executives had cut headcount in anticipation of AI, while only 2% had made large cuts after AI delivered in production. Orgvue found 23% decided on general assumptions without analysing what the eliminated roles actually did.
What is AI work redesign?
AI work redesign means changing how a process runs before you add AI to it: mapping the steps, deciding which decisions an agent can own, and defining where a person verifies or intervenes. The tool goes into the redesigned workflow, not alongside the old one. It is the step that separates AI transformation from expensive automation.
Should a small or mid-sized company cut jobs because of AI?
Not as a first move. A growing company cannot absorb a reversal the way a large enterprise can, and one bad workforce decision can destabilise a team built over years. The better sequence is to redesign the workflow, let AI take the routine part, elevate people into judgement and relationship work, and only then look at what headcount you actually need.
What is an AI readiness assessment?
An AI readiness assessment is a short review of your data, your workflows, and your goals that shows where AI will pay off first and where it will not. Catalys AI runs one as an AI readiness checklist for growing companies at catalys.ai, and follows a Diagnose, Redesign, Install approach so tools land inside a redesigned workflow rather than on top of the old one.