AI Workloads Push Tech Employees to 90-Hour Workweeks Despite Promises of Shorter Schedules

AI Making Jobs Harder Despite Efficiency Promises

Tech leaders have spent years promoting AI as a labor-reducer that would usher in four-day workweeks and dramatically improve work-life balance. Yet current and former employees at companies leading the AI development demands—including OpenAI, Anthropic, Meta, and Google—are reporting a starkly different reality. Workers describe grueling work schedules that often stretch beyond 90-hour workweeks during intense release periods, with weekend shifts and crisis-driven culture becoming the norm rather than the exception1

1

2

2

.

A former OpenAI technical employee told the BBC they consistently worked at least 70 hours most weeks before leaving the company last year. The person described regular crisis meetings, weekend work to ensure systems weren't broken, and super cut-throat performance reviews that could result in sudden layoffs. After moving to another AI startup, their schedule improved to 50-60 hours weekly, except during code sprints2

2

. This contradiction in AI productivity is particularly striking given that OpenAI itself urged other companies earlier this year to experiment with four-day workweek trials, claiming AI would soon accelerate human labor enough to make shorter schedules viable1

1

.

The AI Arms Race Drives Relentless Pressure

The punishing schedules stem largely from the AI arms race, as companies rush to improve models, build computing infrastructure, and add AI features to products simultaneously. At both OpenAI and Anthropic, workers reported that sprints can stretch for many weeks and exceed 90 hours in a seven-day period2

2

5

5

. This creates enormous pressure on engineers, researchers, and product teams to move quickly and fix problems as they emerge, with little consideration for employee burnout or sustainable work practices.

At Meta, the situation involves forced reassignments that employees have dubbed being "drafted." One current and one former employee described how workers were abruptly moved onto urgent AI teams with little meaningful choice. "They just move you over," the former employee explained. "You can't say no—or if you do, you have to quit."1

1

2

2

These teams work on AI systems for software engineering and infrastructure that measures how well models can perform work typically done by people. Workers described late nights, weekend work, and feeling on call even during off hours. The former Meta employee characterized the work as "endless," noting that teams are "literally working in teams of people trying to replicate humans doing jobs."1

1

AI-Driven Efficiency Gains Create New AI Workloads

The human cost extends beyond those directly building AI models. Former Google employee Amin Shali left the company in May 2024 after witnessing how AI priorities affected internal engineering systems. He reported that Google had shifted key resources—including processing units and memory storage—to AI projects, leaving other teams dealing with frequent disruptions that often required engineers to work through the night. Since leaving, Shali said his sleep and health have improved significantly, and he criticized how heavy AI tool usage at large companies "creates a bad culture with excess pressure on engineers."1

1

5

5

Paradoxically, even as companies demonstrate real productivity improvements from AI tools, these gains aren't translating into reduced hours for employees. Anthropic reports that more than 80% of code merged into its codebase was authored by Claude as of May, with typical engineers merging eight times as much code per day during Q2 2026 compared to 20244

4

5

5

. OpenAI says its Codex tool has become the primary AI work tool across every department, including Legal, Finance, and Recruiting4

4

.

Understanding the AI Productivity Paradox

Research from UC Berkeley helps explain this AI productivity paradox. An eight-month study following hundreds of workers at a US tech company found that employees using AI worked at a faster pace, handled more tasks, and worked later into the day. Rather than creating free time, AI tools led workers to expand what they attempted, with work creeping into periods that previously served as breaks. Workers also had to spend additional time checking AI-generated output for accuracy1

1

4

4

. A separate report from the AI Work Institute found that while some employees save as much as 11 hours per week thanks to AI, they spend approximately 6.5 of those gained hours maintaining AI workflows and fixing mistakes3

3

.

Neil Thompson, an innovation scholar at MIT, noted that companies are unlikely to give workers more free time when AI creates efficiency gains. "People assume that 20% less work means four-day weeks," Thompson explained. "But new work emerges."1

1

5

5

Instead, those gains often lead to new tasks, more oversight, higher expectations, and job insecurity as companies consider whether fewer workers can accomplish the same output.

This disconnect raises fundamental questions about who benefits from AI-driven efficiency gains. While executives like Meta CEO Mark Zuckerberg claim AI allows "far fewer employees do more than they ever could have before," and Anthropic CEO Dario Amodei warns of broader job losses as AI becomes more capable, the current reality shows increased workloads rather than reduced hours2

2

5

5

. The pattern suggests companies are using productivity improvements to pile more onto employees' plates rather than creating the promised four-day workweek or better work-life balance that AI was supposed to deliver.