Automating the Joy Out of Work: Why Workers Need a Binding Say

For as long as automation has had a sales pitch, the pitch has been the same. The machines will take the drudgery. They will fill in the forms, file the reports, chase the invoices and reconcile the spreadsheets, and in doing so they will hand back to human beings the parts of work that are interesting, creative and their own. Strategy, design, ideas. The good stuff.

A paper presented at the 2026 CHI Conference on Human Factors in Computing Systems, held in Barcelona from 13 to 17 April, suggests that the pitch may have the order of events backwards. Its title asks the question bluntly: “Are We Automating the Joy Out of Work? Designing AI to Augment Work, Not Meaning”. Its authors are Jaspreet Ranjit and Swabha Swayamdipta of the University of Southern California, Ke Zhou of Nokia Bell Labs and the University of Nottingham, and Daniele Quercia of Nokia Bell Labs and the Politecnico di Torino. Their central finding, in the words of the abstract, is this: “Worryingly, we find that tasks that workers associate with a sense of agency or happiness may be disproportionately exposed to AI.”

Not the tedious tasks. The ones people like.

The authors go further. A central risk of AI, they argue, is not mass unemployment but what they call mass demoralisation: “a loss of meaning and ownership in day-to-day work”. As models generate more of the early-stage output, they write, the visible creative steps can start to feel machine-made, while the human contribution becomes rushed, under-resourced and difficult to see.

That is a different kind of warning from the one that has dominated the debate about AI and jobs. It is not about whether people will have work. It is about what work will be like for the people who keep it. And it raises two questions that most organisations deploying AI have not yet seriously asked. What happens to a person's relationship with their craft when the parts of the job they found most fulfilling are the first to be handed to a machine, and what is left is the emotional and relational upkeep? And if the danger is demoralisation rather than displacement, who should decide which parts of a job get automated?

How the researchers mapped meaning onto ten thousand tasks

The study starts with the US Department of Labor's Occupational Information Network, known as O*NET, which breaks every American occupation into short task statements. The researchers narrowed this universe to tasks performed mainly on a computer, carried out daily or weekly, and ones that workers on the research platform Prolific were highly familiar with. That produced a representative sample of 171 tasks across 22 occupations.

They then surveyed two groups. The first was 202 workers who actually performed those tasks. Each rated their tasks against dimensions drawn from a scoping review of research on meaningful work: whether a task felt pointless or valuable, whether it called for novel ideas, gave them freedom or made them happy, and whether it built relationships or required emotional awareness or in-person contact. The second group was 197 developers who build AI systems, who were asked what traits they would design into an AI tool for each task.

To scale the analysis beyond 171 tasks, the team prompted language models to simulate the ratings, checked how closely the simulated ratings matched the human ones, and, finding reasonable agreement, generated annotations for 10,131 tasks across 512 occupations and 19 sectors. They report that only 0.87 per cent of tasks showed a large gap, of two points or more on a five-point scale, between model and human ratings, and that their main patterns held when they reran the analysis using only the human ratings.

To decide which tasks counted as “exposed” to AI, the researchers used a patent-based measure developed by Ali Akbar Septiandri, Marios Constantinides and Quercia, published in PNAS Nexus in 2024, which matched descriptions of tens of thousands of AI patents filed with the US Patent and Trademark Office against O*NET task statements. Tasks above a threshold on that measure were classed as likely to be exposed. Restricting the analysis to computer-based work left 3,179 tasks in the likely-to-be-exposed group and 2,349 in the comparison group.

The parts of the job that machines reach first

The comparison between those two groups is where the paper becomes uncomfortable. Tasks likely to be exposed to AI were more strongly associated by workers with novelty and creativity, with positive feelings and with autonomy. The researchers say directly that this “challenges the usual narrative that automation mainly targets routine tasks”. Tasks less likely to be exposed were associated with emotional awareness, in-person interaction, relationship building and social connection.

