Aug 27, 2025 · 59m · cheeky-pint
A Cheeky Pint with Cognition CEO Scott Wu
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
Stripe CEO Patrick Collison sits down with Cognition co-founder and CEO Scott Wu over pints of Guinness to discuss the creation of AI software engineer Devin, the future of coding, high-density founder networks, and the transition toward autonomous agentic workflows.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
John as informed peer 4.2 Guest teaching 3.6 Guest disagreement 1.1 John pushing back 1.8
speaking balance: gold is John, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 22:18 Rejecting the generic AI computer-use narrative
Scott directly reframes Patrick's question about foundation labs obsoleting coding agents, labeling it the 'nihilist computer use take' and detailing why real-world software engineering cannot be solved without custom evaluation environments.
Hardest push from John ▶ 7:29 Challenging the claim that founding was easier in the pastPatrick refuses Scott's assertion that being a founder is inherently harder now than during the PC or social media eras, citing intense competition faced by Dell and Facebook.
Biggest teaching moment ▶ 15:30 Deconstructing essential vs accidental complexity in codeScott educates the listener and host on how engineering time is overwhelmingly consumed by routine scaffolding rather than architecture, explaining why agentic automation targets accidental complexity.
John holds their own ▶ 30:41 Detailed analysis of usage-based economics in AI agentsPatrick lays out Stripe's architectural roadmap for autonomous commerce and usage-based billing infrastructure, demonstrating deep domain command.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | John as informed peer | Guest teaching | Guest disagreement | John pushing back | Why |
|---|---|---|---|---|---|---|
| Scott Wu's Math Background and Competitive Origins | 3 | 4 | 1 | 0 | Friendly biographical opening where Patrick asks about Scott's competitive math background and gives an arithmetic prompt, which Scott immediately answers while clarifying the difference between calculation and puzzle-solving. | |
| Dropping Out, Addepar, and the Math Competition Founder Cohort | 3 | 3 | 0 | 0 | Scott recounts leaving high school early, working at Addepar, and dropping out of Harvard alongside a tight-knit peer group of math competitors who became AI founders. | |
| The Changing Dynamics of Young Founders and Startup Maturity | 5 | 2 | 2 | 4 | Patrick posits that young founders serve as an industry biomarker, but Scott offers an alternative take that company building has simply grown more mature, prompting Patrick to push back that early tech eras were not easy either. | |
| The Moneyballification of Poker, Chess, and Competitive Gaming | 4 | 3 | 1 | 1 | Scott introduces his theory on the moneyballification of competitive domains like poker, chess, and Smash Bros, with Patrick contributing parallels from RTS gaming. | |
| Defining Cognition, Devin, and the Asynchronous Agent Paradigm | 4 | 3 | 1 | 2 | Patrick sets up the contrast between inline IDE autocomplete and Cognition's asynchronous agent model in Slack or Jira, which Scott elaborates upon. | |
| Enterprise Adoption, PR Velocity, and Essential vs. Accidental Complexity | 5 | 4 | 1 | 3 | Patrick challenges whether the sync versus async distinction will collapse as IDEs add agentic capabilities, leading Scott to explain the software engineering divide between essential and accidental complexity. | |
| Enterprise Permissions, Code Migrations, and Developer Productivity | 5 | 3 | 1 | 3 | Patrick presses on enterprise security fears regarding autonomous database edits, and later asks how CTOs can truly verify productivity gains beyond noisy PR metrics. | |
| Specialized Engineering Agents vs. Frontier AI General Intelligence | 5 | 5 | 2 | 3 | Patrick brings up the nihilist view that frontier foundation models will inevitably absorb specialized coding tools, prompting Scott to argue that domain messiness and RL benchmark construction provide defensibility. | |
| Cognition's Internal Junior Dev Benchmark and Model Evaluations | 4 | 4 | 1 | 1 | Scott explains Cognition's internal junior dev evaluation framework and discusses value capture across infrastructure, labs, and application layers. | |
| Economic Infrastructure for AI Agents and Autonomous Commerce | 5 | 2 | 1 | 3 | Patrick explains Stripe's agent billing infrastructure and challenges Scott on whether Cognition should pivot to a consumer virtual assistant based on internal emergent DoorDash usage. | |
| Digital Trust, Anti-Bot Protections, and Delegated Agent Identity | 5 | 4 | 1 | 1 | The conversation shifts to web bot-blocking and trust, where Scott proposes replacing binary bot bans with cryptographic user delegation. | |
| Hiring Elite Generalists and the 8-Hour Agent Interview Process | 4 | 4 | 1 | 1 | Scott details Cognition's 8-hour agent building interview and predicts software engineers will stop looking at syntax and code within two to four years. | |
| Retro User Interfaces and the Product Innovation Lag in AI | 5 | 4 | 1 | 2 | Patrick notes how retro text-box interfaces remain in modern AI compared to early mobile innovations, and Scott explains why product design lags behind instantaneous model distribution. | |
| Evaluating AGI Timelines and Incremental Capability Expansion | 4 | 4 | 1 | 1 | Scott discusses realistic AGI timelines and provides an exhaustive behind-the-scenes narrative of acquiring Windsurf over a single weekend. | |
