Microsoft Copilot Harness: 6 Moves Before December 1

TL;DR: The GitHub Copilot harness in Copilot Studio now has a Microsoft-wide name: the Microsoft Copilot harness. It is the same engine behind Copilot Code and the GitHub Copilot app, and it is now in Microsoft Foundry. The name changes, Copilot Credits billing doesn't, and six governance moves belong before December 1.

The engine under your most capable Copilot Studio agents just got a new name, and two billing changes around it land on December 1. On October 28, 2026, Microsoft introduced the Microsoft Copilot harness, the GitHub-built harness that already powers Copilot Studio, Copilot Code, the GitHub Copilot app, the Copilot CLI and Visual Studio Code, and opened it to developers in Microsoft Foundry.

The Microsoft Copilot harness is the orchestration layer around an AI model: it gives an agent its instructions, context, tools, sandbox and controls, and decides what happens next. In Copilot Studio it is the harness that Microsoft Learn still calls the GitHub Copilot harness, built for reasoning-heavy, multi-step work and billed on Copilot Credits.

Our view that Copilot Studio harness choice is a design decision, not a toggle still holds. What changes is the scope: the harness is now Microsoft's standard engine for agentic work, and it deserves one governance model.

What happened to the GitHub Copilot harness in Copilot Studio?

The GitHub Copilot harness in Copilot Studio is now part of one Microsoft-wide harness with a new name, the Microsoft Copilot harness. Microsoft says the GitHub engineering team built it, Microsoft adopted it company-wide, and it is standardizing its Copilot experiences on it.

Three details matter for IT leaders:

• It is multi-model by design. Microsoft says the harness works across more than 20 frontier models plus bring your own key, and can route work to the right model for the task.
• Microsoft runs the agent loop and sandbox. You bring the instructions, skills, tools and model choice, with support for open standards such as Model Context Protocol (MCP) and agent skills.
• Documentation will lag. At the time of writing, Copilot Studio pages on Microsoft Learn still say GitHub Copilot harness, so expect both names for a while.

The new name also fixes a real problem. In 2toLead's steering committee conversations, the word GitHub led finance teams to ask whether every maker needed a GitHub Copilot seat. They don't. Microsoft states that the harness in Copilot Studio is not the GitHub Copilot service, and that customer data isn't sent to or processed by that service when agents run in Copilot Studio.

What is changing with the Microsoft Copilot harness?

The name and the reach of the harness change now, while Copilot Studio billing and your existing agents stay as they are. Here is each change, when it lands and what it means for you.

Does the new name change Copilot Studio billing or licensing?

No. Agents and workflows on the Microsoft Copilot harness in Copilot Studio still use Copilot Credits, and that usage isn't included with a Microsoft 365 Copilot license.

The billing rules Microsoft set in August still apply:

• Billing starts when you start building. Creating an agent with natural language, previewing, testing and generating evaluations all consume credits. The standard harness, by contrast, starts billing after publish.
• Credits cover the full experience. That includes model tokens, tools such as knowledge sources and MCP servers, and the harness itself.
• Harness choice is permanent. You choose at creation, and an agent can't be transferred between the standard harness and this one.

What does change on December 1 is the default around it. New Microsoft 365 Copilot Business licenses bought through Cloud Solution Provider partners arrive with usage-based billing switched on and a default spending limit of 4,000 Copilot Credits per user per month. Partner Center lists the harness alongside Copilot Cowork and Work IQ APIs as services that policy covers. Our view on who should own the Copilot Credits meter covers the decision, and our December 2026 Copilot licensing guide covers the dates.

Why does one harness across Microsoft matter for IT leaders?

One harness means one engine, one cost model and one set of controls across Copilot Studio, Copilot Code, the GitHub Copilot app and Foundry. Improvements Microsoft makes reach every surface the harness powers, and so do the governance questions.

For CIOs, that turns a Copilot Studio governance problem into a platform one. Low-code makers in Copilot Studio, app builders in Copilot Code and pro developers in Foundry can now reach the same reasoning engine. If each group gets its own budget, approval path and definition of a good agent, you pay three times to learn the same lessons. Our comparison of Copilot Chat, Cowork, Code and Autopilot maps which experience does what.

• Nonprofit and education: Partner Center scopes the December 1 default to commercial purchases and excludes education, nonprofit and GCC low. You will switch usage-based billing on deliberately, which is the right moment to set limits before the first agent runs.
• Government: Confirm the Foundry preview is available in your cloud before you plan a pilot.
• Healthcare and financial services: Add Microsoft's data-handling statement to your privacy impact assessments, so reviewers don't stall an agent because of the GitHub name.

Where does the old harness name still live in your organization?

It lives anywhere your team wrote GitHub Copilot harness since August: policies, queries, budgets, training and risk reviews. Use this checklist to find each one, update it and give it an owner.

One item deserves care. Microsoft's governance guidance filters Power Platform inventory on a harness value of GitHub Copilot. Microsoft hasn't said whether that value will change, so build reports that match both names and treat an empty result as a warning, not good news.

