New Copilot Studio — CLI Agent Overview

What this series is about

New Copilot Studio is not just a UI redesign. It is a rebuild of the AI core that brings the “harness design” principles Anthropic, OpenAI, and GitHub have validated in coding agents into an enterprise agent platform.

This series explains that change in three parts: concepts → build → hands-on. You will understand it much more accurately if you read every capability not as “a feature Microsoft added,” but as “industry principle → Microsoft implementation.”

⚠️ This series focuses on publicly discussable concepts, authoring patterns, and pitfalls. Preview-stage features, screens, and timelines are all subject to change.

📎 Complete slide deck (Parts 0–3)

This is a 60-slide deck that consolidates the entire Parts 0–3 series into one file. It includes the article screenshots and hands-on attachment guidance, so you can use it directly for internal sharing or presentations.

The data and skill packages used in the hands-on exercises can be downloaded from Part 3, Chapter 2 · Add knowledge sources and Part 3, Chapter 4 · Import the skill package, respectively.

Why it was rebuilt — at a glance

Classic Copilot Studio was essentially a conversation flow design tool. Makers had to draw every path in advance, and as scenarios became complex, branches exploded and the flow broke down during multi-step, long-running work.

New Copilot Studio rebuilds the AI core (orchestrator) on top of a coding harness, shifting the maker’s job from “drawing flows” to “designing the harness.” The core mental model is simple: people write the “what,” and the agent handles the “how.”

Classic and New coexist side by side. New is opt-in, and there is no forced migration.

Structure of the three parts

Part 1 · What’s New (Concepts)

Starting with the industry principle of harness engineering, this part walks through the agentic loop, skills, tools (MCP), sub-agents, memory, and workflows. It also covers the Classic → New mapping table and the structural runtime behavior of the CLI agent.

Part 2 · Creating an agent (Build concepts)

This part organizes the six components of an agent—instructions, skills, tools, knowledge, connected agents, and memory—in the order concept → authoring method → pitfalls. It goes especially deep on instructions and skills.

Part 3 · Creating an agent (Hands-on)

You will build a telecom sales assistant from start to finish. The agent analyzes a SharePoint Excel file, creates an HTML dashboard, and sends it in an email with the company’s design applied. You assemble it in the order instructions → skill → package → tools → test → deploy.

Key terms at a glance

Next → Part 1 · What’s New (Concepts)