Cognition brings Dioxus team into Devin's runtime work after $2B raise

Why it matters

Cognition is putting fresh capital into the execution layer around Devin. Kelley's team gives it deeper control over the virtual machines, computer interaction, and testing that determine whether generated code works.

Jonathan Kelley (@jkelleyrtp), the former Cloudflare systems engineer who turned a college project into the Rust framework Dioxus, is bringing his team into Cognition to work on the infrastructure surrounding Devin.

Dioxus announced the team move on September 9th. Cognition says the Dioxus engineers will work on Devin's virtual machines, computer-use capabilities, and testing systems. Cognition says it had already used Dioxus extensively while building Devin, including a Dioxus terminal renderer that Cognition rebuilt for the coding agent's command-line interface.

Cognition on X

Kelley gets an unusually direct resolution to the problem facing many open-source founders: Dioxus can keep shipping without having to force a commercial product onto its community. Cognition gets engineers whose code was already embedded in its product and whose recent work had moved into cloud development environments, virtualization, and operating-system support.

The announcements do not describe the move as an acquisition or disclose its financial terms. Cognition and Kelley consistently describe the Dioxus team as joining Cognition.

A college project became Devin infrastructure

Kelley started Dioxus as a side project while studying at Franklin W. Olin College of Engineering. By the time Dioxus entered Y Combinator's Summer 2023 batch, he was presenting himself as its solo founder and pitching a simpler way for small teams to ship applications across web, desktop, and mobile from one Rust codebase.

Before Dioxus, Kelley worked as a systems engineer at Cloudflare on the 1.1.1.1 DNS service and WARP networking product, according to his GOSIM speaker profile. That systems background became increasingly relevant as Dioxus grew beyond a user-interface framework.

Kelley wrote in Dioxus' account of the Cognition deal that stronger AI-generated Rust code and growing demand for cross-platform libraries pushed Dioxus toward a cloud coding harness called SkyVM. Dioxus says SkyVM supports macOS, iOS, Android, Linux, and Windows, with virtual-machine snapshotting, forking, and rollback capabilities. Dioxus also claims new virtual machines can spawn or resume in under 50 milliseconds; that performance figure has not been independently established.

The teams met after Dioxus discovered that Cognition had rebuilt an older Dioxus terminal renderer for Devin's command-line interface. For Kelley, the overlap was larger than a library integration. Dioxus had started building the execution machinery needed to let AI agents write, run, and test software across operating systems. Cognition needed that machinery around Devin.

Cognition is investing below the model layer

The timing puts the integration inside a rapid expansion at Cognition. On September 8th, Cognition said it raised over $2 billion at a $48 billion valuation, with Andreessen Horowitz and Accel leading the Series E. Dioxus announced that its team was joining Cognition the following day.

That timing matters for coding agents. Generating a plausible patch is one part of the job. Devin also needs an isolated environment, access to development tools, a way to operate software interfaces, and tests that can expose a bad change before it reaches a customer.

Cognition's own SWE-bench technical report described testing and environmental feedback as central to Devin's ability to correct mistakes during longer tasks. The Dioxus integration targets that execution loop rather than adding another model or chat interface.

Cognition has continued investing at both levels. RuntimeWire reported September 10th that Cognition shipped SWE-2, a lower-cost coding model for Devin, based on Cognition's own performance and cost comparisons. Bringing in Dioxus gives Cognition additional control over what happens after a model produces code: provisioning the machine, interacting with applications, running the result, and checking whether it works.

Kelley's team also brings experience with the cross-platform edge cases that become expensive when an agent has to do more than edit a repository. Mobile simulators, desktop applications, browser rendering, operating-system APIs, and native dependencies all create failure points that a coding model cannot solve through text generation alone.

Dioxus keeps its open-source work

Cognition says it will continue supporting Dioxus and its related repositories, including Blitz, Taffy, and Subsecond. Cognition plans to increase investment in Dioxus-Native and Blitz, a native HTML and CSS renderer.

Kelley acknowledged that joining Cognition will leave the Dioxus team with less time devoted solely to the original framework. Nico Burns, Dioxus' lead engineer on Blitz, will work full time on maintaining and improving Dioxus, Kelley wrote. Cognition has also positioned Devin as a development tool for the open-source projects themselves.

The arrangement removes a commercial question Kelley had carried since Dioxus became a full-time venture. Kelley said Dioxus raised a total of $3.5 million, with backing from Y Combinator and Khosla Ventures, and credited Futurewei Technologies for supporting the work. Dioxus had explored paid deployment tools, but Kelley now says the Cognition integration removes the immediate pressure to monetize the framework.

That is a favorable outcome for a founder whose adoption came through freely available code. Kelley can keep the projects open, put his systems work into a heavily financed coding agent, and give Dioxus engineers a customer for infrastructure they had already started building. Cognition, meanwhile, brings a dependency and its maintainers closer to Devin at a point when execution quality is becoming as consequential as model quality.