Memory Engineering: Going Beyond Context Engineering
summary
Claims that scaling beyond prompt and context engineering requires a persistent data→memory pipeline where ingested data becomes embeddings, lands in a database, is organized, retrieved (not just vector search but hybrid retrieval) and fed back into LLMs; cites MIT-style finding that 95% of orgs don’t realize GenAI value while 5% do because their agents have short-term, long-term, shared memory. Demonstrations include semantic techniques like semantic cache and retraining embedding models on agent traces; workflow agents require only execution memory, deep-research agents need fuller persistence.
AI-written summary · about
chapters (10)
- 0:00Introduction and speaker welcome
- 1:15The GenAI divide and memory's role
- 4:27Prompt engineering is maximization
- 8:45Limitations of prompt engineering
- 9:06Context engineering is optimization
- 13:46Memory engineering is continuation
- 16:55Types of agent memory
- 18:21Memory engineering lifecycle
- 21:30Key takeaways and definitions
- 22:34Closing the GenAI divide with memory
related talks
25mWhat does it take to build long-term state for AI agents?
15mMemory-Augmented AI Agents: Enhancing Long-Term Reasoning and Context Awareness
46mThe Anatomy of Memory in Humans and AI Agents
22mGenerative AI application stack and providing long term memory to LLMs
28mAgentic Memory
18mArchitecting Agent Memory: Principles, Patterns, and Best Practices
31mBuild Intelligent AI Agents With Context Engineering
54mThe Anatomy of Memory in Humans and AI Agents