Deming, Drucker, Ohno, Ackoff, Schein, Mintzberg, Porter, Simon, Christensen, and Buffett walk into a room . . .
Put those ten in a room and ask them one question — what should we make of AI? — and the first thing you’d get is an argument. They disagreed in life about almost everything: what a strategy is, where value comes from, whether managers plan or improvise, what a company is even for. You would not get consensus.
But listen long enough, past the disagreement, and something quieter emerges. Underneath the different vocabularies, most of them are pointing at the same place — and it is not the place the current AI conversation is looking.
Let me walk the room.
Ohno wouldn’t sit down. He’d ask to see the floor. His whole method — genchi genbutsu, go and see — was a refusal to trust the report over the reality. He’d watch the work actually happen, ask “why” five times, and distrust any dashboard that claimed the process was fine. Drop AI on a process you’ve only ever seen on a slide, and Ohno would already know how that ends.
Deming would refuse to blame the tool or the worker. A bad system, he insisted, defeats a good person every time — and most failure is systemic, not individual. He’d say AI is an amplifier with no opinion of its own: aim it at a sound system and it compounds the soundness; aim it at a broken one and it produces faster, more confident breakage. The question is never the algorithm. It’s the system the algorithm inherits.
Ackoff would warn that you can do the wrong thing righter. Optimize each part and you can still starve the whole, because the performance of a system lives in the interactions between its parts, not the parts themselves. AI, he’d note, is extraordinarily good at optimizing parts — which is exactly what makes it dangerous to a system nobody has looked at whole.
Schein would go quieter and more uncomfortable. His life’s work was the gap between what an organization says it does (espoused values) and the assumptions it actually runs on (the real culture underneath). AI, he’d observe, is a merciless exposer of that gap — it executes the declared process at speed and reveals, in public, that the declared process was never the real one.
Mintzberg would distrust the strategy deck on principle. He spent his career documenting what managers actually do versus what the plan says, and arguing that real strategy emerges from practice rather than descending from intent. He’d want to know what people are really doing with AI on the ground — not what the transformation roadmap claims.
Simon — who helped invent artificial intelligence — would bring the sharpest epistemic caution. Bounded rationality was his great insight: decisions get made against a simplified model of reality, never reality itself. He’d point out that this applies to the AI’s designers too. The system optimizes a declared objective, and the declared objective is a model of the real one. When the model and the reality diverge, the machine pursues the wrong goal with perfect efficiency.
Christensen built his entire theory from a failure question — why do capable, well-run companies fail? — and traced it to a cause no forward framework contained: the very competencies that made them win blinded them to what came next. He’d ask what job you’re actually hiring AI to do, and warn that the strength you’re most proud of is the thing most likely to hide the disruption.
Buffett would be the skeptic in the corner, asking whether AI changes the intrinsic economics of the business or merely the story around it. Distrust the market’s narrative, examine the underlying thing, stay inside your circle of competence — his instincts all point to the difference between what’s declared and what’s real.
That’s eight of the ten reasoning from the same direction: start from the operating reality — the floor, the system, the interactions, the real culture, the actual practice, the true objective, the failure, the underlying business — not from the declared plan.
The other two lean the opposite way, and it’s worth being honest about it. Porter built the most influential forward instrument in strategy — Five Forces, the value chain — designed to analyze toward an intended competitive position. He’d map AI’s effect on industry structure with real rigor. And Drucker, for all his wisdom about asking “what is our business, really?”, is remembered for Management by Objectives — a discipline that starts from intent and builds toward it. Both would have sharp, useful things to say. Both, by instinct, start from where you want to be rather than from where the work is actually failing.
I’ll be clear that this is how I’d map them, not a measurement — Simon, Buffett, and Schein are genuine judgment calls, and a knowledgeable reader could argue one or two either way. But the pattern holds: the overwhelming weight of the century’s best management thinking reasons from operating reality first.
And here is the thing I keep coming back to. Most of them saw the right diagnosis — and then, when they turned the diagnosis into a tool, several reverted. Porter’s insight became a static 2×2. Drucker’s became a cascade of objectives. Even Ohno’s go-and-see calcified, at times, into a snapshot. Diagnosis right; instrument forward. The map they saw and the map they drew were not always the same map.
