No One Was Fired, the Positions Melted Away

Local 34's "Artificial University" Report, the Debates of Tech Workers Around the World, and an Organising Proposal for Turkey

October 18, 2026 at 11:55 AM75 min readAuthor: Oğuz Demirkapı

No One Was Fired, the Positions Melted Away

Local 34's "Artificial University" Report, the Debates of Tech Workers Around the World, and an Organising Proposal for Turkey

The report that occasioned this piece: Artificial University: The Effects of AI on a Clerical and Technical Labor Union — Local 34-UNITE HERE!, August 2026, 36 pages. Local 34 is the union of Yale University's clerical and technical workers. The report is a text prepared not by the employer but by the workers' own organisation: it brings together members' testimonies, the union's membership data and findings on the university's artificial intelligence (AI) investments. This piece introduces the report, highlights its strengths and tries to complete, from a class perspective, the places it leaves unfinished.

Update (15 October 2026): Local 34, together with its sister unions Local 33 and Local 35, held one of the largest university rallies in decades on Beinecke Plaza. Contract negotiations have been going on for 14 months, and alongside pay and benefits the demands include limits on the use of AI. Union negotiator Bob Proto said that if the deadlock continues, a strike could come onto the agenda, as it last did for Yale workers in 2003.

Dear Young Comrades,

When you imagine how artificial intelligence will do away with jobs, a scene probably comes to mind. One morning a company lays off thousands of people, the newspapers run headlines, and photographs are taken of people carrying boxes out the door.

Local 34's report describes a different scene. In it, no one was fired. No one is carrying boxes out the door. It is just that no one was hired to replace the person who retired. The desk of the person who resigned stayed empty. And those who remained were asked, "Have you tried Clarity?"

And within two years the union's membership, for the first time since 2020, began to fall fast.

This scene was not described to us by a management consultancy report, a university press release or a technology company's "jobs of the future" presentation. It was described by the workers' own union. That detail stands at the centre of this piece.

Why Local 34?

Local 34-UNITE HERE! is the union of Yale's clerical and technical workers: administrative assistants, library staff, clinical receptionists, research assistants and laboratory technicians. About 78 per cent of its members are women. For more than forty years it has fought in New Haven to raise these jobs to a middle-class standard of living. As the report itself reminds us, that standard was won through five strikes in twenty years.

In the debate on AI we usually hear two voices: the voice of the companies transforming work and the voice of the states trying to regulate them. The voice of the worker who loses their job, or whose job is transformed, is almost never heard. Local 34's report fills this gap:

It collects its own data. In March 2025 the union surveyed its members, receiving 2,455 responses, and tracked its membership month by month. The evidence of passive automation came not from the employer's records but from the union's.
It lets its own witnesses speak. The body of the report consists of accounts by members speaking under their names and job titles. One does not give their name, for fear of retaliation. Even that says something about the power relationship in the workplace.
It reads the employer's documents through workers' eyes. The task force report, training slides and internal correspondence are obtained through information requests and interpreted from a class perspective.
It takes knowledge to the bargaining table. The report was prepared ahead of the 2027 contract negotiations and fed into the big rally of 15 October 2026.

Dear young comrades, this is called the working class producing its own knowledge. In this piece, when we say "Yale" we mean the employer, and when we say "Local 34" we mean the organised voice of the workers. The real subject of the piece is the second.

This piece will proceed through three questions:

In whose hands? Whose are the technology and the capital of this transformation?
Under whose control? Who decides how AI enters the workplace and whose work it takes?
To whose benefit? To whom does the resulting "efficiency" go, and who pays the price?

First, Briefly: What Does the Report Say, and What Do We Say?

What does the report say?

In August 2024 Yale set aside 150 million dollars for AI and launched its own AI platform, Clarity. Clarity is in fact a "wrapper" that offers ChatGPT, Claude and Gemini through a single interface.
Yale says "using AI is not mandatory". Workers, however, describe being under constant pressure, and in some units having to report their AI use to their manager.
Despite Article 12 of the collective agreement, the university did not give the union 90 days' notice and did not bargain over the effects of AI. The matter is in the grievance process and may go to arbitration.
AI sped up some tasks. Others it slowed down and lowered in quality. Correcting the faulty texts AI produces has become a new workload.
The union's membership peaked at 4,112 in April 2025 and fell to 3,933 in August 2026. The university first froze hiring, then introduced a retirement incentive. It also rejected the union's demand for a minimum staffing floor.
The name the report gives this process: passive automation. AI does not fire the worker; it makes it possible not to hire anyone in their place.

What do we say?

The report names the mechanism correctly. Passive automation is the production, without lay-offs, of what Marx called relative surplus population. For capital it is the cheapest and quietest route.
The most valuable part of the report is showing that Yale is not only an employer that uses AI. Through its board and its investment fund, Yale is also a capital that finances AI. The employer and the investor are the same.
The report's weak side is that its proposed solution is class-collaborationist. It calls for a "responsible transition" and "labour peace", and accepts the development of AI as "seemingly inevitable". Yet that "labour peace" at Yale, as the report itself reminds us, was won through five strikes in twenty years.
The missing demand is this: if the efficiency gain is to go to workers, its concrete form must be a shorter working day at the same pay. Not fewer positions.

What are tech and service workers around the world debating?

In Part XII of the piece we show that Yale is not the only example. Around the world the debate revolves around five headings:

An AI clause in the collective agreement: In Europe, AI provisions are increasing in the 108 agreements examined. In the US, journalists, game developers and actors have won notification, bargaining and consent provisions.
The legislative front: In Britain the union confederation, the TUC, wrote its own draft AI bill. The European Parliament called for an EU directive on algorithmic management.
The core of computing: In the US against Meta's lay-offs of thousands, and in India against TCS's lay-off of 12 thousand, tech unions are raising their voices.
The invisible lower floor: Data labelling workers in Kenya and content moderators from nine countries are organising. One link in this chain is also in İzmir: the organising of TikTok moderators in Çağrı-İş and the dismissals.
The question of independence: "partnership" with the monopolies, or organised countervailing power?

What do we propose for tech workers in Turkey?

In Part XIII of the piece we present an organising proposal built on the lessons we draw from Local 34's experience and from the debates around the world. In summary:

Stance: A struggle not against AI, but over the ownership and control of AI. The tech worker is both the person who builds AI and the person who is meant to be replaced by it. This double position is both a danger and a specific strength.
Four layers: (1) AI committees and record-keeping in the workplace, (2) an "AI and Technological Change" clause to go into collective agreements, (3) an organisational structure that gets past the branch-of-activity barrier and combines unions, professional chambers and cooperatives, (4) a "counter-expertise" network that puts the tech worker's technical knowledge at the service of the class.
Model clause: The piece contains a ten-point draft text that could be put directly into an agreement: advance notice, protection of positions, conversion of the efficiency share into shorter working time, a ban on surveillance through AI data, a ban on using a worker's knowledge to train a system that will replace them, and the right of conscientious objection.

Now let us turn to the details.

I. The Architecture of the Artificial University: Where Did the 150 Million Dollars Go?

In January 2024 Yale provost Scott Strobel set up an "Artificial Intelligence Task Force". The task force's aim was to ensure that Yale was at the "forefront" of the AI revolution. Its recommendations were published in June 2024. And in August 2024 the 150-million-dollar investment was announced:

Buying around 450 GPUs for model training,
Developing the Clarity platform,
Hiring faculty specialising in AI and "computational research support analysts",
Distributing seed funds for research and teaching,
Organising cross-departmental "innovation" events.

One thing is missing from this list: the existing staff.

The report stresses this plainly. Yale presents itself as New Haven's largest employer, but the task force report says almost nothing about how AI will affect the university's workers. The only mention is a single sentence from the Operations Leadership Team saying that AI should help limit the "growth of institutional administrative costs".

Dear young comrades, read that sentence again. Who is the "administrative cost"? It is the people sitting at the desk, answering the phone, writing the letter, organising the calendar. The task force report does not speak of the workers because, in the report, the workers appear as a cost item.

In our checkout piece, Who Took Away the Chair?, we showed that the item "staff sufficient not to disrupt service" in BİM's annual report sits between rent savings and plain decoration. Yale's task force report is the university version of the same logic. A supermarket chain's cost table and an Ivy League university's speak the same language.

Clarity: An AI platform without its own AI

The report draws attention to a detail that looks technical at first glance. Clarity is not an independent model. It is a wrapper that offers ChatGPT, Claude and Gemini through an interface "exclusive to Yale".

