Most enterprise AI dies somewhere between the demo and production. Philippe Rambach, the first Chief AI Officer at Schneider Electric, breaks down the enterprise AI strategy that gets past the pilot stage: five stage gates, one shared platform, and a flat refusal to treat AI as innovation at all.Rambach spent fifteen years running sales and P&L inside Schneider Electric before the board handed him AI in mid 2021, a year before ChatGPT existed, and that business background shapes everything about how he built the function. Schneider Electric is a forty billion euro energy technology company whose AI organisation now spans four hundred core specialists and roughly nine hundred people at any given moment, split evenly between internal efficiency and customer facing products. He argues that industrial AI stalls because companies manage it as research instead of product development, that a learning module sitting on a chip costing a couple of dollars delivers more building energy savings than any large language model could, and that new token economics have quietly reversed his build vs buy rule in the last six months. Speaking with host Arjun Jain, he also explains why Schneider Electric shut down its own internal ChatGPT after eighteen months rather than run two tools in parallel.👉Why Schneider Electric deliberately moved AI out of the innovation lab and into a five gate product development process with quarterly portfolio reviews and joint business sign off.👉How the hub and spoke model actually works, with four hundred core AI specialists plus an equal number embedded in business squads that own delivery from idea through to production support.👉What the enterprise AI build vs buy rule looks like in practice, and why new token pricing pushed Rambach back toward building things he would previously have bought off the shelf.👉How Schneider Electric built its own internal ChatGPT called JoeChatGPT in December 2022, ran it for eighteen months, then killed it completely instead of letting people choose between two tools.👉Why a small learning model embedded in a low cost room controller beats a large language model, delivering up to twenty percent additional energy savings with no LLM anywhere in the loop.👉What single question Rambach asks before approving any agentic AI project, and why answering yes to it usually means the team should go find a different approach. Subscribe to Human by Design for weekly founder conversations and follow Arjun Jain on LinkedIn https://www.linkedin.com/in/arjunjain/ for daily insights.00:00 - Meet Schneider Electric's First Chief AI Officer 05:57 - Taking Enterprise AI Out Of Innovation 07:09 - Inside The 400 Person AI Hub 13:37 - Enterprise AI Build vs Buy Explained 17:37 - Why Schneider Electric Killed Its Own ChatGPT 23:29 - Avoiding AI Vendor Lock In 27:47 - What Industrial AI Actually Looks Like 41:07 - The Question That Kills AI Agents 48:02 - Will AI Cut Your Workforce 52:09 - Physical AI Is Already Here#PhilippeRambach #SchneiderElectric #ArjunJain #HumanByDesign #ChiefAIOfficer #EnterpriseAI #IndustrialAI #AIatScale #EnterpriseAIStrategy #AIBuildVsBuy #AIAgents #DigitalTransformation #AITransformation #HowToScaleEnterpriseAI #WhyAIPilotsFail #AIInManufacturing #EnergyManagement #PhysicalAI #AIOrgDesign #TokenEconomics

The man whose team laid the foundations of ChatGPT explains why the systems passing today's Turing tests might be nothing more than fancy text editors - and why real intelligence, what he calls physical AI, only begins when machines leave the screen and start acting in the world.At fifteen, Jürgen Schmidhuber set himself one goal - build a machine smarter than himself and then retire - and he has spent five decades chasing it. He has been called a father of modern AI, and for good reason: his Munich team in 1990 - 91 published the roots of today's trillion-dollar AI boom - the first Transformer (the "T" in ChatGPT, as an unnormalized linear Transformer), Pre-Training (the "P"), neural network distillation now central to systems like DeepSeek, and deep residual learning that underpins LSTM and the most-cited neural nets of both the 20th and 21st centuries. Today he directs AI research at KAUST, where a sovereign-scale supercomputer backs his next bets. Across this conversation with host Arjun Jain, he argues that the universe itself may be a short computer program, that almost every breakthrough in the history of deep learning traces back to a handful of underfunded academic labs and one relentless trend - 10x cheaper compute every five years - and that today's chatbots are not real intelligence at all. His case for physical AI and embodied robots reframes what artificial general intelligence will actually demand, and why the real race has barely started.👉Why Schmidhuber believes the entire universe could be generated by a program just ten lines long, and what observation would prove him wrong.