Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the technical architecture and development process behind Cursor's Composer model in this 12-minute interview with Sasha Rush, one of the engineers who built this specialized AI coding system. Learn how the team designed Composer using reinforcement learning and mixture-of-experts architecture to achieve fast, agentic coding capabilities that developers can trust and iterate with in real-world scenarios. Discover why speed was identified as a core requirement for coding models and understand the architectural decisions that enable Composer's agent-based approach to software engineering tasks. Examine how reinforcement learning and mixture-of-experts techniques allow the model to specialize for practical coding applications, and gain insights into the distributed infrastructure and training methodologies used to develop Composer at scale, including the use of Ray for distributed computing. Understand how developers are currently using Composer in practice and explore predictions for the future of agent-based coding models and specialized AI systems in software development.
