The $1/Hour Worker: Four Robotics CEOs on Humanoids at Home, China’s Threat, and the End of Dangerous Jobs

概览

This episode is a robotics-focused interview series recorded at the Makina conference in Paris, framed by Jason Calacanis as “AI in the real world.” The discussion moves from industrial quadrupeds to home humanoids, warehouse robots, supply-chain sovereignty, labor economics, and defense risks.

The clearest commercial theme is that robotics is moving beyond demos. Anybotics and Boston Dynamics emphasize inspection robots already deployed in hazardous industrial environments, while Agility Robotics describes Digit moving toward broader warehouse workflows. 1X presents Neo as both a home humanoid and an open platform for developers, teleoperation, and future embodied AI.

A recurring conclusion is that robots are valuable first where they can do dull, dirty, dangerous, remote, or repetitive work better than humans. The guests repeatedly distinguish current, narrow deployments from longer-term visions of general-purpose autonomy, world models, and eventually robots helping build robots, data centers, chip fabs, and other physical infrastructure.

分段落总结

[00:00] Paris Robotics Special

[事实] Jason opens from Paris at the Makina conference, describing the setting as focused on “AI in the real world.”

[事实] The episode is structured as a set of interviews with leading robotics companies rather than a single roundtable discussion.

[推测] The framing positions robotics as the next practical frontier after software AI, with emphasis on where robots already create measurable value.

[00:55] Anybotics and the Quadruped Form Factor

[事实] Jason introduces Dr. Peter Funkhouser, co-founder and CEO of Anybotics, and describes the company’s four-legged robot as an industrial inspection solution.

[事实] Funkhouser says four legs provide mobility, balance, stability, and the ability to climb stairs or move through difficult environments where people can go.

[事实] He argues that humanoids may fit narrow spaces and eye-level work better, while four-legged robots are better suited for industrial facilities with enough room and demanding ground conditions.

[03:12] Industrial Inspection as the First Strong Use Case

[事实] Anybotics’ robots are described as expensive industrial systems, costing in the low hundreds of thousands of dollars with service costs on top.

[事实] Funkhouser says the goal is not simply labor replacement, but “superhuman” inspection using thermal cameras, acoustic microphones, gas sensors, AI, and other sensors.

[事实] He says downtime in critical assets can cost customers hundreds of thousands of dollars per hour, so saving even minutes or hours can pay for the robot.

[推测] The business case depends less on replacing an inspector’s wage and more on preventing rare but very costly failures.

[04:41] Battery, Docking, and Mission Frequency

[事实] Funkhouser says the robot can run missions of about two hours and return to a docking station to charge.

[事实] Some customers run missions 40 times a day because they need information at specific high-risk moments, such as when an electric arc furnace activates.

[事实] He says customers want hands-free autonomy and mainly care about the data and insights, not the robot itself.

[05:24] Edge Compute Versus Cloud Analysis

[事实] Anybotics uses both onboard and cloud compute.

[事实] Funkhouser says real-time functions like obstacle avoidance and data-quality checks must run on the robot because connectivity cannot be guaranteed.

[事实] He says cloud systems are useful for contextual analysis, historical downtime analysis, and broader interpretation after data is collected.

[06:11] Harsh, Remote, and Explosive Environments

[事实] Funkhouser cites offshore oil, offshore wind transformer stations, Norway’s cold, desert heat, dust, humidity, and explosive atmospheres as relevant deployment environments.

[事实] He says Anybotics has a robot designed not to create sparks, making it suitable for oil, gas, chemical, and methane-risk settings.

[事实] Jason and Funkhouser discuss gas leaks and dangerous oil-and-gas environments as places where robots should be sent instead of people.

[推测] Anybotics’ strongest near-term market is infrastructure where human access is dangerous, expensive, or operationally disruptive.

[08:26] From Detecting Problems to Fixing Them

[事实] Jason asks when inspection robots will carry tools to fix leaks or other problems after detecting them.

[事实] Funkhouser says customers ask for this, but reliable real-world repair in explosive or harsh environments is not ready today.

[事实] He describes early steps such as closing levers and opening cabinets, while future systems may need bimanual manipulation or even more arms.

[推测] Manipulation is presented as the next major capability after mobility and sensing, but reliability requirements make it much harder than demos suggest.

[09:14] Supply Chain, China, and Data Trust

[事实] Funkhouser says zero percent of Anybotics’ robot is sourced from China, based on historical local sourcing and customer requirements.