The sector picture sharpens the point. The heaviest exposure of creative, high-agency work fell in Arts, Architecture and Engineering, Computer and Mathematics, and the Life, Physical and Social Sciences. The examples the paper draws from O*NET are telling: an art director “formulating basic layout design”; an actuary “constructing probability tables for natural disasters”. These are first passes, the authors note, of the kind that models can now credibly generate, with humans refining the output afterwards. Meanwhile, in Education, Sales, Community and Social Services and Healthcare, the work that remained less exposed involved activities such as guiding students through tangled academic and personal problems, or “presenting offers while preserving relationships”.

The authors read this as a reversal of the expected trajectory. Generative systems, they observe, already draft text, suggest layouts, start campaigns, run simulations and infer likely emotional responses. In their data, these uses line up with tasks that workers describe as meaningful precisely because they reflect taste and authorship. AI is not confined to chores. It is becoming entangled with the moments where people feel they add something of themselves.

There is a second finding that matters for anyone building the tools. Workers and developers disagreed about what an AI system should be like. Workers wanted systems that were straightforward, tolerant and practical. Developers said they would design systems that were polite, strict and imaginative. The single biggest gap was between straightforwardness and politeness, and it was widest in Production, Architecture and Engineering, and the sciences, where vagueness can mean wasted materials or flawed data. “In high-risk contexts,” the authors write, “the ethical choice may well be the blunt one, and the useful thing to say may be short.” Developers, they suggest, tend to romanticise creativity as a frontier capability, while many workers in structured, routine jobs simply want a tool that spots anomalies and sticks to the rules.

Put the two findings together and the problem comes into focus. The AI that is arriving is being built to be imaginative, and it is landing on the imaginative parts of the job.

Why exposure is not the same as replacement

It is worth being precise about what the study measures. “Exposure” does not mean a task has been automated, or even that it will be. The authors define it as tasks that current or near-term AI could plausibly perform or substantially speed up. It is a statement about technological reach, not about deployment decisions inside any particular firm, and the paper separates automation, where AI performs a task end to end, from augmentation, where humans keep primary responsibility.

The researchers are also candid about their limitations. The study covers only US occupational tasks. Prolific participants tend to be more technologically literate than average, so low-wage and non-digital work may be underrepresented. O*NET task statements are short summaries that different workers may picture differently. The scaling to 10,131 tasks relies on language-model ratings, which the authors treat as an approximation of human responses rather than as ground truth, and which they found to be more uniform and optimistic in some sectors. Even the creativity finding has a wrinkle. When the analysis was rerun on human ratings of the original 171 tasks alone, humans and models agreed on six of the seven significant dimensions; the exception was “requires novel ideas or creativity”, where the authors suggest human annotators tended to underestimate the novelty of tasks.

None of this undoes the core pattern. But the study is best read as a map of where pressure is likely to fall, not a ledger of what has already been lost. What actually happens to people's work is decided later, by managers, vendors and product teams.

From mass unemployment to mass demoralisation

Why should the location of the pressure matter so much? Because meaning in work is not spread evenly across a job. It clusters.

The sociologist Richard Sennett, in his 2008 book The Craftsman, described craftsmanship as the basic human impulse to do a job well for its own sake, and argued that this impulse extends far beyond the workshop, to programmers, doctors and artists. Craft, in that sense, lives in the moments of making: the first sketch, the opening paragraph, the elegant solution to a modelling problem.

If those moments are exactly where AI exposure is concentrated, then a job can be automated in a peculiar way. The worker keeps the job title, the salary and most of the hours. What disappears is the part that made them want the job in the first place. In the paper's own examples, the basic layout and the first probability table are precisely the drafts a model can now produce; what is left to the human is refinement. This is the scenario Ranjit and her colleagues have in mind when they describe a loss of ownership. The creative steps that others can see now appear to come from the machine, while the human's contribution moves into the background, squeezed for time and harder for anyone to credit.

There is already experimental evidence that this dynamic has psychological costs. In a 2025 paper in Scientific Reports, Suqing Wu, Yukun Liu, Mengqi Ruan, Siyu Chen and Xiao-Yun Xie of the School of Management at Zhejiang University reported four online experiments with a total of 3,562 participants. Working with generative AI improved immediate performance on professional tasks. But that performance boost did not carry over when people subsequently worked alone, and the switch from AI collaboration to solo work was accompanied by significant decreases in intrinsic motivation and increases in boredom, even as participants' sense of control rose. The authors' summary is succinct: collaboration with generative AI “enhances immediate task performance but can undermine long-term psychological experiences of human workers.”

