Ship faster with Claude Code and Cowork in Microsoft Foundry | DEMSP388

Caroline Matthews demonstrates how long-running coding agents can be built and guided in Azure AI Foundry using Claude Code and Cowork, focusing on practical patterns like evaluation-first workflows, safe autonomous execution, and multi-agent loops that can plan, test, and recover across larger codebases.

Overview

This Microsoft Build 2026 session covers practical “agentic coding” workflows inside Microsoft Foundry, showing how Claude can take on longer tasks across a codebase (reading existing code, planning across files, running tests, and recovering from failures).

Key themes highlighted in the session description and chapter list include:

• How the Claude Code harness works in practice inside Microsoft Foundry
• Context management approaches for long-running tasks
• Patterns engineers use to keep agents on track, including:
  • Building evaluation systems before coding
  • Using generator + reviewer agent setups (example: a simple “bill split” function)
  • Treating an evaluator as the judge/tester
  • Using loops, scheduled tasks, and parallel agent execution
• A comparison of traditional operation vs Auto Mode for safer autonomous execution
• A demo scenario where a “picker” agent selects a design option and other agents test/fix issues autonomously

Session metadata

• Event: Microsoft Build 2026
• Session code: DEMSP388
• Level: Intermediate
• Format: Demo
• Speaker: Caroline Matthews