| AI Industry Consolidation and Cognition's Cultural Buyout Offer | 5 | 3 | 2 | 2 | Patrick explores regulatory workarounds like 49% licensing deals and asks about Cognition's cultural buyout package for incoming employees. | |
| CEO Growth, Peer Networks, and Information Diets | 4 | 3 | 1 | 3 | Patrick inquires into Scott's CEO learning curve and information diet, teasing him about receiving automated algorithmic feeds instead of running proactive agent summaries. | |
| Live Arithmetic Demonstration: Card Games and Big Numbers | 2 | 6 | 0 | 0 | Patrick hands Scott a random four-digit target, 6843, and Scott rapidly computes the arithmetic chain using nine cards live on the podcast. |
Statements from this episode (34)
Wu: Addepar hired four high schoolers including himself and Alexandr Wang
“You know, funnily enough, there were four of us who started the same at the same time as high schoolers. And it was myself. Alexander Wang was actually another one. We started on the same day. Eugene Chen who's now running Phoenix Dex, and then Sreenath, RA. W…”
Wu: Founders of Perplexity, Pika, and Decagon competed together in math
“Johnny Ho, who's one of the co-founders of Perplexity, for example. Demi Guo, who started Pika. You know, a lot of these, Jesse Zhang, who started Decagon. You know, a lot of us were actually competing in these math and programming competitions in the same yea…”
Collison: Big tech incumbents leave fewer AI opportunities on the table
“Clearly, all the large companies these days, they're very aware, they're very connected with the ecosystem. If you look at Asatia or Mark Zuckerberg, they are very aware of everything that's going on AI, and they're paying a lot of attention to it. And so, yea…”
Wu: Startup playbook maturity gives experienced founders an advantage
“Maybe harder is not the right word. It's more just that the space is a bit more mature, and there's more of a playbook and like more existing knowledge. Yes. There's obviously something unique with every business, but a lot of the details of, you know, here's …”
Wu: Maturing competitive fields shift from intuition to mathematical optimization
“I think for a less mature space, when people don't know what the right questions to ask are, or how to even kind of think about it, like the, what is the right frame of reference? Then I think there's something about just having a really sharp intuition and co…”
Wu: Cognition recently acquired IDE tool Windsurf
“We've been going for the last year and a half, and most recently just acquired Windsurf, and so, you know, Devon, the agent in Windsurf, the IDE, but at a high level, you know, we really want to build the future of software engineering.”
Wu: Devin performs on average like a junior engineer
“We like to call Devon a junior engineer today. There are some things that an AI, of course, is way, way better than all of us at, you know, especially encyclopedic knowledge and just pulling facts and things like that. There are some things that it's, you know…”
Wu: Devin is deployed at thousands of companies including Goldman and Citi
“Devon is deployed in thousands of companies all over the world. You know, we work with some of the biggest banks in the world, like Goldman and Citibank, all the way down to, you know, startups with two or three people.”
Wu: Devin merges 30% to 40% of pull requests in successful orgs
“Typically in a successful org, Devon is merging something in the range of, like, 30 to 40% of all the pull requests that come through.”
Wu: Developer tools will combine synchronous IDEs with async agents
“And so I think this merged experience that comes up is basically something where for anything that actually needs you in the loop, where you can go and make the decision and you're looking at the high level strategy or deciding what you want to build, you're i…”
Wu: Cognition strongly advises users against giving Devin production database access
“So we pretty strongly recommend that people using Devon don't give it, you know, prod database access, for example.”
Wu: Enterprises measure 8x to 15x productivity gains using Devin for migrations
“I think in practice what we see with folks, especially in these kind of like enterprise migrations is, is, you know, when folks measure internally, they see something like an eight to 15 X gain for a lot of these use cases with Devon because, Yeah, as you're s…”
Wu: General AI intelligence alone will not solve messy software engineering
“Software engineering in the real world is so messy, you know, and there's all sorts of these things that come up. And I think in practice, you know, most disciplines look like this. And I would say the same thing about law or medicine or, and so on. And so whi…”
Wu: Solved AI benchmarks shift bottleneck to defining real-world benchmarks
“The thing is when that happens, I don't think what we end up with is just pure ASI end of humanity, human knowledge work or whatever. I think the thing that we end up in is basically a point where the hard question is, all right, now what is the benchmark, rig…”
Wu: Cognition evaluates AI models using an internal 'junior dev' benchmark
“From our perspective, we have a lot of benchmarks internally. You know, the biggest is one that we call junior dev, which we might need to upgrade to senior dev pretty soon, but it is basically the ability to do a variety of just random real world junior dev t…”
Wu: Latest Claude and GPT models outperform predecessors on internal benchmarks
“Yeah, I mean, both of them are, the two of them are better at this benchmark than any of the models that we've seen before this week.”