Which points at the piece none of them named — not because AI is a new kind of problem, but because it is the same problem moving faster.
Every one of these thinkers built on an unstated precondition: that the operating reality their system would act upon had been validated — that the floor Ohno walked, the system Deming trusted, the business Buffett analyzed, was the real one. That precondition has always been the fault line. The real division was never between technology eras — it was between two methodologies that have competed the whole time: the equation-based approach, which optimizes the declared model and trusts the forward map, and the agent-based approach, which starts from the operating reality and traces it. Equation-based work, deployed onto a reality nobody validated, has failed the same way in every era — mainframe, client-server, ERP, cloud. The technology changed. The failure did not.
AI does not change that equation. It does not introduce a new failure mode; it accelerates the existing one. What used to take a year to surface — the slow, forgiving reveal of the gap between the declared process and the real one — now surfaces in weeks, at scale, with an autonomous amplifier compounding the error before anyone steps in. The disease is identical. The onset is faster and the blast radius is larger. That is the whole of what AI adds: not a new problem, but a brutal acceleration of the one that was always there.
Which is exactly why the step no framework on the list makes explicit finally has to be made explicit: before you switch on the amplifier, confirm that the reality it will amplify is real. In a slower era you could skip it and correct as you went, because the failure gave you time. It no longer does.
I didn’t arrive at this through trial and error. I walked into a defence maintenance operation in 1998 with a method already in hand — Strand Commonality™ — and it was that method that generated the question. The operation was delivering next-day parts 51% of the time against a 90% requirement. I didn’t start from the plan. I started from the failure, and the framework pointed me at a question that was nowhere on anyone’s map — what time of day do orders come in? That question walked me out of procurement into the service department, back into procurement, out to the suppliers and the border, and finally into finance, where the program was quietly losing money on every part because two line items each reconciled cleanly and hid the loss in the gap between them. Four departures from the process I was hired to map. Delivery moved from 51% to 97.3% in three months — and the technology came afterward, onto a foundation that was finally real.
The map on the left is the line to a destination. The map on the right is a search starting from failure. Only the second one identifies and removes the obstacles to a successful outcome.
I want to be precise about what that is and isn’t. The giants have cases too — Ohno had Toyota, Christensen had the disk-drive industry, Deming had the shop floors of postwar Japan. But their cases illustrate a principle. Mine traces a mechanism — backward from the failure, across the couplings, to the specific obstacle no one declared. That difference — illustration versus trace — is not a claim to be smarter than any of them. It’s a claim about where the work begins and how far it runs.
And it is why the other two pieces of the work exist. What stays constant across every technology era — mainframe to client-server to ERP to cloud to AI — is that readiness, not the technology, decides the outcome. I have called that Invariant Physics™. The mechanics of how a platform amplifies the operating conditions it lands on, for better or worse, is Implementation Physics™. Phase 0™ is the discipline that validates those conditions before the amplifier is switched on.
That’s the step the room was circling for a century without naming, because in a slower world the failure gave you time to catch it (or at least try to). Ohno assumed the floor was real because he walked it himself, and if he was wrong, the process hopefully corrected over months. The equation-based failure was always there; it was simply survivable — although they didn’t always survive, as FoxMeyer Drug did not. The consequences of AI don’t change the failure — they compress the reaction time and, in a complex global environment, can put survivability itself at risk. In short, there is more on the line today than in previous technology eras. The floor is now a data model, the amplifier runs autonomously, and the same error that once took a year to compound now compounds in weeks. That’s all Phase 0™ is: the assumption their eras let them skip, turned into a discipline for an era that moves too fast to skip it.
So if those ten walked into the same room and argued their way, eventually, to the thing underneath the argument, I don’t think they’d say something none of us had heard. I think they’d say what they always said — understand the reality before you act on it — and then look at what AI does to a reality nobody validated, and recognize it instantly. Not a new failure. The oldest one, accelerated. The same equation-based mistake they each warned about in their own vocabulary, now running faster than the correction can catch it.
-30-
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
Ten Legendary Business Thinkers Walk Into a Room: What Would They Say About AI?
