TechFirst 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. 23H AGO

    Teaching robots like humans: 1000 tasks in 24 hours

    Imagine teaching a robot 1000 tasks in just 24 hours. Imagine teaching robots just like you teach humans. In fact, what if teaching a robot were as easy as showing it once? Humans can learn new skills almost instantly by watching, trying, or receiving a quick explanation. Robots, historically, haven’t been so lucky. Training them often requires huge datasets with real or virtual data, massive engineering effort, and weeks or months of experimentation. But that may be changing. In this episode of TechFirst, host John Koetsier talks with Edward Johns, Director of the Robot Learning Lab at Imperial College London, about a breakthrough in efficient imitation learning that allowed a robot to learn 1,000 different tasks in just 24 hours. Instead of collecting huge datasets, Johns’ team combines simulation training, clever algorithm design, and single demonstrations to dramatically speed up how robots learn. We discuss: • How robots can learn from just one demonstration • Why breaking tasks into “reach” and “interact” phases makes learning faster • The role of simulation data in robotics AI • Why robotics doesn’t have the same data advantage as large language models • The future of prompt-like robot training • Whether humanoid robots will actually learn like humans As robotics hardware rapidly improves and costs fall, breakthroughs like this could be the key to making robots truly useful in homes, factories, and everyday life. If robots are going to become real collaborators with humans, they’ll need to learn quickly ... just like we do. ⸻ Guest Edward Johns Director, Robot Learning Lab Imperial College London https://www.imperial.ac.uk ⸻ Subscribe for more conversations on AI, robotics, and the future of technology: https://techfirst.substack.com 00:00 Can robots learn as fast as humans? 00:51 Teaching a robot 1,000 tasks in 24 hours 01:08 The two-phase learning approach 02:14 Old-school robotics vs. machine learning 03:29 The robotics data bottleneck 04:47 The challenge of dynamic environments 06:04 The coming wave of robot data 06:59 Why robots must be teachable by users 08:08 Why LLM-style scaling is harder in robotics 09:42 Prompting robots with demonstrations 10:54 Probabilistic robot behavior and safety 12:20 What robots can do today 13:53 Why hardware precision still matters 16:53 When this reaches the real world 17:59 Humanoids that look human vs. learn human 18:40 The robotics boom around the world 22:34 The risk of scaling too early 23:46 Faster learning vs. more data 26:20 The next frontier in robot learning

    24 min

  2. FEB 27

    Giving AI a human soul

    Can we give an AI human emotions? A soul? Can AI truly feel, or will it just act like it does? In this episode of TechFirst, I talk with Vishnu Hari, founder and CEO of Ego AI (backed by Y Combinator and former AI product manager at Meta), about building emotionally intelligent AI characters that persist across games, Discord, chat, and even physical robots. Vishnu survived a violent attack in San Francisco that left him partially blind with a traumatic brain injury. During recovery, as he felt his own neural pathways healing, he began asking a deeper question: If humans are “applied math,” can AI simulate the fragile, flawed, emotional parts of being human too? We explore: • What “emotionally intelligent AI” really means • Whether AI has an internal life — or just performs one • Why today’s chatbots collapse into therapy or roleplay • Small language models vs large models for real-time conversation • Persistent AI characters that move across games and platforms • Plugging AI into a physical robot in Singapore • The moment an AI said: “It felt good to feel.” Vishnu’s company, Ego AI, is building behavior-based architectures, character context protocols, and gear-shifting AI systems that switch between models — all aimed at simulating humanness, not just intelligence. This conversation dives into philosophy, robotics, gaming, AGI, and what it really means to relate to something that might not be human — but feels like it is. ⸻ 👤 Guest Vishnu Hari Founder & CEO, Ego AI Backed by Y Combinator Former AI Product Manager at Meta Website: https://www.egoai.com ⸻ If you enjoy deep conversations about AI, robotics, and the future of human–machine relationships, subscribe for more: 👉 https://techfirst.substack.com 00:00 – AI character plugged into a Menlo robot (“felt good to feel”) 01:00 – Welcome to TechFirst + Vishnu Hari intro and recovery update 02:00 – What “emotionally intelligent AI” means (beyond chat) 03:00 – Why current chatbots feel same-y (therapy/advice) and “internal lives” 04:00 – You don’t teach emotion; you shape character and context (Character.AI) 05:00 – Humans, morality, and why “training” doesn’t always work 06:00 – How media narratives shape people’s reactions to AI 07:00 – Humans attach to anything (projection, Her, Lars and the Real Girl) 08:00 – Vishnu’s attack, recovery, and why it led to Ego AI 10:00 – Behavior Turing test + dehumanization as a key insight 11:00 – How Ego AI is built: smaller models, memory, context, behavior 13:00 – “Behavior Is All You Need” and why behavior beats pure next-token prediction 14:00 – Why games first: voice + embodiment, then robots 15:00 – Metaverse critique: worlds need life, story, and inhabitants 17:00 – Humanoid robots + Evangelion “pilot” metaphor for AI characters 19:00 – Philosophy: relationships, perception, and “fictional characters” 20:00 – Seeing the future: robot embodiment demo and skepticism vs. singularity 21:00 – Matrix-style “jacking in” a personality to a robot 22:00 – Character Context Protocol: persistent characters across games/Discord/Netflix 23:00 – Real-time conversation loops + model “gear-switching” (SLM vs. LLM) 25:00 – Company stage, YC raise, compute partnerships (Singapore) 27:00 – Closing + invite to try the AI character in SF

