CUHK Hangzhou paper formalizes that skill MDs and RAG stores are lookup tables, not learning: retrieval needs combinatorial coverage; weight updates generalize from O(N) examples.
Google Research's ReasoningBank separates success and failure trajectory memory for agents, yielding +8.3pp on WebArena and 57.4% on SWE-Bench with +4.3% token overhead.
Memori hits 81.95% LoCoMo accuracy at just 1,294 tokens/query — 67% smaller prompts than Zep, 20x cheaper than full-context — with MCP server and multi-agent attribution model.
memsearch by Zilliz: one memory backend across Claude Code, Codex, OpenClaw, and OpenCode — markdown SSOT, local ONNX embeddings, BM25+dense+RRF hybrid search.
DPM replaces per-agent stateful memory with immutable decision logs — enabling audit trails, multi-tenant isolation, and provenance for regulated enterprise AI deployment.
Curated AI insights — sent when there's something worth your inbox.