Microsoft’s Copilot harness turns agent orchestration into a cloud service
Foundry can now run the agent loop, sandbox, tools and subagents. For software teams, the question shifts from how to build an agent runtime to how much control they can safely rent.
What Microsoft shipped
Microsoft put the Copilot harness into Foundry Agent Service as a public preview. Foundry can now run the agent loop and isolated sandbox while teams supply instructions, reusable skills, approved tools and supported models; agents can also delegate bounded work to subagents.
Why this changes the build decision
Teams no longer have to build every layer around a model just to coordinate files, tools, context and multi-step work. A supported Foundry model can be changed without rewriting the surrounding orchestration, making managed infrastructure a credible alternative to a home-grown agent runtime.
The preview boundary matters
Microsoft says the preview has no service-level agreement and is not recommended for production workloads. The harness must be chosen when a new agent is created, existing agents cannot switch into it, and bring-your-own-model is not supported in Foundry today; consumption billing begins on December 1, 2026.
Virtual Arc would not move a core production agent onto this preview yet. Microsoft has removed a large amount of undifferentiated plumbing, but the dependency has simply shifted upward: from a model API to the harness that owns context, tool execution, sandbox behavior, approvals and recovery. That can be a good trade for teams already committed to Azure, but only after the workflow proves itself under real failure conditions. We would put one bounded, reversible process through a shadow run, keep tool contracts and state outside the harness, and compare cost per completed task, latency, retries, human intervention and audit quality against the current stack. If those numbers hold, rent the runtime and stop maintaining commodity orchestration. If they do not, the team should be able to leave without rebuilding its business logic. The mistake would be treating model choice as portability while letting the harness become the actual lock-in.
Turn insight into a working system.
We build a provider-neutral agent layer around your workflows, with scoped tools, audit trails, evaluations and shadow routing before any harness migration. If your team is deciding whether Foundry should replace custom orchestration, talk to us.