What AI Coding Agents Are—and Why Devin Matters Now
AI coding agents are autonomous or semi-autonomous software systems that interpret human instructions, plan multi-step development tasks, write and review code, and integrate with developer tools to deliver working software with minimal human intervention across the full lifecycle. Cognition’s Devin is a leading example of this shift, presented as an “AI software engineer” that can own tickets, work across repositories, and submit pull requests. Cognition reports that Devin is now used inside thousands of companies, with enterprise usage increasing more than tenfold since early 2026. That growth has pushed Devin’s annualized revenue run rate to about USD 492 million (approx. RM2.26 billion). This scale signals that AI coding agents have moved beyond early experiments into recurring, production-grade enterprise software adoption. Rather than treating agents as side tools, large organizations are starting to embed them into core engineering workflows, from maintenance tasks to complex modernization projects.
The $1B Funding Signal: Investors Are Buying Production, Not Hype
Cognition has raised more than USD 1 billion (approx. RM4.6 billion) at a USD 26 billion (approx. RM119.6 billion) valuation, led by Lux Capital, General Catalyst, and 8VC, with participation from investors such as Ribbit Capital, Atreides, and Layer Global. The jump from a reported USD 10.2 billion (approx. RM46.9 billion) valuation in September 2025 to this new level in under a year reflects a sharp re-rating of AI coding agents. What stands out is that the valuation is tied to operating traction. Cognition’s reported 50% month-over-month corporate growth and a USD 492 million (approx. RM2.26 billion) revenue run rate show Devin is already pulling in serious enterprise spend. According to Cognition, “cloud agents have gone from niche to mainstream, and today they are the fastest-growing way to create software.” For investors, this looks less like a speculative bet on models and more like a high-growth enterprise software business with sticky, expanding usage.
What Enterprises Are Actually Building With Devin
Devin’s traction comes from concrete outcomes in production environments, not demos. Cognition lists major organizations such as Citi, Mercedes-Benz, Goldman Sachs, Dell Technologies, Santander, the U.S. Army, and the U.S. Navy among its technology partners, alongside fast-growing startups like Exa, Modal, Eight Sleep, and OpenRouter. These teams are using AI coding agents for real workloads: legacy modernization, security remediation, and day-to-day feature work. Mercedes-Benz, for example, cut an eight-month legacy modernization project down to eight days. Itaú Unibanco reports that Devin now fixes 70% of its security vulnerabilities automatically. Systems integrators including Infosys and Cognizant have threaded Devin into project delivery workflows to accelerate client software development. This pattern explains rapid enterprise software adoption: organizations are starting with painful, high-effort tasks—upgrades, refactors, and bug backlogs—then expanding into more of the software lifecycle as agents prove reliable over time.
How AI Developer Tools Are Reshaping Workflows
Cognition positions the future of software development as “self-driving”: engineers define and structure problems, while AI agents execute most of the work. Inside Cognition itself, the company says 89% of code committed by its engineers is now committed by Devin, with the remainder generated by local agents running in Windsurf. That internal adoption mirrors how customers are restructuring teams. AI developer tools like Devin are moving beyond autocomplete-style assistance toward full workflow orchestration—planning tasks, editing multiple services, running tests, and merging pull requests. Cognition reports that Devin had merged hundreds of thousands of pull requests by late 2025, giving the agent extensive real-world feedback loops. To support this, Cognition evaluates foundation models across more than 100 categories of software engineering tasks and has released the SWE-1.6 model, now the most-used model within Windsurf thanks to its speed and cost efficiency, reaching throughput of up to 950 tokens per second.
The Economics: From Cost Center to Autonomous Throughput
For enterprises, the economics of AI coding agents are starting to resemble a new class of infrastructure. Devin’s reported USD 492 million (approx. RM2.26 billion) annualized revenue shows that organizations are willing to commit serious budget to agent-based development, treating it like a scalable, on-demand pool of engineering capacity. Instead of hiring only more humans to increase throughput, teams can point Devin at backlogs, migrations, and security issues that were previously under-resourced. Price-to-performance is a core selection criterion: Cognition stresses that customers want the best-performing model per task at the best effective rate, leading the company to optimize model choice rather than lock into a single stack. As Anthropic, OpenAI, and others expand their own coding agents and workflow tools, competition will likely push agent costs down and capabilities up, further shifting software development economics toward high-level design and away from manual coding effort.