[事实] He acknowledges China as a leading region for robotics hardware but argues that Chinese devices today are often hardware platforms rather than full inspection solutions.

[事实] He says Anybotics competes through autonomy, inspection intelligence, workflow integration, cybersecurity certification, and trust in sensitive infrastructure data.

[推测] The discussion treats robot sovereignty and data leakage as central issues when robots carry cameras and sensors inside critical infrastructure.

[11:21] Military Use and Weaponization Concerns

[事实] Funkhouser says Europe has a responsibility to build defense technologies, but Anybotics is not currently pursuing military applications.

[事实] He says military robotics would require different communications, autonomy, application software, and teams.

[事实] Funkhouser says he personally dislikes weaponized robots and references a letter with Boston Dynamics and others condemning robot weaponization.

[推测] The interview draws a distinction between defense responsibility and active weaponization, but leaves open pressure from geopolitical competition.

[15:05] 1X Neo, Home Humanoids, and 2026 Shipping

[事实] Jason introduces Bernd, founder and CEO of 1X, whose company makes the Neo household robot.

[事实] Bernd says 1X intends to ship Neo in 2026, but only to a handful of customers at first and with expectations managed.

[事实] He says 1X had different payment models, including an early-adopter full-payment option and a subscription model, and Jason references $500 per month.

[事实] Bernd says the first 10,000 preorders sold out in the first few days.

[18:00] Early Autonomy and Teleoperation

[事实] Bernd says getting a home humanoid in 2026 will be “rough around the edges” and that the robots may fall.

[事实] He says 1X did not initially want to promise full autonomy, but current progress suggests the company may ship something close to a fully autonomous and useful experience.

[事实] He adds that teleoperation or system guidance may still be involved when users want everything to work out of the box.

[推测] 1X is presenting teleoperation not as a failure mode, but as a bridge toward autonomy and a source of useful data.

[19:40] Neo as an Open Platform

[事实] Bernd says 1X plans to allow developers and other groups to build on Neo.

[事实] He describes a platform including a robot operating system, fleet management, data-collection equipment, and tools for fine-tuning models.

[事实] He says 1X will allow other people’s models to run on Neo, while also believing 1X’s own model should be the best.

[推测] Neo is positioned less as a single appliance and more as a hardware layer for an ecosystem of skills, models, and applications.

[22:11] Teleoperation as Remote Presence

[事实] Bernd gives a personal use case: using Neo as a remote avatar to be present in Norway while he is traveling.

[事实] He says teleoperation can let a person walk around, inspect parts, speak with colleagues, and join meetings through the robot.

[事实] He also describes remote power stations where a robot could sit in a closet and be operated when a rare issue occurs.

[推测] Teleoperation may remain valuable even after autonomy improves, especially for rare, expert, or high-context tasks.

[24:22] Data Pyramid and Human-Like Embodiment

[事实] Bernd describes a data pyramid: high-quality teleoperation data at the top, human sensor data below that, egocentric video next, and general video data at the bottom.

[事实] He says general video data is vastly larger than other robotics datasets.

[事实] He argues Neo’s similarity to a human is important because it lets 1X train on videos of humans doing tasks.

[事实] He says 1X is “all in” on pre-training its own models on internet video data.

[29:34] Hard Takeoff and Physical AI

[事实] Jason asks when robots become recursive, teaching and building themselves.

[事实] Bernd says he is extremely sure robotics is less than a decade away from “hard takeoff,” which he defines as robots building robots, data centers, chip fabs, and doing mining and refining.

[事实] Bernd says his current bet is three years, while acknowledging that even ten years would be brief in human history.

[推测] This is the episode’s most aggressive forecast, and it depends on both AI progress and the ability to scale physical robot deployment.

[33:46] Boston Dynamics Moves from Research to Deployment

[事实] Jason introduces Amanda McMaster of Boston Dynamics and reviews the company’s ownership history, including Google, SoftBank, and Hyundai.

[事实] McMaster says Boston Dynamics is no longer a lab or research-and-development company, but is focused on real-world deployment.

[事实] She says Spot has over 500 customers across more than 46 countries and is the most deployed and most utilized mobile autonomous robot.

[35:16] Spot’s Use Cases and ROI

[事实] McMaster says Spot is used heavily for industrial inspection, including acoustic inspection, gauge reading, vibration detection, asset monitoring, and security perimeter work.

[事实] She says customers need to see ROI in under two years.

