Whenever you're ready, walk me through how you'd approach this.
Tell me about memory in agent systems.
Explain how you would design memory for a tool-using agent, including the distinction between conversation state, working memory, episodic memory, and durable semantic memory. Discuss when to write, retrieve, summarize, forget, or correct memories, and how you would evaluate quality while controlling latency, cost, hallucination, and prompt injection risks.