AI Rebounding: Why Companies Are Rehiring the Workers They Cut

Read the list slowly and it stops looking like survey data. It starts to look like a contract.
A substantial pay rise. Guarantees about future job security, in writing. Full transparency about how artificial intelligence will be used in the role. Those are the three conditions most often named by workers who were replaced by AI and then asked to return to the same employer, according to research published by the talent assessment company SHL on 29 October 2026. Fifty-one per cent of returning workers said they would want a substantial pay increase. Forty-two per cent wanted assurances about future security. Forty per cent wanted to know exactly what the machine would be doing alongside them. Among the American respondents, CNBC reported, 30 per cent also wanted something that does not appear on any payslip: an apology or acknowledgement from leadership.
SHL has a name for the phenomenon. It calls it “AI rebounding”, and it defines it as the moment a company that replaced a person with software comes back and asks for the person. In the American half of its sample, nearly one in five workers said it had happened to them.
Most coverage of the boomerang story has looked at it from the employer's side of the desk. The SHL data records what the people on the other side now want, and the shape of those demands is a precise map of what went wrong. The pay demand says the organisation underpriced what they did. The guarantee says it broke a promise it never knew it had made. The transparency demand says it decided about their work using information they were never shown. The apology says something about dignity that no spreadsheet captures.
If firms are discovering the value of people only after removing them, the problem is not really the technology. It is that most organisations do not know, in any rigorous way, what their people actually do.
What one in five actually measures
Before leaning on a number, it is worth knowing how it was made. SHL commissioned the research firm Propeller Insights to run an online survey of 2,036 employed adults aged 18 and over: 1,029 in the United States and 1,007 in the United Kingdom. Fieldwork ran through June and July 2026. CNBC, which reported the American figures on 30 October, gave the margin of error as five per cent.
The headline is that 19 per cent of American respondents said they had personally been laid off or replaced by AI and subsequently asked back by the same employer. In the UK the figure was 13 per cent. Among workers aged 18 to 24, 24 per cent reported first-hand experience, and 36 per cent said someone close to them had been through it. Across the whole sample, 52 per cent said they had encountered AI rebounding in some form, whether personally, through someone they knew, or by hearing about it.
Three caveats matter. The data is self-reported: a respondent who says they were “replaced by AI” is reporting what they were told, not a verified account of why their role ended. The survey records an invitation to return, not a completed rehire. And SHL sells talent assessment tools, so a finding that organisations should assess talent more carefully is one it has a commercial interest in. None of that makes the numbers wrong. It is a reason to read them alongside independent data.
Even with those caveats, two numbers stand out. The first is how few people would refuse outright. Only 8 per cent of workers who had been replaced by AI said they would decline to return under any circumstances; among the Americans, CNBC put the figure at 6.6 per cent. The second is how willing people are to change rather than leave. Forty-seven per cent said that if AI significantly changed their role, they would rather be retrained for something different than lose their job.
Lucy Beaumont, SHL's global senior vice president of product and design, framed that willingness as the most surprising part of the research. “The striking thing about this research is how willing people are to prove what they can do,” she said in the release, going on to argue that employees are asking to be assessed and developed while organisations keep searching external markets for skills they may already have.
That is the vendor's thesis. But the raw numbers support a narrower point that does not depend on anyone's sales pitch. The rebounded workforce is mostly willing to return. It simply wants different terms.
A pay rise, a guarantee and a disclosure
Take the three conditions one at a time, because each says something specific.
The pay demand is the most revealing. When an organisation removes a role because it believes software can do the work, it is making a statement about price: that person's output is worth less than the machine plus whatever supervision the machine requires. Asking the person back concedes that the statement was wrong. The returning worker now knows their contribution was not substitutable at the price assumed, and a demand for substantially more money is a demand that the employer pay for its own valuation error. It is also simply rational. The worker has more information and more leverage than the first time.
Gartner, which forecasts that 30 per cent of employees laid off because AI replaced them will need to be rehired by 2029, says the rehiring will likely come at a higher cost. Tori Paulman, a vice president analyst at the firm, told Computerworld that the figure is an average across organisations of all sizes and that for large enterprises it is closer to 40 per cent. If that is right, a meaningful share of the savings booked from AI-attributed cuts will flow back out as premium salaries.
