NEXT with John Koetsier

John Koetsier
  • 4.5 (15)
  • Tech News
  • Updated Weekly

Deep tech conversations with key innovators in AI, robotics, and smart matter ...

  1. 6d ago

    1,000 humanoid robots. 1 factory.

    What does it actually take to put 1,000 humanoid robots to work? Hexagon Robotics President Arnaud Robert joins me to explain how the company is moving its Aeon humanoid from pilots and demonstrations into real industrial production, including a plan with Schaeffler that could eventually scale to 1,000 robots. And the big lesson is that scaling humanoids isn't about taking one robot doing one task and multiplying it by 1,000. It's about building a multipurpose fleet that can move between jobs as demand changes across a factory. We talk about: • Why humanoids need to be multipurpose to justify themselves • How Aeon gathers training data while doing real work • Hexagon's “sim-to-real-to-sim” training loop • Why customer-specific factory data matters • The difference between a successful pilot and a production-ready robot • The two metrics Hexagon cares about most: cycle time and human intervention • What Hexagon learned from early sensor-fusion and actuator challenges • Why Aeon uses wheels instead of legs — and can still climb stairs • How humanoids could eliminate bottlenecks instead of simply replacing workers • Schaeffler's “robot gym” and the train-deploy-scale path to 1,000 humanoids Robert's view of the factory of the future isn't a lights-out facility with no people. It's a highly autonomous factory where humans and robots work together — with robots handling repetitive tasks and shifting between bottlenecks while skilled workers focus on problem-solving. This is what humanoid robotics looks like when the conversation moves beyond prototypes and into production. 00:00 What does it take to deploy 1,000 humanoids? 01:31 Why 1,000 robots changes the problem 03:02 Why humanoids need to do multiple jobs 03:55 The race for physical AI training data 04:51 Simulation, synthetic and real-world data 06:20 Hexagon's sim-to-real-to-sim loop 08:10 What robot-generated training data captures 10:21 Pilot vs. production-ready humanoids 12:14 The industrial experience advantage 13:12 The two metrics that really matter 15:39 What Hexagon learned from failure 17:35 The humanoid actuator problem 18:47 Why Aeon uses wheels instead of legs 19:59 Yes, a wheeled humanoid can climb stairs 21:28 What the factory of the future looks like 23:42 Using humanoids to eliminate bottlenecks 25:17 Automation exposes the hidden 3% 26:07 Schaeffler's path to 1,000 humanoids 27:18 Train, deploy, scale

