Copilot Studio went generally available with the GitHub Copilot harness on August 3, 2026. Along with it came a term many of us are seeing for the first time, Harness.
In this post lets understand what a harness is and how many types Copilot Studio offers. We will also cover how Skills replace Topics, the licensing, and the gotchas worth knowing before you build.
What is a ‘Harness’?
Microsoft defines a harness as the software scaffolding that lets an agent plan, reason, maintain context, and interact with tools.
- It sits between the agent you design and the AI model that does the reasoning.
- It decides when to call the model and which components to send.
- It calls the right tool next.
Before we compare the different harnesses of ‘Copilot Studio’, lets make sure you have what is needed.
Prerequisites
- Access to Copilot Studio through any one of the following. A Copilot Studio user license, the Copilot Studio authors role, a Microsoft 365 Copilot license, or a trial license.
- Copilot Credits allocated to your environment.
Note: The trial license lets you create and test agents. You cannot publish them.
Lets now look at the three harnesses.
The three Harnesses
Copilot Studio supports three harnesses. Each targets a different kind of scenario.
GitHub Copilot harness
- The most capable of the three, now generally available.
- Takes a goal, breaks it into steps, and adjusts when a step fails.
- Calls tools across connectors, knowledge, MCP servers, and connected agents.
- Creates and edits Word, Excel, PowerPoint, and PDF files natively.
- Supports skills and memory.
- Runs each task in a secure sandbox governed by Copilot Studio.
- Example: an accounts payable agent that reads invoices, matches them to purchase orders, and routes exceptions for approval.
Standard harness
- Powers rule-based agents and agent flows.
- You author the topics, prompts, and paths, so behavior stays predictable.
- Example: an IT onboarding agent that routes laptop requests through approval and provisioning.
Copilot chat harness
- Built for extending Microsoft 365 Copilot Chat.
- Connects your enterprise knowledge so employees get grounded answers in their everyday apps.
- Publishes to internal teams only.
- Example: an employee onboarding agent that answers questions from SharePoint content.
Here is the side by side view so you can pick quickly.
| Consideration | GitHub Copilot harness | Standard harness | Copilot chat harness |
|---|---|---|---|
| Best for | Complex, multi-step business processes | Rule-based agents and structured conversations | Extending M365 Copilot Chat |
| How it works | Reasons through a goal step by step | Follows the topics and rules you define | Connects enterprise knowledge to M365 Copilot Chat |
| Recovers from failures | Retries and finds alternative paths | Follows only the paths you built | Not a focus |
| Works with files | Word, Excel, PowerPoint, PDF | Not a focus | Not a focus |
| Skills and memory | Yes | Not a focus | Not a focus |
| Publishing | Internal teams or external customers | Internal teams or external customers | Internal teams only |
| Billing | Copilot Credits, usage based | Copilot Credits | Consumption or included in M365 Copilot license |
Now lets see where each harness shows up inside the product.
Where to find each Harness in Copilot Studio
Each harness has its own landing surface. Once you know what to look for, telling them apart is easy.
GitHub Copilot harness
- Open Copilot Studio.
- Turn the New experience toggle on, at the top right corner.
Standard harness
- Turn the New experience toggle off. You land on the
What would you like to build?page.
- Notice the New agent flow card. The agent flows run on the standard harness and not on the GitHub Copilot harness.
Copilot chat harness
- From the left navigation select Agents.
- Select Microsoft 365 Copilot from the list of agents.
- Open the Agents tab and click New agent.
Lets now check the harness of agents that already exist.
Checking the Harness of an existing agent
To understand the harness of an agent
- Select Agents from the left navigation.
- Look at the Powered by column in the list.
Now lets look at the biggest authoring change that comes with the new harness.
Skills replace Topics
- If you have built agents before, you know Topics. A topic is a conversation path with trigger phrases, nodes, and branching logic that you author by hand.
- The GitHub Copilot harness does not use topics. Microsoft states that instead of authoring explicit conversation topics, flows, and branching logic, you describe your agent in natural language.
Topics vs Skills
| Aspect | Topics (standard harness) | Skills (GitHub Copilot harness) |
|---|---|---|
| Authored as | Conversation nodes and branches in a designer | Markdown instructions |
| Activated by | Trigger phrases you define | The runtime, based on the skill description |
| Reuse across agents | Tied to the agent you built it in | Create once, add to multiple agents |
| Sharing | Inside the agent | Exported as a Markdown file or a package |
| Extra files | Not applicable | Scripts, templates, and reference documents |
Lets move to the licensing topic.
Do I need a GitHub Copilot license?
- No. The name
GitHub Copilot harnessrefers to the orchestration technology. It does not create a licensing dependency on GitHub. - What you need is access to Copilot Studio plus Copilot Credits, as listed in the Prerequisites section.
Lets look at the currency behind all of this.
Understanding Copilot Credits
Copilot Credits are the common currency across Copilot Studio, and they pool at the tenant level. Because they pool at the tenant, an admin must allocate them to individual environments.
For the GitHub Copilot harness a single credit pool pays for everything.
- The large language model tokens.
- The tools, which includes knowledge sources and MCP servers.
- The harness runtime itself.
There is no separate token meter to track. One currency covers the whole experience.
Gotchas worth knowing
Billing starts when you start building, not when you publish
- The Standard harness starts billing after you publish.
- The GitHub Copilot harness charges credits from the moment you begin building. Natural language authoring, previewing and testing, and generating evaluations all consume credits.
Agents cannot move between harnesses
- An agent created with the GitHub Copilot harness cannot be transferred to the standard harness. The reverse is also blocked. Choose carefully at creation time.
Existing agents are not migrated automatically
- Agents and workflows created before August 3 stay on the current pricing model until September 1. After that date they move to credit pricing. Standard harness agents remain fully supported.
Running out of credits stops everything
- When an environment has no allocated credits, or has used them all, credit-dependent experiences stop.
- End users see agents stop responding.
- Makers cannot author with natural language, preview, test, or generate evaluations.
Ways to check consumption
- Per agent: open the agent and go to the Monitor page in Copilot Studio.
- Per environment or tenant: go to Licensing > Copilot Studio in the Power Platform admin center.
- Before you build: use the Copilot Studio agent usage estimator to forecast volume.
Note: Since testing consumes credits on this harness, run the estimator before you start and watch the Monitor page while you iterate.
Summary
A harness is the runtime between your agent design and the AI model. Copilot Studio gives us three of them.
- GitHub Copilot harness for reasoning-heavy, multi-step automation.
- Standard harness for rule-based agents and agent flows.
- Copilot chat harness for extending Microsoft 365 Copilot Chat.
The key takeaway on authoring is that topics give way to skills. The key takeaway on licensing is that no GitHub Copilot license is required. You need Copilot Studio access plus Copilot Credits, and on this harness the meter starts running the moment you begin building.
For the official reference, see Choose a harness, Skills overview, Overview of usage-based billing, Enforcement policy and the Copilot Credits Guide for current pricing.
🙂
Leave a Reply