The Devin AI software engineer creates websites or software from a single prompt. (Image Credit: Cognition)
We're starting to see artificial intelligence make an impact in the software development world. Cognition, a tech startup, recently announced it launched Devin, an AI software engineer that can create programs and websites with a prompt entered by the user. Devin AI isn't designed to take over human engineer jobs; rather, it is designed to work alongside them. So far, the AI tool has outperformed other AI models, completed Upwork-based jobs, and passed engineering interviews from the top AI companies.
"Devin is a tireless, skilled teammate, equally ready to build alongside you or independently complete tasks for you to review. With Devin, engineers can focus on more interesting problems, and engineering teams can strive for more ambitious goals," the company said.
This tool has a lot to offer and comes with essential features, such as a code editor, shell, and web browser. It learns from mistakes, makes thousands of decisions, and performs better over time. Additionally, Cognition claims that Devin AI is proficient and evaluates tasks on the SWE-bench coding benchmark. During tests against software engineering problems, the AI ran better than other models. It also passed engineering interviews, which had AI and software engineering-related tasks and challenges, conducted by leading AI companies.
More importantly, Devin AI isn't here to overtake jobs. Instead, it works with human engineers to improve productivity and efficiency by offering updates, collaborating on design choices, and accepting feedback. It's also capable of developing/launching apps, learning new tech, and finding/fixing code errors. Remarkably, this AI tool trains AI models and deals with open-source project issues.
Devin is also better at error resolution than other AI models --- fixing 13% versus the 1% its predecessors solved. "Devin correctly resolves 13.86% of the issues end-to-end, far exceeding the previous state-of-the-art of 1.96%. Even when given the exact files to edit, the best previous models can only resolve 4.80% of issues," the startup says.
Devin has worked on freelance job platforms, such as Upwork, and handled coding assignments that range from computer vision model debugging to creating reports.
Anyone ever try these types of "copilots?"
Have a story tip? Message me at: http://twitter.com/Cabe_Atwell