Cognition Brings Devin to the Genesis Mission: 17 U.S. Department of Energy Labs

Cognition has signed a memorandum of understanding (MOU) with the U.S. Department of Energy and joined the Genesis Mission, a federal program that has been called a "Manhattan Project for AI." The initiative was launched by executive order in November 2025. It connects 17 national laboratories, industry and universities, along with supercomputers and the largest body of federal scientific data. The goal is ambitious: to double the productivity of American science within a decade. Cognition's contribution to the plan is Devin, an autonomous AI engineer that carries out engineering tasks end to end and hands finished, merge-ready pull requests to humans for review.

Four areas of work

Cognition starts from a simple observation: scientific progress is bottlenecked by software. Researchers' workflows, data pipelines, simulation codes and the platforms that tie them together are written and maintained by people, and there are not enough of them. The company has said it will take part in four areas:

• Software and data security. Finding and fixing vulnerabilities in scientific codebases, many of which rely on outdated open-source components.
• Legacy code modernization. Documenting, testing and translating decades-old Fortran, C++ and COBOL codebases before the people who wrote them retire and take their knowledge with them.
• Expanding science's engineering capacity. Researchers hand data pipelines, model infrastructure and deployment over to Devin and spend more time on the research itself.
• Cloud modernization. Migrating old applications and infrastructure to modern cloud-native platforms.

The second item is arguably the most down-to-earth and the most needed. Translating and documenting code that is older than half of its current maintainers does not scale well with human labor. For agents, it is a near-ideal task: the volume is large, the rules for checking the work are clear, and the result can be run through tests.

Security as the main selling point

Cognition puts the emphasis on code protection, and the logic is easy to follow. The Genesis Mission's integrated discovery platform will link supercomputers, experimental facilities and decades of federal scientific data. A system like that becomes a target on a national scale, yet it is built on thousands of open-source libraries and often on code that predates modern security practices.

For this job the company has Devin Security Swarm. It searches large codebases for vulnerabilities, checks in an isolated sandbox whether they can actually be exploited, and opens pull requests with fixes for engineers to approve. Cognition ties this approach to CISA's guidance on remediating vulnerabilities in the age of AI. As part of its consortium membership, the company has offered the labs free code scans and remediation reports. It is also offering free compute to teams selected through Genesis Mission competitions (RFAs).

Certifications matter here too. The Cognition platform is going through FedRAMP Class D (High) and is listed in the FedRAMP Marketplace. Devin Desktop and CLI are already authorized at FedRAMP High and are suitable for ITAR workloads and IL4–IL6 levels. Devin is used by engineering teams in the U.S. Army and Navy and at NASA's Jet Propulsion Laboratory, and the company itself participates in the U.S. Office of Personnel Management's (OPM) Tech Force program.

What comes next and what labs should do

For now this is a framework agreement, not a contract. The MOU sets out how Cognition will work with the Department of Energy, the labs and the winning teams once the mission's first projects get underway. Timelines, scope and financial terms have not been disclosed, so it is too early to judge the impact.

The practical takeaway for labs and research groups: free code scans and compute are the cheapest way to test an agent on your own code. It makes sense to start with a narrow task, such as a dependency audit in a single repository, or tests and documentation for one Fortran module. Set your metrics up front: the share of PRs accepted, the number of confirmed vulnerabilities, and the engineering hours freed up. Cognition is accepting applications and questions through devin.ai/government and at [email protected].