A Humanoid Robot in Every Home? It's Closer Than You Think w/ Brett Adcock (at A360 2025) | EP #156
Figure AI founder Brett Adcock explains why he moved from building Archer's electric aircraft to humanoid robots, arguing a humanoid body is the only viable deployment vector for AGI to act in the physical world. He walks through Figure's rapid hardware iteration, full vertical integration, and its first commercial deployments at BMW and a major logistics company, plus what he describes as unlimited demand from Fortune 100 companies. Adcock describes building Helix, Figure's own in-house vision-language-action model after stepping back from relying on OpenAI, and recounts a demo where two robots put away groceries neither had seen before. He lays out the economics (a targeted $20-30K robot price, a market he sizes at roughly half of global GDP) and says Figure 3, redesigned from scratch and about 90% cheaper, is headed into production this year with home alpha-testing starting in 2025.
Adcock argues solving AGI without giving it a physical body is a dystopian dead end, and that a humanoid is the only single platform that can do everything a human can with no hardware changes.
Adcock says first and second-generation hardware always underperforms, so Figure clean-sheet designs a new robot generation every 12-18 months rather than patching old designs.
Full vertical integration, no existing supply chainRobotics▶ 5:18
Figure builds its own motors, actuators, sensors, batteries, and software stack end to end because no commercial supply chain exists yet for humanoid robot components.
Figure's factory and first commercial deploymentsRobotics▶ 5:35
Adcock walks through Figure's design-build-test-integrate pipeline and describes robots running daily at a BMW plant and a second deployment with a major logistics company.
Pricing economics and total addressable marketEconomy▶ 8:10
Adcock holds to a $20,000-$30,000 robot price target at scale and frames the commercial workforce, worth roughly half of global GDP, as the real prize over the home market.
Team, culture, and recruiting for a moonshotOther▶ 10:46
Adcock credits Figure's progress to an intensely aligned team and culture built around one shared mission, deliberately run with a demanding, non-standard work culture.
The three unsolved problems Figure had to proveRobotics▶ 14:21
In 2022 Adcock told recruits the odds were low and that Figure had to solve three unprecedented problems within a few years: human-speed hardware, neural-net-based imitation learning, and generalization to unseen tasks via language.
After stepping back from baselining on OpenAI's systems, Figure built Helix, its own large-scale model, and showed two robots putting away unfamiliar groceries using a single shared neural net.
Adcock says Figure 3, designed from scratch over 18 months, is about 90% cheaper than Figure 2, smaller, and purpose-built for neural-net control, and is entering production this year.
Home vs. workforce robot deployment and demandEconomy▶ 21:54
Adcock contrasts the workforce track (easier, repetitive, higher margin) with the much harder home track, and says demand from commercial customers is effectively unbounded.
Predictions made
openBrett Adcock: At high-volume manufacturing, humanoid robots will sell near $20,000-$30,000.
“I think you're going to see it in the coming years being put into homes just through speech”
Your call:
Numbers that matter
Global GDP is roughly $110-120 trillion; Figure's addressable market is roughly $50-60 trillionHalf of global GDP is human labor, which Adcock sizes as the robot workforce opportunity.
$300/month, $10/day, 40 cents/hourAdcock's comparison: a $30,000 robot costs about the same as leasing a $30,000 car.
$1 million/month burn reached within six months of foundingHow fast Figure's self-funded spending ramped after Adcock committed to fund the company himself.
Helix was trained on only about 500 hours of dataYet it generalized to picking up household objects it had never seen, including unfamiliar toys.
Figure 3 is about 90% cheaper than Figure 2 after 18 months of from-scratch designAlso smaller, lower-mass, and built with sensors, hands, head, and feet designed specifically for neural nets.
BMW's sheet-metal task sped up from 4 minutes to 40 seconds since last summerEngineering progress on Figure's first commercial deployment.
BMW integration took a year end to end; the second customer took under 30 days; a repeat could take under 48 hoursShows how fast Helix cut new-customer deployment time.
100,000 robots of unmet demand from two customers; ~50 Fortune 100 companies could sign within a weekAdcock's description of current commercial demand for Figure robots.
Worth digging into
🕳️ Helix, Figure's in-house vision-language-action model
Trained on only about 500 hours of data, it let two robots put away groceries neither had seen before and spontaneously learned to make eye contact during handoffs, hinting general physical-world intelligence may be closer than assumed.
🕳️ Figure's break from OpenAI to build its own foundation model
OpenAI was an investor and Figure was baselining on OpenAI's AI systems before Adcock decided to go fully in-house with Helix, a strategic split that says something about foundation-model dependency in robotics.
🕳️ No existing supply chain for humanoid robot components
Adcock says there are no vendors yet for motors, actuators, sensors, or battery systems built for humanoids, forcing full vertical integration and potentially keeping competitors out for years.
🕳️ The economics of a $20-30K humanoid robot workforce
Adcock frames the robot as a $300/month lease replacing far more expensive labor; if that math holds at scale it reshapes roughly half of global GDP.