Cognition, an AI startup backed by prominent figures in the tech industry, has recently emerged from stealth mode with the announcement of its fully autonomous AI software engineer called “Devin.” Unlike other coding assistants, Devin is capable of handling entire development projects from start to finish, including writing code, fixing bugs, and executing the final product. This marks a significant shift in the AI-assisted development space, as engineers now have access to a full-fledged AI worker rather than a copilot that can only provide suggestions or write basic code snippets.
Devin’s capabilities have been demonstrated through its ability to handle projects on Upwork, a popular freelancing platform. By accessing common developer tools within a sandboxed compute environment, Devin can plan and execute complex engineering tasks requiring thousands of decisions. Users simply type a natural language prompt into Devin’s chatbot interface, and the AI software engineer takes over, developing a detailed plan to tackle the problem. It then proceeds to work on the project, writing its own code, fixing issues, and providing real-time progress updates. Users have the option to intervene if something doesn’t look right and give commands to the AI to make adjustments. This allows engineering teams to delegate tasks to Devin and focus on more creative endeavors that require human intelligence.
Devin’s capabilities extend beyond basic coding tasks. It can handle a wide range of development projects, from deploying and improving apps/websites to finding and fixing bugs in codebases. It can even set up fine-tuning for large language models using GitHub repositories or learn how to use unfamiliar technologies. In a test challenging AI assistants with real-world open-source projects, Devin was able to correctly resolve 13.86% of the cases end-to-end without any human assistance. This outperformed other AI models, such as Claude 2, SWE-Llama-13b, and GPT-4, which required some level of assistance to resolve issues.
While AI tools for software development are not new, Devin stands out for its ability to handle multiple steps and complete entire engineering projects. The core technology behind Devin remains undisclosed, but Cognition attributes its capabilities to advances in long-term reasoning and planning. The company is currently offering early access to select users and expects to open up broader access in the future. Cognition also hints at the possibility of launching similar AI agents/workers for other disciplines, indicating that coding is just the beginning.
Cognition has received $21 million in funding so far, with support from Peter Thiel’s Founders Fund and industry leaders like Elad Gil and Tony Xu. The startup’s emergence from stealth mode and the introduction of Devin as a fully autonomous AI software engineer have generated significant interest within the tech industry. As AI continues to advance, the role of AI workers overseen by human supervisors/users may become more prevalent, potentially shaping the future of software development and computer work in general.
Overall, Cognition’s Devin represents a major milestone in the AI-assisted development space, providing engineers with a powerful tool to handle complex projects. Its ability to handle multiple steps and deliver end-to-end solutions sets it apart from other coding assistants. As more users gain access to Devin and its capabilities continue to expand, it will be interesting to see how this AI software engineer transforms the way software development is approached.