Figure AI's CEO Brett Adcock recently disclosed that his startup's internal AI team, composed primarily of Google DeepMind veterans, significantly outpaced OpenAI during their collaboration, ultimately leading him to end the partnership. In a candid appearance on the Shawn Ryan Show, Adcock provided rare insight into the $39 billion humanoid startup's internal culture, its aggressive roadmap toward general-purpose robots, and the mechanical breakthroughs that are pushing Figure 03 closer to real-world deployment .

Why Did Figure AI Really Part Ways with OpenAI?

The dissolution of Figure's high-profile partnership with OpenAI surprised many in the robotics industry, but Adcock's explanation reveals a fundamental mismatch in how the two organizations approach robot learning. According to Adcock, the friction stemmed from OpenAI's inability to maintain the daily and weekly presence required to test AI models on physical hardware rather than in simulation environments .

"We just found that team we had internally... we just ran kind of circles around them," Adcock stated, explaining that Figure was effectively teaching OpenAI how to do robot learning while receiving little value in return.

Brett Adcock, CEO at Figure AI

The breaking point came when Sam Altman reportedly called Adcock to mention that OpenAI was considering launching its own internal robotics program. Adcock recalled his response bluntly: "I was just like, this is over... just get out of here." This split forced Figure to double down on its own "omni-model" architecture, which now handles speech, reasoning, and action simultaneously, eliminating the need for external AI partnerships .

How Does Figure 03 Repair Itself Without Human Help?

Beyond the boardroom drama, Adcock highlighted a significant mechanical breakthrough that could reshape how humanoid robots operate in real-world environments. Historically, losing a single motor or communication link meant an immediate system collapse and the need for human intervention. The Figure 03, however, now features autonomous "self-triage" capabilities that allow the robot to diagnose and compensate for hardware failures on its own .

Adcock explained the practical implications of this advancement: "We can lose a knee... lose full comms of the knee. We can stiffen the joint and we can limp off to the hospital." This "hospital" or triage area is part of Figure's "lights-out" Sunnyvale operation, where robots navigate to repair bays or charging docks without human intervention. This capability represents a major step toward truly autonomous robot operations in commercial settings .

Adcock

Steps to Understanding Figure's Path to Mass Production

• Software 2.0 Strategy: Figure deleted over 100,000 lines of hand-coded C++ in favor of the Helix 02 neural network stack, allowing robots to learn new tasks by uploading new "weights" to their brain, much like downloading an app to a smartphone .
• Manufacturing Acceleration: Figure's "BotQ" manufacturing facility is now producing a robot every 90 minutes, with a long-term goal of 50,000 units per year to meet demand from commercial customers including BMW, Brookfield, and major logistics firms .
• Supply Chain Decoupling: Adcock emphasized a "Made in the USA" manufacturing strategy with a total supply chain decoupling from China targeted for summer 2026, positioning Figure to compete independently in the global humanoid market .

Why Are Homes Still Harder Than Factories for Humanoid Robots?

Despite the symbolic success of a Figure 03 appearing at the White House and performing tasks like tidying living rooms and unloading dishwashers autonomously, Adcock remains notably cautious about domestic deployment. He admitted that he still "babysits" the machines when they are around his own children, signaling that the technology is not yet ready for unsupervised home use .

Adcock contrasted the structured environment of a factory with the unpredictability of a home. In a factory, the robot's tasks can be defined on "a piece of paper," whereas a home requires a level of common-sense reasoning and safety that is still being refined. Every toaster and floorplan is different, creating what Adcock described as "chaos" and "entropy" that current robots struggle to navigate. This caution aligns with scrutiny Figure has faced regarding safety culture, including a whistleblower lawsuit alleging that earlier models generated enough force to fracture a human skull .

The distinction between factory and home deployment underscores a critical challenge in the humanoid robotics industry: while robots can perform well-defined tasks in controlled environments, the variability and safety requirements of domestic settings demand a fundamentally different level of artificial intelligence and mechanical reliability. Figure's focus on commercial deployment first, followed by eventual home integration, reflects this engineering reality.

As Figure prepares for mass deployment to commercial customers and races against Chinese competitors like AGIBOT, Adcock's candid revelations about the OpenAI split, self-repair breakthroughs, and the remaining challenges for home deployment paint a picture of a company that is moving rapidly but deliberately toward a future where humanoid robots become as common as smartphones .