What does “AI rebounding” reveal about AI-driven job cuts?
On 9 November 2026, Fortune reported a survey by assessment firm SHL: 19.1% of the American respondents say they were replaced by AI and then asked to come back by the same employer. The survey ran in June and July among 2,036 employed adults in the United States and the United Kingdom, and its results are self-reported.
“AI rebounding” is SHL’s term for an employee whose job was cut or handed to AI and whom the former employer later asks to return. It describes an experience reported by employees, on a sample drawn from the United States and the United Kingdom.
What the SHL survey measures
According to Fortune, 19.1% of American respondents say they were replaced by AI and then asked back, against 16% for the US and UK combined. Among 18-24 year-olds the share rises to 24%. Of the people asked back, 51% wanted a substantial pay rise before accepting, 42% wanted written job-security guarantees, and 40% asked for transparency about how AI is used in their role.
Lucy Beaumont, SVP of product and design at SHL, quoted by Fortune, says companies are making quick and sometimes myopic decisions. That is the firm’s own analysis, and the survey does not prove it.
What the other sources say
Fortune also cites a Robert Half survey from July: 32% of US hiring managers whose organization cut roles because of AI had to reinstate the same or similar roles. On announced cuts, Challenger, Gray & Christmas counts 121,401 AI-attributed job cuts through October 2026, against 54,836 for all of 2025, with only 1,265 of them in October, the lowest month of the year according to its October report. Challenger counts announcements: what happens next inside teams, reinstatements included, falls outside its scope. The three sources use different bases and methods, so they do not add up. Read together, they are mainly a reason to check what each role carried before cutting it.
Why are companies rehiring roles they had just cut?
The reasons Robert Half’s hiring managers gave, as Fortune reports them, say more than the headline percentage. They describe what an org chart does not show, and AI-based reduction plans can underestimate it.
A role holds more knowledge than its task list
According to Robert Half, as reported by Fortune, 40% of the managers concerned cite institutional knowledge or context that AI does not replace. An agent performs a described task very well. It does not inherit the exceptions a person learned over ten years of cases, or the unwritten arrangements with a client or a neighboring team. When the role disappears, those unwritten rules go with it, and you only find them when they are missing. In a finance team, for example, it can be the habit of calling a particular supplier before a payment run. In a support team, it can be knowing which customers must never receive a templated answer. None of that appears in a job description.
Human oversight costs more than planned
Another reason cited is a need for more human oversight and quality control than expected, at 38%. This is the angle we defend: an agent in production needs someone to supervise it, stop it and correct its outputs, and that someone needs the experience to recognize a wrong answer when it looks plausible. If the reduction plan removed the people who knew what a good answer looks like, oversight has nobody left to carry it. Our article on trust zones for classifying AI agents explains how to decide which actions require a human.
Demand can also rise
Fortune also lists higher business demand among the reasons managers gave. A company that cuts roles on an assumption of constant volume ends up understaffed if its activity grows. AI lowered the cost of a task, but workloads can keep growing.
Do these figures apply to a European company?
None of these surveys covers the European Union, and quoting them as EU findings would be misleading. They give a signal to test within your own perimeter, with your own data. Start with the roles your company has already reduced or automated over the past twelve months: for each one, ask the team lead whether the work came back to a person, a contractor or an agent, and note which.
The limits of the measurement
SHL’s results are self-reported: a respondent who says they were “replaced by AI” is interpreting their own situation, and the employer might describe the facts differently. A sample of 2,036 adults across two countries says nothing about a given sector. “AI rebounding” is the firm’s label for what is best treated as an early signal, to be confirmed before it becomes a trend. The honest wording for a board is that a US survey suggests reinstated roles, and that the company should check its own cases. Ask also who holds the answer: a survey of employees tells you how the episode was lived, and only the employer’s records say how many roles were reinstated and at what cost.
The European legal context
In France and several other European countries, a headcount-reduction plan involves informing and consulting employee representatives, and reversing it costs more than a simple recall. Your legal team is the only one able to say what that framework implies for a given project. What can be said without them is that a sizing error takes longer to correct once it has touched employment contracts, which argues for validating the real effect of AI before removing a role.
How do you map what a role carries before reducing it?
Our reading fits in the image of two superimposed org charts: the humans’ chart, and the chart of the agents working beside them. Before removing a box from the first, you need to know what the agents in the second will actually do, and who stays accountable for their results. Whatever binds the company, whether a customer commitment, a regulatory filing or a financial decision, stays under human control.
Break the role into tasks, judgments and relationships
Take one specific role and list what it does over a real period: the repetitive tasks, the trade-offs that call for judgment, the relationships with customers and other teams, the responsibilities the person carries in their own name. Only the first column is easy to hand to an agent, and it rarely covers the whole role. The other three require either human oversight or an explicit handover to someone with the skills. For a claims handler, for example, sorting and entering documents goes to the agent, while ruling on a disputed file and calling the policyholder stay with the person.
Measure real oversight before deciding
On a pilot perimeter, measure how many agent outputs a human has to correct, review or escalate. Record, for instance, over four weeks, how many outputs were reviewed, corrected or escalated, and who handled them. That measurement gives the oversight time to keep in the plan, before the reduction is voted. Our article on measuring adoption by usage rate offers indicators you can use from the first month, and the LOOP™ governance method puts the human at the center of the control loop.
What do rehired employees ask for, and what should you take from it?
Among the people asked back in the SHL survey, more than half wanted a pay rise and 42% wanted written guarantees. Someone who was told a machine does their job comes back with a request for proof: they want to know which part of their role is protected and how the company will use AI. Forty percent of those asked back ask for transparency about AI use before returning. A leadership team that published a transformation plan without a readable usage rule may have to negotiate that transparency from a weak position.
What this changes for change management
A written usage framework lowers the risk of broken trust: which tasks go to agents, which stay human, how a decision is challenged, who answers for an error. The page on enterprise AI governance describes the full approach, and it is best applied before deployment, well ahead of the first mistake.
How does this connect to AI agent projects that fail?
Rehiring staff echoes something market figures have described since 2025: 46% of AI proofs of concept are abandoned before production according to S&P Global (2025), and MIT (2025) estimates that 95% of pilots show no measurable effect on the income statement. Stalled projects and reinstated roles may share a reflex: computing the gain on the isolated task without looking at the whole system of work. Our article on moving from pilot to production details what blocks it, and the page on enterprise AI agents describes what an agent in production requires.
The topic also touches early-career staff. Our reading of Anthropic’s report on junior hiring details the risks to the talent pipeline over five years.
What should you do this week if an AI-linked plan is on the table?
These actions do not depend on any American survey. They apply to any leadership team considering reducing or transforming roles because of AI.
- Pick one role covered by the plan and break it into tasks, judgments, relationships and responsibilities.
- Identify who holds the unwritten knowledge of that role today and how to record it before any change.
- Measure on a pilot perimeter the human oversight time an agent really requires.
- Write the AI usage framework for teams: what is automated, what stays human, who answers.
- Set a review point before any permanent removal, with the indicators that would tell you the plan is wrong.
The AI maturity assessment places your governance level in a few minutes, and the adopting AI in the enterprise section gathers the articles that cover each step.