BRK279 Breakout Tue 17 Nov 15:15 PST
Building a model for software development from the ground up
Yanan Cai, Yang Liu, Faith Xu
Session description
Microsoft's catalogue description
Benchmarks alone can’t predict how a model will perform in a real product. Follow Microsoft AI’s journey building a model for coding and agentic software development, with GitHub Copilot as its proving ground. Explore how model design, training, real-world workflows, and product signals shape iteration across quality, latency, cost, and control—and leave with practical lessons for building specialized production models.