How AI Agent Memory Works

kept by eddie may 07

An interactive essay explains how agent memory systems work through context windows, embeddings, and retrieval architectures to overcome LLM statelessness.

Today's harness is Tomorrow's Prompt
Temporary scaffolding around AI models becomes obsolete as models improve, so developers should build these harnesses cheap enough to discard.
also on AutoGen, CrewAI, Pinecone, #ai-agents
The Limit in the Loop
As AI agents scale, stateless interactions fail, making memory a systems problem requiring active maintenance—write control, deduplication, and purposeful forgetting—rather than simple storage.
also on Weaviate, #software-architecture, #ai-agents, #user-experience
Can agents replace the search stack?
Using a basic BM25 or e5 retriever with an LLM agent can achieve 0.289→0.453 NDCG on Amazon ESCI by reasoning over queries, but this approach fails on passage retrieval where the embedding model already knows best.
also on #retrieval-augmented-generation, #ai-agents
persistent.info: Sierra MCP Gateway Dev Diary
The diary chronicles Sierra engineers’ incremental construction of an MCP gateway, detailing weekly service additions, user adoption, and solutions for secure data access.
also on Pinecone
The State of AI Coding Report 2025
The Greptile report finds AI coding tools increased lines of code per developer by 3.5x and median PR size by 79% in the past year.
also on pgvector, OpenAI