Creative workers are already reporting effects outside the laboratory. A survey by the Society of Authors, the UK trade union for writers, illustrators and translators, received 787 responses in early 2024. A quarter of illustrators, 26 per cent, and more than a third of translators, 36 per cent, said they had already lost work to generative AI. The survey was self-reported and drawn from a membership sample, so it cannot measure the market as a whole. But it captures the sense, among people whose identities are bound up with a craft, that the craft itself is being lifted out of their hands.

What the remaining work asks of people

If the creative core of a job is thinned out, what is left? According to the CHI study, the tasks least exposed to AI are those built on emotional awareness, in-person presence and relationship building. That might sound like a reassuring residue. Machines draft, humans connect. But it is worth looking at what those remaining tasks involve.

In 1983, the sociologist Arlie Russell Hochschild of the University of California, Berkeley, published The Managed Heart: Commercialization of Human Feeling, which introduced the idea of emotional labour. Studying flight attendants and bill collectors, she described public-facing jobs in which managing feelings was part of what the employer was paying for. Flight attendants had to be, in the phrase used by her publisher, “nicer than natural”; bill collectors had to be “nastier than natural”. Hochschild's concern was that this kind of work, done for commercial purposes, could estrange people from their own feelings.

Relational work is not inherently draining. For many teachers, nurses, social workers and salespeople, the relationships are the point of the job, and the CHI data suggest that workers do regard these tasks as meaningful. The danger lies in the mix. A job that once combined making things with tending to people can become one that is mostly tending to people, with the making delegated. The remaining human work then has a distinctive profile: smoothing over the AI's output with clients, absorbing complaints, reassuring colleagues, maintaining relationships that the automated parts of the workflow no longer support. That is emotional labour, and much of it is invisible in the way that a finished design or a well-built model is not.

The Stanford University researchers behind the WORKBank database, Yijia Shao, Humishka Zope, Yucheng Jiang, Jiaxin Pei, David Nguyen, Erik Brynjolfsson and Diyi Yang, found a related signal in a 2025 study of 1,500 workers across 844 tasks in 104 occupations. Their results, they wrote, offer early signals that AI agents may reshape core human competencies, “shifting from information-focused skills to interpersonal ones.” The CHI paper adds a warning that this shift may not be benign. Among its design examples is the risk that teachers, managers and clinicians who use AI for canned replies and templated feedback end up making their communication with students, staff or patients feel more generic. If AI pushes people towards relational work while also hollowing out that relational work with templates, the worker is squeezed from both sides.

When the most fulfilling parts of a job are automated first, then, a person's relationship with their craft is put at risk in two ways. They lose authorship of the work that once expressed their skill and taste. And they are left with a larger share of work whose value is real but less visible and more emotionally costly. Whether that produces demoralisation depends on what else changes, and here a very different body of evidence becomes useful.

Lessons from people who stopped working

In September 2026, four researchers at Google DeepMind, Stephanie C.Y. Chan, Adam Bales, Katherine L. Hermann and Iason Gabriel, posted a review titled “Work, Wellbeing, and Choice: Empirical Lessons for AI Futures”. Their question was not what AI does to work, but what decades of research reveal about people who live without it. They drew on psychological, sociological and economic studies of the unemployed, retirees, lottery winners, financially dependent spouses and citizens of Gulf states with guaranteed public-sector employment, covering OECD countries, China, India and the Gulf.

Their findings are stark in places. Involuntary job loss harms life satisfaction, mental health, physical health and even mortality, and the damage can last: some studies find psychological “scarring” that persists for years after people find work again. Yet the vast majority of lottery winners, between 85 and 89 per cent across studies of American and Swedish winners, keep working, typically using their windfall to buy flexibility rather than to quit. And retirees, who are also not working, often report high and sustained wellbeing.