The second condition, written guarantees about future security, is not about money. It is about the reliability of the employer's word. A worker who has been through one AI-attributed redundancy has learned that their position can be removed on the strength of a forecast.
Organisational psychologists have a vocabulary for this. In 1994 Sandra Robinson and Denise Rousseau published a study in the Journal of Organizational Behavior titled “Violating the psychological contract: Not the exception but the norm”. They surveyed 128 graduate management alumni twice, once at the point of recruitment and again two years later, and found that 54.8 per cent reported that their employer had violated the unwritten set of reciprocal obligations they believed had been agreed. Violation was associated with lower trust and a greater intention to leave. Being dismissed because a machine was supposedly better, then asked back because it was not, is about as clear a breach of that unwritten deal as one can imagine. The demand for a written guarantee is an attempt to move the terms from the psychological contract into the legal one.
The third condition, transparency about how AI will be used, is a demand for information the employer should already have: which tasks the system handles, which it hands off, what its error rate is, who checks the output, and what happens to the human role if performance improves. Many organisations cannot answer with any precision, which is part of how they ended up asking people back.
Then there is the apology, a striking demand for a condition that costs nothing to grant. For a sizeable minority, the injury was not only financial. It was being told that what they did could be done by a machine, then discovering this was untrue and that nobody was going to say so.
The forecasts, the surveys and a misattributed reason
The SHL research sits on top of a growing pile of evidence about rehiring, and it is worth being careful about which piece of that evidence says what, because the numbers are being blended in circulation.
Start with Gartner, which publicised the 30 per cent forecast in September 2026. Paulman's commentary was blunt. Organisations that cut deeply, she told Computerworld, made the error of “believing that work automation was the point, when workforce amplification was the opportunity”, and more than half of Gartner's enterprise clients had been handed a savings target to meet through AI. In comments reported by CIO Dive, she said that “executives are being steeped in an environment of what I would call magical thinking”. CIO Dive also reported Gartner's analysis of more than a million layoffs in 2025, which found that cuts genuinely driven by AI productivity made up less than 1 per cent of the total.
The forecast has critics. Valence Howden, an advisory fellow at Info-Tech Research Group, told Computerworld: “I'm not sure we can substantiate those numbers, since it's much more of a guesswork statement than anything else.” He agreed AI-attributed layoffs had been excessive, but not that anyone could calculate the rate of reversal with confidence. A prediction about 2029 should be cited as one.
Next, Orgvue. The workforce planning company found that 32 per cent of organisations that cut headcount expecting AI-driven savings had since had to rehire staff. The release, distributed in March 2026, was based on a survey of 300 full-time American HR managers conducted between 11 and 25 September 2025 with Watermelon Research. The same survey found something more pointed: 23 per cent of organisations that made AI-linked layoffs admitted those decisions were based on broad expectations about automation rather than a task-level understanding of what the roles involved. Jessica Modrall, Orgvue's chief product officer, summarised the problem in a sentence: “That's not an AI problem; it's a workforce intelligence problem.”
Here is where the circulating story goes slightly wrong. The reasons now widely quoted for why organisations restore roles, that AI needed more human oversight than expected and could not replicate institutional knowledge, are frequently attributed to Orgvue. The published Orgvue findings available to this article do not include a breakdown of rehiring reasons. The clearest source for those reasons is the recruitment firm Robert Half. In a survey of 1,365 Canadian hiring managers, reported by Canadian HR Reporter on 13 May 2026, 34 per cent of those who had cut positions after implementing AI said they had since added those roles or very similar ones back. Asked why, 38 per cent said more human oversight and quality control were needed than expected. Another 38 per cent said business demand had grown. Thirty-seven per cent cited relationship management that AI could not replicate, 37 per cent said the AI tools had not been fully or consistently adopted, and 36 per cent said the roles needed institutional knowledge or context that AI lacked. Robert Half's American survey of about 2,000 hiring managers produced a similar headline: 32 per cent had eliminated a role primarily because of AI and later rehired for the same or a similar position.
