Colossus magazine published a three-hour interview with Scott Wu, founder of Cognition, the company behind the AI coding agent Devin. Founded in November 2023, Cognition hit $445 million in annualized revenue within 18 months, with usage doubling every eight weeks. It's currently raising at a roughly $25 billion valuation. Its clients include the U.S. Army, Goldman Sachs, and Mercedes-Benz.
Wu's founding timeline is brutally tight: his mother died of lung cancer on October 6, 2023, after he spent over a year caring for her in Louisiana. A month later, on November 17, Sam Altman was fired by OpenAI's board. Wu happened to be in New York that day, talking startup ideas with friends. When the news broke, he made an immediate call: he canceled Thanksgiving plans with his father, flew to the Bay Area, and organized a hackathon. Two weeks later, Cognition was born.
Wu is a three-time gold medalist at the International Olympiad in Informatics (IOI) and widely considered the greatest competitive programmer in U.S. history. His IOI teammates include Alexandr Wang (founder of Scale AI), Johnny Ho (co-founder of Perplexity), and Jesse Zhang (founder of Decagon) — their companies are collectively valued at about $80 billion. Cognition co-founder Walden Yan once said: "The gap between me and Scott is like the gap between me and people who don't even qualify for the competition."
The interview also revealed a few nuggets: Gennady Korotkevich, widely regarded as the greatest competitive programmer of all time, now works at Cognition; Wu once challenged investor Peter Thiel to chess and card games to negotiate better funding terms; and a significant share of Devin's sessions are now initiated not by humans, but by automated triggers — the agent finds tasks on its own and fixes them.
On the tech side, Devin's SWE-Bench score jumped from 13% at launch in March 2024 to roughly 90% on the original benchmark and about 80% on SWE-Bench Pro. Wu admitted in the interview that there haven't been any massive AI research breakthroughs in the last two years — what's driving the usage explosion is scaled training combined with solid engineering for real-world scenarios.
As for the future of AI coding, Wu cited research from METR showing that the amount of time AI agents can work independently has grown from seconds to hours, and doubles every few months. He believes that curve will continue, eventually allowing agents to work for days, weeks, or even years. He summed up his mission in one sentence: "If I was born to do one thing, it's to teach AI how to code."