Scott Wu

Scott Wu is Cognition’s co-founder and, as identified in his June 2026 David Senra interview, its CEO. His central bet is that AI should work as a software-engineering coworker: carrying out multi-step tasks in its own environment and returning work for human review. He describes the broader ambition conditionally: if Devin succeeds, it becomes “the way that humans can tell their computers what to do.”[3][7]

Wu introduced Devin in a March 12, 2024 announcement. The team equipped it with a shell, code editor, and browser in a sandboxed compute environment, combining those tools with long-term reasoning and planning. The intended workflow included real-time progress reports, user feedback, and collaboration on design choices—not simply generating a block of code.[4]

Cognition reported that Devin resolved 13.86% of issues end-to-end on a random 25% subset of SWE-bench, a benchmark drawn from real GitHub issues. Devin was unassisted; comparison models were told which files to edit. Those were historical launch results, and Wu later acknowledged that the release was a prototype demonstration rather than an enterprise-ready product.[4][3]

Wu recalled an early success in which Devin set up MongoDB after the team had repeatedly encountered errors. Instead of asking a chatbot for each next step and manually executing its advice, they let Devin run commands and work through the problem. He said he could not sleep afterward—but also described it as their one success at that point and substantially better than an average run.[3]

Unsuccessful early pilots pushed Wu and the team toward repetitive, scoped tasks, particularly migrations and version upgrades. His reasoning was specific: these jobs contain enough variation to defeat a fixed script, yet offer repeated patterns and tight feedback loops that an agent can use to diagnose and correct mistakes. Broad architecture problems required more context and judgment than the early system could reliably supply.[3]

Wu described a migration at Nubank as the team’s first successful deployment. He and the whole team flew to Brazil, sat with Nubank’s engineers, tried parts of the work manually, and learned what Devin needed to read and understand. They then adapted the agent and its orchestration to that particular task. In his account, getting the first customer working meant nearly building the product for one company.[3]

In June 2026, Wu explained Devin as a compound model system that selects models for different tasks and subtasks. Investigating a ticket, reproducing a bug, examining logs, debugging, testing, and preparing a change for review impose different demands. Difficult reasoning may warrant the strongest available model; repetitive work may favor speed, cost, and an easily verified answer. He said Devin dynamically chooses among external providers, Cognition’s own models, and open-source models.[3]

Wu connects provider neutrality to customer value: Cognition should help teams choose models and workflows that produce useful results efficiently. He criticized ranking engineers by token consumption, arguing instead for measuring completed work and output. His preference for an independent company is similarly qualified rather than absolute: “We would sell if we thought it was the most ambitious thing to do.”[3]

On July 14, 2025, Wu announced a signed definitive agreement to acquire Windsurf. His stated rationale combined Devin’s autonomous agent with Windsurf’s IDE and go-to-market organization. In their July 2026 joint retrospective, Wu and Jeff Wang recalled complementary needs: Cognition sought a go-to-market organization, while Windsurf sought engineering capacity. Their cloud agent and IDE also offered distinct product starting points.[5][6]

Wu said employee treatment was a priority in structuring the agreement. He credited his work with Jeff Wang and announced financial participation for all Windsurf employees, waived vesting cliffs, and fully accelerated vesting for work to date. These were the deal commitments he published, rather than a promise about subsequent staffing retention.[5]

In his May 2026 Colossus interview, Wu identified accumulated “soft context” as an important remaining human advantage. An engineer may remember why a system was originally designed one way, which bug appeared months earlier, or how a team’s conventions differ from others’. He distinguished retrieving those relevant memories from merely having a large context window, and discussed continual learning—updating model weights from experience—as an active research direction.[7]

Wu’s optimistic vision emphasizes creativity and self-expression. He credits co-founder Walden Yan with the Minecraft analogy of moving from survival mode to creative mode, where people can concentrate on what they want to make. His stated concern is the friction caused by rapid change: unequal access, inadequate preparation, and cyber attackers adopting powerful capabilities faster than defenders. He argues that education and broad access must keep pace with technical progress.[7]

Before Cognition, Wu competed internationally in programming, worked on financial-software performance, and co-founded Lunchclub. That earlier history supplies context for both his engineering experience and the collaborators he later gathered around AI.[8][7][9]

Wu won International Olympiad in Informatics gold medals representing the United States in 2012, 2013, and 2014. The official competition record lists 600 points and rank 1 among 311 contestants in 2014. He recalled that national competitions and training camps also formed a lasting circle of friends, many of whom later joined him in starting Cognition.[8][7]

Wu recalled working at Addepar on performance engineering and optimization for financial analytics, alongside substantial recruiting work. He then attended Harvard for about two years before dropping out. In summer 2017, he moved to San Francisco to start what became Lunchclub with Vlad Novakovski, whom he had known through Addepar.[7]

According to Wu, the initial scheduling and personal catch-up app gradually shifted toward professional introductions and deciding which new people users should meet. Forbes separately recognized Wu and Hayley Leibson as Lunchclub co-founders in its 2020 Consumer Technology 30 Under 30 list, published in December 2019. Wu recalled leaving Lunchclub in June 2022, following growth and monetization difficulties and a period in which he spent extended time in Louisiana caring for his mother.[7][9]

Wu subsequently explored ideas with Steven Hao and Andrew He and became increasingly absorbed in AI around ChatGPT’s emergence. He recalled that Sam Altman’s November 17, 2023 firing accelerated plans already underway to start a reasoning-focused lab. The group gathered collaborators in a Burlingame hacker house to experiment. Wu resisted presenting the whole career as predetermined: “I can tell the retroactive version of the story where it was all on purpose, but I don’t think that’s quite the truth.”[7]