The distinction matters. The reasons are close to evenly spread, and one of the top two is not a technology failure at all. It is growth. Some rebounding reflects business improving, not AI disappointing.
The Robert Half data also contains a quieter finding that explains a great deal. A companion survey of 1,005 Canadian workers found that employees using generative AI spend around a third of their working time checking the accuracy of its output and refining the final deliverable. Greg Scileppi, president of Robert Half's international talent solutions business, said “human oversight remains essential to ensure accuracy, quality and context”. If a third of the remaining staff's time goes on verifying the machine, the machine did not remove that labour. It moved it, often onto fewer shoulders.
The work that never made it into the job description
So what, precisely, is the gap between what AI can technically do and what organisations need from the people they let go?
The most useful way to see it is to stop thinking about jobs and start thinking about tasks. In 2018 Erik Brynjolfsson, Tom Mitchell and Daniel Rock published a paper in the American Economic Association's Papers and Proceedings that applied a rubric for machine learning suitability to 18,156 tasks in O*NET, the US government's occupational database. Their central finding has aged well. Most occupations contain at least some tasks suitable for machine learning. Few, if any, occupations can be fully automated by it. Tasks within a single job vary enormously in how amenable they are to automation. The authors concluded that the larger effect would be the re-engineering of processes and the reorganisation of tasks, rather than the wholesale replacement of jobs.
That finding is the technical heart of AI rebounding. A layoff is a decision about a job. An assessment of AI is, at best, an assessment of tasks. Concluding that a role can go because software performs most of its visible tasks is a category error, treating a mixed bundle as though it were uniform.
In November 2025 researchers at MIT and Oak Ridge National Laboratory released the Iceberg Index, a simulation of 151 million American workers across 923 job types. It estimated that current AI could technically perform tasks representing about 11.7 per cent of US wage value, roughly 1.2 trillion dollars, and stated plainly that it was not designed to forecast specific layoffs. That distinction, between technical capability over tasks and what happens to jobs, is the one rebounding organisations failed to draw.
And then there is the question of what the people doing the work think. In 2025 a team at Stanford led by Yijia Shao and Humishka Zope published “Future of Work with AI Agents”, which built a database called WORKBank from the responses of 1,500 workers across 104 occupations and the assessments of 52 AI experts, covering 844 tasks. They introduced a Human Agency Scale running from H1, where the AI handles a task entirely, to H5, where human involvement is essential. Workers were positive about automation for 46.1 per cent of tasks, but the dominant level of involvement they wanted, in 47 of the 104 occupations, was H3: an equal partnership. The study then compared what workers wanted with what experts judged AI could do, sorting tasks into zones of agreement and conflict.
The people inside a role are usually the best source on which tasks belong where. They know which parts are rote and which depend on context nobody wrote down. That is institutional knowledge, precisely what Robert Half's hiring managers said AI could not replicate.
The SHL survey suggests that organisations rarely ask. Only 42 per cent of respondents said their manager fully understood their skills beyond their job title; in the UK the figure was 36 per cent. Fifty-four per cent said their employer favoured external hiring over reskilling or promoting internal talent. If most managers cannot see past the title, they are deciding what to automate on the basis of the title, the least informative description of a job that exists.
Redundant one season, essential the next
What does it do to a person to be told they are surplus and then, within the same year, that they are needed?
In 1982 the social psychologist Marie Jahoda argued in “Employment and Unemployment: A Social-Psychological Analysis” that work provides not only income but latent benefits: structure to the day, social contact, shared purpose, recognised status and identity. Losing a job removes them all at once. A worker made redundant because of AI loses them on the basis of a specific claim: that a machine can supply what they supplied.
The rebound adds a second injury. Being asked back is vindication, evidence that the machine could not do the job. But it arrives as a recruitment approach, not an admission of error. The organisation is not saying “we misjudged you”. It is saying “we need you”, a statement about its own situation rather than the worker's worth. That is why the apology figure matters: it is a demand to have the first verdict formally withdrawn, not merely superseded.
