1.1. What GitHub Copilot Actually Is
💡 First Principle: GitHub Copilot predicts the most likely useful text — code, an explanation, an edit, a plan — given the context it can see. It is not retrieving a stored answer and it is not executing a fixed algorithm. Everything else about Copilot is a consequence of that one fact.
Why this matters: if you believe Copilot "knows" the correct code, you will trust it blindly and the exam will punish you for it. If you understand it predicts a plausible continuation, then the need to validate output, the variability between runs, and the importance of good context all follow naturally. The single most testable theme on the GH-300 — "you must validate AI output" — is downstream of this principle.
The mental model that works best here is a knowledgeable but fallible pair programmer. A good human pair partner suggests, drafts, and explains, but you stay responsible for what ships. Copilot fills that seat: fast, tireless, broadly read, occasionally confidently wrong.
⚠️ Common Misconception: Many learners think Copilot is "just smarter autocomplete." That was true at launch. Today it is an agentic platform — Agent Mode, Edit Mode, Plan Mode, MCP, and Sub-Agents can plan and carry out multi-step, multi-file work. Inline completion is now the smallest thing Copilot does.
