Syndicalist Valley
On February 27, 2026, hundreds of well-paid employees at OpenAI, joined by over a thousand at Google, publicly backed Anthropic’s refusal to grant the Pentagon unrestricted access to its AI models and demanded their employers hold similar red lines against autonomous weapons and mass surveillance. Then in April, Mark Zuckerberg announced that Meta would be collecting its own workforce’s keystrokes, mouse movements, and screen contents for model training. In response, over a thousand of his employees signed an open letter to demand an end to the practice (they partly succeeded). But the boldest instance of worker power took place early in the summer. Triggered by Google’s classified deal to deploy its Gemini models inside military networks, Communications Workers Union members at DeepMind’s UK operation voted 98 percent in favor of unionizing, in a bid to block the company from providing AI technology to the US and Israeli militaries.. These internal battles complicate what the public often takes for granted: that the people working at these companies, especially those pulling seven- or eight-figure salaries, see eye-to-eye with their bosses. They often don’t. But who are these workers? What are their demands? And how far will they go to achieve them?
What is driving these collective actions against employers are two nascent tech worker movements. Recent economic, political, and technological changes have wracked the industry, creating different strata of tech workers separated by company business models, AI-adjacency, and geography. As the tech industry’s labor force has diverged, so too have its grievances, demands, and tactics. Understanding which tech workers are taking collective action, and why, requires understanding how the business models underpinning the tech industry have shifted.
Tech workers’ ambition to fight their bosses for a brighter technological future is not new. For many years, both of us have analyzed the US tech industry in an attempt to understand the relations between labor and capital. In the absence of a stronger regulatory state, we’ve been especially interested in the potential for workers to act as levers of accountability. In the late 2010s, this seemed promising, as tech workers started to organize to stop their employers from working with the military, engaging with oil companies, and turning their backs on their own pro-social missions. These actions coalesced into something unmistakable: a tech worker movement.
But beginning in 2022, the movement appeared to be slowing. The conditions that gave rise to it in the late 2010s have radically changed, and the industry is more willing to squash dissent. Across the country, tech workers are increasingly proletarianized and worried about layoffs. With agentic coding becoming popular, workers have lost what used to be considered a rare superpower—writing code—and are now much more fungible on the labor market. Meanwhile, some tech workers have become high-agency hustlers, leaning into the AI hype and using it aggressively in hopes of escaping the permanent underclass.
The tech workers behind the movement of the 2010s no longer feel powerful. And yet, many are still taking collective action. One group of workers has lost their market leverage and, as a result, become proletarianized. Another has ridden the wave of AI hype and is now grasping for one last chance at upward mobility. And while the tech labor movement of the 2010s was driven by a widespread belief that tech could be a force for social good, a similarly technologically deterministic ideology—this time steeped in fear—is spreading across AI labs today. Understanding that these two separate waves of collective action have the potential to join together is key to understanding this new era.
Why Workers Have Lost Their Leverage
Software has always been an attractive business: build it once, and it can operate without much additional labor. This basic dynamic was intensified with the rise of internet-era platform companies, whose products not only had this build-once-and-forever-profit characteristic, but also benefited from network effects whereby the value (and thus hopefully revenue) of a piece of software (or platform) increased as more people used it. Of course, there are caveats to this. A firm may need to constantly improve the product in order to stave off competition; it may need to overhaul its infrastructure to accommodate increases in its userbase. But the basic principle remains: low marginal costs for each additional user.
This means the vast majority of the industry’s corporate employees do not generate any immediate profit for the firm. Software engineering is fundamentally an occupation concerned with automating things. Code written once can be copied infinite times and never needs to be rewritten. Ideally, tech workers never do the exact same work twice. They build something and then move on, spending some time maintaining prior projects but mainly shifting their focus to something new. In other words, tech companies are inherently future-oriented, with most employees not actively involved in their employer’s current revenue.
