Jul 27, 2026 · 43m · sourcery
Inside the Fastest-Growing Category in AI: Scott Wu, CEO of $26B Cognition · Sourcery with Molly O'Shea
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Sourcery, host Molly O'Shea interviews Scott Wu, Co-Founder and CEO of Cognition, discussing Devin's rapid growth to a $26 billion valuation and $500 million ARR. Scott shares insights on measuring AI productivity, integrating the Windsurf acquisition, maintaining model neutrality, and entering an era of software abundance.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Molly holds 18.4% of the talking time here. How this is scored →
Molly as informed peer 3.6 Guest teaching 2.7 Guest disagreement 1.2 Molly pushing back 1.2
speaking balance: gold is Molly, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 28:41 Dismissing the six-month singularity timeline
Scott dismisses the extreme AI doom/singularity timelines popularized in tech circles, arguing that intelligence alone cannot bypass physical and organizational constraints.
Hardest push from Molly ▶ 23:00 Probing the long-term defense of staying independentMolly challenges Scott on why it is strategically superior to remain an independent platform instead of being absorbed into a hyperscaler or frontier lab.
Biggest teaching moment ▶ 40:02 Correcting the Peter Thiel chess rumorScott directly corrects Molly's premise about beating Peter Thiel in chess, clarifying that the actual event was a poker bet over seed round terms.
Molly holds their own ▶ 7:46 Citing enterprise clients and valuation metricsMolly demonstrates strong industry command by citing Cognition's exact enterprise and defense client list alongside their latest revenue and valuation benchmarks.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Molly as informed peer | Guest teaching | Guest disagreement | Molly pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome to Paris & Elevator Recognition | 3 | 3 | 1 | 1 | Molly introduces Scott in Paris and asks about internal adoption and measuring productivity. Scott explains that Cognition writes 95% of its code with Devin and dismisses token count as a valid productivity metric. | |
| Explosive Revenue Growth & Enterprise Adoption | 4 | 2 | 1 | 1 | Molly references Mayfield's category growth research and highlights Cognition hitting half a billion in revenue in under three years. Scott elaborates on why software engineering TAM expands toward every engineer globally. | |
| The Windsurf Acquisition & M&A Strategy | 5 | 3 | 1 | 1 | Molly reels off Cognition's high-profile enterprise and government clients before probing into the Windsurf acquisition. Scott details the 48-hour weekend sprint to acquire Windsurf after Google acquihired part of their team. | |
| Sponsor Segment: Brex Intelligent Finance | 3 | 2 | 1 | 1 | Molly inquires about the operational challenges of merging teams and integrating products without formal advisors. Scott shares how a 40-person startup absorbed a 200-person organization through organic product alignment. | |
| Independence & Public Sector Strategy in D.C. | 4 | 2 | 2 | 2 | Molly asks why Cognition spent Independence Day in Washington D.C. and pushes Scott on the strategic choice to remain independent rather than selling to a lab. Scott outlines public sector opportunity and warns against a single-entity AGI monopoly. | |
| Why Enterprise Clients Choose Third-Party AI Agents | 4 | 3 | 1 | 1 | Molly channels a question from VC Christian Garrett regarding why enterprises choose third-party tools over foundation labs. Scott explains that enterprise customers demand model neutrality and workflow integration rather than raw tokens. | |
| Debunking the Six-Month Singularity Narrative | 4 | 4 | 2 | 2 | Molly brings up Silicon Valley's six-month singularity narrative and community hysteria. Scott rejects the hyper-accelerated timeline by distinguishing between RL-solvable benchmark tasks and ambiguous, long-horizon real-world problems. | |
| Devin's Breakthrough Moment: Fixing MongoDB | 3 | 2 | 1 | 1 | Molly prompts Scott to explain why MongoDB was the foundational demo for Devin. Scott recounts the late 2023 breakthrough when Devin autonomously diagnosed and fixed co-founder Walden's broken database environment. | |
| The Era of Software Abundance & Unlimited Capacity | 3 | 2 | 1 | 1 | Molly asks Scott what fundamental questions are being overlooked in the AI discourse. Scott argues that the transition will be defined by software capacity expansion and ephemeral on-demand applications rather than mere efficiency gains. | |
| Challenging Peter Thiel's Firm to Poker & Future Outlook | 3 | 4 | 1 | 1 | Molly asks what it was like beating Peter Thiel in chess. Scott immediately corrects her, explaining he actually proposed a heads-up poker match against Founders Fund's Napoleon for seed equity terms before Peter shut it down. |
Statements from this episode (16)
Wu: Cognition writes roughly 95% of its code using Devin
“Obviously we use a ton of Devon to build Devon. In fact, we almost exclusively use Devon to build Devon at this point, like, 95% of the code or something's written by Devon.”
