1. What distinguishes short-term (working) memory from long-term memory in an agent?
2. Why is conversation summarization used as the context window fills up?
3. What is the 'lost in the middle' effect in long contexts?
4. How does episodic memory differ from semantic memory for an agent?
5. Why might an agent store memories as embeddings in a vector store?
6. What is a risk of naively appending every turn to the context indefinitely?
7. What role does a retrieval step play in memory-augmented agents?