[事实] Jason notes that these robots collect data through video, vibrations, acoustics, and other sensors in facilities such as infrastructure sites and pipelines.

[推测] Boston Dynamics’ commercial story mirrors Anybotics: the robot is valuable when it produces reliable operational insight at scale.

[37:35] Boston Dynamics Business Model and Reliability

[事实] McMaster says Spot began with a CapEx model, while Atlas will likely use a robot-as-a-service model.

[事实] She says Spot ranges from about $100,000 for the base robot to about $300,000 when fully loaded with services and integration.

[事实] She says Spot’s battery runs about 90 minutes, docks to recharge, and has a mean time between intervention above 3,000 hours.

[事实] She says Atlas has swappable batteries and can replace one while keeping another as backup.

[39:34] Augmenting Labor and Avoiding Dangerous Work

[事实] McMaster says Boston Dynamics has not primarily framed Spot as labor replacement, but as a way to augment human labor and move people toward knowledge-worker tasks.

[事实] She says some human inspection tasks were not actually being done consistently even when assigned.

[事实] She gives the example of finding an air leak that could otherwise cost millions of dollars per day.

[事实] Jason and McMaster discuss the robotics “three D’s”: dull, dirty, and dangerous work.

[41:05] Two Brains: Robot Control and Reasoning

[事实] McMaster says Boston Dynamics thinks of robots as having two brains: one for physical control and one for reasoning.

[事实] The physical-control brain for movement, manipulation, and reliability lives on the robot.

[事实] The reasoning layer for semantic understanding can be in the cloud and may involve partners such as Google DeepMind or other AI companies.

[推测] This division reflects a practical architecture: latency-sensitive control remains local, while higher-level interpretation can be remote.

[42:28] China, Robot Sovereignty, and National Strategy

[事实] McMaster says 100% of the robot is built outside China and Taiwan.

[事实] When asked whether Chinese humanoid robots should be allowed in the United States, she says no, citing safety and data-leak concerns.

[事实] She says the United States needs a coordinated effort to protect IP and build robotics manufacturing in the U.S. or allied countries.

[推测] The episode treats robotics as a strategic industry comparable to semiconductors, not just another consumer hardware category.

[44:01] Boston Dynamics and Military Applications

[事实] McMaster says Boston Dynamics has a public anti-weaponization stance.

[事实] She says the company is focused on industrial markets and is willing to do non-weaponized government work.

[事实] She identifies explosive ordnance disposal as a current and acceptable robotics use case.

[事实] When Jason asks what happens if China deploys armed robots, McMaster says Boston Dynamics would make the right decision if that time came, but today remains focused on commercial applications.

[47:16] Agility Robotics and the Last Two Decades

[事实] Jason introduces Professor Jonathan Hirst, co-founder and chief robot officer at Agility Robotics.

[事实] Hirst says robotics 20 years ago was mostly an unknown in industry, while research groups worked on humanoids, autonomous mobile robots, intelligence, and hardware.

[事实] He says robotics is now breaking through into real-world impact, while university programs and student demand are growing exponentially.

[49:10] Why This Robotics Cycle Is Different

[事实] Hirst says it is easy to make a robot that looks like a person, but hard to make one that does useful things in human spaces.

[事实] He says perception is almost solved, which is a major inflection point.

[事实] He says AI is enabling broader context awareness, helping people see that robots will soon do many useful things.

[推测] The difference from earlier humanoid cycles is not appearance but the combination of perception, usefulness, and deployable workflows.

[50:48] Robot Control Data Is Still Missing

[事实] Hirst says language models are becoming a commodity-like rising tide, but robot-control data does not exist at internet scale.

[事实] He says robots need data for torque commands, motor control, and sensor input, which must be generated somehow.

[事实] He lists learning from demonstration, teleoperation, animation input, motion capture, world models, and sim-to-real transfer as tools for robot learning.

[推测] The discussion pushes back on the idea that language models alone solve robotics.

[52:21] World Models and the Sim-to-Real Gap

[事实] Jason describes world models as simulated environments where robots can practice many iterations without breaking real-world objects.

[事实] Hirst says world models are part of the solution, but there is no silver bullet.

[事实] He says real-world dynamics such as condensation, wave dynamics, imperfect robot modeling, and object variation still make physical practice necessary.

[推测] Simulation can accelerate training, but real robot data remains essential for reliable deployment.

[53:23] No Single Singularity, but Shared Learning

[事实] Hirst says he does not believe in a single singularity moment, but does believe robots will keep improving as more resources and engineering effort go into the field.

