SWE-agent by Princeton NLP connects an LLM to a computer interface to automatically fix bugs and implement features from GitHub issues. It can solve real-world software engineering problems with minimal human input. A landmark research project demonstrating autonomous software engineering capability.