Startup ditches VLAs, runs customer robot deployment on frontier LLM agents
ihorbeaver · x · 2026-10-17
Unruggable's ihorbeaver reports they fully switched to frontier LLMs to control a robot in customer deployment over the past month. Architecture: a "System 1/System 2 on steroids" stack — a high-level reasoning agent based on Opus and Astra (generations ahead of any VLA) plus a fast execution layer of model-written computer vision code that is fast, editable and reliable. Bonus: free in-context learning and continual learning, as the large model steps in to adapt the pipeline whenever new situations arise. "Feels better to be on the right side of the Bitter Lesson." The team also says they hand-assembled every hardware device this year — this time with some help.
More from coding & agent
- How Codex Happened: Dan Shipper's 8,000-word AI-assisted feature for Every — lmoroney · 2026-10-17
- E2B founder: most agent sandbox providers are just Firecracker wrappers on AWS — zeeg · 2026-10-17
- Anthropic ships OpenTelemetry support for Claude Platform, joins GenAI traces standard group — dsp_ · 2026-10-17
- Claude $200 plan heavy use: 4 subagents ran for 30 hours straight — lucasmeijer · 2026-10-17
- Built a working iOS app with Codex overnight: 'no reason to use React Native in 2026' — indragie · 2026-10-17
- Antigravity resets Argon and Gemini quotas for all users after surge in usage — rseroter · 2026-10-17