In the 2010s, this business model was particularly appealing because betting on the future was cheap. Interest rates for nearly the entire decade sat at just about zero, giving employers the financial freedom to spend lavishly on hiring tech workers to build new products; nearly all of their spending went straight to the large salaries that their employees commanded. But when the zero-interest-rate era ended in 2022, tech was disproportionately impacted: With higher interest rates, the cost of spending on speculative projects increased, and the high salaries became harder to justify to shareholders.
Since then, hiring has cooled and an industry long thought to be immune to layoffs has turned to aggressively cutting costs on its largest operating expense: its workers. Layoffs have become the new normal. In 2022, according to Layoffs.fyi, they shot up tenfold from the preceding year to over 165,000, rising to 260,000 in 2023. Since then, the industry has consistently laid off well over 100,000 workers every year.
Economic forces, not ChatGPT, are at the heart of the industry’s recurring layoffs. That does not mean AI has had no effect. We shouldn’t fall for tech companies using AI as a cover for layoffs. There is no strong evidence that AI has increased productivity in tech enough to justify the cuts. But AI has introduced new dynamics that are fundamentally transforming the industry—chief among them is the shift from spending on workers to spending on tokens and their underlying infrastructure.
The firms now at the industry’s frontier, including the AI labs and the hyperscalers, are no longer the asset-light software companies of the past. Whereas the asset-light business models could expand at near-zero marginal costs, the industry as a whole has become extremely asset-heavy. And that is because every prompt answered or every token spent incurs a very material cost: GPU usage, electricity consumption, and the specialized data-center infrastructure connecting it all. Not only has the labor market worsened for tech workers, but they now have to compete with GPUs and data centers for a shrinking share of company spending.
Tech companies outside of the top AI firms also feel the pressure to reduce labor costs. Promises about the productivity gains from agentic coding tools have led to greater enforcement of AI usage, and even employers that aren’t convinced by AI’s supposed benefits nevertheless feel pressure from their boards or investors to take it up. Of course, more spending on AI tokens means less money left for salaries and so—even without any clear sign that AI uptake is paying off—employers are forced to weigh AI against their own workers. And since tech workers work on speculative profits rather than guaranteed current ones, laying them off does little immediate damage to the company. No wonder they feel their positions are very precarious.
The New Locus of Labor Market Power
The new financial environment, laden with layoffs, has changed the very nature of the industry’s labor relations. Gone are the days of the perk-filled 2010s tech job when workers enjoyed high pay and enormous autonomy, not only in how they worked, but even in what they worked on. Today, the tech workers whom we’ve interviewed report an upsurge in performance-based management practices, including surveillance and metrics-tracking. In some cases, tech companies have created internal dashboards to visualize and quantify personal metrics such as lines of code committed or units of software changes published, which many tech workers believe have, at best, a tenuous connection to actual productive value. Occasionally those dashboards are searchable for employees’ stats, which are sometimes even ranked. Tech workers are reporting speedups and work intensification, either explicitly demanded by management or implicitly suggested by metrics collection.
AI alone is not the primary reason for these speedups. Against predictions from AI CEOs, the jobs of software engineers in particular are not meaningfully being replaced. Despite employers’ best efforts, knowledge work is historically difficult to surveil and therefore difficult to direct and deskill (that is, to break down into simpler, more standardized tasks that require less individual expertise). Indeed, Emily Mazo’s research shows that software engineers have taken up AI coding tools in many different ways, which demonstrates that these workers still have autonomy in their labor process (even if the use of AI tools in a company is mandatory). What we are seeing instead is that the highly skilled and in-demand coders and data scientists, who previously enjoyed a more relaxed pace of work and a stable labor market, are pressed by management to work much faster because they no longer feel individually powerful enough to refuse.