The authors draw on the work of the social psychologist Marie Jahoda, whose 1982 book Employment and Unemployment, published by Cambridge University Press, argued that a job provides “latent” benefits alongside its “manifest” purpose of earning a living. Jahoda identified time structure, social contact, status and identity, regular activity and collective purpose. The DeepMind review finds that both income and some of these latent benefits matter, and adds cognitive health: retirement has been causally linked to declines in cognitive ability.

To explain why the unemployed and the retired fare so differently, the review identifies three factors that mediate the relationship between work status and wellbeing. The first is agency and choice: whether the exit from work was voluntary, and whether people retain agency over the long term. The second is the availability of alternative sources of work's latent benefits, such as volunteering, hobbies or state-provided employment. The third is social and systemic context, including cultural norms around work and the strength of safety nets.

The evidence on each is striking. Between 20 and 30 per cent of retirees perceive their retirement as forced, and across countries those who retired voluntarily report significantly better physical and mental wellbeing than those who did not, even after controlling for health, income and other factors. Volunteering confers benefits at remarkably low doses, with gains for older adults observable from about one hour a week. One study cited in the review found that plant closures produced an almost four-fold greater increase in depression among American workers than among European workers, who have stronger safety nets. And the psychological cost of losing work tends to fall harder on men, which the authors attribute to stronger social pressure to conform to norms about working.

“Many policy discussions about AI futures focus on the availability of work or the availability of resources,” the authors write. “However, these discussions overlook the fact that it matters deeply whether people get to choose their path.” Their summary of the policy lesson is equally direct: purely material interventions such as universal basic income “are valuable but highly insufficient”.

Applying the exit evidence inside the job

The DeepMind review is about people who leave work altogether. But its three factors translate remarkably well to what happens when people stay in work and lose part of it. Demoralisation, in the CHI paper's sense, is a kind of partial exit: a withdrawal not from employment but from the meaningful core of a role. Read that way, the review's factors become a diagnostic for when task automation is likely to hurt.

Start with agency. A critical determinant of wellbeing in the review is not whether people work, but whether they chose their situation. Inside a job, the equivalent question is whether workers chose which of their tasks were handed to AI. The Society of Authors survey hints at the distinction: alongside those who chose to use generative AI, 5 per cent of illustrators and 8 per cent of translators said they had used it because their publisher or commissioning organisation asked them to. The CHI paper notes that, in practice, managers, vendors and technical teams, not workers, usually decide which tasks are exposed to AI and how far automation goes. “Managers, vendors, and technical teams make these decisions before workers can consent or take part, especially in roles with low bargaining power,” the authors write. That is the workplace version of involuntary exit.

Next, alternatives. Retirees do well when they find other sources of structure, status and purpose. Workers who lose the creative core of their job need somewhere else for those things to come from: new responsibilities with real authorship, protected time for the work they value, recognition for the judgement involved in steering and correcting AI output. If the time freed by automation is simply filled with more volume, the latent benefits drain away without replacement.

Finally, context and safety nets. The review shows that the same loss lands very differently depending on norms and institutional support. Inside an organisation, the analogue is culture and protection: whether people who push back on a deployment are heard or penalised, whether those whose roles shrink are offered retraining or redeployment, and whether the organisation measures anything other than throughput.

This is an analogy, not a finding. Neither paper tested it, and evidence about the unemployed cannot be transplanted wholesale onto people still in work. But the convergence is hard to ignore. Working from inside jobs, one paper finds that agency, ownership and meaning matter; working from outside, the other finds that choice, alternatives and context matter. Both suggest the damage depends less on what is taken away than on how.

Workers already say what they want automated

The good news is that workers are not, in general, opposed to automation. They are opposed to the wrong automation.

The WORKBank study asked its 1,500 workers which of their tasks they would like AI agents to take on. For 46.1 per cent of tasks, workers expressed positive attitudes towards automation, even after being prompted to reflect on job loss and on how much they enjoyed the work. Their main motivation was to free up time for higher-value work. The researchers then compared those desires with experts' assessments of what AI could currently do, sorting tasks into four zones: a “Green Light” zone where workers wanted automation and the technology could deliver it; a “Red Light” zone where the technology was capable but workers did not want it; an R&D opportunity zone where workers wanted help that the technology could not yet provide; and a low-priority zone.

