Weaviate's year of agent memory: why store-everything RAG breaks three ways
victorialslocum · x · 2026-09-08
After a year building a memory service at Weaviate, the team explains why the common approach—store every message, search with the latest user turn, dump results into context—fails in three ways:
- Cost and context growth: full history compounds on every message, eating tokens, slowing the model, degrading answers.
- Retrieval is reactive: output quality is capped by the query, usually the vague last user message.
- Contradiction and staleness: users change their minds and facts expire; this was the hardest engineering problem, solved by extracting information from every message rather than piling them up.
A hands-on engineering retrospective pointing toward proactive, structured, updatable agent memory.
More from coding & agent
- GlossoGen platform systematically studies when LLM agents evolve incomprehensible languages — EliasEskin · 2026-09-08
- W3C × GS1 Zurich meeting pushes two-layer trust framework for agentic commerce — melnykowycz · 2026-09-08
- Claude shipped 583 PRs to a SaaS in one week — and found a payment bypass on its own — mhmazur · 2026-09-08
- PipesHub launches as an open-source permission-aware context layer for AI over company data — Effective-Ad2060 · 2026-09-08
- VibeGame: 8-agent adversarial team turns one sentence into a full playable game project — jiqizhixin · 2026-09-08
- Stageflow: a configurable multi-stage agent pipeline with per-stage sessions and human gates — tejasghutukade · 2026-09-08