FAQ for agents powered by the Microsoft Copilot harness

This article answers common questions about agents powered by the Microsoft Copilot harness in Microsoft Copilot Studio, including how they relate to agents powered by the standard harness.

Note

The harness used to build this agent uses usage-based billing. Using, building, testing, and evaluating agents might consume Copilot Credits. Learn more in Overview of usage-based billing.

What exactly is the Microsoft Copilot harness?

It's a redesigned way to build agents in Microsoft Copilot Studio, a single consolidated build surface, and an enhanced orchestration runtime that improves reasoning and response quality. Learn more in Agents overview.

What's different about the underlying orchestration that makes the Microsoft Copilot harness more reliable?

The Microsoft Copilot harness provides deeper reasoning and improved response quality compared to the standard harness, particularly when an agent reasons over Microsoft 365 data. Learn more in Choose a harness.

What are skills, and how are they different from tools?

Skills are reusable, task-specific capabilities defined by a name, description, and instructions written in Markdown. The Microsoft Copilot harness activates a skill when a user's request matches its purpose. Tools, by contrast, connect your agent to external services and actions, such as connectors, MCP servers, and REST APIs. Learn more in Skills overview for agents.

What happened to prompts from the agents powered by the standard harness?

The Microsoft Copilot harness doesn't have a separate prompts feature. Reusable, structured behavior is instead packaged as a skill, which your agent can activate when a user's request matches its purpose. For a fully separate, specialized agent, use a connected agent.

What happened to child agents in agents powered by the Microsoft Copilot harness?

Agents powered by the Microsoft Copilot harness don't include the child agent concept. For a specific capability that lives inside your agent, use a skill. For a fully separate, specialized agent, use a connected agent.

For a focused assignment within a request, your agent can also use task agents, where available. Task agents are distinct from maker-authored child agents and don't require you to create and publish a separate agent.

Can I connect agents built with Microsoft Foundry, the Microsoft 365 Agents SDK, or the A2A protocol?

You can use Microsoft Foundry, Microsoft 365 Agents SDK, and A2A protocol agents as connected agents in the Microsoft Copilot harness. To learn how to connect to these external agent types in the Microsoft Copilot harness environment, see:

Why did my settings and tabs move, and where did they go?

The configuration is more consolidated compared to the standard harness. Knowledge, tools, channels, and connected agents now appear as components in the Build tab. Agent-level configuration that used to live on separate pages is grouped into the Agent settings dialog, organized into four tabs: Agent details, AI & behavior, Safety & access, and Greeting & prompts. Learn more in Settings.

Can agents built in the standard harness communicate with agents powered by the Microsoft Copilot harness?

Not directly. Connected agents in the Microsoft Copilot harness can currently only delegate to other agents that were also built in the Microsoft Copilot harness. There's currently no supported way to connect a standard harness agent and a Microsoft Copilot harness agent so they can invoke each other.

What should I do with my existing production agents?

Nothing. Your agents powered by the standard harness keep running as they do today. The Microsoft Copilot harness doesn't change or interrupt any agent you already published using the standard harness. When you're ready to take advantage of the Microsoft Copilot harness, build your next agent there while your standard harness production agents continue to run.

If you would like to upgrade your standard harness agent to a Microsoft Copilot harness agent, learn more in Upgrade agents to the Microsoft Copilot harness.

Where are my topics, and how do I migrate them to the Microsoft Copilot harness?

Topics aren't carried over as topics in the Microsoft Copilot harness. Instead, custom topics can be rebuilt as skills, or workflows during the upgrade process.

You can ask Copilot Studio to upgrade an existing standard harness agent. The upgrade creates a copy on the Microsoft Copilot harness and carries over instructions, tools, and suggested prompts. It also provides a report of what was migrated, what requires attention, and what was skipped. Knowledge sources, Power Fx expressions, and child agents aren't migrated. The upgrade consumes credits.

Learn more in Upgrade agents to the Microsoft Copilot harness.

What features are available in the Microsoft Copilot harness today (for example, telephony or other channels)?

The Microsoft Copilot harness currently supports a smaller set of publishing channels than the standard harness. For example, Microsoft Copilot, Microsoft Teams, the demo website, and web app embedding are available. For a complete list of supported channels, see Available channels for agents.

How do I access the standard harness?

There are multiple ways to access and build agents and agent flows that use the standard harness. Learn more in Build and access standard harness agents and agent flows.

When will the standard harness be deprecated?

The standard harness remains fully supported alongside the Microsoft Copilot harness, and there's no forced switch.

Can I turn off the Microsoft Copilot harness and keep using the standard harness? How can I govern it?

There's no setting that turns the Microsoft Copilot harness on or off. Admins govern it with four kinds of control: who can build agents, how much an agent can spend, what data and connectors it can reach, and who it can be shared with.

  • Who can build agents: Dataverse security roles in each environment
  • Data and connectors: Data policies and advanced connector policies
  • Who agents are shared with: Managed environment sharing controls and environment security groups
  • How much an agent can spend (next section)
  • Monitor and respond: Usage reports, agent inventory, block or quarantine an agent

How can I control how much an agent powered by the Microsoft Copilot harness spends?

You can control how much credit an agent powered by the Microsoft Copilot harness spends.

In the Power Platform admin center, the following cost controls are available:

  • Set environment credit allocations: Reserve credits for an environment. This control applies only to prepaid capacity packs.
  • Set agent limits: Set limits on how much an agent can spend. An agent must consume some credits before you can set limits.
  • Disable draw from tenant pool at scale: Block a set of environments from drawing credits from the tenant. Use this control to stop some environments from overconsuming from the tenant pool.
  • Copilot credit estimator: Use the Copilot Credit Estimator.

Learn more in Manage costs for agents powered by the Microsoft Copilot harness. Learn about upcoming controls in the roadmap.

Can I configure Copilot Studio to always create agents in the classic experience?

No. The new Copilot Studio experience gives you the option to create agents powered by one of the available harnesses.