Jun 28, 2026 · 1h 5m · david-senra
The Future of Software & AI | Cognition’s 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
In this in-depth conversation, Cognition CEO Scott Wu and host David Senra discuss the origins of Devin, the world's first autonomous AI software engineer, examining the competitive mindset, first-principles reasoning, and enterprise deployment strategies transforming the future of software development.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. David holds 24.7% of the talking time here. How this is scored →
David as informed peer 3.9 Guest teaching 3.3 Guest disagreement 0.5 David pushing back 0.8
speaking balance: gold is David, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 58:55 Scott rejects startup exit nihilism
Scott firmly dismisses the defeatist industry consensus that independent startups cannot compete against hyperscalers, rejecting the premise that existing products represent the final state of software development.
Hardest push from David ▶ 25:40 David rejects Scott's mundane email exampleDavid explicitly pushes back against Scott using a bulk-email formatting glitch to demonstrate long-horizon AI, insisting autonomous agents should be pointed at profound multi-year research missions.
Biggest teaching moment ▶ 16:28 Scott teaches single-use software economicsScott educates David on the history of programming abstractions and outlines how AI agents invert software economics by generating disposable code for single-use white-collar workflows.
David holds their own ▶ 24:20 David cites Edwin Land's two-year instant color researchDavid showcases deep biographical mastery by citing Edwin Land giving an engineer two uninterrupted years to solve instant color photography as the true historical analogue for year-long AI task horizons.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | David as informed peer | Guest teaching | Guest disagreement | David pushing back | Why |
|---|---|---|---|---|---|---|
| The Origin of Scott Wu's Competitive Drive | 4 | 2 | 1 | 1 | David opens by digging into Scott's self-description of being 'salty' and draws a comparative parallel to Demis Hassabis at DeepMind. Scott details his lifelong obsession with winning math and strategy competitions as a decision-tree search. | |
| Childhood Competitions and Family Heritage | 3 | 2 | 0 | 0 | David inquires about Scott's early competitive activities and family background. Scott shares stories about competitive programming, Super Smash Bros, and how his father's Go prowess facilitated their family's immigration to the US. | |
| Parental Archetypes and Unconditional Support | 5 | 2 | 1 | 1 | David brings up historical founder biographies, comparing harsh parenting styles like Ellison and Musk to supportive figures like Estee Lauder. Scott clarifies that while his parents did not push entrepreneurship specifically, his mother instilled fierce confidence. | |
| Sponsor Break: Ramp | 4 | 1 | 0 | 0 | Following an ad read for Ramp, David asks whether the fear of losing outweighs the joy of winning, referencing his conversations with Michael Dell. Scott agrees with the psychological framing while noting that heavy experimentation requires enduring frequent losses. | |
| Cognition's Ambition and the Shift to Creative Mode | 3 | 3 | 0 | 0 | David asks about Cognition's founding vision and what winning looks like for Devin. Scott outlines their mission to build a generational hyperscaler and shift humanity from survival mode to creative mode. | |
| The Evolution of Software Abstraction and Single-Use Code | 2 | 5 | 0 | 0 | David asks how Devin evolves from current developer workflows to a universal interface. Scott educates David on historical layers of software abstraction and introduces the concept of disposable, single-use software dynamically written by AI. | |
| Exponential Curves and the Accelerated AI Horizon | 3 | 4 | 0 | 0 | Scott predicts the major interaction problems will be resolved in five years. David prompts him on the human inability to grasp exponential curves, which Scott illustrates through evolutionary biology and his parents' transition from 1960s China. | |
| First-Principles Thinking and Unassisted AI Task Horizons | 2 | 5 | 1 | 1 | David asks what Scott uniquely understands about AI compared to industry peers. Scott demurs on having unique genius and reframes the conversation around first-principles extrapolation of task-horizon benchmarks. | |
| Deploying Autonomous Agents on Long-Horizon Missions | 6 | 3 | 1 | 3 | David pushes back when Scott uses a bulk-email formatter as an example of long-horizon AI, offering the historical counterexample of Edwin Land giving a scientist two years to invent instant color photography. Scott enthusiastically adopts the mission-based framing. | |
| Sponsor Break: AppLovin | 4 | 3 | 1 | 1 | Following an AppLovin ad, David asks about Devin's origins, explosive revenue growth, and the initial viral backlash upon launch. Scott recounts their breakthrough moment setting up MongoDB and explains why early skepticism never deterred them. | |
| SWE-bench Benchmarks and the AI Coworker Paradigm | 5 | 4 | 1 | 0 | David cites Jeremy Stern's profile and asks about SWE-bench benchmarks. Scott explains why launching early with a 13% benchmark allowed a late-starting startup with 'no right to win' to stake out the coworker paradigm before big tech. | |
| Finding Product-Market Fit in Enterprise Migrations | 4 | 5 | 0 | 1 | David asks about Cognition's shift toward Fortune 500 enterprises. Scott explains diagnosing agent failure modes and uncovering PMF in high-friction, repetitive enterprise code migrations like Nubank's Java upgrades. | |
| Deploying Forward Engineering and Accelerating Enterprise Cycles | 3 | 4 | 0 | 1 | David asks how Cognition converts enterprise leads and whether whole teams still travel on-site. Scott explains shrinking traditional 18-month adoption cycles down to three months via private cloud security and forward deployed user guidance. | |
| Sponsor Break: Deel | 4 | 5 | 0 | 0 | Following a Deel ad read, Scott explains Cognition's 'Switzerland' posture, dynamic compound model orchestration, and why hyper-focused startups can beat foundational labs by caring more about real-world enterprise engineering messiness. | |
| Startup Focus vs. Big Tech and the Path to Independence | 5 | 3 | 1 | 1 | David references Jimmy Iovine's experience battling Spotify at Apple to illustrate the power of founder focus. Scott dismisses startup exit nihilism, arguing that incumbent tech giants cannot monopolize all dynamic industry niches. | |
| The Pursuit of Potential and Rejecting Exit Nihilism | 6 | 2 | 1 | 2 | David asks about acquisition offers and quotes Richard Branson, Steve Jobs, Mark Zuckerberg, and Todd Graves to celebrate founder purity. Scott insists selling makes no sense when the core motivation is testing one's ultimate potential. |
Statements from this episode (22)
Scott Wu compares building a company to a tree search algorithm
“It's the way even building a company, it feels the same. It's just like you're calculating the moves. You're thinking about, okay, if you do this and then this happens and then you do that and here are the different moves and you're like calculating out what c…”
Scott Wu's childhood home displayed math trophies instead of family photos
“We didn't have like, you know, growing up, like we didn't have like pictures of our parents, like on the walls. You know, we just had old math competition trophies.”
