Devin: The World's First AI Software Engineer Is Here 172

Cognition ai Labs has unveiled the world's first AI programmer: Devin. Devin is the world's first AI programmer capable of coding, creating websites, and software from a single prompt. It has successfully passed practical engineering interviews at leading artificial intelligence companies and even completed real work on Upwork. Devin is an autonomous agent that solves engineering tasks by using its own shell, code editor, and web browser. Its capabilities include the following:

1. Can learn to use unfamiliar technologies.

After reading a blog post, Devin ran ControlNet on Modal to generate an image with a hidden message for Sara.

2. Devin can build and deploy applications end-to-end.

Devin created an interactive website simulating the Game of Life! It incrementally added features requested by the user, then deployed the application to Netlify.

3. Devin can autonomously find and fix bugs in code repositories.

Devin helped Andrew maintain and debug his open-source competitive programming book.

4. Devin can train and fine-tune its own AI models.

Given only a link to a research repository on GitHub, Devin was able to fine-tune a large language model.

5. Devin can resolve bugs and feature requests in open-source repositories.

Simply provide a link to a GitHub issue, and Devin will complete all the necessary setup and context gathering.

6. Devin can contribute to mature production repositories.

This example is part of the SWE-bench benchmark. Devin resolved a bug in logarithmic calculations in the sympy Python algebra system. Devin set up the code environment, reproduced the bug, and coded and tested the fix on its own.

What sets Devin apart is its ability to think ahead and plan complex tasks. It can make thousands of decisions, learn from mistakes, and improve over time. In addition, it has all the tools a human engineer needs at its fingertips, such as a code editor and browser.

Core Features of Devin:

What exactly can Devin do?Pretty much anything you throw at it. Whether it's learning new technologies, building and deploying applications from start to finish, or finding and fixing those pesky bugs in code, Devin has you covered. It can even train its own AI models and resolve issues in open-source projects.

• Long-term reasoning and planning: Capable of executing engineering tasks that require complex decision-making.

• Using development tools: Mastery of developer tools such as shell, code editors, and browsers.

• Active collaboration: Able to report progress in real time, accept feedback, and make design choices together with users.

• Learning new technologies: Capable of quickly mastering and applying unfamiliar technologies.

• Building and deploying applications: Completes app construction and deployment from start to finish.

• Autonomously discovering and fixing bugs: Independently identifies and resolves issues in code.

• Training and fine-tuning AI models: Sets up and fine-tunes large language models.

• Handling open-source repository issues: Tackles bugs and feature requests in open-source projects.

• Contributing to mature production repositories: Resolves complex issues in production-grade repositories.

• Completing real-world tasks: Capable of completing real work on platforms such as Upwork.

In the SWE-Bench benchmark, agents are required to resolve real GitHub issues found in open-source projects such as Django and scikit-learn.Devin correctly resolved 13.86% of the issues without assistance, far surpassing the previous state-of-the-art of 1.96%. Even when given the exact files to edit, the previous best model could only resolve 4.80% of issues.

Devin represents a major leap forward in artificial intelligence technology. By automating routine tasks and enabling engineers to focus on more complex problems, it paves the way for a new era of software development innovation. So whether you're a seasoned engineer or just getting started, Devin will make your work easier and more exciting than ever.