How Microsoft AI builds coding models optimized for GitHub Copilot | LIVE158
Seth Juarez, Pierce Boggan, Yang Liu, and Pengcheng He walk through how Microsoft AI builds and optimizes code-focused models for GitHub Copilot, covering training, evaluation, performance, safety, and how real-world developer feedback shapes the Copilot experience.
Overview
This session goes behind the scenes on how Microsoft AI develops coding models optimized for GitHub Copilot. It focuses on what differentiates code-focused models, how they are trained and evaluated, and how performance and safety considerations translate into GitHub Copilot experiences in VS Code.
What the session covers
- How code-focused models differ from general-purpose models
- Training and evaluation approaches for coding workflows
- Reinforcement learning informed by real developer workflows
- A multi-stage training pipeline and environment alignment
- Model architecture considerations, including mixture-of-experts (MoE)
- How model quality improvements surface in GitHub Copilot experiences
Resources
Session context
- Microsoft Build 2026 session (LIVE158)
- Speakers: Seth Juarez, Pierce Boggan, Yang Liu, Pengcheng He
Chapters (from the video description)
- 0:00 - Introduction and Microsoft Build AI Model Overview
- 00:00:28 - Introducing the Mai Code Flash Team and Project Goals
- 00:01:00 - Purpose-Built Model for GitHub Copilot and VS Code
- 00:02:48 - Reinforcement Learning for Real Developer Workflows
- 00:03:50 - Multi-Stage Training Pipeline and Environment Alignment
- 00:05:27 - Model Size and Mixture-of-Experts Architecture Explained
- 00:10:20 - Demonstration of Mai Code Flash in VS Code
- 00:13:06 - Interactive AI Agents Demo Combining Multiple Mai Models
- 00:15:16 - Developer Takeaways and Future Model Enhancements
- 00:16:07 - Closing Remarks and Call for User Feedback