Their most pointed finding concerned where the money was going. Mapping the tasks targeted by Y Combinator start-ups, they found that 41.0 per cent of company-task mappings were concentrated in the Low Priority and “Red Light” zones. A large share of the industry's effort, in other words, was aimed at tasks workers either did not care about or actively wanted to keep.

Combined with the CHI paper's developer-worker mismatch, the pattern is clear. Workers do not resist AI. Their preferences are simply rarely the input that shapes what gets built.

A test for choosing which task clusters to automate

If the risk is demoralisation, then the usual test for automation, which asks whether a task can be done faster or cheaper by a machine, is incomplete. It needs a second question: what does this task mean to the people who do it? Drawing on the two papers and the evidence around them, an organisation could reasonably weigh at least four considerations before automating a cluster of tasks.

The first is meaning. Ask the people who do the work which tasks they find meaningful, enjoyable or expressive of their skill, and which they find pointless. The CHI study shows that this can be measured at scale with straightforward survey instruments. WORKBank shows that workers can distinguish clearly between tasks they want help with and tasks they want to keep. Tasks that workers rate low on meaning and high on tedium are the obvious first candidates. Tasks rated high on creativity, autonomy and happiness deserve a heavier burden of justification.

The second is order. The CHI authors propose that, for creative work, the sequence matters as much as the division of labour. Their design heuristics include starting creative fields empty, so that a human sketch, prompt or outline exists before AI suggestions appear; offering several alternatives rather than a single finished replacement; letting workers accept or reject AI content sentence by sentence rather than through a single “replace all”; and allowing a worker to tick a box marked “I want to do this part myself”. The question for an organisation is not simply whether AI touches a creative task, but whether the human still makes the first move.

The third is what replaces the time saved. The DeepMind review's evidence on alternative sources of meaning suggests that automation which frees time is only beneficial if that time goes somewhere that sustains structure, status and purpose. For relational work, the CHI paper suggests using AI to prepare briefs and summaries so the human arrives better informed, assigning it to silent roles such as note-taking during live interactions, and reserving protected time for direct human engagement. Its proposed measures of success are telling: workers reporting that AI helps them “show up prepared” rather than “speak for me”, and that it does not “take away the good parts” of the job.

The fourth is measurement. The CHI authors recommend tracking, during deployment, whether workers still feel in control and whether the output “still feels like my work”, alongside conventional measures of speed and quality. They also suggest logging the level of assistance AI provides, from suggestion to draft to execution, and recording who made the final decision. Over time, they argue, workers' self-reported skills and confidence in their core craft should remain stable or improve. If those indicators fall while productivity rises, the organisation has learned something its dashboards would otherwise hide.

None of this means creative tasks should never be automated. Some workers will welcome help with the very tasks others prize, and the CHI data show wide variation between sectors. The point is that meaning should be treated as a real cost in the decision, not as a sentimental afterthought.

Giving workers a binding say

That leaves the second half of the question: what voice should workers have? The CHI paper's answer is explicit. Beyond designing interfaces that let individuals retain control within a task, the authors argue that organisations should give workers some say over which task clusters are candidates for automation. Without such mechanisms, they warn, “AI exposure risks deepening existing power imbalances.” They point to remedies from research on algorithmic management: direct involvement of workers and their representatives through formal consultation and collective bargaining over AI deployments, worker or union technology representatives with access to information about how systems allocate and evaluate work, and transparency tools that help workers understand and contest automated decisions.

There are working precedents. The most prominent came from Hollywood. The 2023 Minimum Basic Agreement between the Writers Guild of America and the major studios established that AI cannot be considered a writer and that material it produces cannot be treated as literary material. Under its terms, a writer may choose to use AI with the company's consent, but the guild's own summary states plainly that “the company can't require the writer to use AI software (e.g., ChatGPT) when performing writing services.” Companies must disclose when material given to a writer was generated by AI. In effect, the writers bargained over which parts of their craft would remain theirs. That agreement has since been succeeded by the 2026 contract, ratified on 24 April 2026 with 90.38% of votes in favour and running to 1 May 2030, which “preserves all of the protections negotiated in 2023”. It adds one more: companies must give the guild written notice when they license writers' work to train a commercial generative AI system, after which the guild can request a discussion about the licence, including any payment to writers. It does not establish payment for training.

