There is a particular irony baked into the current AI boom that is difficult to ignore. The companies spending hundreds of billions of dollars building tools designed to reduce human labour are, according to their own staff, working those same humans to the bone. We are talking about weeks that stretch to 90 hours, weekend check-ins just to make sure nothing has collapsed, and performance reviews described as so brutal that colleagues vanish seemingly overnight.
The Promise vs. The Reality
The pitch from tech executives has been consistent, confident, and frankly quite appealing. A Google engineering director said four years ago that AI would deliver a four-day work week by 2025. That deadline has come and gone. OpenAI went a step further earlier this year, formally urging companies to begin trialling a four-day work week at full pay, framing it as a necessary preparation for a future where AI accelerates so much human activity that the traditional five-day grind becomes obsolete.
It was a bold suggestion. The problem, as a BBC investigation has revealed, is that OpenAI never actually trialled that arrangement internally. A former technical employee who left the company last year told the BBC that the work culture inside OpenAI bore little resemblance to the relaxed, AI-assisted utopia the company was recommending to others. Crisis meetings were common. Weekend work was routine. Performance reviews were described as “super cut-throat,” with colleagues departing the company without much warning.
“You go in on Saturday or Sunday just to catch up or make sure things aren’t broken,” the former employee said. That is not a description of a post-scarcity, AI-powered leisure economy. That is a description of a startup in permanent emergency mode.
When the Salespeople and the Product Don’t Match
Anthropic has been especially vocal about what its AI can do. The company has boasted that its coding tool and chatbot, Claude, can work independently for seven hours at a stretch without any human input. A full corporate workday, handled by a machine. The implication is obvious: if Claude can do seven hours of work on its own, surely that frees up human hours elsewhere.
Meta’s Mark Zuckerberg has spoken about this year being the point at which AI “starts to dramatically change the way that we work,” arguing that AI tools allow far fewer employees to accomplish more than was ever previously possible. He is not wrong that fewer employees are doing the work. Meta has since laid off roughly one in ten of its workforce. Whether that counts as AI delivering a better quality of life for those workers is, at minimum, a debatable point.
Anthropic’s chief executive Dario Amodei has been more candid about the trajectory, warning that as AI tools grow more capable, widespread job losses become increasingly likely. It is a rare moment of honesty in an industry that often prefers aspirational language about “augmentation” and “partnership” between humans and machines.
The Gap Between the Executive Floor and the Engineering Desk
What makes this story more than just a tale of corporate hypocrisy is what it reveals about the current stage of the AI race. These companies are not operating from a position of comfortable, confident leadership. They are sprinting. The crisis meetings, the weekend grind, the sudden departures after harsh reviews, these are the hallmarks of organisations under enormous competitive pressure, not organisations coasting toward a leisurely four-day week.
Workers building and using these AI tools are not experiencing reduced workloads. They are, by multiple accounts, working harder than they ever have before. The AI is not yet replacing their effort. It may be adding to it, raising the bar for what is considered acceptable output and compressing the time in which that output must be delivered.
A Structural Tension at the Heart of the Industry
There is a structural tension that the industry has not yet resolved. AI tools may well increase individual productivity, but in a competitive market, increased productivity tends to raise expectations rather than lower working hours. If one engineer can now do the work of two, the company does not necessarily hire half as many engineers and let everyone go home early. It hires the same number, expects twice the output, and uses the surplus to outpace rivals.
This is not a new phenomenon. It echoes debates about automation going back to the industrial revolution, where labour-saving machinery often intensified work rather than reducing it in the short term. The gains, when they eventually came, were won through collective bargaining, regulation, and cultural shifts, not through the goodwill of the companies selling the machines.
What Workers Deserve to Know
The gap between executive messaging and employee experience matters beyond Silicon Valley. Millions of workers across every sector are being told that AI will change how they work, with vague reassurances that the change will be positive. Some of that may prove true over time. But the people closest to these tools, the ones writing the code, training the models, and shipping the products, are not yet living in that future.
They are living in the present, and the present involves long hours, high pressure, and a creeping sense that the technology they are building may eventually come for their jobs too. Dario Amodei said as much himself.
The four-day work week is a genuinely attractive idea. It has research behind it, with trials in multiple countries showing that productivity holds steady or improves when working hours are cut. But it will not arrive as a natural byproduct of AI development, not if the internal culture of the companies building that AI is any indication. It will require deliberate choice, policy, and probably some pressure from the people doing the actual work.
So here is the question worth sitting with: if the engineers building AI cannot achieve a four-day work week, what realistic chance does everyone else have of getting one anytime soon?
