Nikhil KamathHumanoids Cost as Much as an SUV Now | Nikhil Kamath x Brett Adcock | WTF Online Ep 2
CHAPTERS
Figure AI office tour + Brett’s mission: “do everything a human can”
Nikhil opens with provocative questions about kids and safety, then Brett introduces Figure AI from inside the office with multiple robot generations visible. He frames Figure’s core goal as building general-purpose humanoids that can operate in human-designed environments.
From Midwest farm to startups: Vettery → Archer → Figure
Brett recounts his background and entrepreneurial path, from an AI recruiting marketplace to electric aviation, then into humanoid robotics. He highlights the throughline: building complex hardware-software systems at scale.
When flying taxis become practical: tech vs certification bottlenecks
Nikhil probes the real timeline for electric flying taxis in India and elsewhere. Brett argues the technology is flying today, but certification, safety, and policy are the main gating factors; he predicts paying passengers within ~5 years.
Electric aviation’s longer arc: short hops now, regional/long-haul later
They discuss whether electric air travel expands beyond city hops to routes like SF–LA and eventually long distances. Brett believes aviation broadly electrifies over time, but longer-range electrification requires different aircraft designs and tech.
Why Brett moved from aircraft to humanoids: the “ultimate meta problem”
Brett explains that an aircraft is essentially a flying robot, sharing core components like motors, batteries, sensors, and control. He argues humanoids are the most general form factor because the world’s tools and spaces were designed around the human body.
Are we ready for humanoids? Sci‑fi vibes, real constraints, near-term progress
Nikhil describes the uncanny, potentially dystopian feeling of humanoids in daily life. Brett says it’s not scary to the team because they know the limits, but acknowledges the world may feel unready; he emphasizes rapid progress with neural-network-controlled tasks.
What’s inside a humanoid: motors, battery, onboard compute, sensors, cameras
Brett breaks down the robot’s physical and compute architecture, then builds up to perception. He emphasizes high-rate onboard control loops, force/torque sensing, and camera-based perception without lidar.
Onboard vs cloud intelligence: fast reflexes local, “big brain” offboard
Nikhil questions whether compute limits force reliance on networks. Brett argues core control must remain onboard for latency, but cloud reasoning can complement; Figure already runs commercial deployments fully onboard while still benefiting from fleet training.
Energy efficiency + long-horizon autonomy: from today’s gaps to human-like shifts
Nikhil explores whether humanoids can become more energy-efficient than humans. Brett says robots are currently less efficient, but improving mechanically and computationally; the bigger benchmark is long-horizon task completion with robust recovery from failures.
Next AI form factor: voice-first interfaces + new AI-native devices
They pivot to how humans will interact with AI beyond phones and computers. Brett predicts voice becomes the dominant interface once models reach higher quality and richer context; he expects AI-native devices and humanoids to replace pre-AI hardware paradigms.
Data flywheel: fleet learning, real-world data vs synthetic, and transfer across tasks
Nikhil challenges whether robot learning is disadvantaged versus internet-scale text. Brett argues physical-world data is vast and essential for true intelligence; Figure collects human demonstrations to train Helix and sees surprising transfer between unrelated tasks.
Kids, safety, and design language: avoiding both menace and “goofy toy” aesthetics
Nikhil presses on whether Brett would leave kids alone with a humanoid and whether robots need faces/expressions. Brett says they’re not at unattended-kids safety yet; Figure prioritizes utility and trustworthy industrial design over human mimicry, while acknowledging ‘Westworld’ is feasible.
Industry landscape: hardware-only vs software-only bets, teleop skepticism, and China debate
Brett critiques the field’s split between hardware sellers and AI-only labs, arguing full-stack integration is required. He is skeptical of competitor demos that rely on teleoperation and rejects the idea that China is ahead on the hardest part: general-purpose autonomy.
Commercial deployments + the first real problems: factories now, homes sooner than expected
Brett describes real commercial operation: Figure 02 working daily in an auto context, primarily to learn reliability and operations. He also reverses his prior view and now believes the home is nearer-term than expected, largely constrained by data collection rather than feasibility.
Why Figure ended its OpenAI partnership + how AI money becomes revenue
Nikhil asks about the terminated OpenAI partnership and broader AI monetization. Brett says Figure left because internal robotics-model execution was stronger and hiring was hindered by perception of outsourcing; he expects AI and humanoids to drive massive long-term GDP impact despite slow near-term revenue realization.
Investing, what to build, and society after humanoids: abundance vs purpose crisis
They close on investing ideas, startup advice for Indian builders, and the societal consequences of synthetic labor. Brett argues components, data, compute, and supply chain are big opportunities; long term, goods/services prices collapse toward abundance, creating both freedom and existential questions about purpose and jobs.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome