Cognition's AI Software Engineer 'Devin' Emerges from Stealth

Cognition, a recently formed AI startup backed by Peter Thiel’s Founders Fund and industry leaders such as former Twitter executive Elad Gil and Doordash co-founder Tony Xu, has launched a fully autonomous AI software engineer named ‘Devin’.

Autonomous AI for Complete Development Projects

Unlike other coding assistants, Devin is designed to manage entire development projects from start to finish. This includes writing code, fixing bugs, and executing the project. This is the first offering of its kind, and it has demonstrated its capabilities by handling projects on Upwork.

A New Shift in AI-Assisted Development

The introduction of Devin signifies a major shift in AI-assisted development, offering engineers an AI worker capable of fully managing their projects, as opposed to a co-pilot that merely suggests code snippets. However, as of now, Devin remains non-public, with access granted only to a select group of customers.

What Devin can do?

Devin can access common developer tools in a sandboxed environment to execute complex engineering tasks requiring thousands of decisions. The user simply types a natural language prompt into Devin’s chatbot-style interface, and the AI takes over, developing a plan to tackle the problem. It then starts the project using its tools, writing its own code, fixing issues, and reporting on its progress in real-time.

A New Paradigm in Software Development

Cognition suggests that Devin’s capabilities allow engineering teams to delegate some of their projects to the AI, enabling them to focus more on creative tasks that require human intelligence. This could be a glimpse into the future of software development, where AI workers overseen by human supervisors undertake most tasks.

Capabilities and Performance

Devin is capable of handling a variety of tasks, from deploying and improving apps/websites, finding and fixing bugs in codebases, to more complex tasks like setting up fine-tuning for a large language model using a GitHub research repository link. In the SWE-bench test, Devin was able to resolve 13.86% of cases end-to-end without human assistance, outperforming other AI models.

Undisclosed Core Technology

The specifics of the technology behind Devin are currently undisclosed. Cognition has not shared whether it is using its own proprietary model or a third-party model. However, it notes that the work is the result of its “advances in long-term reasoning and planning.”

Looking Forward

Cognition is currently offering early access to Devin only to select users while it ramps up capacity. Broader access is expected in the future. The company hints that this is “just the beginning”, possibly indicating plans to apply its advances in reasoning to launch similar AI workers for other disciplines. The company has so far received $21 million in funding.