According to news on March 13, a start-up company Cognition recently launched an artificial intelligence assistant called Devin, which is designed to assist software engineering teams in completing coding and other development tasks. Unlike existing AI coding assistants, Devin can be programmed to execute end-to-end software projects, including deploying applications, fixing bugs, learning new technologies, etc., while humans play the role of supervision and guidance.
Devin can execute multi-step workflows based on user needs while staying on track. Engineers can observe its progress in real time and issue instructions to make corrections when errors are discovered. This allows teams to outsource some work to AI assistants and focus on more creative work themselves.
According to the demonstration, Devin can handle a variety of tasks including deploying websites, debugging code, generating hidden information images, and training computer vision models. In the software engineering benchmark test, it can complete 13.86% of the cases independently, which is much higher than other large language models.
Although the core technical details have not been disclosed, Cognition said that Devin stems from the progress of its long-term reasoning and planning research. The tool is currently in the internal testing stage, and interested users can apply for early trial use. Wider access may become available in the future.
Cognition hints that coding is just the beginning, which means its artificial intelligence assistant may be extended to more areas. The company plans to leverage the cross-domain reasoning advantages of artificial intelligence to explore empowering multiple industries.
The emergence of Devin brings a new artificial intelligence collaboration experience to software developers. By supervising AI systems to handle tedious tasks, engineers can focus on innovative tasks, which is expected to increase productivity. However, this technology is still in its early stages, and its maturity and effectiveness need to be further evaluated by the industry.