Memory plugins for agents are technically just RAG-based documentation systems

Current agent memory implementations rely on RAG pipelines that convert transcripts into vector snippets for prompt injection. Instead of true memory, these systems function as retrieval tools for project documentation, often failing to provide agents with deep contextual understanding.