Grouping AI-agent memories beats fancier retrieval on tight budgets, a post argues

RSM-full groups related memories as they arrive, then brings those groups back together when an AI agent needs them, according to a post. The author argues for organizing memories before optimizing search, rather than stuffing old interactions into prompts or retrieving isolated chunks. The post reports that on AMA-Bench at roughly 4,000 prompt tokens, RSM-full retained 83% of full-history quality while using 32% of the token cost.