AI Employees Put In 100 Hour Work Weeks As Pressure For Results Rises

The battle in tech to capture the optimal benefits of AI - both for the thrill of being first - and to out-maneuver the competition and thereby grabbing first mover advantage and bigger margins - is leading to 80 to 100 hour work weeks at AI companies. 

The pay is commensurate with the effort and the excitement as AI surges to meet the expectations of investors and tech executives. If it all seems a tad self-indulgent and self-congratulatory, that is the nature of the industry, now in what some argue is the third wave (dot-com, mobile phones and now AI) of the Silicon Valley phenomenon. JL

Bradley Olsen and Meghan Bobrowsky report in the Wall Street Journal:

Inside Silicon Valley’s biggest AI labs, top researchers and executives are regularly working 80 to 100 hours a week. Microsoft, Anthropic, Google, Meta, Apple and OpenAI have said they see their work as critical to a seminal moment in history as they duel with rivals and seek new ways to bring AI to the masses. The time between research breakthroughs and product launches has compressed. Startups have included the expectation of an 80-hour-plus workweek in their employment contracts. The breakneck pace is concentrated among a subset of workers trying to improve core AI models or take emerging capabilities and embed them into new products. Credit-card transaction data shows a surge in Saturday orders from San Francisco-area restaurants for delivery and takeout from noon to midnight.

Josh Batson no longer has time for social media.

The AI researcher’s only comparable dopamine hit these days is on Anthropic’s Slack workplace-messaging channels, where he explores chatter about colleagues’ theories and experiments on large language models and architecture.

Batson is among a group of core artificial-intelligence researchers and executives who are facing a relentless grind, racing to keep pace with a seemingly endless cycle of disruption in pursuit of systems with superhuman intelligence.

Inside Silicon Valley’s biggest AI labs, top researchers and executives are regularly working 80 to 100 hours a week. Several top researchers compared the circumstances to war.

“We’re basically trying to speedrun 20 years of scientific progress in two years,” said Batson, a research scientist at Anthropic. Extraordinary advances in AI systems are happening “every few months,” he said. “It’s the most interesting scientific question in the world right now.”

Executives and researchers at Microsoft, Anthropic, Alphabet’s Google, Meta Platforms, Apple and OpenAI have said they see their work as critical to a seminal moment in history as they duel with rivals and seek new ways to bring AI to the masses.

Some of them are now millionaires many times over, but several said they haven’t had time to spend their new fortunes.

The competition for AI talent kicked into high gear when Mark Zuckerberg began poaching top AI workers from rivals, offering multimillion-dollar pay packages. That showed how the work of a relatively small cluster of researchers and executives was one of the world’s most precious resources. Now companies and the workers themselves are seeking to wring as much work as possible from these individuals each and every day.

“Everyone is working all the time, it’s extremely intense, and there doesn’t seem to be any kind of natural stopping point,” Madhavi Sewak, a distinguished researcher at Google’s DeepMind, said in a recent interview.

Madhavi Sewak of Google’s DeepMind spoke on her work schedule and other topics in a video on YouTube from a recent Imagination in Action conference.

Certain startups have included the expectation of an 80-hour-plus workweek in their employment contracts, people familiar with the matter said. But many companies haven’t had to do so, as top AI employees are motivated by the intense competition with rivals and their own curiosity about new model possibilities.

At Meta, members of TBD Lab—new hires tasked with developing the company’s AI models—work in person near Zuckerberg’s desk at the company’s Menlo Park, Calif., headquarters. The company on Wednesday cut roughly 600 jobs from its AI division.

While round-the-clock schedules have been common among startups during various boomtimes in Silicon Valley’s history, the need for extreme hours at some of the world’s biggest companies has been relatively rare.

Working 0-0-2

The most-intense periods for many come while working on models or new products, when time working extends beyond the “9-9-6” schedule. That stands for 9 a.m. to 9 p.m., six days a week. One startup executive jokingly referred to the schedule as “0-0-2,” meaning midnight to midnight, with a two-hour break on weekends.

The breakneck pace is concentrated among a relatively small subset of workers at different companies, many of whom are trying to improve core AI models or take emerging capabilities and embed them into new products. These small teams often work at all hours, even after many lesser-paid colleagues have gone home, workers said. Although many said the grind has worn them down and prevented them from spending time with family and friends, they also said repeatedly that they work long hours by choice.

“You have all these good ideas, and you know it’s a competition against time,” Sewak said. “You don’t want to let ideas go unexplored, so when you have free time, you’re working on” and chasing other new ideas.

All around the valley, companies have begun to accommodate employees who are in the office almost nonstop. Some serve food on weekends. And some ensure staffing is always in place. Companies appoint “captains” for short rotations to closely monitor model outputs, or for extended, multiweek periods in which they are overseeing product development, according to AI workers.

Corporate credit-card transaction data from the expense-management startup Ramp shows a surge in Saturday orders from San Francisco-area restaurants for delivery and takeout from noon to midnight. The uptick far exceeds previous years in San Francisco and other U.S. cities, according to Ramp.

Aparna Chennapragada, Microsoft’s chief product officer for AI experiences, said that while she has been through past moments of extreme urgency in tech, this one is different.

Aparna Chennapragada of Microsoft commenting at a June tech gathering. David Paul Morris/Bloomberg News

In previous boom times, such as the dot-com boom in the late 1990s or the accelerated move to mobile when the iPhone arrived, user adoption happened over a decade or more. With AI, 90% of the Fortune 500 are already using AI products after just a few years, she said.

‘Nerds are having our moment’

The time between research breakthroughs and product launches has compressed from several years in the past to “the gap between Thursday and Friday” in the AI age, creating immense demand that every company is rushing to meet, she said.

Chennapragada has taken to calling the extra, unimposed responsibility a “second shift” she faces as a manager. To deal with it, she has built some of her own AI tools aimed at speeding up her work, including a browser extension. Every time she opens a new tab, the extension asks her how she could be using AI to do whatever she is doing differently, she said.

Although industry chatter has focused on 9-9-6 work schedules, or even 24/7, she has steered Microsoft employees toward AI as a solution for enormous workloads.

“That 24/7 is not you,” she said. “It should be your AI.”

Josh Batson of Anthropic compares his intense workload to what he experienced during the pandemic. Anthropic

Anthropic’s Batson said he feels that it is hard to plan or schedule work because of how AI models are advancing and how hard it is to know ahead of time how they will behave. The process more closely resembles evolution than engineering, he said.

“You often don’t exactly know what you’re going to get out of training until you’ve got it,” he said. “And you don’t exactly know what it’s going to do until you test it. And even then, you don’t know exactly what it’ll do until it’s deployed in the wild.”

He compared the intensity of the experience to his work in a stand-up testing lab during the Covid-19 pandemic, when he coordinated with a small group of international scientists to understand the trajectory of the virus.

His company’s stated concern for building ethical, human-aligned AI fuels his work ethic, Batson said. He and co-workers are trying “to see if we can increase our understanding at a faster rate than the models are changing, and I think we are hitting that goal.”

Google’s Sewak said that she has enjoyed seeing top AI researchers be paid for all of their smarts and work, even if many continue to be motivated by their work alone.

“I’m very excited that nerds are having our moment,” Sewak said. Most people, though, haven’t yet managed to change their mode of living, she said.

“I see no change in anyone’s lifestyle. No one’s taking a holiday…People don’t have time for their friends, for their hobbies,” or for the “people they love,” Sewak said. “All they do is work,” she said.