1.3. Reflection Checkpoint

Key Takeaways

• Copilot is a probabilistic assistant, not a deterministic tool or an all-knowing oracle — the same prompt can vary, and fluent output can still be wrong.
• Capabilities form a ladder of autonomy from inline completion up to Agent Mode; the skill is matching the lightest capability to the task.
• Copilot is a layer across surfaces (IDE, CLI, GitHub.com, Mobile), all gated by a GitHub identity and by plan, and each surface offers different context and tools.
• Real productivity is the suggest-evaluate-adapt loop, with the developer's judgment and validation as the control system.

Connecting Forward

Phase 2 dives into the highest-weighted domain on the exam: Copilot's features. Everything there — enabling Copilot, the interaction surfaces, the CLI, the agentic capabilities, and org policies — is an instance of the ladder and the surfaces you just learned. Keep the suggest-evaluate-adapt loop in mind; it reappears when we discuss code review and responsible operation.

Self-Check Questions

• Why does "validate AI output" follow logically from the fact that Copilot is probabilistic, rather than being an arbitrary rule?
• Given a task that spans several files and needs a command run at the end, which rung of the autonomy ladder fits, and what makes the lower rungs insufficient?
• If a feature works on GitHub.com but a developer can't find it in their IDE, what two factors should you check first?