  2. Sep 23

    He runs 100s of AI agents ... here’s how

    What does it actually look like to work with hundreds of AI agents? In this episode of NEXT with John Koetsier, I chat with Steve Ancheta, founder and CEO of Zig.ai, an AI-native platform for relationship-driven sales. Steve explains why what looks like a single AI agent to a user can actually be a swarm of hundreds of specialized agents working behind the scenes, handling copywriting, follow-ups, CRM updates, next-best actions, orchestration, and more. We also dig into how Steve personally uses AI agents as force multipliers across his work, why he still writes important investor and customer messages himself, and why simply giving an agent access to documents isn’t enough. Steve shares his approach to building an effective agent harness, including two critical pieces many people overlook: governance and grounded truth. We also discuss OpenClaw, NanoClaw, cloud-based agents, agent security, human-in-the-loop workflows, critical thinking, AI productivity, and the shift from reactive agents that wait for instructions to proactive agents that start doing useful work on their own. Topics include: - Why Steve believes many beginners shouldn’t build agents yet - How hundreds of specialized agents can operate as one system - AI agents as force multipliers rather than human replacements - Why human taste and connection become more important as AI improves - How Steve uses agents for planning, operations, scheduling, and analysis - The dangers of AI-generated documents nobody actually reads - Agent governance, guardrails, and permissions - Giving agents a reliable source of truth - Why vector databases alone aren’t a magic solution - Balancing human interaction with AI-assisted work - The move from reactive AI agents to proactive AI agents - Why humans should remain in the loop for external actions Guest: Steve Ancheta Company: Zig.ai 00:00 Why beginners may not be ready for AI agents 00:24 Working with AI agents every day 01:34 Steve Ancheta’s journey into AI 03:00 The early days of agent workflows 04:36 OpenClaw, Hermes, and the agent boom 05:37 The security problem with early AI agents 06:00 Why Steve moved agents to the cloud 07:00 Local agents vs. cloud-based agents 08:00 How quickly agent technology is evolving 09:00 How Zig.ai uses hundreds of AI agents 10:00 Why one “agent” may actually be an entire swarm 11:00 Specialized agents and orchestration 11:30 What Steve personally uses AI agents for 12:00 Why he still writes important emails himself 13:00 AI agents as force multipliers 14:00 Why AI won’t eliminate the human element 15:00 Human connection and taste matter more than ever 15:50 Steve’s favorite uses for AI agents 16:00 The danger of AI-generated documents 17:15 Agents for planning, operations, and forecasting 18:00 Using agents to find problems across a company 18:40 AI as a chief of staff 19:25 Why beginners probably shouldn’t build agents yet 20:00 What an agent harness actually does 21:00 Zig.ai is fundamentally a data company 22:00 Giving AI agents a reliable source of truth 23:00 Governance, guardrails, and agent permissions 24:00 Why vector databases aren’t enough 24:35 How AI agents change the way we work 25:20 Balancing AI time with human interaction 26:00 Why Steve deliberately spends time away from technology 27:00 The next phase: proactive AI agents 28:00 Keeping humans in the loop 29:00 When AI agents start giving us tasks

  3. Sep 10

    2035 Homes: 1 Humanoid Robot or 10 Specialized Robots?

    What will the home of 2035 actually look like: One humanoid robot that does everything, or 10 specialized robots, each handling one specific task extremely well?In this episode of NEXT with John Koetsier, I talk with Amber Atherton, serial entrepreneur and investor at Patron, about a very different vision for the future of home robotics.Instead of assuming that every household will eventually have a general-purpose humanoid assistant, Amber argues that consumer robotics may scale first through highly specialized devices that quietly take over specific routines and chores.Think skincare. Hair. Gardening. Cleaning. Wardrobe management. Kitchen tasks. Pet care.In other words, the future home may be full of robots without looking like it’s full of robots.We discuss why Roomba is still such an important example of successful consumer robotics, why specialization often wins in consumer markets, what people may actually be willing to pay for useful home robots, and how falling hardware costs plus better AI models could create a new wave of robotics startups.We also get into the bigger platform question: if homes eventually contain dozens of physically intelligent devices, who owns the operating system underneath them?Google DeepMind? NVIDIA? A new robotics platform? Or an ecosystem we haven’t seen yet?And, of course, we debate whether humanoids ultimately win anyway.This episode is cross-posted from my Humanoid Daily podcast, where I cover the companies, technologies, investments, and ideas shaping the humanoid robotics boom.Topics include:• Humanoid robots vs. specialized robots • What home robotics could look like by 2035 • Robots for hair, skincare, gardening, and household chores • Why Roomba remains such an important robotics success story • Consumer robotics business models • Pricing home robots • AI foundation models and physical AI • Robotics manufacturing and supply chains • Robot training data • The future operating system for physical AI • Google DeepMind, NVIDIA, 1X, Figure, and Apptronik • Whether the future home needs a humanoid at allNEXT with John Koetsier explores what’s coming next in technology, business, AI, robotics, and the future of the world around us.

  4. Aug 28

    Quantum computing + AI + agents = AGI?

