Devin is a new AI model that aims to do the work of a Software Developer. Many AI assisstants can help developers write code by adding to existing code and writing isolated bits of code. But Devin does much more.
And it currently holds the record on the SWE-Bench coding benchmark.
What Devin can do
- Make a development plan
- Read API documentation online
- Write code
- Run code
- Read errors
- Add print statements to better understand the errors
- Learn from the print statements, make code changes, try again
- Deliver a progress report
- Fine tune itself with new knowledge
Devin can make plans, read documentation, run and debug code. It currently outperforms all other current AI models on the SWE-Bench coding benchmark which analyzes how well a model can resolve real-world GitHub issues.
Watch the AIGrid’s video for a full rundown including several demos.