[事实] He says humans learn from little data, while robots currently require far more data and examples.

[事实] He notes that robots have a long-term advantage: once one robot learns a skill, that learning can be uploaded to other robots of the same type.

[推测] Robotics progress is framed as compounding and distributed rather than one sudden breakthrough.

[55:19] Digit’s Current Warehouse Workflows

[事实] Hirst says Agility’s Digit currently handles reasonably scoped multipurpose workflows such as picking up bins and totes and carrying them around.

[事实] He says the humanoid form factor is justified when the robot needs two arms, whole-body control, balance, and the ability to work in narrow human spaces.

[事实] He says Digit’s broader value is versatility across use cases such as each picking, bin filling, carrying, palletizing, and depalletizing.

[56:16] Digit V5 and Safety Without Barriers

[事实] Hirst says Digit V5 is coming later in the year and is intended to step out of a work cell without needing a physical barrier between robot and person.

[事实] He says Agility’s work with Amazon showed that doing the task was not enough; the robot also had to meet safety requirements.

[事实] He says safety required a bottom-to-top redesign across robot systems.

[推测] For warehouse humanoids, safety certification may be as important as raw capability for scaling deployments.

[57:37] Robot Economics and Robot-as-a-Service

[事实] Hirst says robot costs are coming down and may eventually be in the range of cars.

[事实] He says Digit V5 can work about 20 hours out of 24 because of fast charging.

[事实] Jason calculates that 20 hours per day over five years can create roughly 40,000 hours of work.

[事实] Hirst says Agility offers both CapEx and robot-as-a-service models depending on customer preference.

[60:12] Automation, GDP, and Where Humanoids Fit

[事实] Hirst says many factories already have more robotic systems than human workers when including AMRs, conveyor belts, and industrial robot arms.

[事实] He says more automation is needed to increase U.S. GDP because population growth is not the driver.

[事实] He argues humanoids are useful for walking into human environments and doing human workflows, while fully specialized automation may be better for fixed 24/7 tasks.

[推测] Humanoids are presented as a flexible layer for legacy and human-centered environments, not a replacement for every form of automation.

[62:30] Last-Mile Delivery and Future Applications

[事实] Hirst says Agility explored package delivery with Ford as one of its first use cases, including a video of Digit leaving a vehicle and delivering a package to a porch.

[事实] He says that use case is on the roadmap but is not the best first market.

[事实] He names retail, grocery stores, hospitals, construction sites, and front-door package delivery as future human-environment opportunities.

[63:31] The Three D’s and World-Positive Work

[事实] Hirst says picking things up and putting them somewhere else is a huge use case that can free people from dull, dirty, and dangerous jobs.

[事实] He hopes future generations will look back at some current jobs the way people now look back at coal mining in the 1900s.

[推测] The moral case for robotics in the episode rests on moving humans away from physically harmful and low-quality work.

[64:17] Robotics Careers and Skilled Trades

[事实] Hirst says robotics is a major opportunity for young people because times of change favor students who can learn new systems.

[事实] He recommends core engineering skill sets as broadly useful.

[事实] He says there will also be non-PhD paths, including robot operators, robot assembly, manufacturing, deployment, and maintenance.

[推测] The episode frames robotics as both a high-end research field and an emerging skilled-trade labor market.

[65:37] Robot Form Factor, Sci-Fi, and Safety

[事实] Jason asks why robots do not have four or six arms like science-fiction characters.

[事实] Hirst says one arm is not enough for large objects, two arms are useful, but additional arms add complexity and need a real use case to justify them.

[事实] Hirst names WALL-E and Baymax as favorite science-fiction robots and connects Baymax to the importance of safeguards.

[事实] He says industrial safety requires supervisory circuits, emergency stops, and systems designed so robots cannot harm people.

播客点评/总结

[推测] The episode is most valuable as a snapshot of robotics becoming commercial: the strongest sections are the concrete details on inspection robots, pricing, uptime, docking, ROI, safety requirements, and warehouse deployment.

[推测] Its main limitation is that several bold forecasts, especially around home humanoids and “hard takeoff,” are presented in interview format without outside validation. Those claims are exciting but should be read as founder outlooks rather than established facts.

[推测] The episode is best suited for listeners interested in robotics startups, industrial automation, embodied AI, labor economics, and technology geopolitics. It is less useful for someone looking for a technical deep dive into robot-control algorithms or safety certification standards.