In September 2026, Cognition raised more than $2B at a $48B valuation, nearly doubling its $26B valuation from May 2026, when it raised more than $1B. The latest round was led by new investors Andreessen Horowitz and Accel alongside existing investors Founders Fund, General Catalyst, and Avenir.

Previously, Cognition raised more than $400M in September 2025 at a $10.2B valuation in a round led by Founders Fund. Other investors include 8VC, Lux Capital, Bain Capital Ventures, Bessemer Venture Partners, Kleiner Perkins, and Nvidia.

Cognition announced a $21M Series A led by Founders Fund in March 2024. Early backers included Patrick and John Collison, Elad Gil, Sarah Guo, and Tony Xu. Founders Fund subsequently led a $175M round in April 2024 at a $2B valuation.

Cognition’s original pitch for Devin was an AI engineer you could hand work to through Slack. Tag Devin in a thread, explain the task, and let it investigate the codebase, make changes, run tests, and return with a pull request. Developers could answer questions in chat and request revisions through GitHub comments, much as they would with a remote teammate.

The initial wedge was well-defined work: small bug fixes, unit tests, targeted refactors, documentation updates, and repetitive code migrations. Developers assigned work and reviewed the result while Devin operated in its own environment with a shell, browser, and code editor. Detailed requirements and clear instructions for checking the output helped it succeed.

As models improved at understanding repositories, using tools, and completing longer tasks, Devin expanded into changes spanning multiple files and recurring engineering workflows. Auto-Triage investigates bugs and incidents, Security Swarm finds vulnerabilities and prepares fixes, and Automations triggers work from Slack, GitHub, Linear, or a schedule. Playbooks provide reusable instructions for jobs such as database migrations.

Devin 2.0, launched in April 2025, added an interactive cloud IDE where developers could inspect an agent’s work, edit files, and run tests. Teams could run multiple Devins in parallel, while planning tools, codebase search, and automatically generated documentation helped them decide what to delegate. The July 2025 acquisition of Windsurf added a desktop IDE, bringing interactive coding and delegated cloud work into the same product lineup.

In June 2026, Cognition introduced Devin Desktop, built on Windsurf, with an Agent Command Center for managing local and cloud agents, pull requests, and project context. Spaces lets related agents share context, while developers can open the editor to debug or make changes themselves. Support for external agents such as Codex CLI and Claude Agent through the Agent Client Protocol makes it a workspace for coordinating agents from multiple providers.

Cognition combines third-party models from OpenAI and Anthropic with its own SWE models, which it post-trains on open-source models for software engineering. It develops these alongside its coding harness—the tools, context, execution environment, and feedback that let a model complete development tasks. Following the SWE family’s origins at Windsurf, Cognition launched SWE-1.5 in October 2025, emphasizing fast execution through Cerebras inference at up to 950 tokens per second. April 2026’s SWE-1.6 focused on less repetitive reasoning, fewer unnecessary steps, and more parallel tool use.

SWE-1.7, released in July 2026, built on Kimi K2.7 and targeted longer asynchronous tasks. September’s SWE-2 built on Kimi K3, using additional reinforcement learning to improve task completion while controlling inference cost, with different reasoning levels for more demanding work. Specialized models handle supporting tasks: SWE-1-mini powers inline suggestions, swe-grep retrieves codebase context, and swe-check supports fast code review. DeepWiki adds repository documentation and codebase Q&A, helping developers understand a system before asking an agent to change it.

Cognition sells subscriptions to individual developers and engineering teams, combining recurring subscription fees with charges for additional usage. Devin offers a free tier, a $20-per-month Pro plan, and a $200-per-month Max plan with higher usage allowances. Teams pricing starts at $80 per month plus $40 per month for each full developer seat, while Enterprise pricing is negotiated.

Paid plans include usage allowances, with additional usage available at API pricing. Consumption varies with the model selected, task size, and reasoning required. Revenue can therefore grow as customers add developers, move to higher tiers, or delegate more work to agents. Customers can also run third-party agents inside Devin Desktop, with usage billed directly by those providers rather than Cognition.

Cognition combines self-serve adoption by developers with enterprise sales. Enterprise plans add centralized administration, single sign-on, dedicated account support, and deployment options for larger organizations. The acquisition of Windsurf expanded Cognition’s presence in developers’ daily workflows, complementing Devin’s cloud agents with a desktop development environment.

Cognition competes most directly with Cursor in its positioning as a dedicated AI coding platform for professional developers. Both built their businesses around the developer workflow, combining models from outside providers with their own coding tools and infrastructure. Cognition initially emphasized delegating work to Devin, while Cursor started with an AI editor. Cognition’s acquisition of Windsurf and Cursor’s expansion into cloud agents have brought their offerings closer together.