Professional worth is partly a private feeling and partly a social fact, dependent on the people one works for recognising one's value. An AI-attributed redundancy declares publicly that one's contribution was replaceable. The rehire contradicts that, but rarely as publicly. The redundancy may have come with an all-hands announcement. The rehire comes with a phone call.
The age profile in the SHL data sharpens this. Nearly a quarter of respondents aged 18 to 24 said they had personally been through AI rebounding. For people at the start of a working life, a professional identity is still being assembled, and an early verdict that their work could be done by software arrives before they have accumulated the track record that might let them shrug it off. Being recalled months later does not erase that first lesson about how much an employer's judgement of them is worth.
There is also the question of what the worker now knows about the organisation. Beaumont, speaking to CNBC, said companies were realising they had “pulled the plug too quickly” on human employees, and that returning workers were “going to be a lot more demanding of what they expect from an employer”. She added: “Culture is so difficult to change, and once you've impacted it negatively, it's so difficult to recover from.” A returning worker has seen how the organisation behaves when it believes it can do without people. That shapes how much discretionary effort they offer and whether they document what they know or keep it close as insurance.
That creates an uncomfortable paradox. The value that caused the rehire was largely tacit. A rational returning worker, having been disposed of once, has every incentive to keep it that way, because writing everything down makes one easier to replace. The cycle of dismissing and rehiring may leave organisations less able to capture the knowledge they would need to automate responsibly next time. Broken trust does not simply cost goodwill. It changes how information moves.
Joseph Fuller, a professor of management practice at Harvard Business School, told CNBC that AI would create “permanent tumult” within many organisations. For workers, that means a career in which the value of their skills is periodically re-litigated, on evidence they are not shown. The demand for transparency reads as a request to be included.
Automating before looking
Does the rebound pattern show that companies are automating roles before genuinely assessing the human capability inside them? The evidence points strongly towards yes, with an important qualification.
Orgvue's finding is the most direct: nearly a quarter of organisations that made AI-linked layoffs admitted nobody had examined the work at task level before removing the people who did it.
Beaumont's own framing in the SHL release was that AI rebounding “is what happens when a company automates a role before it has genuinely assessed” the capability of the people in it, and she warned that without that assessment, “AI adoption decisions become guesswork, which is expensive and erodes trust”.
Incentives explain why the assessment so often does not happen. Fuller described the logic to CNBC: “in order to spend money on a new technology, we have to find some savings in the budget”. Headcount is the most legible line on the cost base, and Paulman's savings-target observation describes the same mechanism from the inside. When the target comes first, the assessment becomes a formality designed to reach the number.
History offers a warning. In 1997 Wayne Cascio, Clifford Young and James Morris published a study in the Academy of Management Journal examining employment-change decisions at major US corporations. They found that firms which engaged in pure employment downsizing did not show significantly higher returns than the average companies in their own industries. Firms that combined headcount reductions with restructuring of assets fared better. The AI layoff is a new version of an old bet that did not reliably pay.
The qualification is that not every rebound corrects an assessment failure. Some organisations assessed carefully and then grew. Others used AI as cover for cuts driven by other pressures, so the rebound reflects recovery rather than technological disappointment. Gartner's finding that AI-productivity-driven cuts were under 1 per cent of 2025 layoffs suggests much of what was labelled AI displacement was not. The rebound is real, but it is not a single phenomenon.
Mike Wilkes, enterprise chief information security officer at Aikido Security, put the stakes plainly to Computerworld: “If 30% of AI-driven layoffs must be reversed, that is an enormous error rate for a strategic workforce decision.” Organisations tolerate that error rate partly because its costs fall on people no longer on the payroll when it is discovered.
What a genuine assessment would involve
“Assess human capability before automating” sounds sensible and means nothing until someone specifies what it involves. The evidence above suggests six concrete components.
The first is a task inventory, not a job title review. Every role should be broken into its constituent tasks, including informal ones absent from the job description. The question for each is not “can AI do this?” but “can AI do this to the required standard, with the data actually available, and who checks it?”
The second is asking the people who do the work. WORKBank shows workers can articulate, task by task, where they want automation and where they want to stay involved. Their accounts are not neutral, since people have an interest in appearing indispensable, but they are the only route to the tacit layer, and they can be cross-checked against observation and output data.
