Skip to content

Pick the engine for the job

When you build an agent in Copilot Studio, one choice shapes everything else: which harness powers it. A harness is the engine underneath your agent — it decides how the agent reasons, what it can touch, where you can publish it, and how it bills. Same builder, different engines, and the right one depends on the job.

Unofficial — confirm before you commit

This is community guidance, not a Microsoft prescription. Harness capabilities, packaging, and billing evolve quickly. Use this to start the conversation, then confirm against Licensing & Prerequisites and Microsoft's Copilot Studio documentation.

Where it's heading — and the honest tradeoff

Microsoft is investing most in the GitHub Copilot harness, and new agent capability tends to land there first. Reach for it when the job is a complex, multi-step process a rules script can't handle gracefully: it reasons through the whole goal on its own, retries and finds another path when a step fails, works directly across Word, Excel, PowerPoint, and PDF files, and carries context with skills and memory. That capability is why it's the engine to grow into. Weigh two facts with equal honesty: it is the only harness a Microsoft 365 Copilot license does not cover, and it bills Copilot Credits for all usage — creation and runtime. The standard harness is covered for Microsoft 365 Copilot-licensed employees and costs less for predictable, rules-based work. Pick for the job and the budget — the Credit Estimator models the real net cost of each.

Updated October 2026 — what changed in the harness list

Microsoft's harnesses overview now lists GitHub Copilot, Customer Conversation (new — customer-facing voice and chat with hand-off to a human), and standard. The "Copilot chat" option earlier versions of this page described is no longer listed as a Copilot Studio harness; to extend Microsoft 365 Copilot Chat with your own knowledge, start with Agent Builder.

Pick the engine for the job

Every harness reasons, publishes, and bills differently. Click any card to see how it works and where it shines — or jump to the side-by-side.

Side by side

The same considerations, compared across all three harnesses. Find the row that matters most to your job.

Considerations GitHub Copilot harness Standard harness Customer Conversation harness
Best for Complex, multi-step business processes Rule-based agents and structured conversations Customer-facing, contact-center-style scenarios over voice and chat
How it works Reasons through a goal on its own, step by step Follows the topics and rules you define Multi-agent architecture that can escalate to a human representative
Recovers from problems Retries and finds alternative paths automatically Follows the paths you've built Escalates to a human representative through a defined path
Works with files Creates, edits, and reasons over Word, Excel, PowerPoint, and PDF files Not a focus Not a focus
Skills and memory Yes Not a focus Yes
Publishing Internal teams or external customers Internal teams or external customers Website chat or a phone number
Covered by your M365 Copilot license? No — credits for all usage Yes — for licensed employees No — usage-based credits
Billing Copilot Credits See Copilot Studio licensing Copilot Credits (usage-based)

Two things to take away

  1. The harness is a capability choice, not just a setting. Autonomy, file-handling, and where you can publish are all downstream of which engine you pick — so choose for the job, not the default.
  2. They bill — and are covered — differently. Copilot Credits, Copilot Studio licensing, and M365 Copilot USLs are not the same meter, and only the standard harness is covered by a Microsoft 365 Copilot license (for licensed employees) — the GitHub Copilot and Customer Conversation harnesses never are. Sort out cost and coverage before you scale, not after — see Licensing & Prerequisites and the Credit Estimator.

📚 Learn more.

Was this page helpful?

Found an issue or want to suggest an improvement? Open a GitHub issue  ·  Star on GitHub