He benchmarks new models by having them build their own copilot — $28 of compute
bucolucas · reddit · 2026-10-27
Reddit user bucolucas shares a personal benchmark: every new model gets to build a copilot for itself as a test of capability. This one was specced with Fable 5-1 and implemented with Sonnet/Opus 5.5 for about $28 of compute; the screenshot shows it building a meal/recipe planner for garden planning. He asks whether others have tried having AI build their own tools.
More from coding & agent
- Cursor launches Agent SDK: agents as code with Markdown prompts and TypeScript tools — soleio · 2026-10-27
- Vibecheck, the 'usable formal methods' hackathon, sells out two floors in SF — spikedoanz · 2026-10-27
- How do you verify AI agent outputs beyond unit tests and LLM-as-judge? — Practical_Green5775 · 2026-10-27
- AI agents may soon become the only way to run code, not just write it — IgorCarron · 2026-10-27
- Hermes Agent used as browser QA tester finds 8 issues in a bug-seeded site — techlatest_net · 2026-10-27
- Open-source FableCut lets AI agents edit a real video timeline via remote MCP — FastPresence9799 · 2026-10-27