BREAKING: Brett Adcock, CEO of Figure

---

1. Key Themes

Theme 1: Humanoid Robotics Is Transitioning from Demo to Deployment

The narrative has shifted from "robots as spectacle" to robots as economically viable workers operating in real commercial environments today.

"We just want humanoid robots to work, they're working now & it's pretty simple.. we're seeing robots do everyday things like clean up a living room, do commercial work. It's cool to see that this is gonna happen in the next few years."

The BMW pilot — a small batch of Figure robots deployed and running every day for six months — is the clearest proof point. The learnings from that deployment directly produced Figure's next-generation AI model, Helix 2.

"We refactored our whole approach to how to commercialize the software and AI systems after that."

---

Theme 2: Robotics Is an Intelligence Problem, Not a Manufacturing Problem

The dominant market narrative frames humanoid robotics as a hardware and manufacturing race. Adcock explicitly rejects this framing, repositioning the core moat as AI and software.

"In my mind, this is not a manufacturing problem. This is an intelligence problem."

Figure designs nearly every component in-house — motors, sensors, batteries, kinematics — but treats these as table stakes. The durable competitive position lives in the AI model layer (Helix), which Figure now owns entirely after ending its OpenAI partnership.

---

Theme 3: Vertical Integration as a Strategic Imperative

Full-stack ownership — from CAD files to AI models — is positioned not just as a preference but as a survival requirement in physical AI. Outsourcing any layer creates unacceptable dependencies.

"Without that you're left with the mercy of some vendor. And then if that has an issue, how are you gonna go solve it? If it's got a code problem, do you understand it? Can you QA it? Can you fix it? Can you patch it?"

This logic also explains why Figure ended its OpenAI collaboration. Sharing visibility into its robotics intelligence with a partner that was simultaneously developing interest in the space was an untenable position.

---

Theme 4: The Global Labor Market as the True TAM

Figure's $39B valuation is best understood not against software multiples but against the global labor market — a frame that makes the valuation look conservative if the technology works.

"About 50% of global GDP is human labor... This will build the biggest business in the world.. We will have the ability to ship, if the robots work well, billions of robots in the commercial workforce."

"You'll build like enormous, tens of trillions of revenue. You'll build something massive. Most tech companies trade at 10 or 20 times revenue. This is gonna be a huge business."

The commercial wage market alone is pegged at $30–$40 trillion annually, dwarfing any vertical SaaS opportunity.

---

Theme 5: Physical AGI as the End Goal — Not Task Automation

Figure is not building narrow automation. The stated objective is general-purpose physical intelligence — a robot that can operate in any environment a human can, without pre-programming for specific tasks.

"A robot that can do everything a human can.. almost like a feeling of a human in a body suit that you can talk to, can look at you, reason, visual understanding. And you can drop into any place and just look around, reason and understand."

"We want Figure to be the first place artificial general intelligence is demonstrated in the physical world, not in a chat interface."

---

2. Contrarian Perspectives

Contrarian 1: Figure Outgrew OpenAI — Domain Expertise Beats Brand in Applied AI

The consensus assumption is that frontier AI labs like OpenAI hold the advantage in AI model development. Adcock's experience directly challenges this. His team — built from robot learning specialists with over a decade of domain experience — outperformed OpenAI on the robotics-specific modeling tasks.

"It got to a point where we were just like, our team internally that was designing these models were running circles around OpenAI. We were just way better at this."

"We were better about testing on the robots, training the models, like all of it."

The implication: in applied physical AI, proximity to the hardware and domain-specific training data matters more than general model capability. Partnerships with frontier AI labs may actually slow vertical robotics companies down.

---

Contrarian 2: The $39B Valuation May Be Underprice, Not Overprice

Most observers treat Figure's 15x valuation increase in 18 months as a sign of froth. Adcock argues the opposite — the valuation is a fraction of what the category is worth if execution succeeds, because it's being priced against the wrong comparables.