This detail matters. One of the richest universities in the world, in a 150-million-dollar AI programme, does not produce its own model. It ties a handful of monopolies' models into its own labour process and pays for it. We discussed the class position of these monopolies in Reading Anthropic's 2030 Model Through a Class Lens, Whose Foot Is on the Brake? and The Expropriation of the General Intellect. The university, which ought to be the public institution of knowledge production, is becoming a customer of knowledge in its privatised form. We looked at this order, in which the monopolies "learn from everyone" and teach no one, in Learning from Everyone Is Permitted, Learning from the Monopoly Is a Crime, and at the locking up of knowledge in closed laboratories, through OpenAI's mathematics papers, in The Closed Laboratory of the General Intellect.

II. The Employer Is Also the Investor: Yale's Capital Ties

The most valuable part of the report, in our view, is the section most readers will skip, titled "Yale's Many AI Investments". Here is the picture:

Tie	What?
Jaime Teevan	Elected to the Yale Corporation in 2025. Chief scientist at Microsoft. Led the development of Microsoft 365 Copilot.
David Sze	On the Corporation. Advisory partner at the venture capital firm Greylock Partners.
Greylock	One of the firms Yale's endowment invests in. Closed a 1.5-billion-dollar AI fund in July 2026.
Andreessen Horowitz (a16z)	One of the firms Yale's endowment invests in. Large AI portfolio. According to the report, it donated millions of dollars to the Trump campaign and lobbied for the deregulation of AI.
Benchmark	One of the firms Yale's endowment invests in. In February 2026 it took on Jack Altman, brother of OpenAI CEO Sam Altman, as a partner.
CodaMetrix	Yale Medicine joined its 55-million-dollar funding round in 2023. In the words of the company's president, medical coding "should become largely autonomous."
Clarium	Yale New Haven Hospital backed the 10.5-million-dollar funding round of this company, which automates the hospital supply chain.

Now put this table next to Local 34's membership list. The union's two largest job classifications are "Senior Administrative Assistant" (970 people) and "Clinical Receptionist" (288 people). Yale Medicine has around 1,700 clinical staff. The medical coding CodaMetrix aims to automate is exactly one of the jobs these people do.

So on one hand Yale encourages its workers to hand their work over to AI, and on the other it becomes, as capital, a partner in the companies that will automate that work. The money of the institution where the worker works is invested in the machine that will take the worker's job.

Let us state the conclusion the report does not state openly: this is not a story of "the technology has arrived, we must adapt". Here the technology does not descend from the sky. It arrives as an organised capital strategy, through investment decisions, board seats and fund portfolios. The answer to the question "in whose hands?" is written in this table.

Nor are the names in the table a coincidence. We examined the "techno-optimist" vision of the future of a16z founder Marc Andreessen in From Futurism to Techno-Fascism. We also discussed how the bourgeoisie extends the life of capitalism by using AI for rent, labour control and repression in The Bourgeoisie Is Buying Time with Technology, and We Will Bury Capitalism. Yale's endowment is the monetary counterpart of this ideology.

III. Passive Automation: Producing a Reserve Army Without Firing Anyone

The report's central concept is passive automation. The report defines it like this: companies, universities and governments will deploy AI to do more administrative labour without needing to rehire workers who have quit or retired.

The report also says that AI is not expected to eliminate jobs one for one. What will happen is "passive replacement": employers will have the same work done by fewer workers with AI assistance.

The numbers

Date	Event	Membership
January 2022	—	3,652
28 August 2024	Clarity goes live	Rising
April 2025	Peak	4,112
May 2025	Decline begins	—
June 2025	90-day hiring freeze	Sharp fall over four months
April 2026	Retirement incentive	—
August 2026	—	3,933

In about fifteen months 179 members, or roughly 4.4 per cent of the union, were lost. On its own this may not look like a disaster. But note: no one was fired. This fall took place without a single lay-off, simply by not filling positions as they fell vacant.

Management's admission: the library email

The strongest evidence that passive automation is not a concept invented by the union is in the report's section on the library. In August 2024 Yale Library sent an email to all staff saying it expected everyone to become familiar with AI concepts, tools and uses. Then library management sent another email. According to this one, the library would reduce costs by not filling open positions and by meeting researchers' needs with "evolving technology", that is, with AI.

There is no need to interpret this. The union did not invent passive automation. Management wrote it in its own email.

As Marx said, but quieter

In the first volume of Capital, Marx explains that the accumulation of capital produces, according to its own needs, a "relative surplus population", that is, an industrial reserve army. The machine makes the worker superfluous, and the superfluous worker creates a pressure that pulls down the wages and bargaining power of those in work.

In its classic form this process is visible: the factory closes, workers are fired, unemployment rises. The form we see at Yale is much quieter:

No one is fired. So no event, no news story, no occasion for resistance arises.
There is no legal basis. Because there is no dismissal, the mechanisms that protect the worker do not come into play.
The loss is invisible. What disappears is not a person but a job advertisement that is never posted.

This reserve army consists not of the unemployed waiting at the gate but of positions that were never born. Of people who were never let in through the door, who never sat at that desk.

We showed how unemployment is hidden in statistics in TÜİK: Two Figures, One Country. Passive automation is its counterpart in the workplace. We also argued, in The Bloody Steps of the Pyramid, that Harari-style talk of a "useless class" is not a law of nature but the result of exactly these kinds of administrative decisions. And we showed how natural-disaster metaphors such as the "AI tsunami" conceal these decisions in What Does the "AI Tsunami" Conceal? A tsunami has no perpetrator. A decision to freeze positions has a signature.

A correction: tax or AI?

Here we have to be honest. The official reason for Yale's June 2025 hiring freeze was not AI. According to the Yale Daily News, the university tightened its budget in anticipation of the federal government raising the tax on endowment income. The report writes this too.

What the report leaves undone is separating these two factors. In the membership graph three events (the launch of Clarity, the freeze and the retirement incentive) are marked one after another, and the reader is led to attribute the whole decline to AI.

Our reading is this: the financial squeeze is the justification for cutting positions; AI is the tool. The tax provides a political pretext for reducing positions. AI makes it possible for the same work to go on being done with fewer positions. Without one, the other would not work at this speed. Had the report set out this relationship openly, its claim would have been both more honest and stronger.

IV. "Not Mandatory" Yet Mandatory: The Manufacture of Consent

Yale's official position is clear: using AI is voluntary. The report shows, step by step, how this voluntariness is manufactured.

The Prompt-A-Thon

The main tool of the campus-wide AI campaign is events called "Prompt-A-Thons". At one session on 20 October 2024, unionised IT staff were taught to write prompts for tasks like these:

"Evaluate the effectiveness of a new student support programme."
"Prepare a communications plan for a new university policy."
"Prepare a budget proposal for a new academic initiative."
"Collect and analyse feedback from faculty about administrative services."

These are tasks in these workers' own job descriptions. The worker is being taught how to hand their own job over to the machine.

The presentations contain cute drawings of Yale's mascot, Handsome Dan, teaching, dreaming and working at a computer. But one slide breaks the cuteness:

"The jobs of tomorrow belong to the learners of today. Are you developing your AI skills?"

This is not an invitation but a warning. Those who do not learn will be left behind.

The task force report describes this process even more openly. According to the report, many operations staff were focused on "what not to do rather than what to do" with AI and were waiting for approval. This was "being corrected through discussions and pilots". Caution is defined as a flaw to be corrected.

Reporting requirements and job ads

A worker in the Development (fundraising) office: "Senior management encourages us to use AI in our daily work. We even have to report to them how we use AI."

Another worker in the same office, who conceals their identity for fear of retaliation: "By April 2025, AI use was not only encouraged, but a mandatory and non-negotiable part of my job." This worker was asked to share which parts of their job they had automated with AI for a department-wide "AI Best Practices" document. The document that reached them a few weeks later contained, in their words, "very good examples of how AI prompts can be adjusted to automate significant parts of my job".

Think about the situation again: the worker is being asked to help write the manual for automating their own job.

The final step is job ads. The same office has added AI competence as a requirement in ads for unionised positions: the candidate "successfully engages with the university's AI platforms to enhance productivity as it relates to data analysis" and "liaises with other units and vendors to maintain/enhance software platforms, including our AI models."

Burawoy's question

In Manufacturing Consent, the sociologist Michael Burawoy asked why industrial workers participate so actively in their own exploitation. His answer was that the labour process is organised like a "game", and the worker, while trying to win at this game, reproduces the system.

The picture at Yale is the digital-age counterpart of this question. Hackathon language, slides with mascots and "best practices" documents turn the labour process into a game. The fear of "being left behind" makes joining the game compulsory. The requirement in job ads institutionalises the game. The moment AI competence becomes a condition of being able to sell one's labour power, voluntariness has disappeared.

This picture confirms what we said when discussing Anthropic's 81,000-person survey: surveys and "usage data" do not only measure; they also manufacture consent. We also discussed how the same mechanism works in the companies' own "ethics" and "safety" discourse in The Poverty of Artificial Intelligence Ethics and Anthropic's Table with the Clergy, and the Poverty of Philosophy.

V. What Is Changing in the Labour Process? Four Testimonies

The body of the report consists of members' testimonies. The survey conducted in March 2025, ahead of bargaining for the 2027 contract, received 2,455 responses from members. Of these, 719 (29 per cent) said AI or automation was in use in their unit. The units with the most reports were Development, the Library, the Faculty of Arts and Sciences and Yale Medicine.

The testimonies show four distinct transformations in the labour process.

1. Deskilling: The writer becomes a carrier

Tamera Sternberger, a programme coordinator in the Development office:

"Before Fall 2025, I would be part of large email solicitation campaigns to donors at Yale College, hands-on with drafting, writing, editing, and multiple draft revisions. In most cases, this fundraising work would entail multiple revisions with my manager and would take several weeks of work to complete. Following the introduction of AI in the Development Office, I am now rarely asked to do any drafting, editing, or revision, since this work is now almost exclusively done by AI by my managers. I hardly write anymore. Yet writing and editing were at least half of my job."

Sternberger's new task? The logistics of student events: moving things, taking equipment and materials to the rooms where events are held. The report sums it up like this: the "creative parts" of the job have been "automated away."

Another worker says the same: "I was hired as a staff writer. The amount of actual writing I do has collapsed because of AI."

In Labor and Monopoly Capital, describing how monopoly capitalism transformed the labour process, Harry Braverman showed that the conception part of work is handed to management and the machine, while the execution part is left to the worker. That is the essence of deskilling. Sternberger's story is almost a textbook example: the part of the work that thinks, writes and designs passed to the machine; what was left for the worker was what her body could carry. We looked at what writing with AI does to labour and authorship in A Debate on Writing with Artificial Intelligence, and at the expropriation of mental labour in The Liberation of General Intellect.

2. Intensification: Three people's work on one person

Carolyn Kyrtopolous, an administrative assistant in the Dean's Office of the Faculty of Arts and Sciences:

"Writing a biography by hand is time consuming work, and usually takes approximately two hours per biography. Now, with the use of Clarity, each biography could take a fraction of the time, and four biographies could be completed in the span of two hours of work for one [clerical and technical worker]."

More striking is the scheduling task:

"This scheduling task involves coordinating around 40 or so [calendars], and usually takes several dozen hours for about three administrators to complete. I was now tasked with completing this task entirely by myself. In several one-to-one and group meetings, [my manager] encouraged me to use AI to complete this task… While Clarity failed to complete the task to the required specification, [my manager] again brought up using Clarity to do this scheduling task for next year's annual meetings."

Here we see not efficiency but intensification. By Marx's distinction, a machine can raise surplus value in two ways. Either it raises productivity, so that more is produced in the same time. Or, by closing the gaps in the working day, it makes more labour be expended in the same time. In the scheduling example the machine could not do the work. But the decision to load the work onto one person did not change. Here AI works not as a productivity tool but as the justification for intensification.

The report also says where this process is heading: the spread of AI in the faculty means fewer workers are needed for the same work, but at the same time the pressure on workers to check enormous amounts of AI text is increasing. "If this trend continues, administrative labour will gradually turn into one big exercise in checking AI output."

3. Invisible repair labour: The bill left for those who do not use AI

Jennifer Ongley, assistant to the chair in the Department of Physics, does not use AI on ethical grounds: its environmental impact, the plagiarism of art and science, and the threat to job security. But AI enters her work anyway:

"A colleague asked me to edit a document she wrote with AI that was supposed to provide instructions for admitted students to book their own travel for a campus visit. I was frankly appalled at the document she shared with me. Instead of clearly laying out the specific travel options, it was a confusing jumble of word salad that somehow said very little while using way too many words. I gave up after making a few edits to the first page and sent it back to my colleague with the suggestion that she scrap it and start from scratch on a new draft, exactly the thing she was trying to save time on by using AI."

Another colleague used AI to make a poster. The AI introduced errors that could not be removed, the poster was redone from scratch, and the job took twice as long.

The most important part of Ongley's testimony concerns the finance chatbot. Management steers workers to use this bot for questions about finance policies instead of checking the website or contacting the Finance Support Center. The bot opens with a warning: answers may be inaccurate, always compare them with the policies, and if in doubt ask your manager or the Support Center. When Ongley tried it, the bot gave an answer that contradicted policy. Ongley's concern is twofold: the university may find the bot "good enough" and stop keeping competent staff, and workers may be disciplined for following the bot's wrong instructions.

Here we see a very important mechanism. The efficiency gain shows up in management's tables. Its cost is passed to the worker in two forms:

Unpaid repair labour: The labour that corrects faulty output appears in no efficiency table.
Risk transfer: If the bot gets it wrong, the one responsible is not the bot but the worker who used it.

We discussed the question of who gets the bill when AI agents make mistakes, through the FTC investigation in the US, in The Agent Escaped. Who Gets the Bill? The answer there was the same: risk flows down, profit flows up.

4. Handing the hierarchy over to the machine: "Ask Clarity, not your manager"

Dana Greene, a senior administrative assistant in the School of Public Health, had to attend a mandatory "AI Training" meeting on 29 October 2025. Afterwards her manager brought up AI use in the fortnightly meetings, in one of every two meetings. In one-to-one meetings, too, she told Greene to ask Clarity first the questions she would normally ask her manager (for example, about the rules on departmental travel).

The report asks a pertinent question: for instance, who would be responsible for a costly error in booking travel if a clerical worker was told to ask AI for information and the AI gave them faulty information? The university has not yet implemented any protections for workers who make errors due to AI outputs.

This fourth transformation is the handing over to the machine of the relationship between worker and manager itself. The manager leaves the task of guidance to the machine, while leaving the responsibility with the worker. We saw where this road can lead in An AI Boss Fired a Human for the First Time: in the Luna/Andon case, in a workplace managed by AI, the proposal to dismiss a human worker was made by a language model. Where today they say "ask Clarity first", tomorrow they may say "let Clarity decide".

The second threat: Work shifting to managers

The report has one more finding that may be overlooked but is very important. According to accounts from the Development office, non-union managers are now doing, with AI assistance, work that unionised clerical workers used to do. The report defines this as a second threat, separate from automation: work is shifting both to the machine and to managers outside the bargaining unit. Both violate the agreement, but both are hard to monitor.

This means redrawing the class relationship against the organised worker. AI is also becoming a tool for taking work out of the bargaining unit.

VI. Surveillance: The Subject the Report Mentions in Passing

There is another detail in Ongley's story that may escape attention. In summer 2024 her manager gave her Plaud.ai's "NotePin": a lapel-worn recording device that records conversations, transcribes them and summarises them. Ongley was asked to use it in faculty meetings. She refused, and her manager allowed her to go on taking minutes by hand.

The report mentions surveillance only in passing. It notes that the Prompt-A-Thon presentations did not address the risks of automation or the "potential for increased surveillance". But surveillance is in fact everywhere in the piece as a whole:

Workers are asked to report how they use AI.
Recording devices are brought into meetings.
Workers' ways of doing their jobs are recorded and standardised in "best practices" documents.

The third point is especially important. When knowledge about how a worker does their job is passed to management, that knowledge is no longer the worker's but management's. What Taylor did with the stopwatch, prompt libraries and usage reports do today. We call this mental Taylorism. We discussed digital labour control in detail in An Epistemological View of Arif Koşar's Analysis of Labor Supervision.

We discussed wearable recording devices in the workplace in Whose Eye Are the Glasses?; the law around union data and labour control in The Aydo Software Operation and KVKK and the 80s; and how the digital panopticon is built from the top down in our piece on the Palantir manifesto. We showed how AI companies can monitor their own users in Anthropic's Activist Monitoring System and "We Stopped Them All": A Monopoly's Grammar of Self-Absolution. The models behind Clarity are these companies' models too.

VII. Women's Labour and the City: Whose Desk Is Emptying?

An important contribution of the report is that it takes the picture to the scale of gender and the city.

78.3 per cent of Local 34 members are women.
13.5 per cent of working women in New Haven are in office and administrative support jobs. This is the second largest occupational group for women in the city.
In 2024, 6,144 people in New Haven worked in office and administrative support. That is 9.65 per cent of all workers in the city.
893 members of Local 34 live in the city itself.

The report also cites a July 2026 projection by the US Bureau of Labor Statistics. According to this projection, job opportunities for office clerks, administrative assistants, secretaries and medical transcriptionists will steadily decline. A researcher quoted in a June 2026 piece by Ben Casselman of the New York Times also describes these "back office" jobs as "anchor jobs" for families and households, that is, a ticket into the middle class.