Cognition started with an unusually large founding team of nine people
“Our founding team is, it was a pretty big founding team. It was nine people. And most of us had already founded our own companies before we had done different things.”
Scott Wu predicts humans will soon stop interacting directly with code
“We're not always, I mean, we're not gonna be interacting with code for that much longer, you know, or programs might not be the right, like level of abstraction, but I think what is always true probably is that it will be the human computer interface.”
Cognition works with Goldman Sachs, Mercedes, and the US government
“We work with a lot of the big, biggest companies across the world, you know, we work with Goldman Sachs, we work with Mercedes, we work with, you know, a lot of areas of the U S government at this point and so on.”
Engineering teams using Devin ship software ten times faster
“And in practice, what it looks like is that teams use Devon to ship 10 times faster and to do 10 times more.”
AI agents will generate single-use code to automate white-collar workflows
“I think what we're gonna get to
Is we're going to get to a point where you are just giving your instructions to that agent and the agent on the back end, you know, you don't even have to look at this, but on the back end, the agent is going to figure out, oka…”
Scott Wu predicts intent-to-execution AI development will be solved in five years
“But I honestly, I mean, I think we'll have solved most of that over the next five years or so.”
Scott Wu predicts society will forget living without AI within ten years
“But I think we'll kind of, Undergo the same period with AI where five years from now, it's going to be insane to think about all these things that we're going to have. 10 years from now, we're going to have forgotten that we ever lived without them, honestly.”
The uninterrupted task horizon for AI agents doubles every few months
“And that's just doubled every, you know, every couple of months basically. And now we're talking about like, Hours of work. So, so basically an AI can just take a task that would have taken humans hours of work to go do.”
Scott Wu predicts AI systems will eventually work unassisted for a year
“And yeah, I think that's true, and I think we will get there.”
Long-horizon AI agents execute autonomous missions rather than narrow tasks
“When you're talking about seconds, you're literally talking about just like a specific command, right? When you're talking about hours, you're talking about giving it a task and having it do the task, right? And I agree with you. Obviously bulk email is not, y…”
The first automated task Devin successfully performed was setting up MongoDB
“The first task that Devon did was it like set up MongoDB for us.”
Scott Wu confirms initial Devin launch videos were real, unstaged runs
“All of those like demos that we showed, you know, in, in, in our launch announcement were like actual runs of Devon that we had done ourselves and like run into and been like, holy shit, this is insane.”
Devin scored 13 percent on SWE-bench against a 3 percent baseline
“At the time the best known was like Three or four percent or something like that, but obviously, yeah, 13% still means you fail, you know, 87% of the time.”
Cognition's first commercial success was an enterprise migration for Nubank
“Our first success ended up being with a company called Nubank, biggest bank in Brazil, you know, by market cap at the time, and the use case was like one of these big migrations, and we had a kind of a custom Devon that was like extremely, extremely optimized …”
Cognition ignores individual hobbyists to focus on production engineering teams
“And so, so we don't focus at all on like, you know, individual hobbyists who are just like trying to make a cool thing or something like that. We focus on like, Real teams who are building real products that they want people to use and getting output out of th…”
Cognition's entire team flew to Brazil to onboard its first enterprise customer
“Well, so NewBank, I mean, the first one, honestly, our entire team was the Ford deployed team. Like we all flew to Brazil.”
Scott Wu admits AI agents did not work generally during early deployments
“Let's be real here, ok? Agents did not work generally, ok? And so, there was a lot of like, how do you make it work for you very specific?”
Cognition aims to complete enterprise customer deployments in three months
“We're saying that these cycles typically take 12 to 18 months. We try to get deployed with folks, you know, in like three months.”
Devin dynamically routes tasks across OpenAI, Anthropic, Google, and internal models
“Devin, rather than being pegged to one model and saying, oh, we're only going to serve you GPT for this, or we're only going to serve you OPFIS for this, Devin can use any of the different models it has in its arsenal, which include all of these models from An…”