But not all tech workers have lost bargaining power. Where the previous era’s prestige hierarchy placed the seasoned engineer at its apex, this new era has elevated the AI researcher with dizzying compensation. Competition over talent is high and salaries in the seven or even eight figures are common. The market leverage these workers hold over their employers is, if anything, comparable to that of the executive class of the platform era (which is why Meta spent billions on hiring AI talent, dolling out tens of millions on individual hires). Yet this group has its own distinctive constraints. Whereas millions of tech workers were needed to build the empires of last generation’s giants, the new AI titans need only tens of thousands of AI researchers. This group is also heavily visa-dependent, composed largely of Chinese and Eastern European immigrants whose domestic political involvement is limited—even as their employers are very politically active. The new elite is powerful, but much smaller and more politically isolated.
Another faction of tech workers is trying to ride the AI wave as well. They are largely located in San Francisco, which has become the gravitational center of the AI transition in a way that even Silicon Valley at its peak was not. They form a peculiar intermediate stratum: SF-based software engineers who are not AI researchers but who remain close enough to the frontier to believe they will escape the proletarianizing pull. These are the 996-ers diving headlong into agentic coding tools, the high agency strivers working punishing hours to escape the permanent underclass. They are the corporate employees jockeying to become the company’s AI evangelist, the administrators of the automation tools—in an attempt to outrun automation by wielding it against other tech workers. These workers have made a strategic decision to try to get as much as they can out of the AI boom while it is still possible to do so. By contrast, the vast majority of tech’s far more distributed workforce—employees in Seattle, Austin, or Bangalore—faces the proletarianizing dynamic more fully, without the proximity to the AI frontier that might offer escape.
So even as the gains of the AI transition have yet to be realized, the new financial pressures on the industry are driving tech workers to position themselves in quite different ways—making some far more likely than others to engage in collective action.
From Californian Ideology to the New Rationalist Movement
To understand what motivates tech workers to organize in this new environment, it’s useful first to understand what motivated tech workers to organize in the past. The tech worker labor movement of the 2010s and early 2020s was empowered not only by that era’s tight labor market, which gave workers significant leverage, but by an industry culture that had long cultivated the belief that its employees were, above all, changing the world for the better. The twin ideologies of techno-solutionism and techno-utopianism, entwined with the libertarianism of the late-twentieth-century tech industry, helped attract workers who could have otherwise become doctors, architects, or lawyers. Google’s first motto was “Don’t be evil.”
In their influential essay “The Californian Ideology”(1995), Richard Barbrook and Andy Cameron describe the formation of an outlook they saw pervading Silicon Valley. In the 1960s and ’70s, hippies and the New Left brought to the nascent tech industry a utopian belief that new technology would lead to the creation of an “electronic agora”—an egalitarian space for free speech—which would support their ideals of democracy, self-fulfillment, and social justice. Their orientation against the state, in particular against the military-industrial complex, melded well with another important influence: the New Right, which subscribed to a libertarianism that opposed the state’s interference with the free market. Both groups saw in computing and the internet “an optimistic and emancipatory form of technological determinism,” where both social liberalism and economic liberalism would be realized by this new technology. Activists, artists, and entrepreneurs could use technology to create, share, and sell, without the gatekeeping or interference of big corporations or the government. Together they could create a freer and fairer world.
The tech workers of the ’70s through the 2010s inherited this ideology. They believed that, without sacrificing anything in the way of potential earnings or working conditions, they could contribute to the social good. So when tech workers discovered that their bosses were pining for contracts with the US Department of Defense, ICE and CBP, and the Chinese government, thus ditching their end of the social contract (Google dropped “Don’t be evil” in 2018), they felt betrayed and decided to fight back. They challenged company leadership at company meetings and online, and organized to write open letters and participate in walkouts, as 20,000 Googlers did the same year.
The public backlash against the tech industry during the first Trump administration and the rightward shift of tech oligarchs during the second have washed away much of the labor force’s techno-optimism. Many of the collective actions that targeted social harms like government contracts led to retaliation from the companies against organizers. Tech companies in the cloud and AI industries became dependent on both government contracts and permissive state regulation to build data centers and allow unchecked AI growth. Few workers entering the tech industry today would think that platform companies of the 2010s were a positive force for democracy or public good.