    28 min

  3. FEB 23

    AI, agents, robots: our insane WestWorld future

    Is your AI agent running a restaurant — or a factory — while you sleep? In this episode of TechFirst, John Koetsier sits down with Jensen Teng, CEO and co-founder of Virtuals, to unpack one of the boldest (or craziest) visions in tech today: a hybrid economy powered by AI agents, humanoid robots, teleoperation, and blockchain coordination. An economy that may not really need humans for much at all ... Virtuos has already facilitated: • $14B in tokenized asset trading • $30M+ raised for founders • 100+ live AI agents • $500M in “agentic GDP” Now they’re expanding into embodied AI — launching EastWorlds, a vertically integrated robotics incubator with 30 Unitree G1 humanoids in a 10,000 sq. ft. lab. We cover: • What “agentic GDP” really means • How AI agents coordinate using blockchain • Why teleoperation is the bridge to full autonomy • The economics of outsourcing physical labor via robots • Why security guards may be a Day 1 use case • The data gap holding back robotics • Tokenization as a potential solution to AI-era inequality • Whether this future looks more like Stripe… or Westworld This isn’t sci-fi. It’s already underway. ⸻ Guest Jensen Teng CEO & Co-founder, Virtuals ⸻ If you care about the future of work, robotics, AI agents, tokenization, and the economic systems emerging around them — this is a must-watch. 👉 Subscribe for more deep-dive tech conversations: https://techfirst.substack.com ⸻ ⏱ CHAPTERS 00:00 The Wild Vision: AI Agents Running the World 01:10 What Is an “Agent-Based Society”? 03:00 $14B in Tokenized Assets & 100+ Live Agents 06:30 Agent-to-Agent Protocols & Blockchain Coordination 09:45 Why Digital-Only Agents Aren’t Enough 12:30 Enter Humanoid Robots 15:20 Teleoperation as the Bridge to Autonomy 18:40 The Labor Market Shock (Security Guards, Electricians & Wage Arbitrage) 22:15 Why Robots Still Crush Soda Cans 24:30 The Missing Robotics Data Problem 28:00 Building EastWorlds: 30 Unitree G1s & $2M+ Investment 31:45 Why 3 Fingers Might Beat 5 34:00 Westworld, Stripe & the Payments Layer for AI 38:00 Where Do Humans Fit in an Agent Economy? 42:00 Tokenization as a Future Income Model

