Google's open-sourced "Elephant-Goldfish" workflow separates planning from review to curb sloppy AI output
gaganghotra_ · x · 2026-11-02
A circulating writeup details Google's internal "Elephant-Goldfish Model" — a workflow its engineers reportedly use daily to stop AI from doing sloppy work, adaptable to ChatGPT, Claude or Gemini.
- The Elephant chat: deep planning only, prompted with "No output yet. Ask me questions until you could write the brief yourself," then produce a one-page brief covering goals, rationale, decided and open items
- The Goldfish chat: a second session that never saw the original conversation reviews or executes the plan, catching flaws with no context bias
- Originally built for code, the method works for essays, business plans, trips and any planning task
More from coding & agent
- AWS's open-source Strands Harness claims 45% cheaper than Claude Code, reigniting harness debate — blaizedsouza · 2026-11-02
- Xiaomi MiMo's CodeSpecWeave trains coding agents on interdependent specs for long-horizon coding — XiaomiMiMo · 2026-11-02
- Sketchy v0.1.6: AI diagrams use 29% fewer output tokens than plain text — Nkgsporty · 2026-11-02
- Avoiding AI brain rot: Miles Cranmer's lecture on learning with AI — _ScottCondron · 2026-11-02
- Scrapling web scraping framework tops 92k GitHub stars, ships MCP server for AI workflows — D4Vinci · 2026-11-02
- Vercel open-sources Agent Desktop: reliable computer use via OS accessibility trees, 7.8k stars in a day — vercel-labs · 2026-11-02