Cognition

AI Software · Tech

Mentions

Podcasts

Episodes

Podcast consensus

Points of consensus

▶Cognition's core product is 'Devon', an AI agent positioned as an AI software engineer that has achieved top results on the SWE-bench benchmark.Feb–May 2026

▶The company heavily utilizes its own Devon agent for internal software development, with the agent handling a significant and growing percentage of code contributions.Apr–May 2026

▶Cognition is in direct competition with other AI developer tools, most notably Cursor, a position solidified by its acquisition of Windsurf.Feb–Apr 2026

▶The company is pursuing a vertically-integrated technical strategy by training its own proprietary AI models and utilizing specialized hardware from Cerebras.Apr 2026

Points of debate

▶The intended scope of Devon's capabilities is unclear; while primarily marketed as a software engineering tool, it is also used internally for non-coding tasks like ordering from DoorDash, suggesting a broader general-purpose agent ambition.May 2026

▶The primary user interaction model is described as 'asynchronous delegation' for entire tasks, yet engineers also work with multiple Devon instances simultaneously, suggesting a potential blend of autonomous and co-pilot workflows.Apr–May 2026

▶There is a tension between Devon's advanced capabilities and its current safety limitations; it is trusted to merge hundreds of pull requests into production but is not recommended for access to production databases.Apr–May 2026

▶Cognition's market position is viewed differently across sources; some see it as a direct competitor to specific developer tools like Cursor, while others group it with a wider field of coding assistants like Claude Code and Codex.Apr 2026

Key themes

▶Devon as an Autonomous Software EngineerFeb–May 2026

Cognition frames its core product, Devon, not as a simple co-pilot but as an autonomous AI software engineer. It operates on an 'asynchronous delegation' model where users assign high-level tasks through project management tools, and Devon handles the execution, a capability supported by its leading results on the SWE-bench benchmark.

This positions Cognition to capture a different market segment than traditional code completion tools, targeting full task automation which could lead to higher contract values but may also face greater scrutiny regarding reliability and security.

▶Aggressive Internal Adoption ('Dogfooding')Apr–May 2026

Cognition extensively uses its own product, with its engineering team primarily using Devon to write code instead of typing it manually. The agent is responsible for a significant portion of the company's codebase, merging hundreds of pull requests monthly and accounting for approximately 25% of all PRs.

This 'dogfooding' strategy serves as a powerful validation and marketing tool, but also tightly couples the company's development velocity to the performance of its own AI, creating a high-stakes feedback loop.

▶Strategic Market Positioning and CompetitionApr 2026

Cognition is positioning itself against a range of competitors, from direct rivals like Cursor (bolstered by the Windsurf acquisition) to broader coding assistants like Claude Code and Codex. The company differentiates itself by focusing on 'reasoning' and training its own proprietary models on specialized hardware.

By building its own models and focusing on reasoning, Cognition is making a capital-intensive bet on creating a defensible moat against competitors who may rely on more general-purpose foundation models.

▶From Benchmarks to ProductionMay 2026

The company validates Devon's capabilities through both public benchmarks like SWE-bench and its own internal 'junior dev' benchmark designed for real-world tasks. This is complemented by its integration with enterprise tools like Linear, enabling its use in practical software development workflows.

The creation of an internal, real-world benchmark suggests Cognition believes public benchmarks are insufficient to capture the full scope of software engineering, indicating a focus on practical, end-to-end task completion over isolated coding challenges.

Source episodes

Sentiment over time

Apr 2026

5 bullish, 3 neutral(8 claims)

Cognition's acquisition of Windsurf has positioned it as a more direct competito...

May 2026

2 bullish, 5 neutral(7 claims)

Cognition recommends that customers do not give the Devon agent access to produc...

Changes over time

Founding Era

Co-founders bring experience from building early versions of the AI code editor Cursor and being an early engineer at Scale AI.

Team & Vision Formation

The company is established as an 'AI lab that focuses on reasoning,' deliberately hiring former founders or those with entrepreneurial ambitions.

Product Debut & Validation

Cognition creates and launches Devon, which quickly establishes credibility by achieving leading results on the SWE-bench benchmark.

Intensive Internal Adoption

The company turns Devon inward, with engineers using it to build the product itself, reaching a point where it handles approximately 25% of all pull requests.

Market Positioning & Expansion

Cognition acquires Windsurf to compete more directly with Cursor and integrates Devon with project management tools like Linear for customer use.

Future Outlook & Ambition

CEO Scott Wu projects Devon will handle over half of the company's pull requests by year-end and expresses extremely high valuation expectations for the company.

Suggested prompts

How does Cognition's 'reasoning'-focused approach technically differ from competitors like Cursor or code-generation models from OpenAI, and how is this reflected in its performance on the internal 'junior dev' benchmark? &nearr;What are the primary security and reliability risks associated with the 'asynchronous delegation' model, and how does Cognition's recommendation against production database access impact Devon's ability to handle complex, end-to-end tasks? &nearr;How scalable is the internal 'dogfooding' model, and what challenges might arise as Cognition transitions from its own controlled environment to diverse and complex enterprise customer codebases? &nearr;Given the use of Cerebras hardware and proprietary models, what is Cognition's long-term strategy for managing computational costs and maintaining a performance advantage over competitors leveraging more commoditized foundation models? &nearr;

Key concepts

Devon (AI Agent) 5 ep Competition (Cursor, OpenAI, etc.) 4 ep Software Engineering Productivity 4 ep Internal Use / 'Dogfooding' 2 ep Benchmarks (SWE-bench, 'junior dev') 2 ep Integrations (Linear, Jira, Slack) 2 ep AI Reasoning 1 ep Hardware (Cerebras) 1 ep Proprietary Models 1 ep User Interaction / Delegation 1 ep

Notable quotes

“Devon, an AI software engineer from Cognition, has achieved leading results on the SWE-bench benchmark.”

Unknown · No Priors Ep. 62 | With Cognition CEO and Co-Founder Scott Wu

“The user interaction paradigm for Cognition's Devon agent is asynchronous delegation, where a user assigns a task at a ticket or project level through an interface like Slack, Jira, or Linear.”

Scott Wu · A Cheeky Pint with Cognition CEO Scott Wu

“He let something slip backstage where he said, you have said that the expected value of Cognition is $8 trillion.”

Host · Lux Capital CEO Brandon Reeves & Cognition CEO Scott Wu at Semafor World Economy

“Cognition's acquisition of Windsurf has positioned it as a more direct competitor to developer tool company Cursor.”

Elad Gil · Elad Gil on the 2000 Dot-Com Crash & the Coming AI Shakeout

Report last updated: May 10, 2026

Get started free

Back to Entities Intelligence Report