User Signal Dies at the Retrieval Boundary
summary
Claims a concrete fix: utility-ranked memory that re-ranks retrieval by outcome history, not just semantic similarity, raising helpful memories and lowering harmful ones. The system is a runtime layer called Agent RX that updates context without retraining, demonstrated on a product SQL demo that turns failed searches into useful tool calls. Reported gains include a benchmark lift from ~66% to ~76% (skills ~80%) and agentic task gains from ~35.7% to ~61.3%.
AI-written summary · about
chapters (6)
- 0:00Agent failures at retrieval
- 1:24Retrieval core problem
- 3:09Evals versus agent action
- 4:57Agent RX and utility score
- 9:59Utility-based memory retrieval demo
- 14:59Self-improving agent systems
related talks
40mAgentic Retrieval: Building Self-Optimizing Search Systems
01h 30mOptimizing Retrieval for Agentic Systems
37mWhen better retrieval makes agents worse
14mMake Agents Remember the Right Context at the Right Time
15mMemory-Augmented AI Agents: Enhancing Long-Term Reasoning and Context Awareness
23mYour Agent Has Amnesia: Why Mobile AI Agents Need Memory to Be Useful
21mPersonalize Search Results with OpenSearch Agentic Memory
22mThe Power of RAG at Scale: A Year of Explosive Growth and Observab...