Moonshots with Peter Diamandis Brett Adcock: Humanoid Run on Neural Net, Autonomous Manufacturing, $50T Market #229
412 snips
Feb 11, 2026 
Brett Adcock, founder of Figure and builder of general-purpose humanoid robots. He talks about shifting from C++ to full-body neural nets with the Helix stack. Covers sensors and tactile fingertips, room-scale autonomy, fleet learning as a moat, manufacturing robots at scale, and plans for commercial and home deployments.
AI Snips
Chapters
Books
Transcript
Episode notes
Replace Heuristics With Retrainable Policies
- Remove hard-coded heuristics and embrace retrainable neural policies to iterate faster and reduce brittle code.
- Replace expensive C++ maintenance with models you can retrain and deploy fleet-wide.
LLMs Don’t Substitute Embodiment Models
- LLMs provide semantic grounding but lack embodied physics and real-time control knowledge.
- Embodied robots need distinct models trained on physical contact, dynamics, and high-dimensional motor states.
Harden Systems Through Commercial Deployments
- Run robots in commercial customer sites to harden fleet ops: safety, maintenance, and reliability.
- Use leases and close customer partnerships to scale deployments responsibly.