What should IT leaders do before December 1?

Make six moves in the next five weeks: find every harness agent, set environment boundaries, cap each agent, review spending policies, update the language and use the free Foundry window on purpose. Together they turn a rename into a governance reset.

1. Find every agent on the harness. Use Power Platform inventory and its count-by-harness sample query, or Agent Inventory V2 in the Copilot Agent Kit, which already covers these agents. Our breakdown of the Copilot Agent Kit September 2026 release explains what it finds.
2. Classify each environment by purpose. Label it exploration and prototyping, test and validation, or funded production. For a hard prepaid boundary, allocate a fixed amount and turn off Draw from the available capacity in my tenant.
3. Cap every exploration agent. The Power Platform admin center has no tenant-wide default agent limit, so apply your development limit to each new agent through a recurring or automated check. Keep the add-on capacity assignment setting with tenant admins, because opening it to environment admins gives them allocation control across every environment.
4. Review spending policies in the Microsoft 365 admin center. Under Copilot > Cost Management, give every policy a business owner and decide whether Auto-apply new services should stay on. It is enabled by default, so new services inherit the policy.
5. Update the language and publish one harness standard. Work through the checklist above, then publish a one-page standard with approved use cases per harness, the approver and the credit threshold that triggers review. Our guide to what an Agent PMO looks like in practice shows who should own it.
6. Use the free Foundry window on purpose. Harness runtime in Foundry Agent Service is free until December 1, although model inference still bills. If pro developers want to test, give them one scoped pilot with a named owner and an end date now, rather than finding five ungoverned ones in January.

How do you measure the value of a Microsoft Copilot harness agent?

Measure hours returned against credits consumed, per agent, per month. Because the harness bills from the first build step, the value case has to cover building, testing and running, not just production use.

Here is an illustrative worked example based on Microsoft's own accounts payable scenario; replace the figures with yours. A manufacturer processes 2,000 supplier invoices a month, and 12% become exceptions, so 240 a month. An accounts payable specialist spends 25 minutes on each one, finding the purchase order, chasing the buyer and logging the outcome. That is 100 hours a month.

A Microsoft Copilot harness agent that reads each invoice, matches it to the purchase order and drafts the exception for approval cuts the human part to 8 minutes. That is 32 hours, so the team gets back 68 hours a month. At a $55 loaded hourly cost, that is $3,740 a month, or about $44,900 a year.

Now set the boundary. Cap the agent's monthly run cost at no more than a third of that value, about $1,250, converted to Copilot Credits at your contracted rate. Budget the build separately. If makers spend 80 hours at $75 an hour building and testing, the net monthly return of about $2,490 pays back that $6,000 in about two and a half months, before build credits.

After 30 days, compare the credits shown on the agent's Monitor page in Copilot Studio with the hours actually saved, and adjust the limit. Report the result with the Copilot adoption metrics that prove value, so finance sees one consistent story.

A new name doesn't change the meter. It changes how many people can reach it.

What do IT leaders ask about the Microsoft Copilot harness?

Is the Microsoft Copilot harness the same as GitHub Copilot?

No. It is the orchestration engine GitHub built and Microsoft now uses across its Copilot experiences, not the GitHub Copilot developer subscription. Copilot Studio makers don't need a GitHub Copilot seat, and Microsoft states customer data isn't sent to the GitHub Copilot service when agents run in Copilot Studio.

Do existing Copilot Studio agents have to change?

No. Standard harness agents remain fully supported with no forced switch. Agents can't be transferred between harnesses, so moving one to the Microsoft Copilot harness is a rebuild that should be justified by a clear benefit, such as better quality or long-running work.

How is the Microsoft Copilot harness billed in Copilot Studio?

It is billed in Copilot Credits from the moment makers start building. Creating, previewing, testing, evaluating and running agents all consume credits, covering model tokens, tools and the harness itself. This usage isn't included with Microsoft 365 Copilot licenses.

Can developers use the Microsoft Copilot harness outside Copilot Studio?

Yes. Since October 28, 2026, it has been in public preview in Microsoft Foundry Agent Service. Developers select it when they create an agent, and harness runtime billing in Foundry starts on December 1, 2026.

How can 2toLead help?

2toLead helps organizations in healthcare, nonprofit, education, financial services, professional services, government and manufacturing govern Copilot agents and prove their value. In a fixed-scope, two-week Harness Governance Sprint, our Microsoft MVP team will:

• Inventory every Microsoft Copilot harness agent and the environments that hold it
• Classify environments and set capacity boundaries and agent-level limits
• Give every Copilot Credit spending policy a business owner before December 1
• Update your harness standard, intake form and maker training with the new name
• Build a value scorecard for your top three agents in hours and dollars

For the source, read Microsoft's announcement of the Microsoft Copilot harness.

For everything else landing before March, our Microsoft 365 quarterly newsletter for November 2026 maps each deadline to an owner.

Find more practical guidance on the 2toLead Insights hub, or book a Harness Governance Sprint.

No hype. Just practical, governed AI that actually gets adopted.