"Most tech companies trade at 10 or 20 times revenue. This is gonna be a huge business."

With $30–40 trillion in global commercial wages as the addressable market, even a low single-digit market share would imply revenues and enterprise value that make $39B look like an early-stage entry price. The risk is execution, not valuation math.

---

Contrarian 3: Commercial Demand Is Already Ahead of Supply — The Bottleneck Is the Robot, Not the Market

The conventional startup risk framing assumes demand must be created. At Figure, the opposite problem exists: commercial customers are ready to deploy robots that don't yet exist at scale.

"I could put so many robots into commercial customers today if they were already ready."

This inverts the typical go-to-market risk profile. The business risk is entirely on the supply and technology side, not adoption. For investors, this is a meaningful signal that market timing is not the concern — technical execution is.

---

3. Companies Identified

---

4. People Identified

---

5. Operating Insights

Insight 1: Use Early Commercial Deployments as Model Refactoring Events, Not Revenue Events

The BMW pilot wasn't primarily a revenue milestone — it was a forcing function for rethinking Figure's entire AI commercialization approach. The six-month real-world deployment exposed gaps that a lab environment couldn't replicate, directly producing Helix 2.

"We refactored our whole approach to how to commercialize the software and AI systems after that."

Takeaway for operators: Treat early customer pilots as structured learning loops with explicit commitments to refactor based on findings. The commercial value is secondary to the data and insight.

---

Insight 2: Set Ruthlessly Specific Reliability Benchmarks Before Claiming Product-Market Fit

Adcock refuses to declare the home robotics use case ready without a specific, unambiguous bar. This kind of precision prevents premature scaling and keeps the engineering team anchored to real-world utility.

"I wanna be able to put a robot into a home and be able to do seven to 10 hours of work successfully without failures with no human intervention, and do that every day forever."

Takeaway for operators: Define your minimum reliability threshold in concrete operational terms — not user satisfaction scores or demo success rates — before committing to broad deployment or consumer launch.

---

Insight 3: Treat IP Security as a First-Principles Operating Decision, Not a Compliance Checkbox

Figure's security posture — phones covered on entry, restricted facility areas, tinted glass after a drone surveillance incident — reflects a founder-level conviction that IP is the core asset and must be protected with the same intensity applied to product development.

"What we're doing is like a very high IP risk. We really think carefully about our engineering, CAD & software, making sure it's very secure from a cybersecurity perspective and internal security perspective."

Takeaway for operators: In deep tech, the security posture should be set by the CEO based on competitive exposure, not delegated entirely to IT or legal. The threat model should inform physical facility design, not just digital infrastructure.

---

6. Overlooked Insights

Overlooked Insight 1: Figure's Hardware Reliability Curve Is a Compounding Signal, Not Just a Progress Update

The fault rate progression across hardware generations — Figure 1 faulting every hour, Figure 2 faulting daily, Figure 3 faulting weekly across the full fleet — is a quietly significant data point. It suggests a compounding improvement curve in physical reliability, which is historically the hardest axis to move in robotics. Investors and observers focused on software benchmarks may be underweighting this signal.

"Figure 1 ran for about an hour before faulting. Figure 2 reduced fault rates to roughly once a day. Figure 3... sees faults across the fleet on a weekly basis rather than daily."

---

Overlooked Insight 2: Figure May Have Indirectly Accelerated OpenAI's Robotics Ambitions

A detail buried in the OpenAI discussion: Adcock suggests that OpenAI's growing interest in robotics was part of why he terminated the partnership — not just that Figure had outpaced them technically.

"I think there was also like some interest as OpenAI was watching us getting into robotics. And so I fired him."

The implication is that the collaboration gave OpenAI front-row visibility into how Figure was solving physical intelligence. The termination may have been as much about competitive information containment as it was about model quality. This raises a broader strategic question for any deep tech company co-developing AI with a frontier lab: at what point does your partner become your competitor?