1. First Principles of AI-Assisted Development

Before any feature, policy, or prompt trick, you need a working theory of what GitHub Copilot is. Almost every wrong answer on the GH-300 traces back to one of two flawed mental models: treating Copilot as a deterministic tool that always returns the "right" answer, or treating it as a magic oracle that knows everything. This phase replaces both with an accurate model — a probabilistic assistant that reasons over the context you give it, across several surfaces, with a human always in the loop. Get this right and the rest of the guide becomes common sense.