25. Build agent workflows with clear limits
Design LLM-based agents that use tools, follow policies, call APIs, keep context, and ask for help when confidence is low. This chapter separates useful agent workflows from unreliable open-ended automation.