Cognition AI Built a Coding Agent With a 15% Success Rate. Now It Is Worth $25 Billion.

In March 2024, Cognition AI posted a demo of Devin, billed as the world's first AI software engineer, and the internet split immediately. Investors were thrilled. Engineers were not. Within weeks, Answer.AI published independent test results: out of 20 tasks assigned to Devin, only three were completed successfully, suggesting a devin ai performance review closer to 15 percent. A YouTube channel accused Cognition of using pre-configured files in its demo. Critics described it as one of the clearest examples of AI hype outrunning AI reality.

Two years later, Cognition AI is in talks to raise new funding at a $25 billion valuation. That is not explained by Devin suddenly achieving a 90 percent success rate. It is explained by a 72-hour acquisition sprint that left Cognition holding an asset neither OpenAI nor Google could properly claim. The story of how a mocked startup went from $1 million in annual revenue to a $25 billion ask tells you something specific about how the AI coding market actually works, and why the companies that seem to be losing sometimes end up with the most interesting positions.

From $400 Million Last Year to $25 Billion: What Changed at Cognition

Cognition AI was founded in August 2023 by Scott Wu, Steven Hao, and Walden Yan. All three are competitive programmers who won gold medals at the International Olympiad in Informatics. The company raised a $175 million Series A before Devin shipped, backed by investors who were betting on the founders as much as the product. That early capital gave Cognition the runway to absorb the criticism that followed the 2024 launch.

The business underneath all the criticism kept growing anyway. Devin's annual recurring revenue grew from $1 million in September 2024 to $73 million by June 2025, a 73-fold increase in nine months. The customers paying that bill include Microsoft, Dell Technologies, and Cisco. Those are not companies that pay for demo videos. They are paying because Devin is completing real tasks in production environments, even if those tasks are narrower and more controlled than the original demo implied.

In September 2025, Founders Fund led a $400 million funding round that valued Cognition at $10.2 billion. The round brought in Lux Capital, 8VC, and Elad Gil. At that valuation, Cognition was already one of the more expensive bets in the AI coding category. Less than eight months later, it is reportedly seeking to raise at $25 billion, more than double, on a trajectory that the Windsurf acquisition accelerated beyond what Devin's organic growth alone would have justified.

Post-acquisition, Cognition AI's combined ARR is reportedly well above $100 million, with enterprise ARR doubling quarter over quarter. Scott Wu described the combined vision when announcing the Windsurf deal: "Both teams share a singular focus on taking coding agents from impressive demos to reliable engineering infrastructure that ships products at scale." The phrase "impressive demos" was not an accident. He was acknowledging the early narrative directly, not pretending it did not exist.

The SpaceX Deal Created the Pricing Window Cognition Needed

To understand why $25 billion is being discussed right now, and not three months from now, you need to look at what happened two days before the Cognition news broke. On April 21, 2026, SpaceX announced a deal to acquire Anysphere, the company behind the AI coding tool Cursor, at a price of up to $60 billion. That single transaction established a new reference point for what the category is worth, and Cognition moved immediately.

Cursor's growth numbers explain the $60 billion figure. The company went from $100 million in ARR in January 2025 to $500 million by June 2025, crossed $1 billion by November 2025, hit $2 billion by February 2026, and is forecasting more than $6 billion by year-end. The spacex cursor deal values Anysphere at roughly 30 times its current annual revenue. That multiple tells you how much the market expects growth to continue.

Cognition is at a different scale: $100 million-plus in ARR versus Cursor's $2 billion. At $25 billion, Cognition is asking for 42 percent of Cursor's valuation on roughly 5 percent of Cursor's revenue. That gap implies the market believes Cognition can close the ARR distance faster than Cursor can pull away, which is a real bet, not a given. But the SpaceX deal also changed the competitive structure. With Cursor committed to SpaceX, there is one fewer independent player at the top of the market. Scarcity creates pricing pressure, and Cognition is raising into exactly that dynamic.

The broader shift in enterprise behavior is also supporting the numbers. AI coding agents, tools that handle entire engineering tasks rather than just completing lines of code, are moving from experimental to operational inside large organizations. Enterprises that evaluated Devin skeptically in 2024 are now deploying it alongside other tools in workflows where the bar is task completion, not suggestion quality. That behavioral shift is what drove the $1 million to $73 million ARR trajectory in the first place, and the Windsurf acquisition gave Cognition a second distribution channel to accelerate it further.

