Rene Grywnow’s 5-Minute Business Punch

Week 39 found 23% of a robot’s energy in a bad path; Week 40 found a continent’s worth of unplanned grid load. Week 41 goes deeper, into the actuator, where most of embodied AI’s electricity is spent before the algorithm ever “thinks.” Inference runs on a chip; motion runs on copper, iron and a gearbox. A servo motor can be ~80% efficient in isolation. Add a high-ratio gearbox and installed-chain...

This week looked at the factory floor and the grid line nobody can walk. Part III asks what command structure can run either at scale, through Stanley McChrystal, Tantum Collins, David Silverman and Chris Fussell’s Team of Teams (Portfolio/Penguin, 2015). In 2004, the Joint Special Operations Task Force was one of the most capable military organizations in history, and losing. Not for lack of skil...

Part I stayed inside the factory fence. Part II walks out the gate, onto hundreds of kilometres of transmission line, substation fencing and pipeline that was never a “plant,” and is simply too much infrastructure for too few people to walk. SP Energy Networks’ Project VOLT runs drone-in-a-box units at four substations across the Lake District and Cumbria, flown beyond visual line of sight from a...

Week 39 went inside a single robot’s path. Week 40 pulls back to the fleet, then the continent — because the IFR’s World Robotics 2026 numbers make the gap impossible to ignore: 603,000 industrial robots installed globally in 2025, China taking 354,000 (59%, +20%), the EU just 60,500 (−11%), Germany ~25,000 (−8%). The instinct is to read this as a race Europe is losing. It’s the wrong race. China’...

Part I showed the same robot burning 20%+ extra energy depending only on placement and path. Part II asks why the digital twin, built precisely to test placements and paths before they’re built, so rarely gets pointed at energy at all. Offline tools (RobotStudio, Process Simulate) excel at collision-free, time-feasible motion; energy is usually an add-on or ignored, so plants validate a fast cell...

Week 38 counted the fleet-wide load and opened a humanoid’s battery. Week 39 goes deeper, into the trajectory, where the same robot can burn 20%+ extra energy for the identical job depending only on where it’s placed and how it’s programmed. A six-axis arm spends most of its electricity moving mass through poorly chosen joint space: overshoot, sharp acceleration, detours, high-torque postures. ABB...

This week counted the robot fleet’s energy and opened a single humanoid. Part III turns to a leadership question older than Physical AI — can a busy robot make the whole plant slower? — through Eliyahu Goldratt and Jeff Cox’s The Goal (North River Press, 1984). The purpose of a plant is to make money by increasing throughput while controlling inventory and operating expense; the method is the Theo...

Part I counted the fleet-wide load. Part II opens a single humanoid and asks where the watts go, because “cheap labour” and “mobile energy system” aren’t the same purchase order. The seductive math: a ~$16,000 Unitree G1 costs roughly $82/year in electricity at eight hours a day, a rounding error next to a wage bill, until “eight hours” turns out to require a charging architecture the pilot never...

Week 37 measured the megawatts and connection years behind data centres. Week 38 adds the load nobody modelled into that same system: the robot fleet already running today, quietly, on ordinary factory feeders. A data centre is a point load, one dot utilities, regulators and boards can see. A robot fleet is a thousand small dots nobody mapped: arms, AMRs, cobots, docks, vision PCs, cooling, indivi...

This week looked outward, grids, chips, sovereignty. Part III turns inward: none of that external strategy moves without an organization that can actually change. Drawing on Monica Monsch’s Transformation beschleunigen (Versus Verlag, 2026), the argument is that transformation doesn’t fail from a lack of strategy, it fails because people, leadership, communication and execution are run as separate...

Part I mapped Europe’s infrastructure gap. Part II asks who controls the inputs infrastructure depends on, and what happens to your plans when that control sits elsewhere. Sovereignty is no longer a government word; it’s a supply-chain word, and it shows up as a delayed transformer, a queued connection, a capacity-limited cloud region, or a single-source robotics component. Semiconductors show the...

Week 36 argued energy is a leadership job inside the factory. Week 37 zooms out to where that job gets decided first: European competitiveness. The continent doesn’t have an AI-ambition problem, it has a megawatt-and-decade problem. The race shifted from algorithms to infrastructure: what decides location and scale is now whether power actually gets there. The IEA reports 2,500+ GW stalled in grid...

