New Copilot Studio — Part 2: Creating Agents (Build Concepts)
The six elements of an agent — instructions, skills, tools, knowledge, connected agents, and memory.
▶ Part 2 in one sentence — Part 2 organizes the six components you need to understand when building a real agent from the Part 1 principles—harnesses, skills, tools, memory, and context engineering—from concept to authoring method to pitfalls. The concrete hands-on work is covered in Part 3.
⚠️ The features, screens, and timelines in this document are based on preview and may change (subject to change).
0. Introduction — from Part 1 to Part 2
▶ Point — Part 1 said that the maker’s job changes from drawing flows to designing a harness. Part 2 explains how to assemble that harness in practice. Remember one core mental model: people write the “what,” and the agent handles the “how.”
0.1 The six elements of a build
In New Copilot Studio, one agent consists of six parts. They map one-to-one to the chapters in Part 1.
| Element | One-line definition | Part 1 basis |
|---|---|---|
| Instructions | Always-on top-level rules (constitution) | Chapters 1–2 |
| Skills | Reusable instruction bundles pulled out only when needed | Chapter 3 |
| Tools | Hands and feet for taking real action outside the agent | Chapter 4 |
| Knowledge | Knowledge the agent reads (Excel, documents) | Chapter 9.4 |
| Connected agents | Collaborative agents that delegate specialized work | Chapter 5 |
| Memory | Maintains context across conversations and work | Chapters 6, 9.6, 9.7 |
0.2 If the six elements were a new employee
The six elements are like assigning work to a capable new employee. You provide the environment and let the new employee (agent) decide how to work.
| Element | Analogy | Role |
|---|---|---|
| Instructions | Work rules | Top-level rules that must always be followed |
| Skills | Work manuals | Documents that describe how to perform specific tasks (opened when needed) |
| Tools | Tools | Means for actually executing something |
| Knowledge | Reference documents and library | Knowledge to read and use as evidence |
| Connected agents | Specialist colleagues | Collaborators to delegate deep work to |
| Memory | Work notes | Stores context from conversations and work |
Core mental model: People write the “what,” and the agent handles the “how.” Instructions and skills focus on what; the core handles how—code, paths, libraries, and so on (Part 1, 1.4).
0.3 How to read this document
- Chapters 1–6 cover the six elements in the order concept → authoring method → pitfalls. They explain instructions (Chapter 1) and skills (Chapter 2) in the most depth because these determine build quality.
- Chapter 7 synthesizes how the six elements interact in a single request.
- The hands-on build of a real agent from start to finish is in Part 3. Part 2 is the conceptual foundation for understanding that hands-on work.
What this part covers
| # | Topic | Key point |
|---|---|---|
| 1 | Instructions | Always-loaded top-level rules — short and clear |
| 2 | Skills | Reusable instruction packages loaded only when needed |
| 3 | Tools & knowledge | Hands and feet (connectors, MCP) and readable knowledge (RAG → code) |
| 4 | Connected agents & memory | Context isolation and two-layer short-term/long-term memory |
| 5 | Synthesis & checklist | How the six elements interact in one request + pitfalls |
Read them in order for the most natural progression, or open the topic you need first.