Devin – The first fully autonomous AI software engineer agent | AI tool set

What is Devin

Devin is the world’s first fully autonomous launch by artificial intelligence startup CognitionAI Software Engineer Intelligent Bodyhas strong programming and software development capabilities, and can assist or complete software development tasks in multiple aspects. In SWE-bench benchmarks, Devin’s performance in solving practical problems far exceeds AI models such as GPT-4 and Claude 2.

Although Devin’s development company Cognition is officially established only two months ago, its team members have rich experience in cutting-edge AI work and have multiple gold medals in the International Informatics Olympics (IOI) competition. It has won the Founders Fund led by Peter Thiel. $21 million in Series A financing.

Devin’s main functions

• Learn new technologies independently:Devin is able to expand its skill set by reading documents and code to learn techniques it is unfamiliar with.
• End-to-end build and deploy programs:Devin can understand the entire software development process, from front-end design to back-end deployment, and even including publishing and launching applications online. This means it can build websites, games, or other software projects from scratch and handle related workflows.
• Search and fix bugs independently: Devin has excellent debugging capabilities, can detect and fix errors in the code, and can be found and solved by problems that developers themselves do not notice.
• Training and fine-tuning AI models:Devin not only handles routine programming tasks, but also helps train and fine-tune other AI models, showing in-depth application capabilities in the field of artificial intelligence.
• Fix open source library: For issues in the open source community, Devin can understand and resolve, such as fixing known bugs or implementing new feature requests.
• Contribute to mature production databases:Devin is able to contribute to mature production libraries, such as fixing known bugs or adding new features.

Performance comparison of Devin

In the SWE-bench benchmark (requiring agents to solve the actual GitHub problem found in open source projects such as Django and scikit-learn), Devin is able to correctly handle 13.86% of the problems. This result is significantly higher than 1.96% of previous technical levels, showing Devin’s huge advantages in understanding and solving practical programming problems.

Comparing other AI models: Devin far outperforms other well-known AI models such as GPT-4 and Claude 2, which usually have lower accuracy rates in the same tests.

How to use Devin

Devin is still in the internal testing, please visitCoginition’s official websiteSee more information, users who want to get the first experience can fill in itDevin’s internal test application form。