What are AI credits, and why does typing a few words consume thousands of them? Every GitHub Copilot interaction assembles a layered prompt before anything reaches the model. Most developers never see this process, which makes slow or irrelevant responses hard to diagnose. In this session, Ken Muse and Mickey Gousset open up the pipeline: how VS Code builds these prompts, how the prioritization engine decides what stays in the context window and what gets silently dropped, and what real-world patterns look like in the debug logs. You'll leave knowing how to engineer your interactions for maximum signal and minimum waste.