AI software engineering agents
Cognition
Cognition builds Devin, an AI software engineering agent that lets engineering teams delegate tasks such as bug fixes, migrations and testing. Working in its own development environment, Devin inspects repositories, edits code, runs tests and returns pull requests for review. Cognition acquired Windsurf in 2025 and subsequently unified the products under Devin Desktop, combining a full IDE with an Agent Command Center for local and cloud agents, pull requests and shared context. DeepWiki turns public GitHub repositories into documentation users can ask questions about.
Founded in 2023 by Scott Wu, Steven Hao and Walden Yan, Cognition is led by Wu as CEO. Its research includes Kevin-32B, developed with Stanford researchers to train CUDA kernel generation through multi-turn reinforcement learning. The model iteratively revises code using compilation errors, correctness checks and runtime feedback. Retaining earlier code and evaluation results while replacing lengthy reasoning with summaries limits context growth across successive attempts.
In July 2026, Cognition reported that revenue run rate had grown from $73 million to over $500 million, and its team from 44 to 350, over the year of Windsurf integration. Separately, Cognition raised over $1 billion at a $26 billion valuation in May 2026, led by Lux Capital, General Catalyst and 8VC.
4 talks
Newest first
The State of Model Routing — NVIDIA, Cognition, OpenRouter
Walden · Carter · Tanay · Alex Atallah · Nav
AI Engineer World's Fair 2026 · 48:17
3 speakers at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here
- The Making of Devin
Start with Scott Wu’s React name-game demonstration to see how Devin takes a concrete application from conference-speaker data through building and deployment.
Scott WuAI Engineer World's Fair 2024
- How Forward Deployed Engineering is done at Cognition
Learn how Jia Wu describes the technical, business, and customer-facing skills involved in selecting and delivering enterprise agent projects.
Jia WuAI Engineer World's Fair 2026
- The State of Model Routing — NVIDIA, Cognition, OpenRouter
Use this panel discussion to examine how evaluation, agent behavior, and prompt tuning enter model-routing decisions.
Walden · Carter · Tanay · Alex Atallah · NavAI Engineer World's Fair 2026
- Devin 2.0 and the Future of SWE
Scott Wu outlines Cognition’s view of rapidly increasing AI software-agent task horizons and Devin’s progression from reliable, repetitive code migrations to broader autonomous engineering.
Scott WuAI Engineer World's Fair 2025
Messages from the stage
Making delegated engineering repeatable
Scott Wu describes planning, repository-aware pull requests, and verification alongside Playbooks and remote environments for repository setup, linting, and CI. These talks connect task execution with the supporting workflows needed for asynchronous collaboration.
Measuring deployment through customer outcomes
Jia Wu contrasts token-usage targets with measurable business results and describes how engineers embedded with customers connect deployment feedback to product development.
Routing across capability and resource constraints
The joint panel featuring Walden discusses frontier versus smaller models, context compaction, and cache-aware routing as parts of balancing application quality, inference cost, and hardware utilization.
Affiliations reflect each recorded session, not necessarily current employment. The model-routing session is a joint panel with participants from Cognition, NVIDIA, and OpenRouter.