A live, technical look under the hood of Copilot Studio's newest runtime, the GitHub Copilot harness, and what it actually changes when you build agents with it.

Copilot Studio now run agents on one of three "harnesses", including a new one, since August 2026, that moves an agent from "following rules" to "planning and solving multi-step tasks across systems." This is a real capability shift, but it comes with a different billing model and a decision you can't undo once made. What is a harness, and when does the classic Standard harness still win?

In this session, you will learn how to:
- Understand what a harness is, and how the three harnesses in Copilot Studio differ
- See a live, multi-step task solved end-to-end on the GitHub Copilot harness: Research, decide, create a document and trigger a workflow
- Recognise the licensing gotchas: usage-based Copilot Credits from the moment you start building, and that an agent can't switch harness once created
- Ask the right governance questions before recommending this: Permissions, data access and approvals
- Apply a simple framework for choosing the right harness

This session is for developers and consultants who want to see how the harness actually behaves, not just what the marketing says, plus the licensing and governance details worth knowing before you recommend it.

Note: the GitHub Copilot harness is a Copilot Studio runtime, a different product from the GitHub Copilot coding assistant, despite the shared name.