From Code Completion to Autonomous Software Engineering Agents

Explore the evolution of AI from simple code completion tools to sophisticated autonomous software engineering agents in this 41-minute conference talk. Learn about SWE-bench, a groundbreaking benchmark built from real GitHub issues that has become the industry standard for evaluating AI's ability to resolve complex software tasks in large codebases. Discover the current state of agentic AI in software engineering, understand the limitations of today's models, and gain insights into how far the field still needs to progress toward truly autonomous AI developers. Engage with hands-on demonstrations of SWE-agent, a simple yet powerful agent framework specifically designed for software engineering tasks but adaptable across various domains. Examine the fundamental principles behind AI agents and their practical applications in real-world software development scenarios. By the conclusion, develop a comprehensive understanding of the current frontier of agentic AI in software engineering, recognize the challenges that lie ahead, and learn how to experiment with AI agents in your own development workflows.