Karrot 基于 AWS 构建特征平台（两部分系列）

Two months after publishing this article, The Wall Street Journal covered the same trend in its article AI workers are putting in 100-hour workweeks to win the new tech arms race. If you'd like to keep up-to-date with the tech industry – and stay months ahead of mainstream media – subscribe to The Pragmatic Engineer.

Hi, this is Gergely with a bonus, free issue of the Pragmatic Engineer Newsletter. In every issue, I cover Big Tech and startups through the lens of senior engineers and engineering leaders. Today, we cover one out of four topics from this week’s The Pulse issue. To get articles like this in your inbox, every week, subscribe here.

“996” stands for “from 9am to 9pm, 6 days a week”, and used to be a common work pattern at Chinese tech companies until it was officially banned starting from 2021. Such extreme working hours have long been rejected in the US. It is even outlawed in Europe because excessive hours tend to lead to burnout and other health issues, longer term.

Despite that, more AI startups are adopting something similar to the 996 work pattern, including Cognition, which expects staff to put in 80+ hours per week. Indeed, the CEO, Scott Wu, was unapologetic about the company’s hardcore culture in a post he shared:

“People have asked about our culture and recent employee communications. Cognition has an extreme performance culture, and we’re upfront about this in hiring so there are no surprises later.

We routinely are at the office through the weekend and do some of our best work late into the night. Many of us literally live where we work.

We know that people who joined Windsurf didn’t expect to join Cognition and while we’re proud of how we work, we understand it’s not for everyone”.

There are several other cases of AI startups mandating grueling hours for workers:

• Lovable: job descriptions by the company detail “Long hours, high pace. Candidates must thrive under high urgency, with AGI timelines approaching.”
• Replit: CEO, Amjad Masad, posted photos of the whole team grinding in the office at midnight, preparing to ship something big.
• xAI: Zeeshan Pate, an engineer at Elon Musk’s AI startup, shared that the team was “grinding in the office day and night”.
• CodeRabbit: just today, the team was still shipping at 2:30am in the San Francisco headquarters
• Icon (ad maker): the founder and CEO shared that the business “only hires [the] top 0.01% engineers with no life”, and expects them to work all week: “why do 6 [days of work] when you can do 7”.
• Google’s AI unit: Google cofounder Sergey Brin told staff in the tech giant’s AI unit: “60 hours a week is the sweet spot of productivity”.

These extreme hours are justified by a sprint to achieve Artificial General Intelligence (AGI) in just a matter of months. That’s because plenty of AI professionals believe that when this point is reached, it will be “game over” for most companies in the segment, with a new, solidified, status quo in place, and AGI being able to improve itself with more resources. From then on, companies with AGI will dominate the industry. This is the incentive for using every means possible (including employees’ labor) to get to AGI – and quickly! Personally, I don’t buy this simplistic prediction about AGI and what might happen when or if it’s reached, but it is the driving force behind the thinking of many founders with commercial pressures.

It’s nearly three years since ChatGPT was released, and there are still no signs of AGI, even though LLMs continuously improve. But what about the exhausting work patterns that are meant to be in place for a few months; could they stay in place for years, and become standard?

Asking or demanding staff to put in very long hours is a recipe for making the pace of work slow down, and for individuals to burn out. We live in an economic system in which companies try to “extract” as much as possible from employees across all industries, so why don’t employers in other sectors also make staff work 80-hour weeks?

In many countries, regulations mandating sensible working conditions are one reason, and unions advocate for this. Another is that the downside of long working hours soon becomes visible:

• Staff take more time off for sickness
• Productivity drops
• More employees quit for places with shorter working weeks

Plus, mandating long working hours automatically excludes many strong potential candidates who…

• have family duties outside of working hours
• live further from the office, and for whom the commute would be too long
• prioritize personal, non-work, activities outside of core hours

Of course, if you hire young professionals, burnout won’t occur so fast – and some people can work for years like this. Plus, if they don’t have families or a busy social life, they may be comfortable with working what looks like “crazy” hours.

