people
scott wu, steven hao, walden yan, and the people around cognition — who they are, what they wrote, and what they claim.
the founders
- scott wu · ceo · scott_wucofounder and ceo. born in baton rouge to chinese-immigrant parents — 1996 in his own words to colossus, 1997 per wikipedia; the discrepancy is unresolved. mathcounts national champion 2011 after competing against middle-schoolers as a third-grader; ioi gold 2012/2013/2014, first overall in 2014; harvard ~two years then out, with an icpc 2016 gold along the way; codeforces legendary grandmaster peak 3350. worked at addepar as a high schooler, co-founded lunchclub (cto, 2017–2022, a16z-backed, forbes 30 under 30), and reportedly spent time at scale ai before cognition. 'i learned to program when i was nine years old, writing little programs in basic on a ti-83.'
- steven hao · co-foundercofounder and cto. mit math/cs; ioi 2014 gold (6th place) after a 52nd in 2013; icpc world finals 6th with mit; a january 2015 jane street trading internship; then an early engineer at scale ai (~2018–2022, autonomy and sensor-fusion ml per speaker bios). secondary profiles also list meteor, d.e. shaw, dropbox, and quora — individually unverified. he keeps a deliberately low public profile: no known blog, no solo interviews. the quietest founder on the cap table.
- walden yan · co-foundercofounder and cpo. ioi 2020 gold for the us; mit primes researcher in crypto/ml; harvard cs/econ — left for cognition, telling bloomberg he 'hasn't yet had the talk' with his parents about his status at the school. reported: a thiel fellowship, an engineering stint at anysphere (cursor) pre-cognition, co-founder of deepreason (web3 security), managing partner of inverted — all on community bios, not primary sources. he narrated the launch's upwork demo and later wrote the company's most-cited essays, 'don't build multi-agents' and its 2026 revision.
the writing
cognition argues in public through wu's launch posts and funding memos and yan's context-engineering essays — plus the unsigned team voice that ships the monthly update cadence. the blog is where the claims get made and, occasionally, walked back.
- 2024-03-12introducing devin, the first ai software engineer · essay, cognition blog · scott wuthe launch manifesto — 'the first ai software engineer,' a sandboxed shell/editor/browser, a 13.86% swe-bench claim footnoted to a 25% unassisted subset, and a 'hire devin' cta that made the labor framing literal.
devin is a tireless, skilled teammate, equally ready to build alongside you or independently complete tasks for you to review.
by solving reasoning, we can unlock new possibilities in a wide range of disciplines—code is just the beginning.
we even tried giving devin real jobs on upwork and it could do those too!
viral launch tweet; hn front page the same day - 2024-03-15swe-bench technical report · document, cognition blog · the cognition teamthe swe-bench technical report — methodology (570 of 2,294 issues, unassisted) plus the eval harness and devin's edits released on github. the paper trail behind the launch number.
- 2024-12-10devin is now generally available · post, cognition blog · the cognition team'devin is now generally available' — $500/month flat, no seat limits, slack-primary with an ide extension and api; the '16x faster timelines' claim and the first named customers (nubank, vesta, ramp) are company-reported.
- 2025-04-03devin 2.0 · post, cognition blog · the cognition teamdevin 2.0 — the agent-native ide experience, interactive planning, parallel devins, and the $20 entry plan that collapsed the $500 flat price into pay-as-you-go acus.
- 2025-05-05deepwiki · post, cognition blog · the cognition teamdeepwiki — free ai documentation for any public github repo; the code-understanding surface that grew into devin wiki and devin search.
- 2025-06-12don't build multi-agents · essay, cognition blog · walden yanthe essay that popularized 'context engineering' as the discipline over prompt engineering — full-context single-threaded agents over parallel sub-agents, with the concession that the field is 'still playing with raw html & css.'
share context
actions carry implicit decisions
the most-cited cognition essay; jason liu's notes circulated the talk version - 2025-07-14cognition's acquisition of windsurf · document, cognition blog · scott wuthe windsurf acquisition announcement, written as wu's internal all-hands memo — the deal terms for employees made legible, the 'bricklayers to architects' thesis to the nyt, and the labor-relations promise the august buyouts would be measured against.
