Moonshots EP 229: The Humanoid Takeover — $50T Market, Figure’s Full Body Autonomy, and Robots in Dorms

Summary

Peter Diamandis and Dave visit Figure AI headquarters in San Jose for an in-depth interview with CEO Brett Adcock. The episode centers on Figure 3 running Helix 2, their fully neural-net-driven autonomy system that replaced the last 109,000 lines of C++ code. Brett demonstrates room-scale autonomous operation (4+ minutes uncut) and draws a sharp line between genuine neural net autonomy and the tele-operation or open-loop replay that most competitors use in demo videos. Key revelations: Figure designed its hardware around the neural net (not the reverse), the robot has more possible states than atoms in the universe (360^40 degrees of freedom), and Brett quietly founded a new AI lab called HARK working on next-gen multimodal models for embodied intelligence. The episode includes a tour of Figure’s 300-400K sqft facility with hundreds of robots on test stands and production lines.

Key Segments

• [00:01-00:05] Figure HQ tour: Figure 1 through Figure 3 evolution, palm cameras, 30lb weight reduction, 90% manufacturing cost cut
• [00:05-00:10] Helix 2 deep dive: all neural net stack, removed final 109K lines of C++, full-body bimanual autonomy
• [00:10-00:15] OpenAI partnership history: collaboration didn’t work, Figure’s team “ran circles around them,” parted ways
• [00:15-00:22] Industry reality check: most competitors use tele-operation or open-loop replay, nothing over 1 minute of continuous neural net operation exists elsewhere in the world
• [00:22+] Development velocity, consolidation prediction (far fewer than 10 companies globally will survive), HARK AI lab reveal

Notable Claims

• Brett states there is nothing over 1 minute of continuous neural net operation on a humanoid robot anywhere else in the world — everything else is cut footage or tele-operated
• Figure designed Figure 3 hardware specifically to run Helix, inverting the traditional hardware-first approach
• Brett predicts the humanoid industry will consolidate to “far fewer than 10” companies globally
• China has 150+ robot companies but Brett characterizes most as not solving general robotics

Guests / Panelists

Peter Diamandis (host), Dave (DB2), Brett Adcock (CEO, Figure AI)

RDCO Mapping

• Neural net vs code paradigm: The shift from hundreds of thousands of lines of C++ to pure neural nets mirrors the broader software transformation. Strong parallel to AI coding agents replacing traditional development.
• Tele-operation exposé: Brett’s blunt call-out of competitors faking autonomy with tele-operation is a strong Sanity Check angle on hype vs reality in robotics demos.
• HARK AI lab: New entity worth tracking — Brett founding a separate AI lab for embodied intelligence models suggests the robotics-AI stack may diverge from general-purpose LLMs.