The third is measuring oversight cost before declaring a saving. Robert Half's one-third figure is the number most automation business cases omit. If a system needs human verification, the verification is part of the system, and it needs people with enough domain knowledge to notice errors, frequently the very people being considered for redundancy.
The fourth is parallel running. Engineers do not decommission a working system on the strength of a demonstration of its replacement. They run both, compare outputs, and switch only when the new one has proved itself. Applied to AI and labour, that means operating the system alongside the existing team, measuring error rates, escalations and customer outcomes, and deciding headcount on the evidence. It delays savings. It could also head off a good share of the expensive rehiring that Gartner forecasts, because the failures would surface while the people who could fix them were still employed.
The fifth is treating undocumentable knowledge as a finding rather than an obstacle. If a role holds knowledge that cannot be written down or transferred to a system, such as the history of a client relationship or the judgement about when an exception is genuinely exceptional, then it is not yet a candidate for elimination. An assessment that keeps hitting that wall is telling the organisation something true about where the value sits.
The sixth is redeployment as the default. Forty-seven per cent of SHL's respondents would rather retrain than lose their job. An organisation that automates tasks and redeploys the people who did them keeps the knowledge, avoids the rehiring premium and keeps the psychological contract intact. It forgoes the immediate headcount saving, and that trade-off should be made explicitly.
A seventh point sits across all six: the assessment has to include the people on the receiving end of the work, not only those doing it. Paulman, in the comments reported by CIO Dive, observed that “customers just don't really like talking with AI, particularly over voice”. A task can be technically automatable and still be one that customers, patients or clients will not accept from a machine. No amount of internal analysis captures that without asking them.
None of this is exotic. The barrier is not knowledge but sequence. When the savings target precedes the assessment, the assessment becomes a search for justification.
Where the law already asks the question
Some legal systems already require employers to pause and look before they cut, at least in principle.
In the United Kingdom, section 188 of the Trade Union and Labour Relations (Consolidation) Act 1992 requires employers proposing 20 or more redundancies at one establishment within 90 days to consult with employee representatives. Where 100 or more dismissals are proposed, consultation must begin at least 45 days before the first dismissal takes effect; for 20 to 99, at least 30 days. The consultation must cover ways of avoiding the dismissals, reducing the numbers affected, and mitigating the consequences, and it must happen before final decisions are made. Applied seriously, that is a statutory window in which to ask whether the technology really can do the work. The weakness is that consultation concerns process, not evidence. Nothing requires an employer to show the AI performs as claimed.
Germany goes further on the technology itself. The Works Council Modernisation Act, which came into force on 18 June 2021, amended section 80(3) of the Works Constitution Act so that a works council assessing the introduction or use of artificial intelligence can bring in an outside expert without having to prove that the expert is necessary. The same law strengthened information and consultation rights over AI in work processes. In effect, it gives employee representatives a statutory right to their own technical assessment of what the system can do, before the consequences land.
The United States offers a contrasting example of disclosure without assessment. In March 2025, following a directive from Governor Kathy Hochul, the New York State Department of Labor added a checkbox to its WARN notice filings asking whether “technological innovation or automation” contributed to the job losses being reported, and requiring employers who tick it to name the technology. According to an analysis by the law firm Hunton Andrews Kurth published in May 2026, more than 160 companies had filed WARN notices in the system's first year, and not a single one had attributed layoffs to AI or automation. Disclosure after the decision, with no penalty attached, is not assessment before it.
Taken together, the three approaches point towards a principle. If an employer dismisses people because a technology can do their work, that is a factual claim, and factual claims can be tested. The UK duty could require the evidence. The German model shows workers can be given the means to scrutinise it. New York shows what happens when the question is asked but nobody has to answer it.
The terms of the second contract
Return to the SHL list. A pay rise, a guarantee, a disclosure and, for nearly a third of American respondents, an apology. Each is a demand for something the first decision withheld.