European law offers a thinner but broader floor. Article 26(7) of the EU Artificial Intelligence Act, adopted in 2024, requires that, before a high-risk AI system is put into service or used at the workplace, “deployers who are employers shall inform workers' representatives and the affected workers” that they will be subject to it. That is a duty to inform, not to consult or obtain consent, and it applies only to systems classed as high-risk. It is also not yet in force for most workplace tools: the Digital Omnibus on AI, Regulation (EU) 2026/1744, which took effect on 27 July 2026, deferred the high-risk obligations for Annex III systems, a category that includes employment and worker management, from 2 August 2026 to 2 December 2027. In Germany, works councils have long held co-determination rights over the introduction of technical systems capable of monitoring employees' behaviour or performance, a provision that lawyers widely expect to cover many AI tools. The limits are visible too: in January 2024, the Hamburg Labour Court rejected a works council's application concerning employees' use of ChatGPT, partly because staff used it voluntarily through private accounts in a web browser rather than as an employer-installed system.

These mechanisms share a weakness when it comes to demoralisation. Most were designed to address surveillance, pay or credit, not meaning. The EU provision is triggered by risk classification, not by what a task means to the person who does it. The German right is anchored in monitoring. Even the WGA's protections, including the 2026 notice requirement on training licences, focus on credit and compensation, although by keeping AI out of the definition of a writer, they also protect authorship. A workplace that took the CHI findings seriously would go further: it would give workers a formal role in deciding which task clusters are automated, informed by evidence about which tasks they find meaningful, with a right to revisit the decision once the effects are known.

That voice matters most precisely where it is weakest. The CHI authors note that the decisions about exposure are made before workers can consent, especially in roles with little bargaining power. The DeepMind review cites evidence that, where safety nets are weak, pre-existing wealth becomes a key buffer against the shock of job loss. A model of AI deployment that consults only those with strong unions would protect screenwriters and leave behind most of the people whose tasks appear in the CHI data.

What a craft needs to survive

It is tempting to treat all of this as a soft concern, a matter of morale to be fixed with a workshop. The evidence suggests otherwise. Meaning is not a perk attached to work; it is part of what makes people willing to invest skill and care, and part of what keeps them well. The DeepMind review shows how much is at stake when people lose work altogether. The CHI paper suggests a quieter version of that loss may already be underway inside jobs that look perfectly secure.

The two papers point to the same remedy from opposite directions. What protects people is not the absence of change but the ability to shape it: to choose, to find meaning somewhere else when the old sources move, and to be supported while they do. For organisations, that means treating which tasks to automate as a decision with human costs, made with the people who bear them, rather than a technical question settled by whatever a vendor's model can do.

Ranjit and her co-authors end on a note of possibility. The outcome they describe, they stress, is shaped by design choices, which can in principle be revised: systems could speed up exploration and drafting while people keep authorship, make the final judgements and sustain the relationships that make work meaningful. That is a more humane division of labour than the current pitch implies. But it will not arrive by default. Left to the logic of capability alone, AI will keep flowing towards the tasks it can reach, and those turn out to be the ones people love. Keeping the spark in work will take a deliberate decision about who gets to say which parts of a craft stay human.

