One-prompt chatbot use has fundamental limits, applies to AI for science too
Afinetheorem · x · 2026-11-15
Follow-up to his earlier point: the scaffold-and-context approach applies to real work and AI for science as well — "type one prompt into a chatbot" has fundamental limits.
Related event: Scaffolding Beats One-Shot Prompts: Same Model, Vastly Different Results(2 posts)→
More from coding & agent
- Ex-county GIS tech: an LLM-written GDAL script would have saved his 200GB tile nightmare — dreamwieber · 2026-11-16
- Qdrant workshop gives a simulated fly brain persistent trauma memories at AI Dev NYC — DeepLearningAI · 2026-11-16
- Explainer animation built entirely with code, no video editor needed — iamfakhrealam · 2026-11-15
- Alibaba's Qoder quietly adds stealth Qwen4-Max-Preview — one-prompt 3D Earth in 10 minutes — iamfakhrealam · 2026-11-15
- Two ready-to-use prompts to make coding agents fix happy-path debt and implicit state machines — Yamapama · 2026-11-15
- Garry Tan: Only ~1% of engineers let agents ship features — it should grow 100x — garrytan · 2026-11-15