The Real Reason Cognition Is Worth $25 Billion Has Nothing to Do With Devin

Cognition's current valuation is built on three things: Devin's organic ARR growth, the SpaceX-Cursor pricing anchor, and the Windsurf acquisition. The third is the most unusual and the least discussed. Understanding it requires going through the full sequence of what happened in July 2025, because the speed and structure of that event created advantages that Cognition did not design for and could not have planned.

In May 2025, OpenAI agreed to acquire Windsurf for $3 billion. It would have been OpenAI's largest acquisition to date. Windsurf, formerly known as Codeium, had built a fast-growing AI coding IDE with $82 million in ARR, more than 350 enterprise customers, and hundreds of thousands of daily active users. For OpenAI, the logic was straightforward: Windsurf would give it a direct presence in the developer IDE market, competing head-to-head with Microsoft's GitHub Copilot.

Then the structural problem emerged. OpenAI's partnership agreement with Microsoft includes provisions that give Microsoft rights to the intellectual property of companies OpenAI acquires. OpenAI refused to grant those rights to Windsurf's technology, arguing that Windsurf's codebase and methods were a core competitive asset against the exact product Microsoft was trying to protect through GitHub Copilot. The disagreement was fundamental, not a negotiation gap. The deal collapsed on July 11, 2025, because of that openai microsoft ip rights conflict.

On the same day the OpenAI deal died, Google moved. Google offered a $2.4 billion package to hire Windsurf's CEO Varun Mohan, co-founder Douglas Chen, and a handful of senior R&D staff into Google DeepMind. This looked like an acquisition from the outside but was structured as an acquihire. Google paid for the talent and the technical knowledge. It did not acquire Windsurf as a company. The product, the brand, the trademark, the intellectual property, and the 210 remaining employees were all still part of a company that had just lost its founders and most of its leadership.

That left Windsurf in an unusual position. It had strong ARR, a large user base, a working product, and a full team, but no executive leadership and no parent company. The window for someone to step in was narrow and obvious.

Cognition made its first call on the afternoon of July 11, after 5 PM. The definitive agreement was signed by Monday morning, July 14. Seventy-two hours from first call to closed deal. What Cognition acquired through the windsurf acquisition cognition includes the full product and brand, all intellectual property and trademarks, and all 210 remaining employees. Cognition also provided fully accelerated vesting to the entire Windsurf team, an unusual term that signaled how urgently it wanted to retain people who could have left quickly given the chaos of the previous 72 hours.

The combined result is a company that no competitor had expected. Cognition brought autonomous agent capabilities through Devin: the ability to receive a task, plan it, write and test code, and complete it without continuous human direction. Windsurf brought daily workflow presence through its IDE: the tool developers actually open every morning for editing, completion, and code review. Together, those are the two layers that Cursor is still in the process of building toward and that GitHub Copilot addresses separately.

The irony of the whole sequence is precise. OpenAI's Microsoft constraint, and Google's decision to buy people rather than a business, created exactly the strategic position Cognition needed to justify a valuation that its organic growth alone could not have supported. Cognition's $25 billion is built partly on Devin's real traction, partly on the Cursor pricing anchor, and partly on being the only buyer in the room at the right moment.

Devin vs. Cursor vs. GitHub Copilot: Three Strategies for Who Controls AI Coding

The AI coding market has three dominant strategies competing against each other right now, and they reflect different theories about where developer leverage will ultimately concentrate.

Cursor built its position as an IDE-first product. Developers use it daily for editing, code completion, and chat-based assistance. The product creates habitual usage at the individual workflow level, then extends upward toward autonomous tasks. With $2 billion in ARR and a path to $6 billion by year-end, it is the clear revenue leader in the category. Its current weakness is that it does not yet own the autonomous end-to-end task layer that enterprise buyers are starting to demand for the highest-value engineering work, the kind where a company hands off an entire feature or migration rather than asking for a suggestion on the next line.

GitHub Copilot has the largest installed base and the deepest enterprise distribution, embedded into Microsoft's existing contract relationships with large organizations. Its structural ceiling is the same Microsoft IP clause that killed the OpenAI-Windsurf deal. Any partner or competitor that gets close enough to acquire a meaningful IDE or agent company runs into that IP friction. Copilot is pervasive, but constrained in how it can grow through external acquisitions.

Cognition started from the opposite direction. The original Devin premise was that an ai software engineer should be capable of handling a complete deliverable: accepting a task, planning it, writing and testing the code, and returning something that works. That positions Devin for high-value, high-autonomy enterprise use cases where the customer wants output, not assistance. Adding Windsurf gives Cognition the daily workflow presence it previously lacked, alongside the agent capability that Cursor is building toward.

