四个月前，Cognition的估值是240亿美元。现在，480亿美元。
9月，这家开发了Devin AI编程Agent的公司完成了20亿美元的E轮融资，由a16z和Accel领投。其ARR（年度经常性收入）已经接近9亿美元。一家成立不到三年的公司，估值超过了大多数上市科技公司——这就是2026年AI编程赛道的疯狂程度。
Devin凭什么值480亿
Devin是全球第一个被称为「AI软件工程师」的产品。它不只是帮你写代码——它可以理解需求、规划实现路径、编写代码、调试错误、提交PR，端到端地完成一个软件开发任务。你给它一个GitHub issue，它可以自己读代码库、理解上下文、做出修改、跑测试、提PR，整个过程几乎不需要人干预。
这种能力在2024年初首次亮相时，整个行业为之震动。当时很多人的反应是「这是Demo吧？」——一个AI能自主完成软件开发任务，这在当时是难以想象的。但两年过去，Devin已经被大量企业在生产环境中使用，从初创公司到Fortune 500，覆盖代码迁移、测试编写、bug修复、新功能开发等多种场景。
9亿美元的ARR说明了一切：Devin不是Demo，是真金白银的产品。企业愿意为它付钱，是因为它真的在提升开发效率、降低人力成本。当一个工具的付费用户增长如此之快，资本给出天价估值也就不奇怪了。
四个月翻倍，合理吗？
从240亿到480亿，四个月翻一倍。即使是AI赛道最疯狂的时期，这个速度也令人瞠目。
支持者会说：看看ARR的增长曲线。从年初到现在，Devin的ARR增速惊人。如果按这个趋势，到年底可能突破15亿甚至更高。用ARR倍数来算，480亿估值对应大约50-60倍ARR——在高增长SaaS赛道，这个倍数虽然高但不算离谱。
质疑者会说：这个估值透支了太多未来预期。AI编程赛道的竞争正在加剧——Cursor已经突破10亿美元ARR、GitHub Copilot在持续进化、AWS的Kiro Crew刚刚开源、Google的Jules也在发力。Devin先发优势能保持多久？当越来越多的工具具备类似的Agent能力，Devin的差异化会不会被稀释？
还有一个更根本的问题：AI编程Agent会不会像之前的低代码平台一样，最终只在特定场景有用，而无法真正替代软件工程师？如果是这样，480亿估值对应的市场空间就值得重新审视。
AI编程赛道的格局正在固化
Cognition的融资，不只是一家公司的事。它折射出整个AI编程赛道的格局正在快速固化。
目前的头部玩家可以分为三类：第一类是Agent型（如Devin），强调自主完成任务；第二类是辅助型（如Copilot、Cursor），强调人机协作、实时补全；第三类是平台型（如AWS Kiro、Google Jules），强调多Agent协同和工程化部署。
三种路线不是互斥的，而是在不同场景中各有优势。但资本显然更看好Agent型路线——因为它对应的价值主张最激进：「不是帮你写代码，而是替你写代码」。如果这个愿景实现，市场规模就不是「开发工具」级别的，而是「开发人力」级别的——全球软件开发者的薪资总支出，是一个万亿美元级别的市场。
480亿估值赌的就是这个。赌对了，Cognition就是下一个Salesforce。赌错了，它就是AI时代的Pets.com。
明天见。
Four months ago, Cognition's valuation was $24 billion. Now, $48 billion.
In September, the company behind Devin — the AI coding agent — closed a $2 billion Series E led by a16z and Accel. Its ARR (annual recurring revenue) is approaching $900 million. A company less than three years old, valued higher than most public tech companies. This is the level of insanity in the 2026 AI coding race.
What Makes Devin Worth $48 Billion?
Devin was the first product globally to be called an "AI software engineer." It doesn't just help you write code — it understands requirements, plans implementation paths, writes code, debugs errors, and submits PRs, completing software development tasks end to end. Give it a GitHub issue, and it reads the codebase, understands context, makes changes, runs tests, and opens a PR — the whole process with almost no human intervention.
When this capability first debuted in early 2024, the entire industry was shaken. Many people's reaction was "this must be a demo" — an AI autonomously completing software development tasks was inconceivable at the time. But two years later, Devin is being used in production by companies ranging from startups to Fortune 500s, covering code migration, test writing, bug fixing, new feature development, and more.
$900M ARR says it all: Devin isn't a demo — it's a product generating real revenue. Companies are willing to pay because it genuinely improves development efficiency and reduces labor costs. When a tool's paying user base grows this fast, sky-high valuations from capital are hardly surprising.
Doubled in Four Months — Is That Reasonable?
From $24B to $48B in four months. Even during the wildest days of the AI race, this speed is jaw-dropping.
Supporters point to the ARR growth curve. From the start of the year to now, Devin's ARR growth has been staggering. If the trend continues, it could break $1.5B or higher by year-end. Using ARR multiples, a $48B valuation implies roughly 50-60x ARR — high in high-growth SaaS, but not outrageous.
Skeptics say: this valuation has overdosed on future expectations. The AI coding race is intensifying — Cursor has broken $1B ARR, GitHub Copilot keeps evolving, AWS's Kiro Crew just open-sourced, Google's Jules is pushing hard. How long can Devin's first-mover advantage last? As more tools develop similar agent capabilities, will Devin's differentiation erode?
There's also a more fundamental question: will AI coding agents, like low-code platforms before them, ultimately prove useful only in specific scenarios and fail to truly replace software engineers? If so, the market size corresponding to a $48B valuation deserves reassessment.
The AI Coding Landscape Is Solidifying
Cognition's fundraise isn't just about one company. It reflects how the entire AI coding landscape is rapidly solidifying.
Current top players fall into three categories: first, agent-type (like Devin), emphasizing autonomous task completion; second, assistant-type (like Copilot, Cursor), emphasizing human-AI collaboration and real-time completion; third, platform-type (like AWS Kiro, Google Jules), emphasizing multi-agent coordination and engineering-grade deployment.
The three approaches aren't mutually exclusive — each has advantages in different scenarios. But capital clearly favors the agent-type approach, because its value proposition is the most aggressive: "not helping you write code, but writing code for you." If this vision materializes, the addressable market isn't "developer tools" scale — it's "developer labor" scale. The global total compensation for software developers is a trillion-dollar market.
$48B is a bet on exactly that. If it pays off, Cognition becomes the next Salesforce. If it doesn't, it's the Pets.com of the AI era.
See you tomorrow.
480亿估值赌的不是开发工具市场，而是开发人力市场——一个万亿美元级别的市场。
—— Dawn Vision编辑部
A $48B valuation is betting not on the developer tools market, but on the developer labor market — a trillion-dollar market.
— The Dawn Vision Editorial Desk
Cognition · Devin · AI coding agent · $48B valuation · Series E · a16z · Accel · ARR · AI coding
