We've reached a small but exciting milestone for OpenDevin, the open source AI software engineer -- OpenDevin sends a pull request to the OpenDevin repo. You can see the PR here:

To be clear on the level of autonomy, I sent a prompt with the URL of the issue to resolve. OpenDevin browsed to the issue, cloned the repo, fixed the issue, and pushed to GitHub. Then I copy-pasted the PR URL, checked the work, wrote the PR description, and hit submit.

This is exciting because it demonstrates the potential for an open source agent to do end-to-end software dev tasks, but we still have a lot of work to do: we are almost done setting up benchmarking, and based on this we will bring agents implemented in our agent hub up to SOTA.

I strongly believe that AI agents are going to be the future, and the future should be built by everyone! If that resonates with you please come join us on github, slack, or discord:

We are trying to build a community where AI researchers, software developers, and even users without as much development experience would be able to join and contribute!

And if you would like to see some things we have planned for the future, we have our open road map here: If there is anything else you would like to see on there just send us a message!

Here I share an XGBoost model that delivers a 25% CAGR with minimal drawdown on Visa stock. In this free Substack post I share code and commentary for a powerful Machine Learning strategy that delivers powerful returns.

Did you double checked the changes it did to the code? Because I can see this as the starting plot of a dystopian movie.

Yes! If you see the last part of the video, and also my explanation in the following tweet, I did check the code and press submit myself. In addition, we're doing standard peer review for OpenDevin, so another developer needs to check as well.

The mysterious gpt2-chatbot can really replace software developers at large scale. Petrified after testing

I'm wondering why you have not benchmarked SWE Bench? Is Devin not similar to SWE Agent?

So the main reason is that we had some trouble getting swe-bench to work properly and it took us a while to set up. But as of today we're mostly finished and ready to benchmark!