A new generative artificial intelligence startup called Cognition AI Inc. is looking to disrupt coding with the launch of a new tool that can autonomously create code for entire engineering jobs, including its own AI models.

That tool’s name is Devin, and it takes the premise of GitHub Inc.’s and Microsoft Corp.’s Copilot developer tool much further, as it can carry out entire jobs on its own, rather than simply assist a human coder.

In a video (below) attached to a blog post announcing Devin, Cognition Chief Executive Scott Wu demonstrates how users can view the model in action. They can see its command line, code editor and workflow as it goes step-by-step, completing comprehensive coding projects and data research tasks assigned to it.

As soon as Devin receives a request, it will set about searching the internet for educational content that can teach it how to complete the assigned task. It can even debug any problems it encounters during the process, though users – who are basically just overseers and editors – can intervene if there’s any need to do so.

The launch of Devin is a big shift in the nascent AI-powered coding industry, with its promise of giving engineers their own automated worker that can carry out complete projects, as opposed to a copilot that would merely assist human coders with snippets or suggestions. For now, Devin is only available in private preview and only a few select journalists such as Bloomberg’s Ashlee Vance have had access to the tool.

Wu told Bloomberg that teaching AI to be a programmer is a “very deep algorithmic problem” where the system is required to make complex choices and look several steps into the future to determine what it should do next. “It’s almost like this game that we’ve all been playing in our minds for years, and now there’s this chance to code it into an AI system,” Wu explained.

According to Wu, Devin can access standard developer tools including a code editor, browser and shell. It can run these within a sandboxed environment to plan and then carry out extremely complex engineering tasks that require thousands of decisions to be made.

Human users simply outline the project in natural language prompts via a chatbot-style interface, and Devin does everything asked, according to the startup. It begins by creating a detailed, step-by-step plan to complete the assigned task and then gets started using its developer tools, just as a human coder would do, albeit much faster. It can write its own code, fix issues, test and report on its progress in real time, so users are always kept informed about its progress.

Should human users decide that something looks off, they can add another prompt into the chatbox and ask the AI to fix the problem. In this way, engineering teams will be able to delegate certain projects to Devin and focus their energy on more creative tasks for which human intelligence is still better suited.

Vance explained that he asked Devin to create a basic Pong-style game and create a website from scratch, and it completed those tasks in less than 20 minutes. It can also handle much more complex tasks, though those might take longer to complete.

Holger Mueller of Constellation Research Inc. said it was only a matter of time before generative AI crossed the chasm from merely assisting humans to taking over completely, albeit with human supervision. “Cognition has become one of the first to do this with Devin, which looks like an extremely promising application,” Mueller said. “We are now moving into the new phase of AI that can enable automated software operations, where software is essentially able to write itself.”

It’s somewhat ironic that Wu, a software engineer himself, is launching a generative AI tool that is ultimately designed to replace human software engineers. Computer scientist Silas Alberti, who also had early access to Devin, told Bloomberg: “It’s less like an assistant helping with code and more like a real worker doing its own thing.”

As the AI era marches on, the “learn to code” slogan that was once suggested as an alternative to humans who lose their jobs to AI is looking more outdated than ever. Devin’s creators believe it will eventually be able to perform many low-level coding jobs instead of human coders – and do them much more quickly.

Cognition now has funding in its pocket too, with the startup recently closing on a $21 million Series A raised led by Founders Fund. It’s now looking to expand capacity and extend early access to more select users, and encourages companies that want to explore its capabilities to apply via email.

Image: kenshinstock/Freepik