Wu: Seat-based SaaS pricing makes no sense for AI agents
“Seats don't really make sense when it is like, the AI themselves are arguably seats as well, you know, like they're doing a lot of the labor too. And then on the other side, I think you know, usage obviously just goes so naturally with the cogs themselves beca…”
Wu: Cognition uses Devin to order DoorDash and Amazon packages
“Devon is obviously entirely focused towards software engineering, but like we order our DoorDash on Devon, you know, we order our Amazon packages with Devon, and it's like, there are pieces of that that turn out to work nicely anyway.”
Wu: Agent access will shift to delegated identity and user attribution
“What we will probably need to see a lot more of over time is basically, like delegating access, if that makes sense. Like, making it more clear that an agent can do something on your behalf, and, you know, in some sense, you're attaching some of your reputatio…”
Wu: Cognition's core engineering team grew from 19 to 30-35 post-Windsurf
“With Windsurf, obviously, the team count has grown a lot, but actually, you know, with core engineering itself, it hasn't actually gotten all that much bigger. It's gone from 19 to something in the range of like 30 to 35.”
Wu: Candidates interview by building an AI agent in 8 hours
“Our whole interview process, for example for Lotties is basically just having people build their own Devon in, in eight hours and seeing how far they get with it.”
Wu: Language syntax will matter less than product sense in software engineering
“Yeah, I think what we find is, is, and I think this, you know, we'll see this trend generally in software engineering, which is knowing all the little, memorizing all the facts or knowing all the little details or being really good at Syntax of some language o…”
Wu: 21 of Cognition's initial 35 team members were former founders
“And so, yeah, a lot of our team actually are specifically former founders which is kind of a fun one, like, of our initial, kind of, 35, I think. 21 of us have founded a company before and so it's been it's been a very high density of that.”
Wu: Developers will stop using code as primary interface in 2-4 years
“I think that there will come a point, and my guess on this point is probably in the neighborhood of, let's say, two, three, four years from now, where we stop using code as the main interface.”
Wu: The transition to AI will create far more software engineers
“Funnily enough, I mean, I think, if anything, we will have way more software engineers, not fewer.”
Wu: High school and college students should still study computer science
“People often ask us, like, my son or daughter is in high school or has just started college. Should they even be studying computer science? And my answer is always absolutely yes.”
Collison: Everything will pass through a transformer model before consumption
“Everything will pass through a transformer model before it's consumed.”
Wu: AI scales faster because single-player utility lacks hardware barriers
“AI is, I'd say, uniquely different from some of these previous ways, in an important way, which is, you know, personal computer, or internet, or mobile phone. All of these had a, one of two things, or often both. One was a big hardware component, of like, yeah…”
Wu: Freezing AI models today leaves a decade of product progress
“I think you could freeze all the capabilities today and have no new models and no new research come out and there would still be like a whole decade of product progress to make.”
Wu: Sudden AGI singularity won't happen in the next few years
“My honest opinion is, I think there is some, you know, rapid, singularity, superintelligence thing that people kind of talk about. I would guess It's very hard to say. You know, nothing's impossible, but I would guess that that's not something that happens in …”
Wu: Cognition reached out cold to Windsurf leadership on Friday evening
“We reached out to them cold that evening, and got to meet the new Windsurf leadership, you know, Jeff and Graham and David that evening.”
Wu: Google hired Windsurf's core research team while other functions remained intact
“There was a core kind of, like, research and product engineering team that went to Google, and the, all of the other functions were entirely intact, which includes enterprise engineering, infra, deployed engineering go to market, marketing, finance operations,…”
Wu: AI landscape is polarized toward becoming a hyperscaler or bust
“Like maybe one of my hot takes is like, I think for a lot of the big, of course there will be, you know, many medium sized outcomes in AI, but I think In this space, a little bit more so than previous ones, it's a little bit more polarized towards, like, you b…”
Wu: Only a small fraction of Windsurf team took Cognition buyout offer
“Yeah, yeah, yeah, I think for us, it's you know, and most folks have been really excited to come in and do it, and only a small fraction have taken the buyout, but I think from our perspective we just want to make sure it's an opt-in, you know, situation for e…”