Cognition AI has introduced Devin, described as the world’s first fully autonomous AI software engineer.
Cognition AI was created by coding experts Scott Wu, Steven Hao, and Walden Yan, bolstered by a $21 million Series A funding led by Peter Thiel’s Founders Fund.
One of the founders, Scott Wu, described Devin as a “tireless, skilled teammate,” capable of working alongside humans or independently completing entire projects.
Devin is far more sophisticated than OpenAI and Microsoft’s CoPilot. Instead, it’s comparable to an AI agent that develops software from natural language prompts into a finished project rather than spitting out individual code segments.
Cognition AI released a series of video demos detailing Devin’s capabilities across a broad spectrum of software development and engineering tasks.
Some of Devin’s showcased applications from the Cognition AI demo:
- Adaptability to new technologies: After reviewing a blog post, Devin successfully executed ControlNet on Modal to generate images embedded with concealed messages, demonstrating its ability to creatively learn and adapt from inputs.
- End-to-end app development: Devin autonomously developed an interactive website simulating the Game of Life. Devin managed the entire project lifecycle, from incorporating features based on user requests to deploying the application on Netlify.
- Autonomous debugging: Another of Devin’s key features is its ability to identify and rectify bugs within codebases without human intervention. Cognition demonstrated this by showing Devin maintaining and debugging an open-source competitive programming book.
- AI model training: Pushing the boundaries of AI’s role in software engineering, Devin autonomously set up and fine-tuned a large language model (LLM), given only a link to a research repository on GitHub. Yes, this is an AI model capable of autonomously creating AI models.
Devin’s capabilities were rigorously evaluated on the SWE-bench coding benchmark. This challenging test asks agents to resolve real-world GitHub issues in open-source repositories.
Devin’s performance was remarkable, correctly resolving 13.86% of the issues end-to-end, a massive leap from the previous state-of-the-art, which only managed 1.96%.
Speaking to Bloomberg, Wu said, “Teaching AI to be a programmer is actually a very deep algorithmic problem that requires the system to make complex decisions and look a few steps into the future to decide what route it should pick.”
“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.”
This massive improvement demonstrates Devin’s advanced problem-solving abilities and its potential to enhance productivity and efficiency in software development.
Devin’s ability to autonomously execute complex software engineering tasks offers a glimpse into a future in which manual programming is all but extinct.
Nvidia CEO Jensen Huang touched on this at the World Government Summit in Dubai, telling the audience, “It is our job to create computing technology such that nobody has to program. And that the programming language is human, everybody in the world is now a programmer. This is the miracle of artificial intelligence.”
However, AI is also excellent at replacing creative jobs, so where do humans turn?
It’ll be about wielding AI when it’s effective while cultivating critical thinking and creative agility – and probably a fair dose of luck in your chosen career trajectory.