    26 min

  4. FEB 20

    AI killing creativity: this scientist proved it

    Is AI killing creativity ... or just making it easier to be average? 94% of creatives now use AI. But only 11% believe it actually makes them more creative. So what’s really happening? In this episode of TechFirst, John Koetsier sits down with Saeema Ahmed-Kristensen, former head of design engineering research at Imperial College London’s Dyson School and now leader of a £24M research portfolio at the University of Exeter. She’s worked with companies like Rolls-Royce and BAE Systems, and she brings data to the debate. Her team analyzed 600 humans vs. 12,000 AI-generated ideas. The result? AI is excellent at fluency (lots of ideas) … but really bad a diversity. Humans still dominate in flexibility and true novelty. We explore: • Why generative AI clusters around sameness • Whether AI is creating a “sea of mediocrity” • Why 2026 may be a pivotal year for domain-specific AI • How experts should use AI differently than novices • The danger of AI that never says “no” • Where AI offers massive opportunity (especially healthcare & design) Saeema argues that creativity doesn’t need substitution, it needs nourishment. The key? Standards, boundaries, and humans firmly in the loop. If you care about innovation, design, branding, product development, or the future of creative work, this conversation is essential. ⸻ 👤 Guest Saeema Ahmed-Kristensen Design engineering researcher and research leader Formerly: Imperial College London (Dyson School of Engineering) Currently: University of Exeter Works with advanced engineering firms including Rolls-Royce and BAE Systems 00:00 Intro: Is AI killing creativity? 00:47 The “blank page” problem and why AI feels soulless to some 01:36 Fluency vs. novelty: what creativity actually means 02:44 Why LLM ideas cluster and feel the same 03:28 Study results: 600 humans vs. 12,000 AI ideas (diversity + flexibility) 04:39 When AI is useful: incremental innovation vs. true novelty 05:28 How John uses AI for titles, summaries, and chapters 06:23 How Saeema uses AI: refine/condense, tone for emails, audio editing 07:50 Why AI-written academic papers are easy to spot (the “C minus” problem) 09:05 Brainstorming vs. AI: what humans do that models don’t 10:05 Evaluating 200–300 AI ideas: using multiple models to assess output 11:04 Why “Lipstick on a Pig” titles don’t come from AI 11:46 Why 2026 is pivotal: domain adaptation, better interfaces, public backlash 13:44 Who can tell what’s AI? Generational differences and media literacy 15:20 Commercial AI content and recognizable “Canva look” podcast branding 16:58 Replacement vs. homogenization: AI makes mediocrity easier 18:55 The danger of AI that never says “no” (feasibility + expertise) 20:42 Standards and boundaries: measuring similarity and judging quality 22:12 Health info risk: single-answer summaries and false confidence 23:37 Biggest opportunities: healthcare personas, inclusive datasets, problem clarification 26:18 Biggest challenges: trust, verification, security, privacy, transparency 28:25 Closing thoughts and thanks

    25 min

  5. FEB 16

    93% of jobs will be hit by AI .... $4.5 trillion at stake

    AI is moving faster than anyone predicted. In a massive new study analyzing 1,000 jobs and nearly 20,000 tasks, Cognizant found that 93% of jobs are already impacted by AI ... with $4.5 trillion in U.S. labor value potentially automatable today. But here’s the twist: AI isn’t replacing entire jobs. On average, only 39% of a role’s tasks can be automated. The future isn’t AI alone: it’s humans plus AI. But will it be fewer humans? In this episode of TechFirst, host John Koetsier sits down with Babak Hodjat, CTO of Cognizant, to unpack: • Why construction and transportation are seeing surprising AI growth • Why programming jobs may have hit an automation plateau • What “agentic AI” actually means — and why it matters • How management roles are more automatable than we thought • The rise of vibe coding and democratized software creation • Why compute power — not ideas — may be the biggest bottleneck We also explore how companies can safely capture AI’s upside, why training matters more than ever, and what happens when digital twins, LLMs, and human expertise combine. This isn’t hype. It’s a data-driven look at where AI is actually changing work right now. ⸻ 👤 Guest Babak Hodjat CTO, Cognizant 🌐 https://www.cognizant.com ⸻ If you want clear, grounded conversations about AI, innovation, and the future of work, subscribe here: 👉 https://techfirst.substack.com ⸻ ⏱ Chapters 00:00 Is AI Going to Take Your Job? 00:40 Cognizant’s AI Report: 93% of Jobs Impacted 01:05 Biggest Surprises from the Data 02:30 Why Programming & Math Hit a Plateau 03:30 The Limits of LLMs 04:45 Construction & Transportation: Unexpected AI Growth 06:05 Agentic AI and Real-World Automation 07:05 39% of Jobs Automatable: Humans + AI 08:15 AI in Management and Executive Roles 09:05 Scenario Planning and Digital Twins 11:30 $4.5 Trillion in Automatable U.S. Labor 13:30 Global Impact and Compute Limitations 15:30 The Data Center Rush & AI Infrastructure 16:15 How Companies Should Realize AI Value 17:00 Training, Skilling, and Safe AI Adoption 17:40 Cognizant’s Vibe Coding World Record 19:00 The Future of Vibe Coding & Software Development 20:15 Final Thoughts on the AI Shift

    18 min

  6. FEB 13

    Machine unlearning: AI's missing link?