100% of windsurf employees will participate financially in this deal … will have vesting cliffs waived … will receive fully accelerated vesting for their work to date.
there's only one boat and we're all in it together
- 2025-09-08funding, growth, and the next frontier of ai coding agents · essay, cognition blog · scott wu'funding, growth, and the next frontier of ai coding agents' — the definitive metrics post: devin arr $1m→$73m in nine months, lifetime net burn under $20m, >30% combined-arr growth in seven weeks post-windsurf, <5% customer overlap. all company-reported.
we hunkered down in a new york apartment and built the product we always wanted for ourselves.
- 2025-10-29swe-1.5 · post, cognition blog · the cognition teamswe-1.5 — a frontier-size in-house model at 950 tok/s via cerebras, measured on swe-bench pro. the buried lede: the company that made swe-bench famous stopped reporting it a year earlier.
we stopped reporting swe-bench numbers in 2024
- 2025-11-14devin's 2025 performance review · post, cognition blog · the cognition team'devin's 2025 performance review: learnings from 18 months of agents at work' — the self-evaluation format applied to the product itself, claiming deployment at 'thousands of companies.'
- 2026-04-22multi-agents: what's actually working · essay, cognition blog · walden yan'multi-agents: what's actually working' — the revision: multiple agents may contribute intelligence to a task, but writes stay single-threaded. the 2025 absolutism narrowed by a year of shipping.
- 2026-05-27more devins in more places · post, cognition blog · scott wu'more devins in more places' — the >$1b series d at $26b, the 'independent agent lab' positioning, and the engineer-hours framing that turns employees into 'executives' over an army of devins.
at cognition, 89% of code committed by our engineers is committed by devin
- 2026-07-08the ti-83 origin post · post, linkedin · scott wuthe linkedin origin-story post — ti-83 basic at nine, and the claim that the end state was always 'a full operating ai software engineer,' not autocomplete.
i learned to program when i was nine years old, writing little programs in basic on a ti-83 calculator in baton rouge, louisiana.
the ultimate form factor of software engineering was not going to be an autocomplete or a text assistant. it was going to be a full operating ai software engineer
- 2026-09-08do it all with devin · post, cognition blog · scott wu'do it all with devin' — the >$2b series e at $48b led by a16z and accel; andreessen's companion post repeats the 13%→90% production-code arc and notes a16z is backing wu 'for the second time.'
the team
- russell kaplanpresident since january 2024 — pre-launch. stanford cs bs/ms, treehacks founder, tesla autopilot senior ml scientist, founder/ceo of helia (acquired by scale ai in 2020), then scale's director of engineering and ml. he co-led the windsurf outreach with wu and announced the 2026 exec bench. some event bios style him 'co-founder & president' — a styling, not a fact pattern.
- the medalist bench · touristthe founding team's claimed '10 ioi gold medals' are only partially named: the three founders plus neal wu — scott's older brother, 3x ioi gold himself, founding advisor from november 2023, departed for mira murati's thinking machines lab in february 2026 — gennady korotkevich, whose codeforces 'tourist' tier was literally named for his handle when he broke 4000 while at cognition, and andrew he ('ecnerwala'). the about page adds that 21 of the initial 35 team members had previously founded a company.
- jeff wangthe human thread through the windsurf saga — windsurf's head of business since may 2023, interim ceo the day google took the founders, ceo through june 2026, then 'president, new enterprise' at cognition running international expansion and government work. he recounts 'the craziest 72 hours' of the deal and notes windsurf's retained engineering core is 'maybe five or six people' amid agent-driven reorgs every two or three months. windsurf's actual founders — varun mohan and douglas chen — went to google deepmind, not cognition.
- the cognition teamthe extended roster and the unsigned company voice on the blog: swyx (latent space) and investors-turned-employees christian lawless (conversion capital) and emily cohen (neo) joined around the september 2025 series c; the 2026 bench added takumi masai (cognition japan), richard spence (apac, singapore), paul grewal (chief legal and global affairs, ex-coinbase), art l. (global partnerships, ex-brex), and jared palmer (vp engineering). headcount arc: ~10 at launch, an eng team of ~15 each running ~5 devins in early 2025, ~200 by late 2025 per wikipedia, 300–400 across 21 countries per linkedin.