The pay rise is the price of the information the employer lacked: what the person was actually worth. The guarantee is the price of the trust the employer spent: the assumption that it would not dispose of people on a forecast. The disclosure is the price of the visibility it never offered: an account of what the machine does and does not do. The apology is the price of the dignity it took: the public verdict that a person's work was replaceable, now quietly contradicted.
Organisations that want to avoid paying those prices a second time can pay for the information the first time. That means a task-level inventory built with the people who do the work, an honest measure of the oversight AI requires, a period of parallel running before anyone is dismissed, and redeployment as the default rather than the exception. It also means giving workers in advance the transparency that 40 per cent of SHL's returning workers now demand as a condition of coming back. None of it is free. All of it is likely to cost less than a 30 per cent error rate.
The deeper lesson is not that the technology failed. Often the software performs exactly as specified. The failure is in the specification. Organisations described their own work too thinly, through titles and process maps that omitted the oversight, relationships, context and judgement that made it function. Then they automated the thin description and discovered the thick reality only when it was gone.
The workers being asked back know this better than anyone. Their demands are not unreasonable. They are an itemised invoice for a valuation that should have been done before the redundancy letter, not after it.
References & Sources
- SHL. “Workers Replaced by AI Would Demand Higher Pay and Greater Job Security to Return, SHL Research Finds.” Business Wire press release, 29 October 2026 (syndicated via Yahoo Finance). Survey conducted by Propeller Insights, 2,036 employed adults (1,029 US, 1,007 UK), June to July 2026. https://finance.yahoo.com/technology/ai/articles/workers-replaced-ai-demand-higher-120200927.html
- Caldwell, Sophie. “AI-driven layoffs are creating a new class of boomerang employees: companies are realizing they 'pulled the plug too quickly'.” CNBC Make It, 30 October 2026. https://www.cnbc.com/make-it/2026/10/30/ai-rebounding-employees.html
- Sugue, Mara. “SHL survey finds AI-replaced workers are being rehired.” CFOtech US, 29 October 2026. https://cfotech.news/story/shl-survey-finds-ai-replaced-workers-are-being-rehired
- HR Tech Edge. “SHL Study: AI Rebounding Challenges Workforce Planning.” October 2026. https://hrtechedge.com/hr/shl-finds-ai-replacements-are-being-asked-to-return/
- Schuman, Evan. “Layoff remorse: Gartner says at least one in three positions eliminated by AI will be restored by 2029, at a higher cost.” CIO (originally Computerworld), 9 September 2026. https://www.cio.com/article/4220486/layoff-remorse-gartner-says-at-least-one-in-three-positions-eliminated-by-ai-will-be-restored-by-2029-at-a-higher-cost-2.html
- Sharma, Nidhi. “One-third of AI-replaced workers will be rehired by 2029: Gartner.” CIO Dive, 16 September 2026. https://www.ciodive.com/news/AI-replaced-workers-rehired-gartner/830426/
- Gartner. “Gartner Identifies Four Shifts Shaping the Future of Work.” Press release, Stamford, Connecticut, 9 September 2026. https://www.gartner.com/en/newsroom/press-releases/2026-09-09-gartner-identifies-four-shifts-shaping-the-future-of-work
- HRD Canada. “Employers will need to rehire 30% of workers affected by AI layoffs, at higher cost: report.” 2026. https://www.hcamag.com/ca/specialization/transformation/employers-will-need-to-rehire-30-of-workers-affected-by-ai-layoffs-at-higher-cost-report/589966
- Orgvue. “New Research Exposes the Complexity of Deploying AI Systems in the Workforce.” PR Newswire, March 2026, as published by TechIntelPro (survey of 300 US HR managers, Watermelon Research, 11 to 25 September 2025). https://techintelpro.com/news/hr/ai/orgvue-research-32-rehire-after-ai-driven-layoffs
- HR Tech Edge. “AI Layoffs Backfire: 32% of Companies Forced to Rehire After Misjudging Automation.” 13 March 2026. https://hrtechedge.com/ai-layoffs-backfire-32-of-companies-forced-to-rehire-after-misjudging-automation/
- Sellers, Matthew. “'AI correction hires' on the rise as Canadian employers reverse course.” Canadian HR Reporter, 13 May 2026. https://www.hrreporter.com/focus-areas/automation-ai/ai-correction-hires-on-the-rise-as-canadian-employers-reverse-course/394432
- Tiger, Jake. “UK could follow US, Canada AI rehiring wave.” Staffing Industry Analysts, 11 August 2026. https://www.staffingindustry.com/news/global-daily-news/uk-could-follow-us-canada-ai-rehiring-wave
- Cruz, Stephanie. “AI Layoffs Backfire as 32% of Bosses Rehire Roles They Thought Robots Could Do.” International Business Times UK, 2 July 2026. https://www.ibtimes.co.uk/ai-layoffs-backfire-companies-rehire-staff-1806357
- Robinson, Sandra L., and Denise M. Rousseau. “Violating the psychological contract: Not the exception but the norm.” Journal of Organizational Behavior, vol. 15, no. 3, 1994, pp. 245 to 259. DOI: 10.1002/job.4030150306.