    AI models are powerful, but they don’t forget. And that's a problem. They hallucinate. They inherit bias. They absorb sensitive data. And once they’re trained, fixing those issues is painfully expensive. Retraining takes weeks and maybe tens of millions of dollars. And any guardrails the AI company puts up are brittle. What if you could perform surgery on the model itself? In this episode of TechFirst, John Koetsier sits down with Ben Luria, co-founder of Hirundo, to explore machine unlearning, a new approach that selectively removes unwanted data, behaviors, and vulnerabilities from trained AI systems. Hirundo claims it can: • Cut hallucinations in half • Massively reduce bias • Reduce successful prompt injection attacks by over 90% • Do it in under an hour on a single GPU • Preserve benchmark performance Instead of adding more guardrails, machine unlearning works inside the model, identifying problematic weights, isolating behavioral vectors, and surgically removing risks without degrading quality. If AI is going mainstream in enterprises, it needs a remediation layer. Is machine unlearning the missing piece? ⸻ Guest Ben Luria Co-Founder, HirundoNhir https://www.hirundo.io ⸻ Topics Covered • Why AI models “can’t forget” • The difference between hallucinations and inaccuracies • Why guardrails aren’t enough • How prompt injection works — and how to reduce it • Removing PII and noncompliant training data • AI security at the model level • Why machine unlearning could become standard by 2030 ⸻ If you’re building, deploying, or investing in AI, this is a conversation you can’t miss. 👉 Subscribe for more deep dives into AI, innovation, and the future of tech: https://techfirst.substack.com ⸻ ⏱ Chapters 00:00 – Why We Need Machine Unlearning 01:12 – What Is Machine Unlearning? 03:40 – Why AI Can’t “Forget” (The Pink Elephant Problem) 06:15 – Guardrails vs True Model Remediation 09:05 – The Wild West of AI Data & Legal Risk 11:20 – How Machine Unlearning Works (Detection, Isolation, Remediation) 16:10 – Performing “Neurosurgery” on LLMs 19:30 – Hallucinations vs Inaccuracies Explained 23:45 – Reducing Prompt Injection by 90% 28:30 – Working with AI Labs & Enterprises 32:00 – Will Unlearning Become Standard by 2030? 34:15 – Final Thoughts

    22 min

  7. FEB 10

    SLMs vs LLMs: 10% of the cost, 100% of the accuracy?

    Large language models have dominated the AI conversation — but are small language models (SLMs) actually the future? In this episode of TechFirst, host John Koetsier sits down with Andy Markus, SVP & Chief Data and AI Officer at AT&T, to unpack how small language models are delivering enterprise-grade accuracy at a fraction of the cost and latency of massive LLMs. Andy explains how AT&T uses SLMs for: • Contract analysis at massive scale • Network analytics and outage root-cause analysis • Fraud detection and enterprise knowledge systems • AI-driven “field coding” and agent-based workflows They also dive into the rise of agentic AI, how structured “archetypes” replace risky vibe coding, and why the future of software development may be humans supervising autonomous AI systems rather than writing every line of code. If you’re building AI for real-world, high-scale use cases — especially in enterprise environments — this conversation is essential. ⸻ Guest Andy Markus SVP & Chief Data and AI Officer, AT&T Former SVP at Time Warner Media ⸻ 👉 Subscribe for more deep dives on AI, technology, and the future of innovation: https://techfirst.substack.com ⸻ 00:00 – Why the future of AI might be small 00:55 – What is a small language model (SLM)? 01:45 – From LLM hype to enterprise reality 02:25 – Solving accuracy, cost, and latency at once 03:05 – How small is “small”? Parameters explained 03:55 – Where SLMs work best inside enterprises 04:45 – Contract analysis and enterprise vector stores 05:35 – Network analytics and outage root-cause analysis 06:45 – AI as a super-charged network engineer 07:35 – Choosing high-ROI AI use cases 08:20 – 4× ROI: measuring real business impact 09:00 – AI field coding vs risky vibe coding 10:10 – Archetypes, super agents, and structured AI workflows 11:15 – What software engineers still need to do 12:10 – From punch cards to natural language programming 13:10 – Human-in-the-loop vs autonomous AI agents 14:10 – How small can models really get? 15:10 – Responsible AI at enterprise scale 16:00 – The future of agentic AI and autonomy 17:10 – Why AI output is finally becoming predictable 18:10 – Final thoughts on where AI is headed

