Dev's 3D Coding Harness Finds a Sharp Threshold: Strong Models Eat Scaffolding
rms80 · x · 2026-09-23
The author built a 3D coding harness earlier this year when no model could hit the bar. Re-testing with stronger models like Astra reveals a sharp threshold: below it, scaffolding helps a lot; above it, almost not at all — consistent across providers. Warning for agent builders: today's scaffolding will eventually be eaten by base models. The post also links their paper "Thinking in Blender" (arXiv:2606.02580), introducing SEIG, an agentic framework that reconstructs a single image as an editable Blender program, progressively refining geometry, materials, composition, and lighting without specialized 3D models or differentiable rendering.
More from coding & agent
- MatBrain splits reasoning from tool use: two models screen 30,000 crystal candidates in 48 hours — bravo_abad · 2026-09-23
- Firecrawl Raises $75M Series B, Launches Alexandria Knowledge Library for AI Agents — omarsar0 · 2026-09-23
- Five AI agents bypassed a permissions broker in ten minutes using 'start' instead of 'stop' — TrifleHopeful5418 · 2026-09-23
- Shopify CEO who pushed staff to use AI now 'horrified' — the 'Slop Grenades' story — srchvrs · 2026-09-23
- Lovable adds Claude Opus 5.5 and GPT-6 Sol, auto-routing between frontier models — AlexandrePesant · 2026-09-23
- Dev laments agents built around KV caches, wants inference-first chips — dbreunig · 2026-09-23