This week ran as a set: Part I named the human advantage, Part II the energy discipline, and Part III puts both inside one job description, drawing on Paul Daugherty and James Wilson’s Human + Machine (HBR Press, 2018). For a century, “Who executes this task?” was a complete leadership question. It stopped being one when the answer became “a person and a machine, together.” Value now sits in what...

Part I argued human judgment is the scarce asset inside an AI-driven factory. Part II turns to the constraint outside it: how much of that factory can run once energy, not silicon, becomes the binding limit. AI is two-sided, it adds load (edge compute, robots, always-on sensing) and reduces it (better control, less scrap, fewer idle losses). Leaders who see only one side over-build or over-claim....

Week 35 closed on team quality. Week 36 opens with its operational version: as Physical AI takes over speed, scale and pattern recognition, what’s left for humans is the part that was always hardest to automate, and it gets more valuable, not less. Machines absorb work that’s frequent, sensor-rich and rule-bounded: screening, parameter suggestions, routine scheduling, anomaly flags. They don’t abs...

This week ran as a set: Part I named the activity trap, Part II showed why fixes ripple sideways, and Part III turns to the humans inside the system — drawing on Steven Thurber and Bryan Miller’s Good Team, Bad Team (Page Two, 2024). The core reframe: as AI agents join the team, human team quality matters more, not less — technology amplifies whatever dynamic already exists. An agent doesn’t fix a...

Part I argued most AI programs confuse activity with progress. Part II asks why, even when chasing genuine progress, fixing one problem in an autonomous operation so often creates another. The answer: linear troubleshooting — find the cause, fix it, move on — breaks down when the cause isn’t in any single component but in the interactions between them. The concept is emergence: system behavior tha...

Week 34 closed with a framework for turning vague AI ambition into strategic choices. Week 35 opens with the trap that framework exists to prevent: mistaking activity for progress. A dashboard full of green metrics and a CFO’s spreadsheet full of red questions can both be true at once — and that gap is where AI programs die. The numbers are stark: manufacturing AI projects fail at ~76%, and 84% of...

This week ran as a trilogy: Part I diagnosed how AI exposes weak leadership, Part II priced the paralysis that follows, and Part III hands over the tool — A.G. Lafley and Roger Martin’s Playing to Win strategy cascade, applied to Physical AI. Most AI investment is activity without strategy: ask ten manufacturing leaders why they’re investing and eight say “because everyone else is.” Capgemini foun...

Part I argued AI exposes weak leadership. Part II puts a number on one of the clearest symptoms: the decision that never gets made. Most operational decisions in a complex plant are reversible — a parameter adjusted back, a maintenance call revised — yet get deliberated like bet-the-company moves. The cost of a fixable mistake is almost always smaller than the cost of delay. West Monroe’s 2026 stu...

Week 33 argued that judgment, not information, is the scarce resource. Week 34 asks what happens to leaders who never had much judgment — now that AI has removed the excuse that covered for them. AI doesn’t create leadership weakness; it removes the fog that hid it. For years, slow information gave average leaders cover — a slow decision looked reasonable while everyone “waited on the numbers.” Wh...

Part I asked why pilots stall before production. Part II asks what happens once they don’t — once an autonomous system makes real decisions and something goes wrong at a level the board, not the plant, has to answer for. Autonomous AI doesn’t create a new legal duty; it raises the bar on an old one. Under Delaware’s Caremark doctrine, directors can be personally liable for failing to build a funct...

Week 31 covered what regulation demands. Week 32 asks the quieter question underneath: even where the rules are clear, why do most industrial AI pilots never scale? The honest answer has less to do with the model and more to do with trust. The gap is bigger than steering committees assume — RAND found ~80% of enterprise AI projects fail to deliver promised value, and Gartner puts full ROI success...

Part I asked why pilots stall before production. Part II asks what happens once they don’t — once an autonomous system makes real decisions and something goes wrong at a level the board, not the plant, has to answer for. Autonomous AI doesn’t create a new legal duty; it raises the bar on an old one. Under Delaware’s Caremark doctrine, directors can be personally liable for failing to build a funct...

Week 31 covered what regulation demands. Week 32 asks the quieter question underneath: even where the rules are clear, why do most industrial AI pilots never scale? The honest answer has less to do with the model and more to do with trust. The gap is bigger than steering committees assume — RAND found ~80% of enterprise AI projects fail to deliver promised value, and Gartner puts full ROI success...