Wu: Cognition's shipped code increased roughly 7x in six months
“Even in the last six months, I think the number for us was like, our total amount of code shipped is roughly seven X in that time. And so it's like the number of PRs we've shipped, the amount of code”
Wu: Devin customer usage grew 11x to 12x in six months
“Our customers in the last six months or so, I've also grown their usage in something in the range of like, 11 to 12 acts, and of Devin.”
Wu: Nearly all software engineers will use coding agents within three years
“The rough answer for, you know, two, three years from now is roughly every software engineer in the world, and probably a lot more than all the Just the software engineers because I think a lot more people will be able to produce code and produce products and …”
Scott Wu: Coding IDEs and AI agents must share persistent codebase context
“The really good coding IDE products, you know, the tool that you're kind of like sitting in and like using to look at the code itself versus the agent product, which is kind of that almost the engineer that you can delegate. You obviously want this to be very …”
Cognition completed the Windsurf acquisition over a single weekend
“We got in touch with the Windsurf team that evening, so it was Friday evening, and then basically over the course of that weekend we,
Worked out the entire acquisition together. So, by Monday morning, we had the announcement ready to go.”
Cognition Had $70M–$80M ARR and 35–40 People When Buying Windsurf
“Cognition at the time was all of 35 or 40 people and we had a meaningful business with, you know, real traction. We were doing in the range of 70 or eighty million of revenue run rate at the time but it was a very small team. Windsurf, on the other hand becaus…”
Wu: Roughly 30 of Cognition's First ~50-60 Employees Were Former Founders
“Our first, like, you know, 56 people, I think like 30 of us were, had founded a company before this”
Wu: AI will be capable of performing all pure human knowledge work
“I've always felt that there's, in terms of just pure knowledge work, there's, there is nothing that humans do that AIs can't do, you know, partly because Humans are, humans, you know, our brains are just computers, you know, like the AIs are too, right?”
Wu: Rapid model leadership shifts deter enterprises from single-vendor lock-in
“Obviously, the way enterprises work is they want to have long-term partnerships, and frankly, it's just hard to know what's going to happen in the long term, right? Like, who even is going to have the best model in six months, or 12 months, or whatever, and I …”
Wu: Reinforcement learning can solve basically any clearly defined benchmark
“We're kind of getting to the point where you can solve basically any benchmark, right? Because what does it mean to have a benchmark? It means you've already defined the task. You've clarified what success or failure looks like. You've given a bunch of example…”
Wu: AI will solve all intelligence problems, but not in six months
“I really don't think that there are, like, intelligence problems that AI will not solve in the long term. I just don't think of that as, like, it's gonna be all over in six months.”
Wu: Devin's breakthrough first task was autonomously fixing a MongoDB setup
“And, like, Devin just ran a bunch of stuff for a few minutes, and then fixed it. And then, like, sent Walden a message, like, hey, I got it, it's up now. Walden didn't really, he didn't really believe it at first. I think we still have the video recording of t…”
Scott Wu: The real AI unlock in software is capacity, not efficiency
“I think the real unlock is not in efficiency, but in capacity, right? It's like, what can we do if we can build so much more?”
Scott Wu: Single-use and self-driving software will exist
“Like single use software is going to exist. Like self-driving software is going to exist, right?”
Wu: Cognition Offered to Play Thiel's Firm in Poker for 1% Equity
“Since you're such a big fan of poker, you know, we can just play a heads-up poker match for that, and we can have the winner decide whose terms we actually sign. It would have been for, in retrospect, it would have been for, like, one percent of the company or…”