Microsoft has released v1.0 of Microsoft Agent Framework and made the renamed 'AI toolkit for VS Code' (now 'Microsoft Agent Framework') generally available. The release covers a full developer journey for building and deploying AI agents: local development with VS Code tooling, multi-agent orchestration with Agent harness patterns (local shell, hosted shell, context compaction), managed long-term memory in the new 'memory' feature in the Azure AI foundry agent service, a unified 'toolbox' for connecting agents to tools, hosted agent execution with session isolation and sub-100ms startup, full observability via OpenTelemetry-based tracing and an AI Red teaming agent, and one-click publishing to Microsoft Teams and Microsoft 365 copilot.
Table of contents
Step 1: Build Locally with Microsoft Agent Framework + Foundry Toolkit for VS Code Copy linkStep 2: Build Agents That Actually Do Things — Agent Harness & Multi-Agent Composition (Public Preview) Copy linkStep 3: Make Agents Stateful with memory in Foundry Agent Service (Public Preview) Copy linkStep 4: Give Your Agents the Right Tools – Toolbox in Foundry (Public Preview) Copy linkStep 5: Host and Manage Agents at Scale — Hosted Agents in Foundry Agent Service (Public Preview) Copy linkStep 6: Know What Your Agents Are Doing — Observability in Foundry Control Plane Copy linkStep 7: Put Agents in the Hands of Users – Publishing to Microsoft 365 (Public Preview) Copy linkSort: