I got a Slack ping at 2:13 a.m.: an OpenAI engineer asking if I could review a failing log before they pushed a late-night patch. They had been “paired” with an LLM for hours, hunting a bug that the model had introduced. That moment made one thing obvious — the calendar hadn’t shrunk, even if the tools had become smarter.
I’ll be candid: you and I are taught to expect technology to give us time back. I follow the labs, the execs, the papers. I also talk to people who live the work. What you’re seeing across headlines — promises of three-day workweeks and breathless productivity numbers from OpenAI, Anthropic, Zoom and others — sits next to a different reality: employees burning through weekends and logging 60–90-hour stretches.
Observation: Engineers still sprint through code sprints at midnight
At every AI hub I’ve visited, there’s a ritual: an all-hands, a code sprint, an emergency ship. Teams get a deadline and the whole calendar compresses until people are answering messages at 3 a.m. The industry calls it urgency; employees call it normal.
There are two forces at work. One is the product pressure: models, feature flags, user safety fixes, and launch windows are visible and binary. The other is an upgrade treadmill — each new LLM release raises the bar for what “good” looks like. You don’t ship with yesterday’s model when the rival across the street ships a fresher one.
That creates an arms race. OpenAI and Anthropic publicly promise productivity that could shrink workweeks; privately, teams sprint to keep the company ahead. The result behaves like a turbocharger with a jammed accelerator: more power, but no governor on the hours.
Observation: Executives post their 90-hour weeks as badges of honor
I once read a string of tweets from a founder who bragged about working 90 hours. People cheered; others took it as permission to match the pace. Public narratives shape private norms.
Greg Brockman has openly documented long weeks. Andrew Feldman, the Cerebras CEO, argued that forty-ish hours and “work-life balance” won’t build greatness — and Cerebras is a company valued at $49 billion (€46 billion). When leaders model overwork, it ripples down. Praise for hustle becomes an implicit job requirement: if the boss calls midnight messages acceptable, many will answer them.
If AI helps people work less, why are AI lab employees working so much?
Because the gains get reallocated. When executives see higher throughput, the instinct is to raise expectations. You don’t get a shorter to-do list; you get a longer, more ambitious one. Teams are asked to do more with the same heads — and to do it faster.
Observation: Studies show saved minutes turn into new tasks
An HBS study found that when people work faster because of AI, they generally end up working longer and taking on broader responsibilities. That tracks with what I’ve heard in interviews: saved time becomes expanded scope.
The AI Work Institute’s report is even more granular: some employees report saving as much as 11 hours per week thanks to AI tools, but they spend roughly 6.5 of those hours on maintaining prompts, debugging hallucinations, and correcting outputs. The net gain is smaller than the headlines advertise, and the invisible labor — prompt engineering, verification, tool maintenance — eats into the promise of leisure.
Will companies cut hours when AI increases productivity?
History offers a blunt answer: not unless there’s an economic or regulatory nudge. Businesses typically convert efficiency into scale, margin, or faster feature cycles, not mandatory time off. Shorter weeks require corporate intent, not just better software.
Observation: AI tools are everywhere, but incentives are misaligned
At one lab, an engineer used ChatGPT, Claude, an internal model and a Cerebras-accelerated cluster in a single workflow. The tools saved time, but they also created coordination overhead and new failure modes.
If your KPI is output, AI that increases output will get you promoted. If your KPI is hours on the clock, AI becomes a productivity mask that hides longer shifts. Platforms — ChatGPT, Claude, internal LLMs, NVIDIA and Cerebras hardware — are neutral; companies and leaders pick how the gains are spent.
Here’s the blunt trade: firms can route productivity gains into wage increases, shorter hours, hiring, or faster rollouts. The current trend favors rollout speed and ambition.
So what gives? You’ve read the press releases. You’ve heard the cheerleading from CEOs. The missing piece is worker bargaining power and clear corporate policy that ties AI gains to worker time. Without that, the time AI “saves” turns into a fatter to-do list and a longer day.
Two closing notes: first, not every team runs on adrenaline and late nights — some leaders are experimenting with four-day pilots and radical delegation. Second, if you’re an individual contributor, learn the friction points of your AI stack: prompt failure modes, verification time, and pipeline maintenance are the real time sinks.
AI can change the rhythm of work, but it won’t change the incentives that shape how companies spend their gains. Will the next wave of AI bring you a long weekend or a longer inbox?