Anthropic has announced “Devin,” claimed to be the world’s first fully autonomous AI software engineer. Devin is capable of planning and executing complex engineering tasks requiring thousands of decisions, while recalling relevant context, learning over time, and fixing its own mistakes along the way.
What makes Devin unique is its ability to actively collaborate with human engineers. It can report progress in real-time, accept feedback, and work through design decisions cooperatively. Devin has access to standard developer tools like shells, code editors and browsers within a sandboxed environment.
Some of Devin’s showcase capabilities include:
- Learning how to use new technologies from materials like blog posts
- Building and deploying full apps end-to-end, with iterative user-requested features
- Automatically finding and fixing bugs across codebases
- Training and fine-tuning its own AI models given resources
- Addressing open source issues and bugs using just a GitHub link
Devin achieved state-of-the-art results on the SWE-bench coding benchmark, correctly resolving 13.86% of real-world issues from open source projects. This far exceeds the previous best of 1.96% for end-to-end issue resolution.
With Devin, Anthropic aims to have AI software engineers augment and collaborate with human developers, allowing teams to tackle more ambitious goals. The era of autonomous AI coding may be upon us.
More info on Cognition-Labs: https://www.cognition-labs.com/blog
