In December 2025, the autonomous coding agent Devin wrote roughly 13% of the code at the company that built it.
By the time Cognition CEO Scott Wu briefed investors on the Series D round earlier this month, that number had reached 89%.
The startup announced the result on Wednesday, May 27, alongside the round that valued the company at $26 billion post-money. Lux Capital, General Catalyst, and 8VC co-led the financing. Founders Fund, Ribbit Capital, Atreides, and Layer Global joined.
Eight months ago, in September 2025, Cognition closed a $400 million round.
The post-money valuation that round set was $10.2 billion. Wall Street's appetite for independent AI coding companies has since multiplied that number by two-and-a-half times.
Devin Has Become a Real Product, Not Just a Demo
When Cognition unveiled Devin in March 2024, the company billed it as "the first fully autonomous AI software engineer." Skeptics noted the demos cherry-picked easy tasks. The product struggled in customers' hands for most of 2024. The narrative looked like vaporware.
That stopped being the story sometime late last year.
Cognition told investors it has reached $492 million in annualized revenue run-rate, according to TechCrunch's confirmed reporting on the round.
Enterprise usage of Devin has grown 50% month-over-month for the past six months, per the company's own disclosures. The customer roster, confirmed by TechCrunch, includes Mercedes-Benz, NASA, Goldman Sachs, and Santander. The Decoder and other outlets list Citi, Dell, and US military units in production.
The product change that made it work was less about model quality than about how Devin operates. Cognition rebuilt Devin around long-running tasks with explicit checkpoints, parallel subagents, and a chat-based workflow that lets engineers correct it mid-stream. Customers stopped asking Devin to "build me an app" and started asking it to modernize one library, port one service, or close one bug ticket at a time.
VCs Bet Against the Model Makers Eating This Market
Last year, the conventional wisdom was that the model makers would own software engineering. Anthropic's Claude Code, OpenAI's Codex, and Google's coding agent Jules each gained traction. Google acqui-hired the top of Windsurf's team. Cursor's valuation kept doubling. Cognition's bet looked precarious.
Then came the funding rounds.
Cursor closed at $50 billion in April. SpaceX paid ten billion dollars just for the option to buy Cursor outright. Now Cognition has more than doubled its own valuation in eight months on revenue that barely existed a year ago.
| Company | Recent Valuation | Revenue Marker | Lead Investors |
|---|---|---|---|
| Cognition | $26B post-money (May 27, 2026) | $492M ARR, 50% MoM growth | Lux, General Catalyst, 8VC |
| Cursor | $50B (April 2026) | Doubled in 5 months | Oversubscribed round |
| Anthropic Claude Code | Embedded in Anthropic | Not disclosed publicly | Internal investment |
| Google Jules | Embedded in Google | Not disclosed publicly | Internal investment |
The bet from Lux, General Catalyst, 8VC, and the rest is that enterprise customers want a vendor focused on autonomous coding agents as a category, not as a feature of a foundation model.
Cognition's pitch also leans on a structural advantage. The company acquired the remaining pieces of Windsurf in July 2025, after Google's acqui-hire deal pulled the founders out. Windsurf's IDE technology let Cognition wrap Devin in a desktop and editor surface that competing model makers do not control end-to-end.
The Mercedes-Benz Number Is the One Investors Are Talking About
Multiple outlets covering the round have cited a Mercedes-Benz proof point that, if accurate, is the strongest single argument for autonomous coding agents in production: the carmaker reportedly used Devin to modernize a legacy system in eight days that internal teams had estimated would take eight months.
The number originates in Cognition's pitch material. Mercedes-Benz has not publicly confirmed the figure. Cognition has not detailed which system, which programming language, or which team owned the original estimate.
If even half-true, it explains the round. A 30x productivity claim on a real production modernization is the kind of data point that gets a Series D priced at twenty-six billion dollars.
The Counterargument Is Getting Harder to Make
Skeptics still have arguments. Devin makes mistakes. Customer reviews on Hacker News and Reddit through 2025 ranged from awestruck to dismissive. AI coding benchmarks have repeatedly been gamed by labs since 2023, and Cognition has not publicly disclosed how its enterprise customers measure success. Cursor itself recently told developers to "stop coding" and let the agent run unattended, a positioning that prompted a backlash among working engineers.
The 89% internal-code figure also invites the question of what counts as a pull request. A trivial dependency bump and a 5,000-line refactor are both technically pull requests. Cognition has not broken out the mix.
But the dollar figures tell a story even skeptics struggle to dismiss. Revenue does not grow 13x in a year on hype alone. Enterprise contracts at Mercedes-Benz, Goldman Sachs, and the US military do not get signed without internal champions who tracked real outcomes.
The honest assessment: the autonomous coding agent category has graduated from interesting demo to production tool. The remaining debate is about who wins the category.
What Practitioners Should Take From This
For data scientists and ML engineers, three things matter from this round.
First, the foundation model layer is no longer the only place value accrues. Cognition does not train a frontier model. Its $26 billion valuation rests on harness design, agent orchestration, and the workflow that wraps the model.
The same logic applies to Cursor at fifty billion dollars. Building on top of Claude or GPT is no longer a discount to the underlying model company.
Second, the question of whether AI coding agents work in production has answers now. The answers are not uniformly positive, but the existence of nearly half a billion dollars in revenue at 50% month-over-month growth from Fortune 500 customers means the category is real. ML engineers evaluating where to spend integration effort can stop treating Devin, Cursor, and Claude Code as research projects.
Third, the recursion has started. When a coding company's coding agent writes most of the coding company's code, the rate of improvement compounds. Anthropic acknowledged the same dynamic in May, when Andrej Karpathy joined the company with a stated job of using Claude to build the next Claude. The companies that ship working coding agents first will ship the next generation of coding agents fastest.
The Bottom Line
Eight months ago Cognition was a $10.2 billion company telling investors its AI software engineer would write code for Mercedes-Benz one day.
Today it is a twenty-six-billion-dollar company telling them it already does, and that the same AI now writes 89% of Cognition's own code.
The leap is what venture capitalists pay for. Whether the price tag holds depends on whether the autonomous coding agent category settles into a winner-take-most market, or whether Anthropic, OpenAI, and Google succeed in bundling the capability into their own subscriptions before standalone players entrench themselves.
The next data point will come in roughly six months. If Cognition's annualized revenue has not at least doubled again, the recursion thesis breaks. If it has, the next round will be priced higher than Cursor's last one.
As TechCrunch summarized it, the round amounts to "a giant vote of confidence from top-tier VCs that there will be room for independent AI software coding startups." That confidence cost a billion dollars to express.
Sources
- TechCrunch: AI coding startup Cognition raises a billion-dollar Series D at twenty-five billion pre-money valuation (May 27, 2026)
- Bloomberg: AI Coding Startup Cognition Raises a Billion at Twenty-Six Billion Valuation (May 27, 2026)
- The Next Web: Cognition raises a billion at twenty-six billion valuation for AI coding agent (May 27, 2026)
- The Decoder: AI coding agent Devin maker Cognition more than doubles its valuation to twenty-six billion (May 28, 2026)
- eWeek: AI Coding Startup Cognition Hits Twenty-Six Billion Valuation After Massive Billion-Dollar Raise (May 27, 2026)
- StartupHub: Cognition AI Series D funding round (May 27, 2026)
- TechCrunch: More details emerge on how Windsurf's VCs and founders got paid from the Google deal (August 1, 2025)