Notion AI lead: much of what runs on GPUs should run as CPU code
threepointone · x · 2026-10-17
Notion AI lead Sarah Sachs argues that a lot of work currently served on GPUs should actually run on CPUs via code execution.
Her reasoning: executed code brings determinism and observability, while having LLMs reinvent the process each time is token-heavy and slow. Giving agents a local file system to store intermediate thoughts and write short Python scripts has been one of the most important tools for Notion's internal agents, alongside edit and search.
Related event: Notion AI head: move agent work from GPUs back to CPUs(2 posts)→
More from coding & agent
- 10 open-source repos for training small models to do your agent's job — victor_explore · 2026-10-17
- Robotics engineer challenges coding agent hype: 'They can't do visuomotor control' — AlexTensor · 2026-10-17
- Dev uses Google Argon procedural generation scripts to build real 3D game assets — Promptmethus · 2026-10-17
- LLM-agent-free Jev SRE diagnosis pipeline passes 76.2% of 105 diagnoses in 14.6s median — tianyin_xu · 2026-10-17
- TwelveLabs' video intelligence agent Jockey hits GA, plugs into ChatGPT and Claude via MCP — lmoroney · 2026-10-17
- Google demo: Gemini rebuilds ancient ruins into explorable 3D worlds via Antigravity + Blender MCP — jnack · 2026-10-17