The class meaning of this picture is this: the burden of passive automation falls first on women's paid labour. And as women's paid labour erodes, the pressure to push women back into unpaid domestic labour increases. We discussed the invisibility of domestic labour in The Class of Alimony in the Centenary of the Civil Code. AI does not abolish the gendered division of the labour market; it reorganises it.

The city dimension matters too. Library worker Amber Suess was hired through New Haven Works, a programme run jointly by Yale and Local 34 that helps New Haven residents get jobs at the university. So far this programme has placed 60 New Haven residents in unionised library jobs. Suess says the three AI tools the library has tried (Clarity, Consensus AI and a library-specific MCP tool) "directly compete" with her work, and adds: "When the library decides to add this function directly into the chat window on its website, significant parts of my job will be automated."

The report puts this in a striking sentence: workers like Suess feel that their jobs are "one software update away".

Passive automation also closes one of the most important doors through which poor city residents could enter middle-class jobs. A job advertisement that is never posted is the announcement of a chance that is never given.

VIII. The Legal Front: Article 12 and Its Limits

The report's most concrete legal claim is this: Article 12 of the 2022 collective agreement (Memorandum of Agreement) between Local 34 and Yale, the "Technological Change Agreement", requires the university to give the union 90 days' notice before introducing a technology that could "substantially change" the nature of members' work. This period exists so that the union can bargain with the university over the effects of the change on members.

Yale launched Clarity without notice and refused to bargain over the effects of AI. The matter is now in the grievance process and may go to arbitration.

This article is a gain that a trade unionist looking from Turkey should study carefully. But its limits must also be seen:

What is bargained over is not the decision itself, but its effects. The union can bargain not over whether the technology comes, but over how workers will be affected when it does. This puts the union on the defensive from the start.
"Substantial change" is open to interpretation. Whether a tool presented as "voluntary", like Clarity, substantially changes the nature of the work is debatable. The insidiousness of passive automation lies exactly here: each step is small, each step "voluntary", and the sum is a huge change.
The law works only as far as the power relationship does. Yale did not fulfil its notification obligation. The sanction in the union's hands is a long arbitration process. Whether the article really works depends on how strong the union is on the ground.

IX. "A Better Way"? A Comradely Critique of the Report's Solution

The last section of the report is titled "A Better Way". Here the report tells the history of the relationship between Yale and Local 34. There were long and bitter contract struggles, disputes often resolved only by strikes. Then, under President Richard Levin and Vice President John Pepper, the parties built a "problem-solving" relationship. Since 2003 contracts have been signed peacefully, and the Great Recession, the integration of Yale New Haven Hospital and the 2020 pandemic were weathered through this relationship. The report quotes a 2025 speech by Levin: "Both stories begin badly. Both end well."

The report's proposals come from this framework:

The report's proposal	Our assessment
Compliance with Article 12, 90 days' notice and impact bargaining	Concrete and legally sound. The union's strongest card.
Protection of at least 3,933 clerical and technical positions for the next five years	The right structural response to passive automation. This should be the centre of gravity of the bargaining.
Investment in training programmes and higher pay for workers who undergo training	Reasonable, but it again loads the burden of "adapting to AI" onto the worker.
Extending the time spent in the Interim Employment Pool (IEP)	A laid-off member keeps their pay and benefits for 1–15 months (plus a 50 per cent extension). A good protection, but the report itself admits: as jobs shrink, the pool will clog up.
Limiting subcontracting and outsourcing	Critical. According to the report, the new jobs created by the technology are mostly non-union jobs at vendor companies, and laid-off members cannot access them.
Joint working groups to assess the advantages of AI	The right intention, but it stays at the level of consultation, not a veto.

The report's diagnosis is strong. But we see three problems in its proposed solution.

1. What created the peace is forgotten

The report describes the "labour peace" of the Levin era as if it were the product of good-faith cooperation. Yet in the same paragraph there is also this phrase: "after five strikes in twenty years".

Dear young comrades, do not skip over that phrase. What brought Yale to the table was not Levin's courtesy. It was five strikes. The "problem-solving relationship" was the product of the union having proved its strength. As we said in our checkout piece: wherever capitalism looks tame, someone has paid a price to leash it. We described at length how social democracy, at critical moments, tied the class's hands for the sake of "labour peace" in our History of the European Working Class dossier.

The report was written ahead of the 2027 contract bargaining. So at precisely the moment when the union needs to show its strength, what is recalled is not the struggle that brought the gains, but the employer's "legacy".

2. Inevitability is accepted

On the last page of the report there is this sentence: "The continued development of AI appears inevitable."

This sentence is bourgeois narrative, which presents technology as a neutral natural phenomenon, seeping into the language of the union. The development of AI as a technology may be inevitable. But its development in this form, a form that takes the worker's job and writes it up as a gain for capital, is not inevitable. What must be debated is not the technology itself but its ownership and control. We opened this up in The Tearing of the Illusion and The Revolt of Crystallized Labor. AI is the privatised form of humanity's accumulated labour, that is, of dead labour. The problem is not the machine but whom the machine belongs to. We looked for the answer to this question in the Socialist AI Manifesto and in Is a Real Emancipation of Artificial Intelligence Possible?

We also showed that Acemoğlu-style reformism of "choosing the right technology" runs into the same limit in Saving Acemoğlu Does Not Save the Class and The Acemoğlu–Economist Dispute. Every solution that looks for the problem in the direction of technology skips the question of who gives it its direction.

3. The best sentence does not become a concrete demand

The strongest sentence in the report's introduction is this:

"Workers, not the company, should benefit from AI's productivity."

But this sentence does not become a concrete demand in the conclusion. What is proposed is training, higher pay during training and protection of positions.

Yet this question must be asked: if the same work can now be done with less labour, whose is the time saved? In today's order the answer is clear: the time saved is credited to capital, in the form of people who are fired or never hired.

The class's answer, though, should be this: a shorter working day at the same pay. If productivity rises, working time should shrink, not positions. As we said in A Class Look at Bill Gates's AI Warnings: to whom the time saved belongs is a question of ownership. We also showed how Elon Musk's promise of "abundance" skips this question in A Marxist-Dialectical Critique of Elon Musk's Vision of the Future.

X. Two Readings, Side by Side

Management's narrative	A class reading
Using AI is voluntary	Voluntariness turns into compulsion through fear, gamification and job-ad requirements
AI frees workers from boring tasks	AI takes the creative part of the work and leaves the worker with carrying and checking
AI increases efficiency	AI is often the justification for intensification; repair labour does not appear in the efficiency table
No one was laid off	No one was fired because the positions were never opened: passive automation
The reason for cutting positions is tax	Tax is the pretext, AI the tool
Yale is an educational institution	Yale is also an investor in the companies automating its workers' jobs
AI is inevitable, we must adapt	What is inevitable is not the technology but the property relation; and that can be changed
Solution: responsible transition, joint working groups	Solution: a minimum staffing floor, a veto over technological change, a shorter working day at the same pay, organisation

XI. Seen from Turkey

This report comes from a US university. But the mechanism it describes is not at all unfamiliar to us.

Public sector and university administrative staff. In Turkey, in the public sector and at state universities, the "norm staffing" system and not filling vacant positions are already a common method of saving money. Passive automation will sit on top of this infrastructure. The new structures set up to manage AI applications in the public sector could turn this process into an official programme. We examined these structures in Did You Know There Is a Directorate General of Public Artificial Intelligence?, On Turkey's Artificial Intelligence Action Plan (2026–2030) and The National AI Board: Who Is at the Table, Who Is Not? What they all have in common is that workers and unions are not at the table. We also discussed how the "indigenous and national" discourse covers up this picture in The Illusion of "Indigenous and National" Artificial Intelligence. Given the narrow limits of public servants' right to collective bargaining, we must recognise that a protection like Yale's Article 12 does not exist in Turkey's public sector.

Foundation universities and the private sector. The administrative staff of foundation universities and office workers in the private sector are largely unorganised. As we showed in our checkout piece, branch of activity no. 10, which also includes office workers, is one of the largest and least organised branches in Turkey. At Yale the union could at least track its membership and document the decline. In Turkey there is not even an organisation to measure that decline.

The technological change clause. As far as we know, clauses in Turkish collective agreements that require notification of and bargaining with the union before technological change are quite rare. For tech workers this should be a priority demand. We looked at the report the TMMOB prepared for the engineering and architecture professions, and at its shortcomings, in Reading and Developing the TMMOB's Artificial Intelligence Report. First we will look at examples from around the world, and then in Part XIII we will propose a concrete draft.

