New Copilot Studio — Part 1: What's New (Concepts)
Rereading New Copilot Studio through harness design principles. A guide to the seven topics in the concepts part.
▶ Part 1 in one sentence — New Copilot Studio is not a UI redesign; it brings the “harness design” principles Anthropic, OpenAI, and GitHub have validated in coding agents into an enterprise platform. To understand it accurately, read every capability not as a “Microsoft feature,” but as “industry principle → Microsoft implementation.”
⚠️ The features, screens, and timelines in this document are based on preview and may change (subject to change).
What this part covers
| # | Topic | Key point |
|---|---|---|
| 1 | Why was it rebuilt? | Why Classic hit its limits |
| 2 | Harness engineering | What a “harness” is and why it is a coding harness |
| 3 | Agentic loop & skills | How the core runs, and why skills are loaded only when needed |
| 4 | Tools, MCP & sub-agents | How tools are attached, and why fewer is better |
| 5 | Memory, workflows & mapping | Where memory persists, and what Classic concepts become |
| 6 | CLI runtime structure | What the harness looks like in practice |
| 7 | Claude Code comparison & references | Same philosophy, different implementation |
Read them in order for the most natural progression, or open the topic you need first.