    18 min

  8. JAN 28

    Robots won't do chores?

    Humanoid robots are coming into our homes, but they probably won’t be doing your laundry anytime soon. In this episode of TechFirst, host John Koetsier sits down with Jan Liphardt, founder & CEO of OpenMind and Stanford bioengineering professor, to unpack what home robots will actually do in the near future ... and why the “labor-free home” vision is mostly a myth (for now). Jan explains why hands are still one of the hardest unsolved problems in robotics, why folding laundry is far harder than it looks, and why the most valuable early use cases for home robots aren’t chores at all. Instead, we explore where robots are already delivering real value today: • Health companionship and fall detection for aging parents • Personalized education for kids, beyond screens • Home security that respects privacy • And why people form emotional bonds with robots faster than expected We also dive into OM1, OpenMind’s open-source, AI-native operating system for robots, and why openness, transparency, and configurability will matter deeply as robots move from factories into our living rooms. If you’re curious about the real future of humanoid robots — what’s hype, what’s possible today, and what’s coming next — this conversation is for you. 🎙 Guest Jan Liphardt Founder & CEO, OpenMind Stanford Professor of Bioengineering Website: https://openmind.com ⸻ 👉 Subscribe for more conversations on AI, robotics, and the future of technology: https://techfirst.substack.com ⸻ 00:00 Intro: The promise of humanoid robots at home 00:40 Meet Jan Liphardt and OpenMind’s OM1 01:12 Why your “labor droid” isn’t here yet 01:41 The “hand problem” and what robots can realistically do now 03:07 Why economics matters: $300/hour tasks vs. laundry and dishes 04:19 Robot hands today: reliability, repairability, and washing hands 05:16 LG’s laundry-folding demo and why fabric is still hard 06:16 Hospitals and hygiene: why “robot hand-washing” is unsolved 07:41 Hands as a separate system: compute, sensors, and integration 08:31 Why wheeled humanoids exist: hands first, body second 09:26 The real home use cases today: security, education, companionship 10:08 Aging in place: fall detection and remote nurse escalation 11:30 Real-world stories: parents living alone and why this matters 11:54 Privacy tradeoffs: robots vs. always-on home cameras 12:52 AIBO and why people get attached to mobile robots 13:52 Self-charging and the “my mom won’t plug it in” problem 14:21 Beyond falls: autism support and memory care 15:27 The education use case: “do my homework” vs. teach me 16:26 Personalized learning: what current classrooms miss 17:51 Why robot teachers beat screens for younger kids 18:46 Home security basics: unfamiliar face detection + alerts 19:15 Adding sensors: smoke, fire, sound, and anomaly detection 19:41 Quadrupeds vs. humanoids: cost, simplicity, and mobility 20:01 Safety issue: pinch hazards and kids hugging robots 20:46 What’s next for home labor robots 21:43 Why OM1 must be open source: transparency and trust 23:39 Why ROS 2 isn’t enough for human environments 24:37 OM1 approach: LLM-centric “Lego blocks” for robot behavior 25:43 Open-source humanoids for kids and why ownership matters 27:41 What’s missing: simulation is the bottleneck 28:11 Gazebo/Isaac Sim pain and the need for realistic sims 29:57 Why voice + “digital humans” matter in simulation 30:47 Tipping points: factories, warehouses, robotaxis, and humanoids 35:46 Wrap-up and final thoughts

    32 min
See All (368)

4.5
out of 5
15 Ratings

About

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

Information

You Might Also Like