Devin, an AI-based software engineer capable of coding, building apps and more, revealed by Cognition Labs

Devin, a generative artificial intelligence (AI) model that can function as a software engineer, was unveiled by AI startup Cognition Labs. The company claims that Devin has successfully passed hands-on engineering interviews from AI companies and has even completed real-world tasks on Upwork. The AI ​​tool comes with its own shell, code editor, and browser to perform complex engineering tasks like completing end-to-end coding projects, building and deploying websites and apps, and even training and fine-tuning your own AI models.

Cognition Labs unveiled the AI ​​model in a post of X (formerly Twitter) and hailed him as “the first software engineer.” Making the announcement, the startup said: “Devin is the new state of the art in SWE-Bench’s coding benchmark, has successfully passed hands-on engineering interviews from leading AI companies and even completed real-world Upwork tasks.”

The AI ​​model is equipped with its own shell or interface, a built-in code editor for writing and deploying code, and a browser within a sandbox computing environment that allows it to perform complex engineering tasks. IN blog post, the company delved into its capabilities. According to the post and multiple video demonstrations, Devin can learn to use unfamiliar technologies, build and deploy end-to-end applications, autonomously find and fix bugs in codebases, address bugs and feature requests in open source repositories, to contribute to mature production repositories and even train and fine-tune its own AI models.

In addition, Devin AI also scored 13.86 percent on the SWE-bench coding test. Not only did it significantly outperform other major AI models like Claude 2, which scored 4.80 percent, and GPT-4, which scored 1.74 percent, but the company claims it was able to solve the problems without assistance. It should be noted that all other AI models were assisted and told exactly which files needed to be edited.

Although Cognition has made tall claims, they cannot be verified at this time because the platform is not publicly available. The startup also hasn’t released a detailed technical report on the AI ​​model, although it has said it will be released soon. However, if the claims are true, the Devin AI model has set a new standard in the AI-powered code generation space. So far, all coding-oriented models are assistive in nature and can only perform tasks based on prompts and in a limited capacity. However, Devin can not only work autonomously, but also handle end-to-end projects. The pressing question is whether or not it can replace a human software engineer.

Devin is currently in early access, but the developers said that people who want to hire the AI ​​model for engineering work can contact them.