A practical guide to AI agent rate limits: where to throttle, how to separate model limits from action limits, and the production patterns that keep agent systems fast without letting them melt your budget or downstream tools.