Meanwhile, the risks posed by AI came to the fore, which brought the rationalist community center stage. First emerging from blogs such as LessWrong and Slate Star Codex, the rationalist community was dedicated to learning how to think more accurately about the world. An early champion was Eliezer Yudkowsky, who was obsessed with the existential threat posed by AI many years before ChatGPT ever came online. Other writers in the community treat the advance of AI as a coin flip between utopia and extinction. For decades, they have argued that the development of some kind of superintelligence could happen through recursively self-improving models that would pose a terminal risk to humanity if its core utility function (the function it exists to optimize for) is misaligned. When generative AI appeared in 2022, these preexisting ideas of the rationalist movement were suddenly ripe for dissemination; its adherents became key voices in the new field of AI Safety. Everyone in San Francisco began asking each other, “what’s your p-doom?” (the probability of existential catastrophe).
This convergence between rationalist ideas and the accelerating capabilities of AI has formed the basis for much of the dissent within today’s frontier AI labs. Dario Amodei and other researchers at OpenAI protested Sam Altman’s leadership by quitting and starting Anthropic, which set out to be much more sensitive to the concerns of AI Safety. Later, when OpenAI formed a “Superalignment” team, which similarly aimed to keep AI systems aligned to humanity’s benefit, it was quickly dissolved after key leaders resigned, stating publicly that safety had taken a backseat to shiny products. This tension has recurred every few months across the frontier labs. Safety-oriented researchers protest by resigning, sometimes with a public letter accusing their companies of sacrificing safety to the financial pressure of shipping AI models. The same concern with safety has fueled organizing at OpenAI and Google’s DeepMind as hundreds openly protested their labs’ willingness to collaborate with the military. At DeepMind, this dispute escalated into a union drive. What’s common across these cases is the conviction that these labs bear personal responsibility for increasing the probability of ending humanity. Raising the alarm internally—or wielding the tools of labor organizing to slow things down—is simply the most rational way to decrease the odds of extinction.
What we are seeing is not one undifferentiated movement, but multiple movements, with varying goals, whose differences are easily overlooked or occluded. At OpenAI and DeepMind, in-demand workers with high amounts of individual labor market power are speaking up about their bosses enabling the use of the technology they have built for harm. At Meta, the majority of signees of the 1,600-person open letter against screen-recording software were non-AI workers, who felt at risk of imminent layoff, and who were protesting the degradation of their everyday work. These non-AI workers are beginning to resemble the populations that the US labor movement is used to organizing: workers who want to bargain over job security and working conditions. Meanwhile, workers in AI labs have begun organizing around military contracts, AI safety, and other issues beyond the workplace.
Independently, these two movements, still young, have much to win. The AI workers, if united, can act as a bulwark against the profit-driven incentives that have spurred their CEOs to race ahead, often at the expense of building safer, smaller, and less sycophantic models. Other tech workers, disempowered by the labor market, remain in critical positions across the industry that is aiming to disseminate AI to the rest of the economy. Notably, the 996-strivers—that intermediate stratum—are missing in action. They have largely bought into the AI upskilling narrative and are dealing with their jobs’ precarity as individuals, often at the expense of other workers. In the short term, this grind-or-get-left-behind mindset might give tech workers an easier path out since it’s very hard to organize this sector. However, joining forces, seeing where their labor intersects, and combining their demands are where all tech workers will find the most power. Less-precarious AI workers could lend their power to the collective, and more-precarious non-AI workers could bring their numbers. Together they could demand democratic control over what gets built and who gets to use it, as well as basic protections and job security. In the face of millions of dollars of lobbying over AI regulations, a powerful tech worker movement could play a decisive role in the future of AI and the future of labor.
