Microsoft has opened its Copilot harness to developers and enterprise customers through Microsoft Foundry, sharing the multi-model agent engine that already powers Copilot Code, Copilot Studio, the GitHub Copilot app, the Copilot command line and Visual Studio Code. Announced on October 28, 2026, the move lets outside teams build agents on the same orchestration layer Microsoft uses in its own products, instead of building and maintaining a harness of their own.
What A Harness Actually Does
If the model supplies the intelligence, Microsoft says, the harness is everything around it. It is the orchestration layer that gives an agent the right instructions, context, execution environment and controls to do meaningful work. A good harness also supports open standards such as the Model Context Protocol, along with agent skills and agent plugins, so an agent can call tools and follow procedures rather than simply answering a prompt.
That surrounding machinery is the hard part, and it is getting harder as models change. A frontier model that leads today can fall behind tomorrow, and a team that built its own router or agent loop must keep evaluating new models, updating integrations and fixing the plumbing. Microsoft’s pitch is that customers can now lean on its harness instead, and let Microsoft carry that maintenance burden while they focus on what their agent should know and do.
The harness is built to work across model families, with support for more than 20 frontier models plus bring-your-own-key options. Instead of locking a project to one model, it routes work to the right model or models for the task at hand. Microsoft also says it recently ported the harness runtime to Rust, cutting session overhead and time to first token, a detail that matters when agents run many short steps in a row.
Born In GitHub, Tested At Scale
The harness began with GitHub’s engineering team, which built it to power its own multi-model orchestration work. Microsoft says it adopted the harness company-wide because it had the highest satisfaction among its most demanding audience: developers. Because the same harness now runs across Microsoft’s Copilot surfaces, the company argues, it is battle-tested at the scale of millions of users and hundreds of real-world use cases.
What Builders Get In Foundry
In Foundry Agent Service, the Copilot harness arrives in public preview. It gives developers a fully managed way to build agents that reason, use tools and complete complex, long-running tasks. Microsoft runs the agent loop and sandbox, with governance and optimisation built in from the start, while the customer supplies the instructions, skills, tools and model choice.
The preview bundles four building blocks. Reusable skills package instructions, scripts and supporting resources into procedures an agent can apply across tasks. Connected tools link supported Foundry and Model Context Protocol tools, with explicitly configured authentication and approvals. Execution and delegation let agents work with files and code and hand bounded tasks to built-in subagents, while the main agent coordinates the result. And source-controlled agent definitions let teams create and test in Foundry, or manage definitions alongside application code.
Alongside the harness preview, Microsoft announced general availability of long-running resilience for hosted agents. Foundry detects interruptions and restarts the work with its original input, while the application saves progress and intermediate results so the agent can pick up where it left off. For work that takes hours or days, that recovery matters as much as the model itself, though Microsoft warns that conversation history alone is not enough to restore work in progress.
There are preview limits to note. During the preview, teams select the Copilot harness when creating an agent; existing agents are unchanged, and switching the harness after creation is not yet supported. Foundry manages the execution environment’s processor, memory and idle settings. Billing for the harness in Foundry Agent Service begins on December 1, 2026, with consumption-based pricing; harness runtime use is free before that date, while model inference and separately billed tools remain charged as usual.
Why It Matters
The harness matters because the agent race is shifting from who has the best model to who has the best system around the model. Models are becoming interchangeable parts, and the differentiator is the machinery that picks the right model, manages context, runs tools safely and recovers when something breaks. By opening its own machinery, Microsoft is betting that developers will choose the managed route over building that machinery themselves, just as they once chose managed databases over running their own.
It also sharpens a familiar trade-off. A managed harness saves engineering time and brings Microsoft’s improvements automatically, but it ties an agent’s behaviour to Microsoft’s roadmap and preview limits. Teams that need full control over the agent loop can still use hosted agents with their own framework in Foundry, and that choice, between convenience and control, is now the real decision for agent builders. The same tension runs through the wider agent market, from Mistral’s new Responses API for agent builders to Gemini running ServiceNow workflows for enterprises.
Background
Microsoft Foundry is the company’s platform for building and running AI applications, spanning models, agent services and developer tooling. The Copilot harness grew out of GitHub’s work on multi-model coding agents, and its release through Foundry extends that engineering from developer tools into general enterprise agents.
Source: Microsoft announcement, Foundry details
0 Comments