Take a look at Devin, the world's first fully ‘autonomous’ AI software engineer

US-based applied AI lab focussed on reasoning, Cognition, has developed the world’s first fully ‘autonomous’ AI software engineer, Devin.

Stealing the spotlight in the AI world, a US-based applied AI lab focussed on reasoning, Cognition, has developed the world’s first fully ‘autonomous’ AI software engineer, Devin.

Devin, the state-of-the-art on the SWE-Bench coding benchmark, has cleared practical engineering interviews from the leading AI companies and has also completed real jobs. (Representative Photo)

“ Devin is a tireless, skilled teammate, equally ready to build alongside you or independently complete tasks for you to review,” says the team at Cognition. Through the development of the fully ‘autonomous’ AI software engineer, the team says they aim to focus on more interesting problems and engineering teams can strive for more ambitious goals.

Devin, the state-of-the-art on the SWE-Bench coding benchmark, has cleared practical engineering interviews from the leading AI companies and has also completed real jobs, mentions Cognition in an official tweet on X (previously Twitter).

What is Devin capable of doing?

According to the developers, Devin has been equipped with common developer tools including the shell, code editor, and browser within a sandboxed compute environment, all the tools that humans would need for work to get done.

Cognition mentions some of the following as some of the functions that Devin is capable of doing:

I) Can learn how to use unfamiliar technologies

II) Can build and deploy apps end-to-end

III) Can autonomously find and fix bugs in codebases

IV) Can train and fine-tune its own AI models

V) Can address bugs and feature requests in open-source repositories

VI) Can contribute to mature production repositories

VII) Completed real jobs on Upwork

Performance:

“ Devin correctly resolves 13.86 percent of the issues end-to-end, far exceeding the previous state-of-the-art of 1.96 percent. Even when given the exact files to edit, the best previous models can only resolve 4.80% of issues,” stated Cognition in their blog post.