Cognition, the startup behind the AI coding agent Devin, is in talks with investors about a new funding round at a valuation of about $40 billion, according to people familiar with the matter. The figure, if confirmed, would nearly double the company’s valuation in a matter of months and put it among the most valuable startups in the fastest-moving corner of the AI industry.
The company closed a $1 billion round only a few months ago at a $26 billion valuation. The jump to $40 billion in one step reflects the intensity of investor demand for AI software that writes code, a category that has become the hottest in venture capital. Startups in the space have raised enormous sums at rising valuations, betting that software development will be the first business function where AI agents replace human workers at scale.
Cognition makes Devin, an autonomous coding agent that can take a task described in plain language and work through it: reading code, writing changes, running tests, and fixing errors without constant supervision. The product has drawn attention both for its ambitions and for its limits, which have been documented by developers who have pushed the agent into production work. The company says its tools are used by engineering teams at large companies, though it does not publish detailed customer numbers.
The valuation surge across the category is driven by a simple conviction: coding is where AI pays for itself first. Software engineers are expensive, and companies that can automate part of the work stand to save enough to justify almost any software subscription. Investors are funding the companies that can show agents doing real work, and the winners, they believe, will become the application layer of the AI economy.
The competition is intense. Devin competes with coding assistants from OpenAI, Google, and a cluster of startups, including companies whose tools have reached large developer audiences by embedding in existing editors. Cognition’s pitch is different: its agent works more independently, taking longer tasks and executing them end to end, rather than suggesting snippets for a human to approve. The distinction matters to investors because it defines how much of the engineering job can be automated.
The category’s metrics are unusual. Coding agents are sold on developer time saved, a benefit that is real but hard to measure, and the products are priced per seat or per agent rather than per outcome. That pricing is part of the appeal to investors: if agents truly replace seats, the software can command subscription fees comparable to the salaries they displace, a ceiling far above the per-developer pricing of traditional tools.
The funding round, if completed, would give Cognition a war chest to match its ambitions. The company has said it intends to hire engineers, expand its agent’s capabilities, and build the infrastructure to serve more customers. The money would also fund the compute that coding agents consume, a cost that scales with usage and that has become a defining expense for every company in the category.
Not everyone is convinced the valuations are justified. Skeptics note that coding agent revenue remains small relative to the sums being invested, that the technology still makes mistakes a careful engineer must catch, and that the economics of agents could be eroded if model providers squeeze the layer between themselves and developers. The history of developer tools is full of companies that grew fast and then watched incumbents absorb their features.
The history cuts both ways. Developer tool companies have been acquired or copied by the platforms they built on, and a coding agent that runs inside a model provider’s environment could find its most valuable capabilities absorbed by the provider itself. Cognition’s answer is to control more of the stack, including the models it runs, and the funding round would support that integration work. Whether that is enough to keep a layer of independent agents alive is one of the open questions of the category.
The broader signal is in the capital itself. Venture funds that spent two years writing checks for model companies are now writing larger checks for the applications built on top of them, and coding is the category attracting the most. The bets are being made on the assumption that the value created by AI in software development will be captured by the application companies, not by the models underneath. That assumption has not been tested in public markets.
For Cognition, the round would be a validation of its product direction and a bet on its ability to execute. The company’s founders have said they believe agents will transform how software is built within a few years, and the $40 billion figure prices in that transformation happening soon. If it does, the valuation will look cheap; if not, the correction will be painful. Investors in the round are making the same bet the company is: that the future of coding is autonomous, and that Devin will be part of it.
The round would also set the bar for the rest of the sector. Every coding startup in the market is now fundraising against the $40 billion figure, and investors will use it as a reference point when valuing rivals, a dynamic that pushes the whole category upward. That is good for companies with revenue and customers, and harder for earlier-stage startups that must now justify valuations against a much bigger anchor. The Cognition round, if completed, will be remembered less for the money raised than for the number attached to it.
