Prompt like a top expert: set extreme bars, real tools, and outside judges
petesena · x · 2026-10-26
The "Depth" entry in petesena's thread on Felix Rieseberg's ambition guide: make the model work like the best person in the field. Need a site to load in 5 seconds? Ask for 200 milliseconds. Have it use the tools experts actually use, like databases and stats models — pretty tables don't count. And never let it grade its own homework: bring in an outside judge.
Related event: Anthropic Engineer's Four Ambitions Prompting Guide Gains Attention(5 posts)→
More from coding & agent
- LabOps Series: A Weekly Agent That Flags Miscalibration and Contamination in Lab Equipment — nlarusstone · 2026-10-26
- Meta's Muse Agent Can Now Connect to 1Password With Scoped Vault Access — TechPreacher · 2026-10-26
- Why dedicated agent infrastructure like phone numbers keeps custom agents alive — krishnan · 2026-10-26
- "Stop naming your agents": dev argues multi-agent teams are just one LLM in costumes — ziv_ravid · 2026-10-26
- Current LLMs Are Horrible at Type Design in Dependent Type Languages — hargup13 · 2026-10-26
- SKILLs Aren't Token Waste: How Custom SKILL.md Files Teach Agents Your Workflow — ievkz · 2026-10-26