Why AI Work Culture Is Burning Out Your Team—and What to Do About It

The AI Boom's Hidden Toll on Workers

If you've been following the AI industry lately, you've probably seen the headlines about record valuations and breakthrough models. But behind the scenes, a quieter story is unfolding—one about the people actually building these systems. Recent reports from inside companies like OpenAI, Anthropic, and Meta paint a picture of relentless pressure, 90-hour workweeks, and a culture that treats burnout as a badge of honor.

This isn't just a tech-industry problem. As AI reshapes every sector, the expectations set in Silicon Valley are bleeding into other industries. For HR leaders, managers, and anyone responsible for team well-being, understanding these dynamics is no longer optional. It's central to caring for your people.

90-Hour Workweeks: The New Normal in AI

Let's start with the numbers. Multiple current and former employees at top AI companies told reporters that during major development pushes—what insiders call "sprints"—weekly hours can exceed 90. One former OpenAI technical staffer said he routinely worked 70+ hours a week there. After moving to a startup, his hours dropped to 50–60, but even then, product launches meant weekend troubleshooting and constant availability.

This isn't isolated. At Meta, employees describe being "conscripted" into AI teams without much choice, then expected to build automation systems with no end in sight. The result? Deep exhaustion and a sense of being permanently on call.

Why AI Doesn't Save Time—It Creates More Work

A common promise is that AI will free us up for more meaningful work. But a study from UC Berkeley suggests the opposite: AI tools speed up individual tasks, yet employees end up with more tasks overall. Managers use the efficiency gains to pile on new requirements, not to give people breathing room. You still have to review AI output, fix its mistakes, and learn new tools. The workload doesn't shrink; it morphs.

MIT innovation scholars have observed the same pattern: when AI boosts productivity, companies rarely respond by cutting hours. Instead, they raise expectations and add more deliverables. It's a treadmill that only gets faster.

Token-Based Lending: A New Way to Fund AI Startups—and Its Human Cost

Meanwhile, the financial side of AI is also evolving in ways that affect workers. In China, the Bank of China recently launched a "Token Loan" product, lending money to AI companies based on their token consumption—a measure of how much they're actually using large language models. It's a clever way to assess the health of a business that has few physical assets. The bank has already approved 28 million yuan in credit, with 8 million drawn down.

But here's the catch: companies that take such loans are under pressure to show growing token usage. That means they need to keep their models active, keep users engaged, and keep shipping features. Guess who bears that pressure? The engineers and product teams. It's a direct line from financial innovation to workplace stress.

Stock Options and Clawbacks: When Loyalty Doesn't Pay

Another care-related issue is how companies treat employees when they leave. Take the case of a former Xiaohongshu (Little Red Book) employee, Jiang Dong. He was fired just eight days before his stock options were set to vest. His colleague Chen Hao was let go five months before his options matured, and nearly 50 other ex-employees report similar patterns. Legally, the company may be in the clear—a court ruled the dismissal was lawful—but ethically, it stinks.

This isn't unique to Xiaohongshu. As AI companies grow and their valuations skyrocket, options become a huge part of compensation. But the rules often favor the employer. Employees who are pushed out just before vesting lose out on life-changing money. For HR, this is a red flag: if your retention strategy relies on dangling equity, and your terminations conveniently happen before vesting, you're not just being ruthless—you're destroying trust.

AI's Dark Side: Watermarks, Surveillance, and Privacy

Finally, let's talk about the tools themselves. Anthropic recently announced that some Claude models will embed invisible watermarks in text output. The goal is to help detect AI-generated content, which is fine in theory. But it raises concerns: will a simple grammar check by Claude leave a watermark? Could that watermark be used to penalize someone who used AI for minor edits? And who controls the detection tools?

Similarly, OpenAI's new "Computer History" feature for ChatGPT on Mac tracks your clicks, typing, and app switches across your computer—with your permission, but still. It's designed to help ChatGPT remember what you were doing, but it also creates a detailed log of your work habits. For employees, this could feel like surveillance. For employers, it's tempting to use it to monitor productivity. That's a slippery slope.

What This Means for Your Team

So, what should you do if you're leading a team in this environment? First, push back on the 90-hour week culture. Set realistic deadlines and model healthy boundaries. Second, be transparent about how AI tools are used. If you're tracking productivity, explain why and what you're not tracking. Third, review your termination and vesting policies. Make sure they're fair, not just legal.

Caring for your team isn't just about perks and pizza. It's about designing work that doesn't destroy people. The AI boom is exciting, but it doesn't have to be brutal. You can be part of the change.