The historical parallel that keeps coming up in analyst discussions is the 2012 to 2014 mobile development tool consolidation, when several well-regarded independent IDEs and frameworks were absorbed into or displaced by Xcode and Android Studio once the platform companies committed to their developer ecosystems. What is different this time is the speed of capital deployment and the absence of a clear platform winner. None of OpenAI, Google, Microsoft, or any other incumbent has cleanly locked in the AI coding developer base yet. That uncertainty is what makes the current window valuable for independent companies like Cognition.

After $25 Billion, What Cognition Actually Has to Do Next

Closing at $25 billion is still a negotiation, not a fact. The reporting noted that talks are ongoing and terms could change. The first thing Cognition needs to prove is that it can close on those terms, which depends on showing investors a clear path to the ARR growth the multiple implies.

To justify a $25 billion valuation at the multiples the market is currently applying to this category, Cognition needs to grow combined ARR from $100 million-plus today to somewhere between $500 million and $1 billion in the next 12 to 18 months. That is achievable if the Windsurf integration is clean, the combined product retains users from both sides, and enterprise deployment continues accelerating. It is not achievable if the integration creates product friction, Windsurf users migrate to Cursor, or the autonomous task completion layer turns out to require more product work than the 72-hour acquisition timeline allowed for due diligence.

The integration risk is real. Windsurf users bought an IDE. Devin users bought an autonomous agent. Merging those two user expectations into a coherent product is a harder challenge than either team has faced independently. The vision Scott Wu described, taking coding agents from impressive demos to reliable engineering infrastructure, is the right goal. Whether 310 combined employees can execute it at the pace the valuation implies is a separate question.

The longer-term dynamic is what analysts have started calling the vertical integration war in AI development. Whoever controls the full stack from developer IDE to autonomous agent to deployment infrastructure will have disproportionate leverage in how enterprise engineering teams are structured. Cognition currently holds the first two layers through Windsurf and Devin. The third layer, hosting and runtime infrastructure, is where its relationships with cloud providers will become critical. How Cognition manages those relationships, particularly given that Microsoft is simultaneously one of its largest customers and the entity whose IP clause ended the OpenAI-Windsurf deal, is a variable worth watching.

For engineering teams evaluating AI tooling, the Devin plus Windsurf combination is now a single product worth reassessing. Teams that have built structured workflows around their engineering knowledge base will find that the agent layer Cognition is building is increasingly relevant to how that knowledge gets applied. The question is not whether AI coding agents will become standard infrastructure. The question is whether Cognition can close enough of the Cursor ARR gap before the vertical integration race produces a clear winner.

The critics who called Devin a scam in 2024 were not wrong about the product at that moment. They were wrong about the company.

FAQ: Common Questions About Cognition AI and Devin

What is Cognition AI?

Cognition AI is a startup founded in August 2023 that builds autonomous AI coding agents. Its flagship product is Devin, marketed as the world's first AI software engineer. The company is headquartered in the United States and is backed by Founders Fund, Lux Capital, and 8VC.

What is Devin and how does it actually work?

Devin is an autonomous coding agent designed to accept engineering tasks, plan a solution, write and test code, and return a completed deliverable without requiring step-by-step human direction. Unlike code completion tools that suggest the next line, Devin operates over longer time horizons on entire tasks. Early independent testing in 2024 put its real-world success rate at around 15 percent, though the company's ARR growth to $73 million by mid-2025 indicates enterprises found meaningful value in production deployments despite that initial gap.

What happened with Windsurf and why did Cognition end up with it?

In May 2025, OpenAI agreed to acquire Windsurf, a fast-growing AI coding IDE, for $3 billion. The deal collapsed in July 2025 because Microsoft's partnership agreement with OpenAI gave Microsoft rights to intellectual property from OpenAI's acquisitions, and OpenAI refused to share Windsurf's technology with its direct competitor GitHub Copilot. Google then hired Windsurf's founding team for $2.4 billion as an acquihire, leaving the product, brand, IP, and 210 employees without leadership. The company acquired those remaining assets within 72 hours, gaining $82 million in ARR and hundreds of thousands of daily active users at a fraction of the original deal price.

How does Devin compare to Cursor and GitHub Copilot?

Cursor focuses on the IDE layer, building daily workflow habits with code editing and chat, then extending toward autonomous tasks. GitHub Copilot has the broadest distribution through Microsoft's enterprise contracts but is constrained in acquisitions by the same IP clause that ended the OpenAI-Windsurf deal. Devin approaches the market from the agent layer, handling full-task execution, and the Windsurf acquisition now adds IDE distribution. Each strategy addresses a different point of developer leverage.