    What happens when you combine frontier AI, agentic systems, and quantum computing? In this episode of NEXT with John Koetsier, we chat with Mykola Maksymenko, co-founder and CTO of Haiqu, about how AI agents are dramatically accelerating quantum research and potentially scientific discovery as a whole. Maksymenko shares how an AI system was able to reconstruct months of his own PhD research in a fraction of the time, even identifying a bug in one of his formulas. He also discusses experiments involving genomics, molecular simulation, quantum chemistry, and condensed matter physics. The bigger question: do we really need to wait for fault-tolerant quantum computers with hundreds or thousands of logical qubits before quantum computing becomes useful? Maksymenko argues that useful quantum applications are already emerging today, particularly when AI agents help scientists discover algorithms, orchestrate workflows, and handle the complexity of working with noisy quantum hardware. We also explore the risks of increasingly capable AI systems, scientific guardrails, quantum utility versus quantum supremacy, and what happens when researchers can test ideas in a weekend that previously might have required months or years. Topics include: • Agentic AI for scientific research • Quantum utility vs. quantum supremacy • AI-assisted genomics • Quantum chemistry and molecular dynamics • Automating quantum software workflows • AI as a scientific collaborator and educator • The risks and guardrails of frontier AI • Why scientific discovery could accelerate dramatically 00:00 — “I think AGI is here” 00:40 — Meet Mykola Maksymenko 01:01 — Why a quantum physicist started experimenting with genomics 02:18 — Reproducing years of research with AI and quantum computing 03:07 — The convergence of AI, agents, and quantum computers 04:26 — Are we entering the singularity? 04:45 — AI reconstructs six months of PhD research 05:28 — The risks of AI-assisted genomics and scientific discovery 06:46 — Pandora’s box and the race for frontier AI 07:02 — What researchers are doing with the technology now 08:29 — Can quantum computers already do useful work? 09:21 — From quantum supremacy to quantum utility 11:06 — Molecular dynamics and quantum chemistry 12:12 — The minimum viable product of quantum computing 13:00 — Why Haiku moved deeper into agentic AI 13:30 — How much faster can AI make scientific research? 15:47 — AI as a scientist’s educator and collaborator 16:51 — Running research experiments over a weekend 17:16 — What happens next for AI + quantum computing 17:38 — How AI agents operate quantum computers 18:59 — Closing thoughts

  5. Aug 13

    AI can now edit DNA and create deepfake viruses

    AI is moving beyond text, images, and code. Now it’s learning to read and write -- shall we say program -- DNA.In this episode of NEXT, John Koetsier talks with Eric Nguyen, co-founder and CEO of Radical Numerics, about the rapidly emerging world of biological AI.Nguyen and his team helped create Evo and Evo 2, generative foundation models for DNA, and are now working toward what they call “general biological intelligence”: AI systems capable of understanding biology across DNA, gene expression, methylation, proteins, and other biological signals.The potential upside is enormous. These systems could help scientists detect cancer earlier, develop treatments for antibiotic-resistant superbugs, understand disease more deeply, and eventually design countermeasures to emerging biological threats on demand.But the same capabilities introduce serious risks.Nguyen explains how AI could potentially generate biological sequences that retain dangerous functions while evading traditional sequence-matching detection systems ... essentially creating biological “deepfakes.” He also discusses why AI labs need to develop biodefense capabilities alongside increasingly powerful biological design tools.The conversation covers DNA foundation models, AI-generated viruses, biosecurity, pathogen detection, wastewater surveillance, attribution of biological threats, antimicrobial resistance, cancer detection, open-source versus closed-source biological AI, and why biology may be the next major frontier for artificial intelligence.00:00 AI-designed DNA and “deepfake” viruses00:21 AI is moving into biology00:55 Meet Eric Nguyen of Radical Numerics01:20 Why DNA is a language01:39 Teaching AI to read and write DNA02:54 What happens when AI can create biological systems?03:15 The promise and risks of programmable DNA04:32 The medical upside of AI-driven biology04:58 Moving beyond single-molecule drug discovery06:30 Can AI model the complexity of the human body?06:59 Biology has more data than we know how to use08:38 Where all the DNA data comes from09:12 The dangerous side of AI-generated biology10:01 What is a “deepfake virus”?12:41 Could AI make pathogens more dangerous?14:36 Putting AI biodefense on the front lines16:56 The three pillars of biodefense18:22 On-demand treatments for new diseases19:11 How close is this future?21:53 Why biodefense capabilities are falling behind24:36 Should powerful DNA models be open source?25:40 How Evo and Evo 2 were made safer26:49 Why Radical Numerics is keeping Omni closed27:55 AI-generated bacteriophages and superbugs29:23 Balancing breakthrough biology with biosecurity30:53 Using AI to detect cancer earlier32:08 Why cancer detection needs multiple biological signals33:49 “Sensor fusion” for biology34:24 Why a holistic view of medicine matters

  6. Jul 30

    Do we really need 400 humanoid robot companies?

