Scott Wu

CEO and Co-Founder of Cognition, the AI lab that created the AI software engineer Devon.

Claims

As speaker

Mentioned in

Key positions and views

The role of a software engineer is fundamentally shifting from a 'bricklayer' focused on implementation to an 'architect' focused on high-level design and problem-solving.

Sufficient 'base intelligence' already exists in today's foundation models; the primary challenge is now teaching AI agents the specific, idiosyncratic workflows of real-world engineering.

Specialized AI coding agents will maintain a competitive advantage over general models due to the deep, contextual knowledge required for complex software engineering tasks.

AI agents will increase, not decrease, the demand for software engineers by lowering the barrier to entry and expanding the scope of what can be built.

The ideal user interaction with an AI agent is asynchronous delegation, where tasks are assigned at a high level (e.g., via a Jira or Linear ticket) and the agent handles the entire implementation workflow autonomously.

Podcast consensus on Wu

Points of consensus

▶Scott Wu consistently characterizes the current capability of the Devon AI agent as being equivalent to a 'junior engineer' across multiple appearances [41, 68].Feb–May 2026

▶Multiple sources confirm Wu's emphasis on Cognition's unique, high-intensity culture, which is heavily populated by former company founders [5, 18, 24, 53].Feb 2026

▶Wu repeatedly states that Devon is being deployed in large, complex enterprise environments, specifically naming major financial institutions like Goldman Sachs and Santander as customers or partners [13, 33].May 2026

▶Across several discussions, Wu articulates a consistent vision for the future of software engineering, where the role shifts from manual coding ('bricklayer') to high-level design and architecture ('architect') [1, 40, 59].Feb 2026

Points of debate

▶There is a tension between Wu's claim that AI will increase the number of software engineers [6] and his simultaneous prediction that the core function of writing code will be automated and abstracted away, fundamentally changing the role into one of architecture and product management [1, 4, 59].Feb 2026

▶Wu describes Devon's capability as that of a 'junior engineer' [41, 68], which contrasts with claims of it achieving superhuman productivity gains (8-15x) [10], merging 30-40% of all pull requests in organizations [11], and autonomously negotiating with human customer service agents [12].

▶Wu asserts that current foundation models have sufficient 'base intelligence' for AI agents [39], yet also states that future improvements will depend on better foundation models, reasoning, and memory [8], suggesting current capabilities are a limiting factor.Feb–May 2026

▶Wu claims his internal engineering team 'no longer types code manually' and primarily uses Devon [31], but also states that Devon is responsible for 'approximately 25%' of pull requests [52], implying that human engineers still manually code the other 75%.May 2026

Key themes

▶The Transformation of Software EngineeringFeb 2026

Wu consistently argues that AI agents like Devon will fundamentally alter the software engineer's role, shifting it from implementation ('bricklayer') to high-level design and problem-solving ('architect'). This change will make current skills like learning specific programming languages obsolete and ultimately lead to a future where engineers interact with computers via natural language rather than code.

This theme suggests a strategic bet that the value in software development will migrate from execution to ideation, positioning Cognition as a key enabler of this shift and potentially disrupting the entire software development toolchain and labor market.

▶Devon as an Autonomous Agent

Wu details Devon's capabilities as an autonomous AI software engineer that can handle end-to-end tasks, from setup and coding to debugging and deployment. He emphasizes its integration into existing workflows (Slack, Jira, GitHub) and its ability to learn a team's specific context, leading to significant productivity gains and taking over a large percentage of coding tasks.

The focus on full workflow automation and asynchronous delegation, rather than simple code completion, indicates Cognition's strategy is to replace entire segments of the development process, making them a potential substitute for, rather than just a supplement to, junior engineering talent.

▶Cognition's High-Intensity, Founder-Led CultureFeb 2026

Wu portrays Cognition as a company built by and for founders, characterized by an intense work culture, rapid execution (e.g., acquiring a company over a weekend), and a high concentration of former founders on the team. This ethos is presented as a key competitive advantage, enabling them to pivot quickly and attract talent aligned with ambitious, high-stakes goals.

This founder-centric culture may be a significant asset for innovation and speed but could also present scaling challenges and integration difficulties, as hinted by the buyout offered to employees from the Windsurf acquisition.

▶Views on the AI Industry and Technology Stack

Wu believes the AI industry is a 'hyperscaler or bust' environment where value accrues to differentiated layers in the stack. He argues that specialized tools like Devon will maintain an edge over general models due to the need for deep, contextual knowledge of real-world engineering, and that sufficient 'base intelligence' already exists in models to build transformative products.

Wu's position suggests Cognition's strategy relies on building a defensible moat through specialized reasoning and workflow integration, betting that this application-layer expertise will be more valuable than owning the underlying foundation models.

Source episodes

Timeline

2014

As a high schooler, Scott Wu works at Adapar alongside several other individuals who would go on to become prominent founders and leaders in the tech and AI space [64].

November 2023

Cognition originates as a hackathon project, marking the beginning of the work that would lead to Devon [51].

Early 2024

Cognition is officially incorporated as a company [51].

March 2024

Cognition launches its flagship product, the AI agent Devon, to the public [51].

Post-Launch 2024

Wu describes Devon's capability as rapidly progressing from a 'high school CS student' at launch to a 'college intern' and then a 'junior engineer' [41]. Cognition acquires Windsurf over a single weekend to integrate its enterprise, infrastructure, and go-to-market teams [27, 67].

Mid-2024

Wu reports that Devon is responsible for approximately 25% of pull requests at Cognition [52] and states the goal is for it to handle more than half by the end of the year [37].

Suggested prompts

How does Scott Wu reconcile his prediction of an increased number of software engineers with the simultaneous automation of their core implementation tasks? &nearr;What are the primary risks and safeguards Wu identifies for deploying an autonomous agent like Devon, which can interact with external systems, in an enterprise environment? &nearr;Based on Wu's claims, what is Cognition's core defensibility against large, general foundation models entering the AI coding agent space? &nearr;How does the evolution of Devon's capabilities from 'high school student' to 'junior engineer' correlate with the specific technological improvements Wu mentions, such as reasoning, memory, and self-play? &nearr;

Key concepts

Devon (AI Agent) 4 ep Future of Software Engineering 4 ep Cognition (Company Culture & Strategy) 3 ep Productivity Gains 3 ep Enterprise Deployment 3 ep Agentic AI 4 ep AI Industry Analysis 2 ep User Workflow Integration 2 ep Foundation Models 2 ep

Notable quotes

“[1:03:10 -> 1:03:43] my hot take here, which I would give, is I think in terms of base intelligence, we're honestly basically already there. And I think a lot of what we see actually, and what we spend our time on is less so... increasing the base IQ of a model, for example and more about kind of like teaching it all of the idiosyncrasies of real world engineering”

Scott Wu · Inside Devin: The AI engineer that's set to write 50% of its company’s code this year | Scott Wu

“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.”

Scott Wu · A Cheeky Pint with Cognition CEO Scott Wu

“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.”

Scott Wu · A Cheeky Pint with Cognition CEO Scott Wu

“[0:55:31 -> 0:55:57] I think the one that we'll get to over the course of several years is probably something where you don't even look at the code at all, right? And you're actually just looking at your own product and you're just able to look and specify and say, hey, you know, this button should be a little bit rounder, let's do that. And, and by the way, let's add a new tab here. And maybe we should save this information... and having your agent build out those things for you.”

Scott Wu · Inside Devin: The AI engineer that's set to write 50% of its company’s code this year | Scott Wu

Report last updated: May 10, 2026

Get started free

Back to Entities Intelligence Report