2.5. Reflection Checkpoint

Key Takeaways

• Match feature to need on two axes: which rung of the autonomy ladder (inline → Chat → Edit → Plan → Agent) and which surface (IDE, CLI, GitHub.com, Mobile). Most "which feature" questions resolve on these two axes.
• The CLI is a terminal-native agent, not the old command-suggester — plan mode, autopilot, sub-agents, and session memory define it. Authentication and plan/policy gate it.
• Know the look-alikes apart: Edit vs. Plan vs. Agent Mode (autonomy); MCP (external reach) vs. sub-agents (parallelism) vs. sessions (state); instructions files (standards) vs. exclusions (hide files) vs. Spaces (curated context); prompt files (reusable prompts).
• Org controls are admin/governance powers on Business/Enterprise: feature availability, code review policies, audit logs, and REST API seat management. Individual IDEs reflect, not override, these policies.

Connecting Forward

Phase 2 was about what Copilot can do. Phase 3 turns to how to use it responsibly — and you'll see that the validation habit from Phase 1's suggest-evaluate-adapt loop and the cautions sprinkled through this phase (autopilot, code review, generated scripts) are the connective tissue. Responsible use isn't a separate topic; it's the discipline that makes these powerful features safe.

Self-Check Questions

• A developer wants to make a precise change to two known files with full control over each edit. Which mode is correct, and why would Agent Mode be the wrong default here?
• A team needs Copilot to draw on data from an internal service when answering questions. Which capability enables that, and who decides whether it's permitted in an enterprise?
• An admin must extend Copilot code review to PRs from contributors without a Copilot seat. Is that a default behavior or a policy, and what cost factors apply after June 1, 2026?