The link with platform labour. Passive automation in the office and "self-employment-isation" in courier work are two faces of the same tendency: the narrowing of secure employment. One never opens the positions; the other turns the position into a contract and the worker into a "tradesman". We discussed this in detail in our Gig Economy Dossier.

XII. Local 34 Is Not Alone: What Are Tech and Service Workers Debating Around the World?

Local 34's report is not a lone cry. Throughout 2025 and 2026, in different parts of the world, unions, confederations and newly founded workers' organisations published reports on the impact of AI on labour, signed agreements, wrote draft legislation and took action. In this part we map these debates. Our aim is not to present a catalogue. It is to draw a lesson for Turkey from each example and to assess, from a class perspective, where these debates stand. You can also follow our series The Week's Class Struggle in Technology, in which we compile these developments every week.

1. An AI clause in the collective agreement: What was put on the table?

Europe: A breakdown of 108 agreements. On 12 October 2026, the Friedrich-Ebert-Stiftung and UNI Europa, the European federation of service-sector unions, published the results of a three-year study. 108 collective agreements were examined for provisions on AI and algorithmic management. The findings:

55 per cent of respondents know that AI has been addressed in collective bargaining.
Bargaining over AI has been increasing since 2023. The increase is especially marked in Spain.
The most common topics are training, worker participation and data protection.
In Italy, a telecoms agreement provides for a 12-month trial period for a new AI tool. With the energy company Enel, a joint committee monitoring AI projects has been set up.

The study's recommendation matters for us: to frame AI as a significant change at work and to take it onto the bargaining agenda together with the leverage of the strike.

US: Journalists, game developers, actors. According to an Axios round-up from July 2026:

The journalists' union NewsGuild-CWA has explicit AI provisions in 85–90 of its contracts. President Jon Schleuss's summary: "AI is something that can be bargained."
At Politico, management introduced faulty AI reporting tools without the union's consent. The unionised reporters went to arbitration and management withdrew the tools. A won example of the kind of dispute over Article 12 at Yale.
At ZeniMax, the Microsoft-owned games company, the December 2023 tentative agreement with the CWA provides that AI should support workers rather than replace them. If an AI implementation could affect union members' work, the company will notify the union and bargain on request. The same agreement brought 77 contract workers into the bargaining unit. It is no accident that the AI clause was won together with a clause against subcontracting. As Local 34's report also shows, the two threats come together.
The writers' (WGA) strike of 2023 and the actors' (SAG-AFTRA) contract barred AI from replacing workers without consent.

Germany: The first AI collective agreement. The text signed by ver.di and the actors' association BFFS in the film production sector, presented as the first AI collective agreement in Germany, has been in force since 1 March 2025. Under the agreement, an actor's digital copy (replica) may be used only with written, purpose-specific and separately paid consent, and every significant digital change requires renewed approval. It is not the tech sector, but the approach matters: a digital asset produced from the worker's labour cannot become a capital good independent of the worker. This is the creative-sector counterpart of point 6 (ownership of knowledge) in our proposal. We discussed the relationship between copyright and digital labour in The Dialectic of Copyright in the Digital Age.

Lesson for Turkey: An AI clause is not a dream. It has begun to enter agreements in many countries. But note: behind the clause in almost every example stands either a strike (WGA), an arbitration victory (Politico) or a strong union structure (Italy, Germany).

2. The legislative front: Unions writing their own law

Britain: The TUC's "ready-made" AI bill. Instead of waiting for the government, Britain's union confederation, the TUC, wrote its own draft bill. The Artificial Intelligence (Regulation and Employment Rights) Bill, published in April 2024, provides for:

Mandatory risk assessment before and after deployment of "high-risk" AI decisions in the workplace, and a review every 12 months,
A statutory duty to consult unions before a system is deployed and every year,
A public register of the AI decision systems an employer uses,
A personalised right to an explanation of how AI affected decisions about the individual, and a right to human review,
A right of unions to access the member data used in AI decisions,
A ban on emotion recognition technology against workers,
A right to disconnect.

This method is very valuable for us: the union movement writing its own legal text means taking the debate off the agenda of the employer and the state and onto its own.

European Union: The demand for a directive. The European Trade Union Confederation (ETUC) is calling for a dedicated EU directive on algorithmic management and AI at work. Its demands include human oversight in all AI-assisted decisions, setting the rules on data collection and surveillance through collective bargaining, preventing job loss and deskilling, paid AI training, and giving workers a say in R&D and industrial policy decisions. The European Parliament also adopted a legislative initiative report in this direction by 451 votes in favour, 45 against and 153 abstentions. According to the data cited in the report, 51.3 per cent of European workers were subject to algorithmic management in 2024.

But there is also UNI Europa's warning: the EU's "Digital Omnibus" package risks weakening workers' existing legal protections. In other words, every step won on the legislative front is being pushed back with capital's talk of "competitiveness" and "reducing red tape".

Lesson for Turkey: In Turkey there is no specific regulation on AI in the workplace. Apart from the KVKK (the personal data protection law), the legal basis is very weak. Tech unions, professional chambers and lawyers could come together and, as in the TUC example, write their own draft. This draft would be both a campaign tool and an educational text.

3. The core of computing: Those who write the code are organising too

Software developers, for many years seen as "highly paid, privileged workers with no need for a union", are reaching the end of this illusion with the AI wave.

US: Meta and Google. In May 2026 Meta laid off around 8,000 people. The lay-offs fell in the same week as Google I/O, Google's annual product showcase. The Alphabet Workers Union (AWU-CWA), the union of Google and Alphabet workers, called on tech workers to organise in a statement:

"The products that bring enormous profits to companies like Meta and Google could not exist without our labour."

At the I/O event the union collected signatures for its "Googlers for Job Security" campaign. According to the data Axios cites, the unionisation rate in computer occupations in the US is only 4.4 per cent. Among education and library workers it is 36 per cent. Do not forget that Local 34's members could write their report because they have a union, while software developers mostly cannot.

We described in As the Golden Handcuffs Come Undone how Silicon Valley's "good conditions", high pay and stock, are not a privilege but a management technique built to keep the union out, and how AI, by lowering the cost of this technique, takes back the gains. The lay-offs at Meta confirm this thesis.

India: TCS and 12 thousand people. In July 2025 TCS, India's largest IT company, announced that it would lay off around 12 thousand people as part of its "future-ready" transformation and its plan for "deploying AI at scale". The Karnataka State IT/ITeS Employees Union (KITU) said many workers had been forced to resign and that this was unlawful. Data given in the same report show the face of passive automation in India: the net hiring of the six largest IT companies fell by 72 per cent from the first to the second quarter of 2025 (from 13,935 to 3,847). The Karnataka state government called TCS in for talks and launched a survey on the impact of AI on the workforce.

The Indian example is especially important for Turkey. A significant part of Turkey's software sector also relies on exporting services abroad, that is, on outsourcing work. Outsourcing work will be among the first things AI hits.

Sweden: A strike notice at a tech company with no agreement. Another example comes from Sweden. When a year of talks at King, the games company behind Candy Crush, ended without agreement, the white-collar union Unionen and the engineers' union Sveriges Ingenjörer served strike notice on 15 September 2026. The demand is a collective agreement that would increase security at the workplace and give workers more say over their working conditions. Tech companies' attitude of "we are different, we don't need an agreement" is still an obstacle to be overcome even in unionised countries.

4. The invisible lower floor: Those who train the machine

Behind AI's appearance of being "intelligent" there are millions of low-paid workers who train the models, label the data and filter out harmful content. We looked at this invisible labour in The End of the Gilded Illusion: Immaterial Labor in the Age of Artificial Intelligence and an Organizing Guide for the IT Worker in Turkey. These are the most forgotten section in the AI debate. And the liveliest organising of the last two years has come from precisely this section.

Kenya: "They pay us peanuts." In February 2025 AI data workers in Kenya founded the Data Labelers Association (DLA). The association reached 339 members in its first week. Its demands are fair pay, mental health support, legal aid and collective bargaining with data companies. In the words of the association's president, Joan Kinyua:

"We are the workers powering all this technological progress, but they pay us peanuts and don't even recognise us."

Another member describes the retaliation: "Every time we raise our voices, especially the 'task workers' at the lowest level, we are automatically fired."

The Global Trade Union Alliance of Content Moderators. In April 2025, at a meeting in Nairobi attended by workers from nine countries, the first global trade union alliance of content moderators was founded. Its demands are a living wage, long-term employment contracts, humane production standards, real mental health support during working hours, and big tech companies taking responsibility for the harm they cause through subcontractors. In the words of UNI Global Union General Secretary Christy Hoffman: "Companies like Facebook and TikTok can no longer hide behind subcontractors to avoid responsibility."

