Andrew Ng memo argues agent memory needs graphs, not bigger context windows
solyarisoftware · x · 2026-07-27
Andrew Ng’s new 15-page memo lays out a 4-step shift from loops to graphs for agent memory.
- The core argument: bigger context windows do not solve memory; they only delay forgetting until the next session.
- Instead of RAG over static documents, agents need a temporal knowledge graph that merges chat history and structured business data into one queryable layer.
- The key design choice is bi-temporal data: each fact keeps both the time it happened and the time the system learned it, so nothing is overwritten.
- Contradictions are invalidated, not deleted, which preserves answers like “who owned this in April” months later.
- The post also points to the paper’s benchmark results, including a claim that the system beats MemGPT on Deep Memory Retrieval.
The attached paper image shows the title “Zep: A Temporal Knowledge Graph Architecture for Agent Memory”, reinforcing that this is a concrete memory-stack proposal, not a high-level concept post.
Related event: Andrew Ng Advocates Graph-Based Memory for AI Agents(2 posts)→
More from coding & agent
- Alex Townsend posts 200 open problems in numerical linear algebra for humans and AI agents — IgorCarron · 2026-09-11
- Kimi K2.8 Preview rolls out: near-K3 coding performance, 1M context for all tiers — teortaxesTex · 2026-09-11
- Looking for a classifier of software engineering task shapes to pick models per task — StewartalsopIII · 2026-09-11
- Steal this idea: prompt-to-hardware where agents assemble custom devices — paraschopra · 2026-09-11
- Model Is the Least Interesting Part: A Guide to Six Core AI Architectures from RAG to Multi-Agent — goyalshaliniuk · 2026-09-11
- Non-coder builds layered memory architecture: 20k tokens tracks a year of agent conversations — matteoianni · 2026-09-11