👉How 1991 in Munich quietly seeded today's AI - the linear Transformer, pre-training, distillation, and deep residual learning - decades before the compute existed to make any of it useful.👉How one trend, computing power getting ten times cheaper every five years, explains why powerful 1990s ideas only became useful two decades later.👉What the generative adversarial networks of the 1990 artificial curiosity system actually did, with one network inventing actions and another racing to predict their consequences.👉Why he dismisses today's screen-based AI as "nothing" and insists real intelligence must do what a plumber or a seven-year-old with a football can do.👉How self-replicating, self-improving robots could open what he calls an age of abundance that expands far beyond Earth.Subscribe to Human by Design for weekly conversations on the future of intelligence, and follow Arjun Jain on LinkedIn [https://www.linkedin.com/in/arjunjain/] for daily insights.Chapters 00:00 - Building AI smarter than himself 08:25 - Is the universe a program? 16:22 - Why quantum randomness may be illusion 22:18 - Lugano to KAUST and compute 28:00 - Artificial curiosity, world models, and PMAX (1992) vs JEPA (2022) 48:35 - Why compute, not ideas, unlocked AI 50:58 - 1991, the year deep learning opened up 01:30:23 - Japan, GPUs and the CNN story #HumanByDesign #PhysicalAI #EmbodiedAI #ArtificialGeneralIntelligence #AGI #DeepLearning #Transformers #AIResearch #KAUST #FutureOfAI #AIRobots #MachineLearning #ComputationalUniverse #AIpodcast #WhatIsPhysicalAI #IsTheUniverseAComputer #WillRobotsReplaceHumans #HistoryOfDeepLearning #Munich1991

Simon Jagers, Co-Founder of Samotics, has built the world's most scalable sensorless industrial AI platform, raising $44M and claiming roughly 50% of the global condition monitoring market for hard-to-reach assets.Up to 40% of critical machines in heavy industry operate in places where traditional sensors simply cannot survive - inside blast furnaces hitting 1,500 degrees Celsius, submerged in sewage networks, or buried 3,000 metres underground in oil wells. Simon Jagers spent 11 years solving exactly that problem. Starting as a data science consulting firm and pivoting through failed vibration sensor pilots, Simon built Samotics from the ground up - designing custom hardware, proprietary signal processing, and an AI platform that analyses high-frequency electrical data to detect mechanical and electrical faults across tens of thousands of industrial machines worldwide. The platform now serves clients in steel, oil and gas, wastewater, chemicals, and mining - and is embedded directly into ABB variable speed drive firmware as native infrastructure. Simon shared the full journey, including the near-death pivot at Heineken and the decision to burn every bridge, in this candid conversation with host Dr. Arjun Jain.What you will learn in this episode,👉How Simon Jagers built Samotics into the global leader in sensorless AI condition monitoring, covering roughly 50% of the market for hard-to-reach industrial assets👉The Heineken pilot that almost ended the company - and the exact decision that saved it👉Why up to 40% of factory machines have never been monitored, and how Electrical Signature Analysis (ESA) changes that permanently👉The counterintuitive case for keeping humans in the AI loop - and why one engineer managing 2,000 assets is a scale strategy, not a workaround👉Why most industrial AI pilots fail before a single algorithm is written, and the one question every AI team must ask about their data👉Simon's contrarian take on co-founder trust, cultural inertia, and why the best idea must always winIf this episode gave you one insight you can use, subscribe to Human by Design for weekly conversations with founders and operators who are making AI work in the real world. Follow Dr. Arjun Jain on LinkedIn and X for daily posts on AI, entrepreneurship, and what it actually takes to build at the frontier.Chapters:00:00 - Simon Jagers and the Samotics origin story04:30 - From IT sales to industrial AI founder08:45 - The 40% of machines nobody can monitor13:20 - Why sensors burn, drown, and fail in factories17:40 - Turning every motor into an AI sensor20:30 - The Dutch railway that started everything24:00 - The Heineken pilot that almost killed Samotics29:00 - Human-in-the-loop as a scale strategy35:00 - How ESA detects faults without touching the machine41:00 - Predicting machine failure, the art vs. science debate48:00 - Burning the boats - no plan B, no off ramp55:30 - Industries Samotics serves globally01:00:00 - Claiming 50% of a new AI market category01:03:00 - Co-founder trust and company culture#SimonJagers #Samotics #IndustrialAI #PredictiveMaintenance #ElectricalSignatureAnalysis #SensorlessMonitoring #ConditionMonitoring #AIStartup #HeavyIndustryAI #AIDisruption #IndustrialIoT #ManufacturingAI #AITransformation #PredictiveMaintenanceAI #DigitalTransformation #FounderStory #AIFunding #HumanByDesign #DrArjunJain #EnterpriseAI #Industry40 #MachineLearningManufacturing #AIScaleup #IndustrialAutomation #OperationalTechnology