    Why aren’t hundreds of millions of intelligent robots already operating in the physical world? In this episode of NEXT with John Koetsier, John speaks with Seth Winterroth, partner at Eclipse, about the rapidly changing robotics investment landscape, the rise of physical AI, and the race to build the next generation of autonomous machines. They explore whether the world really needs hundreds of humanoid robotics companies, why timing matters as much as technology, and why some robotics startups may need to build the entire stack ... from hardware and embedded software to AI models, evaluation systems, and deployment infrastructure. The conversation also covers Genesis AI, Wayve, Project Prometheus, Apptronik, Figure, 1X, autonomous vehicles, delivery drones, industrial automation, surgical robotics, and the future of robots in the home. Topics include: • Why robots still aren’t widely deployed • The five forces driving robotics investment • Whether 400 humanoid robotics companies are too many • Full-stack versus platform-based robotics strategies • The challenge of achieving reliability and safety • When useful home robots may finally arrive • Why autonomous vehicles are already robots • The industries likely to adopt robotics first • The future of manufacturing, logistics, transportation, and surgery • What the next major robotics inflection point could be Seth Winterroth is a partner at Eclipse, an investment firm focused on companies transforming physical industries. Eclipse has backed robotics and automation companies including Genesis AI, Wayve, MiND Robotics, Foxglove, and Third Wave Automation. Subscribe to NEXT for more conversations about AI, robotics, emerging technology, and the companies shaping the future. 00:00 Why aren’t robots everywhere yet? 00:37 Introducing Seth Winterroth 01:03 The robotics investment landscape 02:00 Seth’s background in applied robotics 03:01 The five forces accelerating robotics 04:04 Are there too many humanoid robot companies? 05:00 Creative destruction in robotics 06:02 How investors pick the winners 06:38 Why robotics companies need velocity 07:00 The danger of being too early or too late 07:51 How Wayve benefited from entering later 08:36 Project Prometheus and the $10 billion seed round 09:05 Why Genesis AI is building full stack 10:01 Full-stack robotics versus foundation-model platforms 11:03 The dirty secret: Where are all the robots? 12:03 Reliability, safety, and deployment friction 12:39 The challenge of vertical integration 14:02 What robotics companies should build themselves 15:03 Capital requirements and the robotics J-curve 16:01 How robotics companies can scale rapidly 16:39 When humanoid robots may reach the market 17:04 Figure, 1X NEO, Unitree, and AgiBot 18:24 A broader definition of robotics 19:04 Are cameras and dishwashers robots? 19:39 Autonomous vehicles as embodied AI 20:23 Why transportation autonomy could be transformational 21:05 Why humanoids still have a long way to go 21:44 One general-purpose robot or many specialized machines? 22:35 What will make home robots successful? 23:39 Finding the minimum viable home robot 24:20 Physical and digital products 25:04 Why the App Store model matters for robotics 26:05 How robots will gain new capabilities over time 26:35 The brutal economics of consumer robotics 27:12 Consumers buy outcomes, not robots 27:48 Robotics beyond humanoids 28:18 Kiva Systems and the modern robotics era 29:05 Why constrained environments win first 30:02 Robotics in automotive manufacturing and logistics 31:02 Eclipse’s robotics portfolio 31:31 When robots will begin walking among us 32:03 The next major robotics inflection point 32:42 Escaping the trough of disillusionment 33:29 Why autonomous transportation is nearly solved 34:04 How self-driving changes cities and society 34:36 The future of robotic surgery 35:01 Delivery drones and regulatory barriers 35:39 Closing thoughts

  7. Jul 14

    Rhoda AI: 1000X less training data required?

