Agent swarm builds ice-flow PDE solver, discovers glacier sliding law beating textbook models
DynamicWebPaige · x · 2026-11-21
Using an agent swarm in Antigravity with Gemini 4 Neon, the author built a differentiable 2D ice-flow PDE solver from scratch over NASA ITSLIVE velocities (2015–2022) and Bedmap3 topography for Thwaites and Pine Island glaciers, mapping basal friction with <0.5% grounded-speed error. Strict cross-basin tests: laws fit on one glacier predicted the unseen neighbor; both basins independently revealed the same 350 m slippery-till vs rough-bedrock transition; bed-geology sliding laws beat every classical law (cutting fast-trunk errors >50%); and NASA's +56 m/yr speedup confirms near-plastic till mechanics.
More from coding & agent
- OpenAI details when Ultrafast speed on GPT-6.2 is worth it for rapid iteration — OpenAIDevs · 2026-11-21
- Paradigm CTO: be unreasonably ambitious with AI — rewrite your whole dependency stack in days — nptacek · 2026-11-21
- Tip: Add Antigravity to your phone's home screen for remote control of all agent threads — amankhan · 2026-11-21
- The sentences your customer-facing AI agent must never say—and how to test for them — rcleffke · 2026-11-21
- Haijun Xia pushes back on 'UI is dead': chat is a UI too, and far from sufficient — sh_reya · 2026-11-21
- Fully local pipeline: open-weight 27B models ship production code with Playwright + vision-loop tests — awitod · 2026-11-21