References and Sources

  1. Ranjit, Jaspreet; Zhou, Ke; Swayamdipta, Swabha; Quercia, Daniele. “Are We Automating the Joy Out of Work? Designing AI to Augment Work, Not Meaning.” Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26), Barcelona, Spain, 13-17 April 2026. DOI: 10.1145/3772318.3791845. Preprint arXiv:2603.14963, 16 March 2026. https://arxiv.org/abs/2603.14963 (full text: https://arxiv.org/html/2603.14963)
  2. Chan, Stephanie C.Y.; Bales, Adam; Hermann, Katherine L.; Gabriel, Iason. “Work, Wellbeing, and Choice: Empirical Lessons for AI Futures.” Google DeepMind. arXiv:2609.11019, 10 September 2026. https://arxiv.org/abs/2609.11019 (full text: https://arxiv.org/html/2609.11019)
  3. Septiandri, Ali Akbar; Constantinides, Marios; Quercia, Daniele. “The potential impact of AI innovations on US occupations.” PNAS Nexus, 3(9), pgae320, September 2024. DOI: 10.1093/pnasnexus/pgae320. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11421150/
  4. Wu, Suqing; Liu, Yukun; Ruan, Mengqi; Chen, Siyu; Xie, Xiao-Yun. “Human-generative AI collaboration enhances task performance but undermines human's intrinsic motivation.” Scientific Reports, 15, article 15105, 29 April 2025. DOI: 10.1038/s41598-025-98385-2. https://www.nature.com/articles/s41598-025-98385-2
  5. Shao, Yijia; Zope, Humishka; Jiang, Yucheng; Pei, Jiaxin; Nguyen, David; Brynjolfsson, Erik; Yang, Diyi. “Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce.” arXiv:2506.06576, 6 June 2025 (revised 1 February 2026). https://arxiv.org/abs/2506.06576
  6. Hochschild, Arlie Russell. The Managed Heart: Commercialization of Human Feeling. University of California Press, 1983 (third edition 2012). https://www.ucpress.edu/books/the-managed-heart/paper
  7. Jahoda, Marie. Employment and Unemployment: A Social-Psychological Analysis. Cambridge University Press, 1982. ISBN 978-0-521-28586-5. https://ideas.repec.org/b/cup/cbooks/9780521285865.html
  8. Sennett, Richard. The Craftsman. Yale University Press, 2008. https://yalebooks.yale.edu/book/9780300151190/the-craftsman/
  9. The Bookseller. “A third of translators report losing work to generative AI systems, SoA survey reveals.” April 2024. https://thebookseller.com/news/a-third-of-translators-report-losing-work-to-generative-ai-systems-soa-survey-reveals
  10. European Writers' Council. “SoA survey UK AI 2024.” April 2024. https://europeanwriterscouncil.eu/soa-survey-uk-ai-2024/
  11. Books+Publishing. “A third of translators report losing work due to AI.” 15 April 2024. https://www.booksandpublishing.com.au/articles/2024/04/15/250317/a-third-of-translators-report-losing-work-due-to-ai/
  12. Writers Guild of America West. “Know Your Rights: Artificial Intelligence.” https://www.wga.org/contracts/know-your-rights/artificial-intelligence
  13. Writers Guild of America. “Summary of the 2023 WGA MBA.” https://wgacontract2023.org/the-campaign/summary-of-the-2023-wga-mba
  14. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), Article 26: Obligations of deployers of high-risk AI systems. https://artificialintelligenceact.eu/article/26/ (Official Journal: https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng)
  15. Littler Mendelson. “Germany: Does the Works Council Have a Say When Employees Use ChatGPT? No, According to the Labor Court Hamburg.” 2024. https://www.littler.com/publication-press/publication/germany-does-works-council-have-say-when-employees-use-chatgpt-no
  16. Bird & Bird. “First Judgement on the Rights of Works Councils when Employees use AI Systems.” 2024. https://cm.twobirds.com/en/insights/2024/germany/erstes-urteil-zu-rechten-des-betriebsrats-bei-einsatz-von-kuenstlicher-intelligenz
  17. Writers Guild of America. “WGA Ratifies 2026 Minimum Basic Agreement With AMPTP.” 24 April 2026. https://www.wga.org/news-events/news/press/wga-ratifies-2026-minimum-basic-agreement-with-amptp
  18. Writers Guild of America. “Summary of the 2026 WGA MBA.” https://www.wga.org/contracts/contracts/mba/summary-of-the-2026-wga-mba
  19. Regulation (EU) 2026/1744 amending Regulation (EU) 2024/1689 (Digital Omnibus on AI), in force 27 July 2026. https://artificialintelligenceact.eu/ai-act-explorer/digital-omnibus/

Tim Green

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

Support What We Do

Discuss...