Devin – artificial intelligence that can do the work of programmers
Cognition has introduced Devin, a new tool developed on the basis of artificial intelligence that can independently perform various engineering tasks.
According to the developers, Devin can plan and execute complex engineering tasks that require thousands of decisions, recall the relevant context at each step, learn over time, and correct errors.
The new AI is also equipped with the usual developer tools, including a shell, a code editor, and a browser in the sandbox of the computing environment.
Of course, the AI can actively cooperate with the user. Devian will report on its progress in real time, accept feedback, and work with the user to make design choices if necessary.
The developers claim that Devin will be able to learn how to use unfamiliar technologies, create applications, find and fix bugs, train and customize its own AI models, etc.
Cognition also evaluated the AI on the SWE-bench, which requires agents to solve real-world GitHub problems found in open source projects.
According to the developers, Devin correctly solves 13.86% of the problems end-to-end, which is significantly higher than the previous state of the art of 1.96%. Previous models could only solve 4.80% of problems.
The AI is currently in early access, but the company can be contacted via Google Form or email: info@cognition-labs.com.