Another powerful incentive for an AI startup to create a long-hours culture: the promise of generational wealth. Consider these two questions:

1. Would you work 6 days a week at a startup, and pull 80+ hours per week indefinitely, for less than you’d earn at a company with a 40-hour work week?
2. Would you work 6 days a week and put in 80+ hour weeks for 1-3 years, then walk away with $10M in compensation?

The answer to #1 is likely “no”, but the answer to #2 is “obviously yes” for most people! And the incredible growth in the AI industry means that #2 seems achievable to many. Take Windsurf: just 10 months after launching the Windsurf IDE, 40 employees from the team were acquired by Google. The founders likely made hundreds of millions, and some engineers may have made $10M+ in compensation! That’s not bad for a few years’ work!

Now, put yourself in the shoes of founders who are set to make not $10M, but a multiple of that. For them, it was absolutely worth putting in the long hours and pushing their team to do the same.

Based on that, we can expect founders to keep pushing staff to spend every waking moment at work, or thinking about work.

While there's the promise of making it big with AI startups, these working hours will likely stay. Within AI startups, founders face pressure to ship fast or be out-executed by rivals. Speed is essential to win in AI, time-to-market is essential, and the most obvious way to attempt to get things done faster is to push people to work more.

For the importance of time to market: just look at startups like Magic.dev. A year ago, it raised $515M in funding and claimed it could support 100M token context windows, at a time when most models could not even support 100K. That meant Magic’s model would be a 1,000x improvement on mainstream LLMs! However, a year later and Google’s Gemini already supports 2M tokens, and Claude added support for 1M tokens. So Magic’s lead is cut to 50-100x – which is still considerable, but much reduced.

At the same time, there have not been updates from Magic, and this 100M context window model is not publicly available for use. If the company does not ship a public-facing product soon, they could see mainstream LLMs catch up in context window length, and thereby lose most of their potential customer base in the dev market.

Still, right now there’s a massive business opportunity to make lots of money with AI products, and for early employees to create generational wealth via generous equity – if their startup executes well, and is later acquired for a huge sum.

But long hours alone don’t guarantee success, and there are signs of this. Cognition is proud of its “extreme performance culture”, but its office culture hasn’t quite led to business success: Devin was one of the least-referenced AI tools in The Pragmatic Engineer 2025 survey.

Another sign that grueling hours don’t automatically generate success is the ad maker, Icon. Despite demanding 7-day working weeks from staff, the company seems to be pivoting to become just another advertising agency, offering to create unlimited versions of ads for $1,000/month. Ad agencies are pretty good lifestyle businesses, but rarely the kind of high-growth ventures valued in the billions!

I expect that things will eventually return to normal in a few years' time, when every tech company will also be an “AI company” and it will be business as usual. But for now, the fear of missing out (FOMO) is so strong across the industry that very long workweeks look set to spread.

Everyone’s situation is different, so figure out if you want or need to work extreme hours. But there will be more of a push from AI startups to weed out applicants who are resistant to very long work weeks. Accepting such a position can be an amazing career boost; yes, long hours have many downsides, but you “gel” better with colleagues, and relationships forged under pressure can last decades. Of course, all the stress can also lead to burnout and health issues.

In the end, just know that this trend is real and likely to stick around. If your workplace employs a more “normal” work pattern, it’s worth knowing that this is not necessarily something you can take for granted across tech!

Long hours are not a new thing in tech, and could be just be as much about excitement about LLMs. A now retired software engineer told me how the above summary brought back memories to her of doing pretty much the same, back 50 years ago:

“Thank you for that brief trip down memory lane. Of course this, current generation is working exhausting hours, this technology is new and exciting.

My generation did the same thing with computers back in the 1970s....

Storing cases of beer in the cooling system beneath the raised tiles.

Eating burgers at 7:00 am having pulled an all nighter after system crashes.

I suppose it is all new to you youngers!”

This was one out of four topics from this week's The Pulse issue. Read the full article here.