Turkey: The TikTok moderators in İzmir. One link in this global chain is directly in Turkey. Around 1,000 moderators work in İzmir for Telus, the outsourcing company that does TikTok's content moderation. According to the account given in Labor Notes by Kıvanç Eliaçık, head of DİSK's international relations department:

The moderators filter videos containing child abuse, suicide, terrorist propaganda and animal torture. Most earn a little above the minimum wage.
The algorithmic performance targets are strict: if 94 per cent accuracy is not met, bonuses are cut. If "wellness minutes" are overused, it is flagged as poor performance. In one worker's words: "We're allowed to see a psychologist, but only during our breaks."
In early 2024 the workers began joining the independent union Çağrı-İş, and over the summer they passed the threshold needed for legal recognition. Instead of negotiating, the company challenged the union's authorisation.
At least 15 workers were dismissed. Many were dismissed under "Code 25", which pays no severance or unemployment benefit. In March 2025 around 30 more workers were dismissed.
The most striking detail is this: the company argued that the workers should be classified not as "communications" workers, for whom union representation is permitted, but as "general office" workers, and the high administrative court upheld this classification.

This last detail is living proof of the branch-of-activity barrier we take up in the next part. In Turkey, the workers who do the heaviest labour of AI are being de-unionised through branch-of-activity classification. The struggle of the moderators in İzmir is one of the first great class experiences of tech workers in Turkey in relation to AI, and it has not yet received the attention it deserves.

5. Conscientious objection: What will we not work for?

Another axis of the debate is what the tech worker will refuse to produce. In the US in 2024, Google workers held sit-ins against the company's cloud and AI contract with the Israeli government (Project Nimbus), and dozens of workers were dismissed. Similar protests took place at Microsoft in 2025, and the protesting workers were dismissed. At Yale, too, Jennifer Ongley refused to use AI because of its environmental impact and plagiarism.

These examples show one thing: the tech worker's conscience is also a class question. If the worker cannot decide what to produce, their labour can be used in military targeting, mass surveillance, or systems that will put their own class siblings out of work. That is why we put the right of conscientious objection into the model clause in our proposal. When this right is defended individually, it ends in dismissal. Only when it is protected by agreement and by organisation can it become a real right. We discussed the limits of individual conscience through the farewell of an Anthropic researcher in The Class of a Resignation: resigning alone is a testimony, but it does not change the order.

6. The question of independence: At the monopoly's table, or facing it?

The last axis is the most controversial. In December 2023 the AFL-CIO, the largest union confederation in the US, announced a "tech-labour partnership" with Microsoft. The partnership includes "open dialogue" with workers on the development of AI, training programmes, and a commitment to neutrality on union organising at Microsoft.

Partnerships of this kind should be read both ways. On one side there are real gains: the ZeniMax agreement was made possible within this neutrality framework. On the other, there is a risk of the union becoming a legitimiser of the monopoly's "responsible AI" discourse. The class collaborationism we saw in Local 34's call for a "responsible transition" takes on an institutional form in these partnerships. We discussed why the monopolies' claim to regulate themselves is not credible in Can Companies Audit Themselves?, and why workers and the public are not at the table in "safety" measures in Who Holds the Key to the Cage?

UNI Global Union's stance is tougher. In a statement ahead of Davos in January 2026, Christy Hoffman said 2026 could be a critical threshold for AI-driven job losses, and stressed that the rising far right and "techno-authoritarian" politics are attacking unions, weakening regulation and concentrating power still further:

"Tech companies say it is up to governments to manage the social costs." "If we choose to wait, that is a decision to let this inequality win."

This warning coincides with the axis we discussed in the Palantir manifesto and From Futurism to Techno-Fascism: the alliance between the AI monopolies and authoritarian politics is the biggest obstacle to the unions' demand for regulation.

7. The map in summary

Country / organisation	What is it debating, what has it won?	Lesson for Turkey
US, Local 34 (Yale)	Documenting passive automation, the Article 12 dispute, the demand for a staffing floor, the 15 October rally	Keeping a position count and collecting testimonies is the first step of organising
Europe (FES-UNI Europa)	AI provisions in 108 agreements, a joint committee and trial period in Italy	An AI clause can enter existing agreements through an additional protocol
US (NewsGuild, Politico, ZeniMax)	Notification, bargaining, withdrawal of tools through arbitration, bringing contract workers into the unit	The AI clause and the anti-subcontracting clause should be demanded together
Germany (ver.di)	The first AI collective agreement: written and paid consent for a digital copy	A digital asset produced from the worker's labour must stay under the worker's control
Britain (TUC)	A draft AI bill written by the union itself	Writing our own legal text, taking the agenda into our own hands
EU (ETUC, European Parliament)	Demand for a directive on algorithmic management; 51.3 per cent subject to algorithmic management	Regulatory gains are being rolled back with talk of "competitiveness"
US (AWU-CWA)	A job security campaign against Meta's lay-off of 8,000	The illusion of the "privileged software developer" is ending
India (KITU)	TCS's lay-off of 12,000, a 72 per cent fall in hiring	Outsourcing work is among AI's first targets; Turkey faces this risk too
Kenya (DLA), the global alliance	The organising of data labelling and moderation workers	The workers on AI's lower floor can be the vanguard of organising
Turkey, İzmir (Çağrı-İş)	The organising of TikTok moderators, the dismissals, branch-of-activity classification	The branch-of-activity barrier directly de-unionises the AI worker

8. What the global debate is missing

This map is hopeful. But looking from a class perspective, we also see a common shortcoming.

In almost every example the demands are protective: notification, consultation, consent, transparency, human oversight, training. These are indispensable gains. But they all take as given that AI will remain in the employer's hands. The unions bargain over how the machine will be used; they do not ask whom the machine will belong to.

Two demands are on the agenda almost nowhere:

Turning the efficiency gain into shorter working time. The time saved should return to the worker not as unemployment but as a shorter working day at the same pay.
The ownership and control of AI. The models produced from humanity's accumulated knowledge being the private property of a handful of monopolies. Public, open models under workers' control. As we said in A Common Product Is Not Common Without Common Ownership: something produced in common cannot count as common without common ownership.

The Local 34 clerical worker, the moderator in İzmir, the data labeller in Nairobi and the software developer in Bengaluru stand at different links of the same chain. The owner of the chain is the same. The next step in the debate must be a programme that connects these links and puts the question of ownership on the table. The proposal in the next part is a step in that direction.

XIII. An Organising Proposal for Tech Workers in Turkey

The workers at Yale were clerical workers forced to use AI. Tech workers in Turkey stand somewhere else: they are the people who build, integrate, test and sell AI, and at the same time the people meant to be replaced by it.

In this part, by "tech worker" we do not mean only those who write code. Network and system administrators, testers, product and project managers, interface designers, technical writers, content and localisation workers, call centre and technical support workers, data labelling workers and freelancers working through platforms are also included. The jobs AI is entering fastest today are precisely these.

This proposal does not come out of nowhere. On the organising of tech workers we have previously published The Computing Worker's Handbook, the organising guide The End of the Gilded Illusion and Class Tasks in the World of Computing. We also recounted the debates during the founding of BilişimSen in On the Threshold of Bilişim-Sen and Coding the Labor of the Future Together. The proposal below is an up-to-date link in this accumulated work, focused on AI.

1. First the stance: What are we against, and what do we want?

An organising proposal must start with the right stance. Three wrong stances need to be eliminated:

Wrong stance	Why is it wrong?	Our stance
Opposition to technology: "Let us reject AI."	It is not possible to stop technology. This stance places the worker in the past and capital in the future.	We are against not AI, but the ownership and control of AI.
Individual adaptation: "Everyone should upskill; whoever learns to prompt wins."	It reduces the problem to the individual's skill. The worker who learns to prompt is not outside passive automation either. At Yale the workers learned AI, and the positions still melted away.	Not individual skill but organised bargaining power.
Technocratic optimism: "Productivity will rise and everyone will win."	How the gain is shared is a question not of technology but of the power relationship.	The efficiency gain must return to the worker as a shorter working day.

Let us sum up the stance in one sentence: We will not be the ones who break the machine, but the ones who demand its key.

2. The tech worker's double position: Danger and strength

The tech worker stands in two places at once in relation to AI.

The danger side: Code assistants and agents enter software work most of all. Entry-level positions are the first to shrink. The "one senior developer and three agents" model is the software version of Yale's "three administrators' work on one person" model. In other words, passive automation works in computing too: new graduates are not hired, and the places of those who leave are not filled. And the target is not only the software developer. As we showed in Against Whom Is 'Physical AI' Arming?, world models target not only the robot itself but also the simulation engineer and the robotics developer who design the robot.

