Dev calls out Anthropic for not shipping built-in memory wiki — because it sells tokens
arieljalali · x · 2026-10-31
Quoting a viral post: devs waste 20 minutes every morning re-explaining their project to Claude Code, even though Karpathy showed the fix a year ago — use AI to build a "second brain" instead of just writing code. Point Claude Code at an Obsidian vault, drop in articles, transcripts, PDFs, and it builds a living wiki that compounds as you feed it more.
The author's take: Anthropic could easily build this wiki/memory into Claude, but doesn't — "because they sell tokens." Re-prompting context is itself a revenue source.
More from coding & agent
- Dev compiles Game Boy CPU instructions into Qwen weights, runs via unmodified HF inference — vincent_koc · 2026-10-31
- Claude Code Mods explained: run JS modules in-session to build a programmable agent runtime — bibryam · 2026-10-31
- Memgraph 3.14's breaking changes will break column-position parsers — release notes double as a regression checklist — JeremyCMorgan · 2026-10-31
- Open-source PostgreSQL MCP Server lets AI assistants query databases in natural language with read-only safety — modelcontextprotocol · 2026-10-31
- zero-core-verify MCP connector adds post-transaction outcome verification via LLM-as-judge — modelcontextprotocol · 2026-10-31
- ComfyUI Agent v0.4.0 adds transactional canvas edits, subgraph support and HF model lookup — FishermanLive8958 · 2026-10-31