- the backersfounders fund led the first three priced rounds — peter thiel personally involved at the seed — with angels elad gil, patrick and john collison, chris re, and sarah guo. 8vc led the ~$4b round with lux co-leading; lux, general catalyst, and 8vc co-led the series d; a16z and accel led the series e — andreessen's 'proud to back him for the second time' closing a lunchclub-era loop — inside a ~30-firm syndicate that includes nvidia, also a devin customer.
interviews and talks
- 2024-03devin's upwork side hustle · youtube'devin's upwork side hustle' — the demo video walden narrated that internet of bugs later took apart frame by frame: the wrong prompt, self-generated bugs, files that didn't exist in the repo. the artifact at the center of the demo-reality gap.
- 2025-08cognition ceo scott wu on acquiring windsurf · cheeky pintjohn collison's cheeky pint interview — a segment literally titled 'i think we have agi'; sixth-grade math competitions with alexandr wang, the windsurf weekend, the 'moneyball-ification of everything.'
- 2025scott wu on the david senra show · david senrathe david senra episode — the nine-person founding team of mostly-repeat-founders, the hyperscaler ambition, and the founder-as-search-algorithm self-description.
- 2026the wu tapes · colossuscolossus's 'the wu tapes' — the long q&a: born 1996 in his own telling, the ioi cohort that became scale, perplexity, and decagon founders, and a claimed '$445 million of revenue run rate in its first 18 months' (self-reported).
- 2025what comes after code · ai & ievery's ai & i — 'cognition's ceo on what comes after code': cs fundamentals, rl for coding agents, and the argument that 'we may already be living with agi.'
- 2026-07devin writes 95% of our code · sourcerythe sourcery podcast at raise summit — 'devin writes ~95% of our code' (self-reported), the single-weekend windsurf negotiation, the dmv as the next target, and the origin story traced to a mongodb error in 2023.
- 2025inside devin · youtubethe 'inside devin' interview — an eng team of ~15 each running ~5 devins, a quarter of prs committed by devin then with >50% expected by year-end, the 'eight pivots' admission, and the devin name origin.
- 2025a golden age of software engineering · hg orbitrussell kaplan on hg's orbit podcast — 'a golden age of software engineering': the software-abundance thesis, 'talent is everything,' and a business he describes as doubling every ~70 days (self-reported).
- 2026the 72-hour deal, retold · neo showjeff wang on the neo show — the 72-hour windsurf weekend from the inside, and the agent-era org chart: reorgs every two or three months as pm/design/qa roles collapse into 'full stack engineers… talking to agents all day.'
- 2025-07the context-engineering notes · linkedinwalden's context-engineering conversations — circulated mostly as other people's notes (jason liu's linkedin summary is the canonical record) rather than a talk of his own.
the philosophy
themes tracked across the writing: agents as coworkers, competitive programming, context engineering, independence, intensity culture, labor-shaped pricing, software abundance, the demo-reality gap.
- 2023–2026when did cognition startthe founding-date drift — august 2023 (wikipedia's body, senra's notes, mungomash) versus november 2023 (wikipedia's infobox, wu's linkedin) versus 'incorporated ~january 2024' (bloomberg's launch piece) versus an outlier january 2023 fan bio. scott wu's own birth year splits 1996 (his words to colossus) versus 1997 (wikipedia).the dossier keeps both dates visible rather than picking one; the working synthesis is that work began in h2 2023 and incorporation landed ~january 2024.
- 2024–2026the numbers are the company's ownthe self-reported ledger — swe-bench 13.86%, '16x faster timelines,' arr $1m→$73m→$492m→~$900m, '89–95% of code written by devin,' the productivity guarantee's own estimator, swe-1.5/2 performance — every load-bearing number is company-published. steven hao's bio is reconstructed from linkedin and speaker bios; walden yan's thiel fellowship and anysphere stint rest on community directories.no external audit exists; the dossier marks each number self-reported rather than reconciling them silently.
- 2025-08the 80-hour memothe culture memo — three weeks after promising 100% financial participation and 'one boat,' cognition laid off 30 windsurf staff and offered ~200 others nine-month buyouts conditioned on six office days and 80+ hour weeks: 'we don't believe in work-life balance—building the future of software engineering is a mission we all care so deeply about that we couldn't possibly separate the two.'wu publicly owned the posture — 'many of us literally live where we work,' 'we're proud of how we work, we understand it's not for everyone' — and bi headlined it 'i live for these moments.' critics read bait-switch; wu reads filter.
AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.