4.2. Lifecycle and Limitations

💡 First Principle: The end-to-end lifecycle is a closed loop — context in, suggestion out, your accept/dismiss feeds back as a quality signal — and every limitation of Copilot is a property of some stage in that loop (the model's knowledge, the context window, the lack of true verification).

Why this matters: "visualize the code suggestion lifecycle" and "describe limitations of LLMs and Copilot" are explicit syllabus bullets. Holding the whole loop lets you answer both: where a step happens, and where the inherent limits come from.

The mental model: an assembly line with inspection stations and a feedback sensor at the end. Material (context) enters, stations (proxy filters, model) transform and inspect it, the product (suggestion) ships, and a sensor (accept/dismiss telemetry) reports back to improve the line.

⚠️ Common Misconception: "Copilot knows my whole codebase and the latest libraries." LLMs have a fixed context window and a knowledge cutoff. Copilot reasons only over the context it's given (plus indexing on Enterprise); it does not inherently know recent APIs or unopened files.