Cognition also competes with OpenAI’s Codex and Anthropic’s Claude Code, which pair their providers’ foundation models with coding tools and execution environments. These products increasingly handle the same work as Devin, from implementing features and fixing bugs to running tests and preparing pull requests.

That relationship is also complementary. Cognition uses third-party models alongside its own, and developers can select OpenAI and Anthropic models inside Devin Desktop. They can also run external agents such as Codex CLI and Claude Agent through its Agent Client Protocol integration, making Devin Desktop a place to use competing agents as well as Cognition’s own.

Cognition seeks to differentiate through its coding harness—the software that gives models codebase context, development tools, execution environments, and feedback from tests—and how it packages that harness for specific jobs.

Auto-Triage investigates incoming bug reports and incidents, while Security Swarm finds vulnerabilities, checks whether they can be exploited, and prepares fixes for review. Teams can also configure repeatable workflows for tasks such as database migrations or resolving issues raised in Slack and Linear.

Cognition complements this with its own SWE models, developed by post-training open-source models for software engineering. For example, SWE-2 builds on Kimi K3 with additional reinforcement learning aimed at improving task completion and cost efficiency. Its model development focuses on specializing existing models for engineering workflows, while OpenAI and Anthropic also invest in pretraining the underlying foundation models.

GitHub Copilot competes through its integration with GitHub, VS Code, and existing enterprise development processes. Its coding agents can take assigned tasks, modify repositories, and open pull requests, bringing delegated engineering work directly into tools developers already use.

GitHub’s distribution and control over source-code hosting and review give it an advantage in adoption. Cognition must give teams a reason to buy and manage a separate platform through better execution, workflow automation, or support for complex engineering environments.

Replit and Lovable let users describe an application in natural language, generate it, and publish it through an integrated platform. Both serve prosumers and business users as well as developers, with adoption spanning prototypes, internal tools, and customer-facing applications.

Their overlap with Cognition is strongest in building new applications from a prompt. Cognition focuses more on professional engineering teams working inside existing codebases, where tasks require understanding established systems, following development conventions, and passing tests and review.

Devin Desktop expands Cognition’s role from supplying a coding agent to providing the environment where developers manage multiple agents. Its Agent Command Center brings local and cloud agents, pull requests, and project context together, while Spaces lets related agents share context.

Support for third-party agents broadens the platform’s usefulness even when teams choose another provider for particular tasks. That creates an opportunity to become a central part of the development workflow, although third-party agent usage is billed by those providers and does not automatically generate usage revenue for Cognition.

Cognition is expanding from individual coding assignments into recurring work across software maintenance, security, and incident response. Auto-Triage investigates incoming issues, Security Swarm finds and helps remediate vulnerabilities, and Automations lets teams trigger work from systems such as Slack, GitHub, and Linear.

These workflows create opportunities to capture spending beyond developers’ coding subscriptions. As customers delegate ongoing maintenance and security work, Cognition can grow usage within an engineering organization without requiring a proportional increase in developer seats.

Cognition has opened offices in London, Madrid, São Paulo, Tokyo, and Singapore, extending its reach beyond its US hubs. Local teams support enterprise sales and deployment, helping customers integrate agents into established engineering processes.

The opportunity is to expand adoption across multinational companies and win regional enterprises, bringing Devin into more engineering teams and a wider range of codebases and workflows.

Cognition can expand from individual developers and pilot projects into organization-wide deployments. Centralized administration, single sign-on, dedicated deployment options, and account support help larger companies adopt Devin across engineering teams.

The opportunity is to move from discretionary developer-tool spending into larger engineering budgets. As companies integrate Devin into shared development processes, Cognition can grow through additional seats and increased agent usage across codebases and departments.

Model dependency: Cognition's autonomous capabilities depend on large language models managed by third parties, which control their performance, availability, and pricing. Any decline in model quality, substantial price increases, or restrictions on access could materially affect Devin's functionality and Cognition's unit economics.

Security concerns: Autonomous AI systems with repository access and code generation features pose significant security and intellectual property risks for enterprise customers. A single vulnerability or data breach involving Devin could undermine customer trust and constrain enterprise adoption, particularly in regulated sectors.

Talent competition: The autonomous coding market is drawing significant investment and talent from established technology companies and well-funded startups. Cognition must continually enhance its AI capabilities while competing for limited AI engineering talent against organizations with greater financial resources and more established research teams.