Posted on August 1, 2026
0
Deming, Drucker, Ohno, Ackoff, Schein, Mintzberg, Porter, Simon, Christensen, and Buffett walk into a room . . .
Put those ten in a room and ask them one question — what should we make of AI? — and the first thing you’d get is an argument. They disagreed in life about almost everything: what a strategy is, where value comes from, whether managers plan or improvise, what a company is even for. You would not get consensus.
But listen long enough, past the disagreement, and something quieter emerges. Underneath the different vocabularies, most of them are pointing at the same place — and it is not the place the current AI conversation is looking.
Let me walk the room.
Ohno wouldn’t sit down. He’d ask to see the floor. His whole method — genchi genbutsu, go and see — was a refusal to trust the report over the reality. He’d watch the work actually happen, ask “why” five times, and distrust any dashboard that claimed the process was fine. Drop AI on a process you’ve only ever seen on a slide, and Ohno would already know how that ends.
Deming would refuse to blame the tool or the worker. A bad system, he insisted, defeats a good person every time — and most failure is systemic, not individual. He’d say AI is an amplifier with no opinion of its own: aim it at a sound system and it compounds the soundness; aim it at a broken one and it produces faster, more confident breakage. The question is never the algorithm. It’s the system the algorithm inherits.
Ackoff would warn that you can do the wrong thing righter. Optimize each part and you can still starve the whole, because the performance of a system lives in the interactions between its parts, not the parts themselves. AI, he’d note, is extraordinarily good at optimizing parts — which is exactly what makes it dangerous to a system nobody has looked at whole.
Schein would go quieter and more uncomfortable. His life’s work was the gap between what an organization says it does (espoused values) and the assumptions it actually runs on (the real culture underneath). AI, he’d observe, is a merciless exposer of that gap — it executes the declared process at speed and reveals, in public, that the declared process was never the real one.
Mintzberg would distrust the strategy deck on principle. He spent his career documenting what managers actually do versus what the plan says, and arguing that real strategy emerges from practice rather than descending from intent. He’d want to know what people are really doing with AI on the ground — not what the transformation roadmap claims.
Simon — who helped invent artificial intelligence — would bring the sharpest epistemic caution. Bounded rationality was his great insight: decisions get made against a simplified model of reality, never reality itself. He’d point out that this applies to the AI’s designers too. The system optimizes a declared objective, and the declared objective is a model of the real one. When the model and the reality diverge, the machine pursues the wrong goal with perfect efficiency.
Christensen built his entire theory from a failure question — why do capable, well-run companies fail? — and traced it to a cause no forward framework contained: the very competencies that made them win blinded them to what came next. He’d ask what job you’re actually hiring AI to do, and warn that the strength you’re most proud of is the thing most likely to hide the disruption.
Buffett would be the skeptic in the corner, asking whether AI changes the intrinsic economics of the business or merely the story around it. Distrust the market’s narrative, examine the underlying thing, stay inside your circle of competence — his instincts all point to the difference between what’s declared and what’s real.
That’s eight of the ten reasoning from the same direction: start from the operating reality — the floor, the system, the interactions, the real culture, the actual practice, the true objective, the failure, the underlying business — not from the declared plan.
The other two lean the opposite way, and it’s worth being honest about it. Porter built the most influential forward instrument in strategy — Five Forces, the value chain — designed to analyze toward an intended competitive position. He’d map AI’s effect on industry structure with real rigor. And Drucker, for all his wisdom about asking “what is our business, really?”, is remembered for Management by Objectives — a discipline that starts from intent and builds toward it. Both would have sharp, useful things to say. Both, by instinct, start from where you want to be rather than from where the work is actually failing.
I’ll be clear that this is how I’d map them, not a measurement — Simon, Buffett, and Schein are genuine judgment calls, and a knowledgeable reader could argue one or two either way. But the pattern holds: the overwhelming weight of the century’s best management thinking reasons from operating reality first.
And here is the thing I keep coming back to. Most of them saw the right diagnosis — and then, when they turned the diagnosis into a tool, several reverted. Porter’s insight became a static 2×2. Drucker’s became a cascade of objectives. Even Ohno’s go-and-see calcified, at times, into a snapshot. Diagnosis right; instrument forward. The map they saw and the map they drew were not always the same map.
