1.2.2. The Suggest-Evaluate-Adapt Loop
💡 First Principle: Productive Copilot use is a loop, not a handoff: Copilot suggests, you evaluate against intent and correctness, then you adapt — accept, reject, refine the prompt, or steer with more context. The developer's judgment is the control system.
This loop is the practical expression of everything in Phase 1. Because output is probabilistic and possibly wrong, you never simply accept; you assess. Because Copilot reasons over context, when a suggestion misses, the fix is usually to change the context or prompt, not to give up.
Notice that validation does not end when you accept a suggestion. Accepted code still goes through the same scrutiny any code would: tests, review, security checks. This is why "validate AI output" appears across multiple exam domains — it is not a one-time gate but a habit woven through the loop.
✅ Best Practice: When a suggestion is wrong, resist the urge to hand-fix silently. First ask why the context led there — a missing import, an ambiguous function name, a stale comment. Steering the context usually produces a better next suggestion and teaches you how Copilot "sees" your code.
⚠️ Exam Trap: Questions sometimes describe a developer who accepts suggestions without review and ships a bug or a vulnerability. The "responsible" answer is almost never "trust Copilot more" — it is to insert evaluation and validation into the loop.
Reflection Question: When a Copilot suggestion is subtly wrong, why is refining the prompt or context often a better move than simply rejecting and retyping the code yourself?
