New trend: extreme hours at AI startups - The Pragmatic Engineer

• Companies like Cognition, Lovable, and Icon are openly demanding 80+ hour weeks, 6-7 days—even Google's AI unit says "60 hours is the sweet spot"
• The economic incentive is powerful: Windsurf's 40 employees were acquired by Google after just 10 months, with some making $10M+
• Time-to-market pressure is real—Magic.dev's 1000x context window advantage shrank to 50-100x in one year as competitors caught up
• Long hours don't guarantee success: Cognition's Devin was one of the least-referenced AI tools despite their "extreme performance culture"
• This will persist as long as the wealth creation opportunity exists, but automatically excludes anyone with family duties or who lives far from the office

AI startups are normalizing work patterns similar to China's "996" culture (9am-9pm, 6 days/week), which was officially banned in 2021. Companies like Cognition explicitly demand 80+ hour weeks, with CEO Scott Wu stating they "routinely are at the office through the weekend" and "many of us literally live where we work." Other examples include Lovable requiring candidates who "thrive under high urgency, with AGI timelines approaching," Icon demanding 7-day work weeks from "top 0.01% engineers with no life," and even Google's Sergey Brin telling AI unit staff that "60 hours a week is the sweet spot."

The driving force is a two-part incentive structure: the belief that AGI will be reached in months (creating a "game over" moment for competitors), and the promise of generational wealth for early employees. The key question isn't "would you work 80 hours/week for less money than a 40-hour job?" but "would you do it for 1-3 years to walk away with $10M?" Windsurf demonstrates this isn't fantasy—40 employees were acquired by Google just 10 months after launching, with some likely making $10M+. For founders set to make hundreds of millions, pushing teams to work extreme hours becomes rational. Time-to-market pressure is intense: Magic.dev's 100M token context window advantage (1000x better than mainstream LLMs) shrank to just 50-100x in one year as Google and Claude caught up.

However, long hours don't guarantee success. Cognition's Devin was one of the least-referenced AI tools in surveys despite their extreme culture, and Icon appears to be pivoting to become just another ad agency. The usual downsides of overwork (burnout, productivity drops, attrition) are being temporarily overridden by FOMO and wealth potential, but this automatically excludes strong candidates with families, long commutes, or outside priorities. A retired engineer notes this isn't new—her generation did the same in the 1970s with early computers. The trend will likely persist for years while the AI opportunity remains hot, not just the "few months" founders claim.