The strength side: The tech worker is the person who knows the system. They are the first to see which data is collected, which metric is tracked, which module automates whose job. Local 34's members learned what Clarity does from the employer's slides. The tech worker in Turkey builds that system with their own hands.

This knowledge is bargaining power. As we said in our call The Revolt of Crystallized Labor: the machine is humanity's accumulated labour. The organising of the hands that operate that accumulation reopens the question of whom the machine will serve.

3. A four-layer organising proposal

Layer 1: The workplace — Seeing, recording, asking

The biggest gap in Local 34's report was the lack of unit-level staffing data. Organising begins by collecting this data.

Draw up an AI inventory. Which AI tools are used in your workplace? By what decision did they come in? Whose model? Which work data goes to these models? No one will give you this list. You will draw it up.
Keep a position count. Has anyone been hired to replace those who have left your team? How has intern and new-graduate hiring changed over the last two years? This is the earliest indicator of passive automation.
Track metric surveillance. Many code assistants offer managers team-level and individual usage dashboards: how many suggestions were accepted, how many lines were generated, in how much time. The moment this data starts to leak into performance evaluation, AI stops being an efficiency tool and becomes a digital stopwatch. This is mental Taylorism in its purest form.
Measure repair labour. Who is left reviewing the code AI produces, debugging it and closing its security holes, and how long does it take? Ongley's invisible labour exists in software too. The load on the senior developer doing code review is growing, but this load appears in no table.
Set up an AI desk in the workplace. Even in workplaces without a union representative, a solidarity group of three to five people can keep these records. Records are the memory of organising.

Layer 2: The agreement — Putting the gain in writing

Yale's Article 12 is in almost no agreement in Turkey. Below we propose a draft that could be added to a collective agreement, workplace regulations or an individual contract. Treat this draft as a starting text. It needs to be developed according to the conditions of each workplace.

MODEL CLAUSE: Artificial Intelligence and Technological Change

1. Advance notice. At least 90 days before introducing any AI system that could affect the nature of the work, the workload, the number of positions or the monitoring of workers, the employer shall give written notice to the union and to the affected workers. The notice shall include the system's name, its provider, its purpose, the data it will use and its expected employment impact.

2. Bargaining. After notice, the parties shall bargain over the system's effect on work, workload and employment. The system shall not be introduced before bargaining is completed.

3. Protection of positions. No worker's employment contract shall be terminated because of an AI implementation. During the term of the agreement, the number of positions in the workplace shall not fall below the number on the date the agreement was signed. Positions vacated through retirement or resignation shall be filled.

4. Efficiency share. The productivity increase achieved through AI implementation shall be passed on to workers, without loss of pay, as a reduction in weekly working time or as additional pay. How the productivity increase is measured shall be determined jointly by the parties.

5. Ban on surveillance. Usage data from AI tools (number of suggestions accepted, amount of code or text produced, usage times, etc.) may not be used as grounds for individual performance evaluation, discipline or dismissal. Systems performing emotion recognition and biometric monitoring may not be used in the workplace.

6. Ownership of knowledge. Documents, code, correspondence and knowledge of work processes produced by workers may not be used, without the union's consent, to train AI systems that would replace those workers' jobs.

7. Ban on algorithmic decisions. Decisions on hiring, promotion, pay, discipline and dismissal may not be left to AI systems. In every case where AI is used in these decisions, the reasons shall be explained to the worker and the right to request human review shall be granted.

8. Liability. A worker may not be held liable for damage arising from the faulty output of an AI system whose use the employer requires.

9. Training. Training on AI shall be provided during working hours, by the employer and free of charge. A worker who undergoes training shall receive a pay rise for the new skill.

10. Conscientious objection. A worker may refuse to work on AI projects involving military targeting, mass surveillance or the violation of fundamental rights. They shall not be subject to any sanction for this refusal.

Why point 7 is necessary is shown by the Luna/Andon case we described in An AI Boss Fired a Human for the First Time. We presented a broader framework for point 10, a proposal for a socialist code of conduct for AI, in Who Holds the Leash on Artificial Intelligence?

There are examples showing these clauses are not a dream (we looked at them in detail in Part XII):

Screenwriters in the US, at the end of the 148-day strike of 2023, wrote into their contract that AI cannot write or rewrite scripts and that companies cannot force writers to use AI. It was the strike, not the table, that brought the gain.
In Germany, section 87 of the Works Constitution Act prevents technical systems suitable for monitoring workers' behaviour or performance from being used without the works council's consent. So a surveillance ban like point 5 has a legal counterpart.
The European Union AI Act classifies AI systems used for recruitment and worker management as "high-risk". It requires an employer to inform workers' representatives and affected workers before using such a system in the workplace. It also bans emotion recognition systems in the workplace. For firms in Turkey that sell software to the EU or work for EU companies, this is a basis to draw on.

Layer 3: Organisational structure — Getting past the branch-of-activity barrier

In Turkey, computing is not a separate branch of activity. Workers at software firms are mostly counted in branch no. 10, Commerce, Office, Education and Fine Arts. This branch has millions of workers. The 1 per cent threshold makes it almost impossible for a tech union to win the authority to bargain collectively. We discussed this structural barrier in detail in The Computing Worker's Handbook. The experience of the TikTok moderators in İzmir (Part XII) shows how this barrier becomes a weapon: the workers' union authority was challenged by counting the workplace in the "general office" branch rather than "communications".

Against this barrier, our proposal is a multi-layered structure that does not lean on a single organisational form:

Organisational form	Its function on AI
Union	Even without the authority to bargain collectively, it provides membership, legal protection and representation. It connects the AI desks in workplaces. It runs a campaign to take the model clauses into individual contracts and workplace regulations.
Professional chamber (chambers attached to the TMMOB)	Professional ethics principles, technical reports and public information. Independent technical assessment of the public impact of AI systems.
Workplace committees	Can be set up in any workplace regardless of the threshold. Inventory, position count, records of metric surveillance.
Cooperatives	A living example of using AI under workers' control. A different answer, in practice, to the question of whom the efficiency gain goes to. In Turkey, the Albatros Computing Cooperative is an example on this path (our piece on Albatros).
Freelancer and platform networks	Software developers, translators, designers and data labelling workers working through platforms. The section where AI is driving prices down fastest. Solidarity for rate floors and contract templates.

In the long run the common aim of this structure should converge on a single demand: the recognition of computing as a separate branch of activity and the abolition of the branch threshold. The AI debate makes this demand current and concrete, because the issue is no longer only pay but the future of the work itself.

Layer 4: Counter-expertise — Opening the tech worker's knowledge to the class

The tech worker has a specific responsibility that sets them apart from other workers. Local 34's members needed a technical expert to understand what Clarity does. The supermarket worker, the courier, the nurse, the teacher and the municipal worker in Turkey are in the same position.

That is why we propose:

Set up a volunteer technical team. A network of tech workers who can offer other unions and workers' organisations support with an AI impact assessment, explaining what the AI systems entering their workplaces do, what data they collect and what work they automate.
Protect the unions' own data. As we saw in The Aydo Software Operation, even union membership data is not safe. Unions and associations should not upload their correspondence, membership lists and internal documents to commercial AI services. Instead they should be supported with open models that can be run locally and with infrastructure they host themselves. We also discussed the state's view of personal data in KVKK's "80s Trend" Warning.
Defend the public and open alternative. It is not inevitable that AI stays in the hands of a handful of monopolies. Open-source models, public computing infrastructure and the people's control over the people's data are part of the long-term goal. We continue to discuss this within the framework of the Socialist AI Manifesto. We looked at how ecosystems that appear "open" are also drawn into the monopolies' orbit in How Should We Read the OpenAI and Hugging Face Incident? We are also taking this subject to ÖYD's Odak 2026 conference, themed "Free AI – The People's Data, the People's Model" (An Invitation to Odak 2026 and Our Submission).

4. Alliances: The tech worker cannot organise alone

Wherever AI transforms work, the tech worker and other workers stand at the two ends of the same system. A developer writes the shift-planning software for a cashier. A data scientist trains a courier's routing algorithm. A product manager designs the chatbot that will replace a call centre worker.

That is why alliance is not a choice but a necessity:

With couriers and platform workers: The two ends of algorithmic management. As we discussed in our Gig Economy Dossier, the one who writes the algorithm and the one managed by it are parts of the same class.
With office, public sector and health workers: As at Yale, the areas AI enters through passive automation. The tech worker's technical knowledge can be a source of strength at these workers' bargaining table.
With teachers, students and academics: Universities and schools are where the manufacture of consent for AI is most intense. Yale's "Prompt-A-Thons" are arriving in Turkey too, as "AI literacy" campaigns.
With rights organisations: With organisations working in the fields of the KVKK and human rights on surveillance, personal data and algorithmic discrimination.