Which points at the piece none of them named — not because AI is a new kind of problem, but because it is the same problem moving faster.
Every one of these thinkers built on an unstated precondition: that the operating reality their system would act upon had been validated — that the floor Ohno walked, the system Deming trusted, the business Buffett analyzed, was the real one. That precondition has always been the fault line. The real division was never between technology eras — it was between two methodologies that have competed the whole time: the equation-based approach, which optimizes the declared model and trusts the forward map, and the agent-based approach, which starts from the operating reality and traces it. Equation-based work, deployed onto a reality nobody validated, has failed the same way in every era — mainframe, client-server, ERP, cloud. The technology changed. The failure did not.
AI does not change that equation. It does not introduce a new failure mode; it accelerates the existing one. What used to take a year to surface — the slow, forgiving reveal of the gap between the declared process and the real one — now surfaces in weeks, at scale, with an autonomous amplifier compounding the error before anyone steps in. The disease is identical. The onset is faster and the blast radius is larger. That is the whole of what AI adds: not a new problem, but a brutal acceleration of the one that was always there.
Which is exactly why the step no framework on the list makes explicit finally has to be made explicit: before you switch on the amplifier, confirm that the reality it will amplify is real. In a slower era you could skip it and correct as you went, because the failure gave you time. It no longer does.
I didn’t arrive at this through trial and error. I walked into a defence maintenance operation in 1998 with a method already in hand — Strand Commonality™ — and it was that method that generated the question. The operation was delivering next-day parts 51% of the time against a 90% requirement. I didn’t start from the plan. I started from the failure, and the framework pointed me at a question that was nowhere on anyone’s map — what time of day do orders come in? That question walked me out of procurement into the service department, back into procurement, out to the suppliers and the border, and finally into finance, where the program was quietly losing money on every part because two line items each reconciled cleanly and hid the loss in the gap between them. Four departures from the process I was hired to map. Delivery moved from 51% to 97.3% in three months — and the technology came afterward, onto a foundation that was finally real.
The map on the left is the line to a destination. The map on the right is a search starting from failure. Only the second one identifies and removes the obstacles to a successful outcome.
I want to be precise about what that is and isn’t. The giants have cases too — Ohno had Toyota, Christensen had the disk-drive industry, Deming had the shop floors of postwar Japan. But their cases illustrate a principle. Mine traces a mechanism — backward from the failure, across the couplings, to the specific obstacle no one declared. That difference — illustration versus trace — is not a claim to be smarter than any of them. It’s a claim about where the work begins and how far it runs.
And it is why the other two pieces of the work exist. What stays constant across every technology era — mainframe to client-server to ERP to cloud to AI — is that readiness, not the technology, decides the outcome. I have called that Invariant Physics™. The mechanics of how a platform amplifies the operating conditions it lands on, for better or worse, is Implementation Physics™. Phase 0™ is the discipline that validates those conditions before the amplifier is switched on.
That’s the step the room was circling for a century without naming, because in a slower world the failure gave you time to catch it (or at least try to). Ohno assumed the floor was real because he walked it himself, and if he was wrong, the process hopefully corrected over months. The equation-based failure was always there; it was simply survivable — although they didn’t always survive, as FoxMeyer Drug did not. The consequences of AI don’t change the failure — they compress the reaction time and, in a complex global environment, can put survivability itself at risk. In short, there is more on the line today than in previous technology eras. The floor is now a data model, the amplifier runs autonomously, and the same error that once took a year to compound now compounds in weeks. That’s all Phase 0™ is: the assumption their eras let them skip, turned into a discipline for an era that moves too fast to skip it.
So if those ten walked into the same room and argued their way, eventually, to the thing underneath the argument, I don’t think they’d say something none of us had heard. I think they’d say what they always said — understand the reality before you act on it — and then look at what AI does to a reality nobody validated, and recognize it instantly. Not a new failure. The oldest one, accelerated. The same equation-based mistake they each warned about in their own vocabulary, now running faster than the correction can catch it.
-30-
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
Share this:
Like this:
Related