2.3. Agentic and Advanced Capabilities
💡 First Principle: The capabilities in this section are what separate modern Copilot from "autocomplete." They share one idea: extending how much Copilot can see (MCP, Spaces, instructions files) and how much it can do on its own (Agent/Edit/Plan modes, sub-agents, code review).
Why this matters: this is the densest, most heavily tested part of the biggest domain. Expect questions that ask you to distinguish similar-sounding capabilities (Agent vs. Edit vs. Plan), recognize what MCP is, and place features like Spaces, Spark, PR summaries, and instructions files. Precision on the differences is the whole game.
The mental model: a workshop full of attachments for the same power tool. The tool (Copilot) is constant; each attachment changes what it can reach or how independently it can run. The skill is picking the right attachment.
⚠️ Common Misconception: "Agent Mode and Edit Mode are the same." Edit Mode applies scoped, multi-file edits you direct turn by turn; Agent Mode autonomously plans, runs tools and terminal commands, and iterates toward a goal; Plan Mode drafts the step plan first. They feel similar in the Chat panel but differ sharply in autonomy.