5. First steps: The next six months

The proposals above may look big. But each begins with small steps:

First month: Draw up the AI inventory in your workplace. Who is using which tool, by what decision?
Second month: Set up an AI desk with three to five people. Start the position count and the records of metric surveillance.
Third month: Adapt the model clause to the conditions of your own workplace. Which points can be demanded today, and which are long-term?
Fourth month: Make contact with a union or a professional chamber. Share your records.
Fifth month: Run a small survey, as Local 34 did. How many people feel forced to use AI? How many positions have been left empty?
Sixth month: Share the results with the public. A report, a blog post, a press statement. Make the invisible visible.

Local 34's members had a union, a contract clause and a forty-year tradition of organising. And still they could not stop their positions melting away. The tech worker in Turkey has none of these. That shows one thing: we have no time to wait.

For practical tools on organising in the workplace and running meetings, see also the Knowledge Commons Guides page.

XIV. Concrete Tasks

Dear young comrades, we do not want this piece to remain an "awareness text". The following points are things that can be done.

For comrades working in workplaces, offices and laboratories

Keep a position count. In the last two years, how many people in your unit have retired, how many have resigned, and how many have been replaced? Put this in a table. This is exactly the most missing evidence in Local 34's report: unit-level data on positions left unfilled. If this data is kept in even a single workplace, it is a powerful organising tool.
Document your job description. Keep a written record of which tasks you did before AI came in and which you do today. Sternberger's sentence "half my job was writing" became evidence because she was able to document her work.
Record repair labour. Note the time you spend correcting someone else's AI output. Invisible labour becomes visible only when it is measured.
Be careful with "best practices" documents. Every document that passes your own way of working to management may become the manual for automating your job. While the knowledge is in your hands, it is bargaining power.
Question AI usage reports. The demand "report how you use it" is a form of surveillance. Ask where this data is kept, who can access it and what it will be used for.

For unions and professional organisations

Get a technological change clause into agreements. Advance notice, impact bargaining and, if possible, a say over the decision itself. Yale's Article 12 can be a starting point.
Put the demand for a staffing floor at the centre. The only structural response to passive automation is to tie the number of positions to the agreement.
Make "a shorter working day at the same pay" concrete. Wherever the productivity increase can be measured, demand that the time saved be turned into shorter working time.
Monitor subcontracted and vendor work. Prevent the new jobs created by AI from moving outside the bargaining unit.
Demand protection against AI errors. No worker should be disciplined because of the wrong output of a tool imposed by the employer.

For the organised movement

Write your own report. What Local 34 did, surveying its members, collecting testimonies and tracking membership data, a union or professional chamber in Turkey can do too. Even in one workplace, one university, one municipality.
Map the capital ties. Draw up for your own employer the table showing the relationship between Yale's board and its investment fund. Who sits on your institution's management? What ties does it have with technology companies? The question "in whose hands?" is useful only when answered with concrete names.

Dear young comrades,

At the start of this piece we said no one was fired. Did you notice how reassuring that sentence is?

If no one was fired, there is no culprit. There is no news story, no photograph, no doorstep. There is only a job advertisement that was never posted, a desk left empty, and a manager asking, "Have you tried Clarity?"

This is also the greatest strength of passive automation: its invisibility. If something cannot be seen, it cannot be objected to. If it cannot be objected to, it cannot be changed.

The real value of Local 34's report is that it makes this invisible process visible. It puts membership numbers on a graph. It records the stories of Sternberger, Kyrtopolous, Ongley, Suess and Greene. It documents the library management's email.

But making visible is the first step. What won Local 34 that "labour peace" was not a report. It was five strikes. And today the one setting the report beside the rally is the same organisation.

The machine is new; the chain is old.

And in the next debate on "how will AI affect jobs?", the question to ask is not "how many people will be fired".

The question to ask is this: Who will sit at that desk again?

Sources

The report itself

Local 34-UNITE HERE!, Artificial University: The Effects of AI on a Clerical and Technical Labor Union, August 2026

The main sources the report draws on

Update from Local 34

"Yale workers demand fair contract, and reins on AI, in largest university rally in decades" — CT Public, 16 October 2026

Reports, agreements and debates from around the world

Examples cited in the organising proposal

Theoretical background

Karl Marx, Capital, Volume I, Chapter 25: "The General Law of Capitalist Accumulation" (relative surplus population, the industrial reserve army)
Harry Braverman, Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century (1974)
Michael Burawoy, Manufacturing Consent (1979)

Related pieces on this blog: A reading route

For young comrades reading this piece, we have arranged our earlier pieces by subject. To start, we suggest the first piece under each heading.

1. The class character of AI: Whose is the machine?

2. AI and labour: Unemployment, intensification, surveillance

3. The monopolies, consent and the "responsible AI" discourse

4. Techno-authoritarianism and the ideology of capital

5. AI policy and surveillance in Turkey

6. The organising of tech workers

7. The alternative: Public, open AI under workers' control

Knowledge Commons | Oğuz Demirkapı | Knowledge Belongs to Everyone This piece is shared under the CC BY-SA 4.0 licence.

Topics:Artificial Intelligence and Technology Law, State and Politics Labour and Organising Economy, Capitalism and Ecology World and Imperialism Surveillance and Digital Rights Unions

Tags:#artificialintelligence #computing #digitallabour #labour #union #organising #class #surveillance #women #monopolycapitalism #usa #turkey

No One Was Fired, the Positions Melted Away

Local 34's "Artificial University" Report, the Debates of Tech Workers Around the World, and an Organising Proposal for Turkey

No One Was Fired, the Positions Melted Away

Local 34's "Artificial University" Report, the Debates of Tech Workers Around the World, and an Organising Proposal for Turkey

Why Local 34?

First, Briefly: What Does the Report Say, and What Do We Say?

I. The Architecture of the Artificial University: Where Did the 150 Million Dollars Go?

Clarity: An AI platform without its own AI

II. The Employer Is Also the Investor: Yale's Capital Ties

III. Passive Automation: Producing a Reserve Army Without Firing Anyone

The numbers

Management's admission: the library email

As Marx said, but quieter

A correction: tax or AI?

IV. "Not Mandatory" Yet Mandatory: The Manufacture of Consent

The Prompt-A-Thon

Reporting requirements and job ads

Burawoy's question

V. What Is Changing in the Labour Process? Four Testimonies

1. Deskilling: The writer becomes a carrier

2. Intensification: Three people's work on one person

3. Invisible repair labour: The bill left for those who do not use AI

4. Handing the hierarchy over to the machine: "Ask Clarity, not your manager"

The second threat: Work shifting to managers

VI. Surveillance: The Subject the Report Mentions in Passing

VII. Women's Labour and the City: Whose Desk Is Emptying?

VIII. The Legal Front: Article 12 and Its Limits

IX. "A Better Way"? A Comradely Critique of the Report's Solution

1. What created the peace is forgotten

2. Inevitability is accepted

3. The best sentence does not become a concrete demand

X. Two Readings, Side by Side

XI. Seen from Turkey

XII. Local 34 Is Not Alone: What Are Tech and Service Workers Debating Around the World?

1. An AI clause in the collective agreement: What was put on the table?

2. The legislative front: Unions writing their own law

3. The core of computing: Those who write the code are organising too

4. The invisible lower floor: Those who train the machine

5. Conscientious objection: What will we not work for?

6. The question of independence: At the monopoly's table, or facing it?

7. The map in summary

8. What the global debate is missing

XIII. An Organising Proposal for Tech Workers in Turkey

1. First the stance: What are we against, and what do we want?

2. The tech worker's double position: Danger and strength

3. A four-layer organising proposal

Layer 1: The workplace — Seeing, recording, asking

Layer 2: The agreement — Putting the gain in writing

Layer 3: Organisational structure — Getting past the branch-of-activity barrier

Layer 4: Counter-expertise — Opening the tech worker's knowledge to the class

4. Alliances: The tech worker cannot organise alone

5. First steps: The next six months

XIV. Concrete Tasks

For comrades working in workplaces, offices and laboratories

For unions and professional organisations

For the organised movement

Dear young comrades,

Sources

Related Posts

As the Golden Handcuffs Come Undone

The Week's Class Struggle in Technology

The End of the Gilded Illusion: Immaterial Labor in the Age of Artificial Intelligence and an Organizing Guide for the IT Worker in Turkey

The Bourgeoisie Is Buying Time with Technology, and We Will Bury Capitalism

The Class Map of the Labour Regime in Trump's America: H-1B, the Green Card and the Fronts It Has Opened

A Debate on Writing with Artificial Intelligence

Newsletter