    Can robots learn from the internet the same way ChatGPT learned from text? In this episode, Andrew Wooten, co-founder of Rhoda AI, explains why his company believes the future of robotics isn’t collecting millions of hours of robot data ... it’s learning from internet-scale video. Instead of relying on traditional vision-language-action (VLA) models that require enormous training datasets, Rhoda’s approach teaches robots physical intuition by predicting the future through video. We also explore why, in Andrew's opinion, warehouses and factories will likely be the first major market for humanoid robots (not homes!), why Rhoda chose a wheel-based humanoid design, how language models fit into physical AI, and how the company’s robots can learn complex tasks with just 8–10 hours of training data instead of 10,000+ hours. If you’re interested in robotics, AI, automation, or the future of manufacturing, this conversation offers a fascinating look at where physical AI is heading. In this episode: * Why warehouses beat homes as the first market for humanoid robots * Why Rhoda chose wheels instead of legs * The biggest limitation of today’s robot AI models * How internet-scale video teaches robots physics * Why predicting the future helps robots manipulate the real world * Edge AI vs. cloud robotics * The role of LLMs in controlling robots * How Rhoda cut robot training from 10,000+ hours to just 8–10 hours * When zero-shot robot learning could become reality Guest Andrew Wooten Co-founder, Rhoda AI Website: https://rhoda.ai 00:00 Why Humanoid Robots Don’t Have Wheels 00:18 Can Robots Learn From Internet Video? 00:42 Best Use Cases for Humanoid Robots 02:10 Why Warehouses and Factories Come First 04:02 The Economic Impact of Robotics 05:00 Home Robots vs. Industrial Robots 06:05 Why Rhoda AI Chose Wheels 08:00 Building a General-Purpose Robot 09:55 Why Full-Stack Robotics Companies Have an Advantage 10:40 The Evolution of Physical AI 12:20 Why Vision-Language-Action Models Fall Short 14:05 Training Robots With Internet-Scale Video 16:05 How Rhoda AI’s Video-Action Model Works 17:25 Edge AI vs. Cloud Computing 19:05 How Robots Develop Physical Intuition 20:40 Predicting the Near Future in Real Time 22:00 Can Robots Build a Subconscious? 23:10 Using Language Models to Control Robots 25:15 Rhoda AI’s Hardware Strategy 26:20 The Biggest Problems With Today’s Humanoids 28:20 When Will Robots Truly Learn on the Job? 30:05 Training Complex Tasks in 8–10 Hours 31:15 Zero-Shot Robot Learning and What Comes Next

  8. Jul 9

    Neo's amazing new hands - just released!

    My instant first look at 1X’s just-unveiled NEO robot hands. Hot take: they're a massive leap toward truly useful humanoid robots. With 25 degrees of freedom, tendon-driven actuation, tactile sensing, force feedback, and near-human dexterity, these hands can do far more than simple gripping. They can assemble LEGO, pick up tiny screws and coins, plug in a USB-C cable, zip jackets, use tools, sort delicate fruit, and even wash themselves. (Washable robot hands! That's not common.) In this video, I break down what makes these hands fundamentally different from traditional robotic grippers, why backdrivability and force transparency matter, and why hands may be the single most important component of practical humanoid robots. Topics covered: * Why most robot hands are still “numb” * How tendon-driven hands improve dexterity * Force sensing and tactile feedback explained * Precision handling of tiny objects * Strength, durability, and IP68 sealing * Why washable robot hands matter for home robotics * Over-the-air upgrades and the future of NEO If humanoid robots are going to become useful assistants in homes, warehouses, and workplaces, breakthroughs like this are what will make it possible.

See All (387)

4.5
out of 5
15 Ratings

About

Deep tech conversations with key innovators in AI, robotics, and smart matter ...

Information

  • Creator
    John Koetsier
  • Years Active
    2019 - 2026
  • Episodes
    387
  • Rating
    Clean
  • Copyright
    © John Koetsier
  • Show Website
    NEXT with John Koetsier

You Might Also Like