- Brynjolfsson, Erik, Tom Mitchell and Daniel Rock. “What Can Machines Learn, and What Does It Mean for Occupations and the Economy?” AEA Papers and Proceedings, vol. 108, 2018, pp. 43 to 47. DOI: 10.1257/pandp.20181019. https://www.aeaweb.org/doi/10.1257/pandp.20181019
- Shao, Yijia, Humishka Zope, et al. “Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce.” arXiv:2506.06576, Stanford University, 2025. https://arxiv.org/abs/2506.06576 ; project page: https://futureofwork.saltlab.stanford.edu
- Smith, Dave. “MIT report: AI can already replace nearly 12% of the U.S. workforce.” Fortune, 27 November 2025, reporting the MIT and Oak Ridge National Laboratory Iceberg Index. https://fortune.com/2025/11/27/mit-report-ai-can-already-replace-nearly-12-of-the-us-workforce/
- Jahoda, Marie. Employment and Unemployment: A Social-Psychological Analysis. Cambridge: Cambridge University Press, 1982.
- Cascio, Wayne F., Clifford E. Young and James R. Morris. “Financial consequences of employment-change decisions in major U.S. corporations.” Academy of Management Journal, vol. 40, no. 5, 1997, pp. 1175 to 1189. DOI: 10.5465/256931.
- Trade Union and Labour Relations (Consolidation) Act 1992, section 188. legislation.gov.uk. https://www.legislation.gov.uk/ukpga/1992/52/section/188
- Working Families. “Collective consultation: when your employer proposes to dismiss over 20 employees.” https://workingfamilies.org.uk/articles/collective-consultation-when-your-employer-dismisses-over-20-employees/
- Luther Rechtsanwaltsgesellschaft. “Newsletter Arbeitsrecht 04/24” (on section 80(3) BetrVG and AI experts). 2024. https://luther-lawfirm.com/fileadmin/user_upload/NL_Arbeitsrecht_04-24_DE.pdf
- Der Betrieb. “Betriebsrätemodernisierungsgesetz verabschiedet: ein Überblick.” 2021. https://der-betrieb.de/meldungen/betriebsraetemodernisierungsgesetz-verabschiedet-ein-ueberblick
- Quackenboss, Robert T., and Michelle Meyer. “New York WARN Act: No AI-Related Layoffs Reported in First Year of Adding AI-Related Disclosure to the System.” Hunton Andrews Kurth, Hunton Employment and Labor Perspectives, 18 May 2026. https://www.hunton.com/hunton-employment-labor-perspectives/new-york-warn-act-no-ai-related-layoffs-reported-in-first-year-of-adding-ai-related-disclosure-to-the-system
- Kaufman Dolowich. “New York Amends WARN Act to Require Disclosure of AI-Related Layoffs.” 4 August 2025. https://www.kaufmandolowich.com/news-resources/new-york-amends-warn-act-to-require-disclosure-of-ai-related-layoffs-by-keith-j-gutstein-esq-and-shiddhartha-uddin-esq-8-4-2025/

Tim Green UK-based Systems Theorist & Independent Technology Writer
Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.
His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.
ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk
Listen to the free weekly SmarterArticles Podcast








