2.2.3. Commands, Interactive Use, and Sessions
💡 First Principle: The CLI's power comes from sessions — persistent, stateful conversations where Copilot remembers earlier steps — rather than isolated one-shot commands. Memory across the session is what lets it tackle multi-step work.
Beyond translating intent into shell commands, the CLI runs interactive agentic sessions. Within a session it can plan (Plan Mode), execute steps, optionally run without per-step confirmation (Autopilot Mode), and even delegate parts of a task to parallel sub-agents for speed. Session memory persists context across interactions, so a later instruction can build on an earlier one.
Recognize these session behaviors for the exam:
- Plan Mode — drafts a strategy before acting, mirroring the IDE concept.
- Autopilot Mode — executes steps without stopping for each confirmation (use deliberately).
- Sub-agent delegation — splits a task so parallel agents handle different parts.
- Session memory — carries context forward across the conversation.
⚠️ Exam Trap: Autopilot trades oversight for speed. The "responsible" choice in a sensitive or production context is usually to keep confirmations on, because Phase 3's validation principle still applies — autonomy does not remove the developer's accountability.
⚠️ Exam Trap: Don't confuse the CLI's session memory with Copilot's repository Memory feature (deduced repo facts used by the cloud agent and code review). They are different mechanisms with similar names.
Reflection Question: When would enabling Autopilot Mode be reasonable, and when would it conflict with the responsible-use principles you'll meet in Phase 3?
