Microsoft launches Copilot Harness in Foundry Agent Service for multi-model orchestration

What happened.Microsoft made its Copilot harness generally available in Foundry Agent Service, allowing developers to orchestrate multi-model autonomous agents across OpenAI, Anthropic, and open-source models.
Why it matters.Building custom agent execution harnesses forces engineering teams to maintain fragile scaffolding around context windows, tool retries, and session memory. Adopting a standardized managed harness allows teams to upgrade underlying frontier models instantly as price and intelligence frontiers advance.
The technical specifications:
- Managed execution runtime: The managed harness decouples model reasoning from operational orchestration, automating context tracking, subagent delegation, and tool execution through Model Context Protocol.
- Multi-model portability: Developers can switch underlying foundation models across disparate providers without rewriting business logic, session state, or tool integrations.
- Enterprise governance: Built-in Microsoft Entra identity isolation protects runtime credentials, while the Foundry Agent Optimizer continuously tunes prompt instructions against production telemetry.
STORYLINE PROGRESSION & CONTEXT
This report represents a verified evolution of prior newsroom intelligence: