Like many Americans, I enjoy the monthly pleasure of calling Verizon to adjust my bill. OK, it’s not really a pleasure. It’s a ritual. Somewhere in Verizon’s billing department there is a machine, or possibly a person, or possibly a person who has become a machine, whose job is to add charges to my account that we did not agree to. My job is to call and remove them. We’ve been doing this dance for months. I assume it counts as a relationship.

This month the mistake was special. Large enough that I noticed it without looking, which is not the usual craftsmanship. So I made the call.

The first voice was an old friend: the classic chatbot. Chipper, dumb as a toaster. You know this bot. Everyone knows this bot. His purpose is not to help you. His purpose is to waste enough of your time that you give up and pay whatever number appears on the bill. He offered me menus. He offered me the website. He offered me everything except the one thing I called for.

So I did what I always do. I asked for a transfer.

And then something new happened.

For years we were trained like laboratory animals: press zero, say “representative,” keep pressing until a human appears. That ritual created its own expectation. After the first bot, the next voice is supposed to be real. So when the chipper toaster handed me to a woman with a warm, conversational voice, patient in the way call-center workers have learned to be, I assumed the old rule still applied.

Except.

I run an AI startup, and when you spend your days building these systems, you develop an ear for them, the way a sommelier can taste the oak. She had quirks. The answers came a beat too smoothly. The empathy arrived on schedule. The evasions followed a script no human would bother to memorize, because humans go off script.

So after a few rounds of pleasant nothing, I asked the modern question:

“Are you a chatbot?”

A pause. Then, in effect: well, yes.

Yes. Verizon had transferred me from a robot to a better robot. Escalation, in the customer service sense. I had been moved up a tier without changing species.

And here is the part I didn’t expect: I was impressed. Impressed by Verizon, a sentence I have never before typed. Not because anything got fixed. Nothing had. But because the vast majority of callers would never have known. The accent, the warmth, the small hesitations, all synthetic, all convincing. And the technology is not slowing down. This year I caught her. In a year or two, maybe I don’t.

Alan Turing proposed his test in 1950. It got settled on a customer service line, in the most tedious corner of American commerce, which is where you’d expect the future to show up first. Nobody wastes miracles on important things.

This is where I’m supposed to worry about Skynet. I mostly don’t. I doubt the machines are coming for us with red eyes and metal skeletons. If they are, they’ll have to get through their first assignment: customer retention.

The real worry is smaller, and worse. I did eventually reach a flesh-and-blood human, who fixed the mistake in about four minutes. But consider what I was calling about. A “mistake.” The kind that always, somehow, runs in Verizon’s favor. Month after month. A coin that lands heads every time is not a coin. It’s a business model.

Now give that business model a better robot. The bots may get better at serving us. I hope so. But they will certainly get better at the other thing: mistakes engineered to slip past us. Subtler. Priced just under the threshold where you’d bother to call. A machine that can almost pass for human can absolutely pass for an accounting error. The remedy is one sentence at the top of the call: you are speaking to a machine. Say it out loud, every time, whether or not I think to ask. Any company that fights that is telling you what it wants the confusion for.

That’s the real worry. Not robots that kill us. Robots that bill us.

After all, they learned from the best.

- Anastassia, co-founder VSTRAT.ai

Subscribe now

This weekend Spider-Man: Brand New Day posted the biggest opening day in box office history: $168 million, dethroning Avengers: Endgame, with $927 million worldwide in one weekend. Marvel Studios now owns the four largest openings ever recorded.

Ten years ago I wrote a bestselling case about how that became possible. The short version: Marvel nearly died. Repeatedly.

A Senate panel in 1954 blamed comics for juvenile delinquency, one psychiatrist testifying that Hitler was a beginner compared to the comic book industry. Marvel survived. A conglomerate that couldn’t tell Superman from Spider-Man bought it. Survived. Corporate raider Ron Perelman leveraged it into bankruptcy while paying himself hundreds of millions in special dividends and junk bond proceeds. Survived. Carl Icahn circled the carcass. Survived him too.

Then a toy executive named Ike Perlmutter and a turnaround CEO named Peter Cuneo rebuilt it from a company that almost missed payroll, licensing characters for scraps. Sony got Spider-Man for $10 million plus points. The first three Spider-Man films grossed about $2.5 billion; analysts estimate Marvel saw maybe $62 million. Blade grossed $131 million. Marvel’s check: $25,000.

In 2004, Hollywood veteran named David Maisel flew out to pitch Perlmutter something radical: stop licensing, build a real studio, make the movies ourselves, finance it with the characters as collateral.

Here’s the part I only learned talking to the principals during COVID. Maisel didn’t fly to some office park. He flew from LA to Mar-a-Lago, where Perlmutter holds court. And midway through the pitch that became Marvel Studios, in wandered a friendly neighborhood member wanting to say hello and talk about his upcoming reality TV show. Donald Trump, promoting The Apprentice, strolling through the founding meeting of the most successful film franchise in human history.

You cannot write this. Fiction editors would reject it as too much.

The pitch worked. $525 million in debt secured by the characters themselves, a studio above a car dealership with no free coffee, unknown actors on long contracts because the characters were the stars, and Iron Man in 2008. Disney paid $4.2 billion a year later, mostly in crisis-priced stock that made the real number closer to $10 billion. You know the rest.

Now look at this weekend’s producer credits, because nothing in this story is not odd. Avi Arad, who quit rather than watch Marvel make its own movies and sold his stock at what became one of the worst personal trades in entertainment history, still collects a credit, contractual sediment from the era his strategy built. Amy Pascal, whom Sony pushed out as studio chief a decade ago, kept Spider-Man producing rights on her way out the door; the company discarded the person, the person kept the character, and the company just had its biggest weekend ever on her film.

And Kevin Feige. Rich beyond counting, famous beyond need, twenty years and thirty-some films in, still making movies. Plenty of people with his position retired to Aruba ages ago.

Step back and consider how strange the whole timeline is. The visitor who wandered through the pitch meeting talking about his reality show went on to win the presidency in one of the great upsets in American history, lose it, then win it again. Meanwhile the character being pitched that afternoon spent those same decades resurrecting its own past, old villains, old timelines, old Peter Parkers, culminating in a movie that just broke the record its own studio set. A reality star ran the free world twice while a cartoon teenager from 1962 out-grossed everything Hollywood ever made. Fiction isn’t stranger than fact. Fiction couldn’t get this greenlit.

And notice the ground they built on, because it was never fair. Marvel created its first blue ocean, the 1960s character explosion, while DC literally owned Marvel’s distribution arm and throttled how many titles it could ship. It built its comeback while Sony owned its crown-jewel character and Fox held the X-Men, its own children licensed out for scraps, watching other studios profit from them. The platform that can strangle you, the rival that owns your best asset, the landlord who is also your competitor: that’s not the exception in building. That’s the terrain. Today’s founders building on AI platforms that might ship their product as a feature next quarter are not facing some unprecedented cruelty. They’re facing DC owning the trucks. Builders build anyway, and the ones who do inherit the record books.

But through every strange turn, one beat never changes: builders build. Not because they need to. Musk sleeps on factory floors he could buy and demolish on a whim. Buffett is still reading filings and trading in his tenth decade. Gates invests in technology companies while holding more money than some countries. None of them has needed a paycheck in living memory. The engine doesn’t ask about need. It’s the same engine that had Feige on set this year, thirty-odd films past the point where anyone would have blamed him for a beach.

Stan Lee wrote the reason into the character himself, the most famous line in comics: with great power comes great responsibility. Everyone quotes it about power. Almost nobody notices what the responsibility is. It’s not to hold the power, guard it, or retire on it. It’s to use it. To build.

That’s the actual lesson under the box office records. Arad had maximum position and traded it away because his judgment said the plan would fail. Feige had judgment, barely any position, and stayed, and built, and kept building through every year people asked when the fad would end. Position is what you hold. Building is what you do. The cap table forgot who was right. The credits never do.

Presidents change, twice over in this story. Owners change, raiders come and go, strategies get abandoned by the people who most opposed and most profited from them. The builders just keep building. Everything else is weather.

As the late, great Stan Lee said, closing every comic book: Excelsior.

Builders, ONWARD!

Subscribe now

We built software, wrote cases, wrote articles, including cases on Blue Ocean Strategy itself. Georges Doriot, the father of venture capital, had passed away long before either of us arrived, but we sat for years alongside people who’d worked directly with him and carried his thinking forward.

One thing I always wanted to do, though, was prove that Blue Ocean businesses were more likely to succeed. The book makes that case too, but the original research was limited by the technology available at the time, and the sample was relatively small.

Along the way, while researching the actual genesis of modern entrepreneurism, I got to speak with Arthur Rock, the man who coined the term venture capital, on a referral from Draper Sr., who’d taken Doriot’s class at Harvard Business School. Rock told me himself he regretted never taking it.

More recently, sparked by a piece of homework my wife was doing for Venture Lab, I built the idea out further inside VSTRAT itself. In v2, you build a strategy, talk it through with our stable of personas (or build your own), analyze it, and then run it against a Monte Carlo simulation to see how it actually performs over time, compared to competitive or marketing-led approaches. That simulation is what convinced me to finally write this.

The result: a portfolio loaded toward Blue Ocean businesses is typically 2-5x more likely to beat the S&P 500 than one built on competitive or marketing-led strategies. Somebody better at math than me could probably tighten that number. But the result repeats, run after run.

I’m finishing VSTRAT v2 now, but I’ll be opening the simulator itself to anyone who wants to try it.

If you’re looking for slow, steady, predictable returns, blue ocean businesses probably aren’t for you. Then again, neither is venture capital or private equity. But if you care about the entrepreneurial ecosystem broadly, this tool makes a strong case that blue ocean is where the real upside lives.

This whole thing traces back to Anastassia’s own homework for Venture Lab, and a genuinely great essay she wrote arguing that venture capital itself was founded and named by people the Blue Ocean community calls noncustomers, funding noncustomers.

- Michael

PS: if you’re managing a portfolio or building a strategy and want to run your numbers through the simulator, or build custom personas for your market, drop me a line or subscribe below.

Subscribe now

Michael Porter wrote Competitive Strategy. Clayton Christensen gave us The Innovator’s Dilemma. Chan Kim and Renée Mauborgne, people I know well having spent years working alongside them, wrote Blue Ocean Strategy. These are the books serious business people cite. The ones that show up on MBA syllabi, airport bookstore displays, and the recommended reading lists of people who want you to know they have recommended reading lists.

Elon Musk’s north star appears to be none of these.

Subscribe now

I think Musk’s bible is a science fiction novel from 1961 that the modern business press has completely ignored as a lens for understanding the most consequential entrepreneur alive.

Stranger in a Strange Land. Robert Heinlein. Published the same year Yuri Gagarin went to space. Currently sitting unread on the shelves of people who would find it extremely useful.

The premise: a human child, Michael Valentine Smith, is born on Mars and raised by Martians. He arrives on Earth as a biological human and a complete alien in every other sense. What he brings back from Mars is the concept that organizes the entire book, a verb Heinlein invented that has since drifted into the culture without most people knowing its origin.

To grok.

To grok is not to understand. Understanding is surface. To grok is to absorb something so completely, with such total focused empathy, that you and the thing become indistinguishable. You don’t analyze the situation. You become it. And once you’ve genuinely grokked something, you can move it. The world bends around real perception the way water finds the lowest point.

Now.

Musk’s AI company is called Grok. Not a coincidence. A man who names his artificial intelligence after a Heinlein verb about total empathic understanding is showing you his intellectual DNA, if you know where to look.

He is building rockets to go to Mars. In the book, Mars is not a destination. It’s a point of origin. A birthplace for a radically different way of seeing the world.

He has renamed a global communications platform and built around it something that functions less like a social network and more like a congregation, loved fiercely by its members and despised by everyone outside it. Smith in the book does exactly this. He builds a movement of grokkers. He is feared. Misunderstood. Accused of running a cult by people who cannot imagine that what looks like arrogance might be a form of perception they simply don’t have access to.

Smith also, not incidentally, establishes a free love commune. Musk obviously loves children and is openly working to repopulate the world with, grokkers?

The standard analysis of Musk is that he lacks empathy. Cold. Transactional. Drops people and projects the way most people drop subscriptions. But what if that reading is exactly backwards? What if the actual model is hyper-focused empathy, total and consuming, deployed completely until the thing is grokked and then withdrawn completely because the next thing needs grokking?

It would explain the obsessive focus followed by apparent indifference that baffles everyone around him. You cannot grok two things simultaneously. When the grokking is done, it’s done.

There is one more piece of this that deserves attention.

Smith in the book is not always gentle. When he perceives something as fundamentally, irreparably wrong with the world, he folds it away. Removes it from existence. Quietly. Completely. Without apparent emotion, because emotion would imply the thing deserved more consideration than grokking had already provided. The Martian does not agonize. He perceives, concludes, acts.

Here is what strikes me as genuinely strange. All of this is hidden in plain sight. Read Stranger in a Strange Land and then watch Musk for a week and the pattern is not subtle. It is not a stretch. It lands with the thud of something obviously true the whole time and you simply hadn’t looked at it from the right angle.

And yet search through the vast library of Musk analysis, the biographies, the profiles, the think pieces, the obsessive Twitter threads, the academic papers, the breathless hagiographies and the equally breathless takedowns, and the book is essentially absent. An entrepreneur who named his AI after a Heinlein verb, who is building rockets to the Martian point of origin, who built a congregation of grokkers on a platform called X, and the connection goes almost entirely unexamined.

Maybe nobody making the analysis has read the book. Maybe that’s the point. You can’t grok what you haven’t read.

Porter, Christensen, Kim and Mauborgne built frameworks for understanding how businesses compete. They’re excellent. They belong on every serious reading list.

But the man currently reshaping multiple industries simultaneously may be running on a different operating system entirely. One written in 1961 by a science fiction author most MBAs have never opened.

Grok it or don’t. He’s not waiting.

Share

Marc Andreessen once said software was eating the world.

Maybe. Once upon a time. Now, it’s AI slop. And it is not nibbling around the edges. It is devouring everything with the grace of a raccoon in a dumpster.

You have seen the headlines. LLMs will program everything. Tech executives high on their own hype have been sprinting to prove it. Take Brian Armstrong at Coinbase. In a move that will live in tech infamy Coinbase laid off actual engineers and proudly announced that non programmer project managers would now be writing code with AI.

The very next day? Catastrophic failure.

Let that sink in. The crypto exchange holding billions decided the future was vibe coded features slapped together by people who could not debug Hello World. Anybody still keeping serious money on Coinbase at this point is not investing. They are gambling with extra steps. When the next outage hits and it will you will have no one to blame but your own refusal to read the room.

This pattern is everywhere. Click view my hotel reservation and nothing happens. The system happily confirms it exists but the button leads to the digital equivalent of a shrug emoji. Plane tickets? Same deal. These are not edge cases. They are the new normal. Features that were clearly never tested because who needs testing when you have vibes and a prompt?

The engineers who used to say no the ones who kept the whole thing from collapsing have been marginalized reorged or fired. Now it is just PMs executives and their shiny new AI toys in the room. And do not forget the army of fake experts who could barely tell a byte from a bite whose entire expertise dates back to when ChatGPT launched.

The AI crapification is accelerating. Bugs are multiplying faster than crypto tokens in a bull market. And we are left with three possible futures.

The public just adjusts like we do with every other piece of crumbling infrastructure. Learn the workarounds when the app dies. Or miraculously the systems improve and the bugs retreat, the AI’s becoming much better. Or companies finally feel the revenue pain their customers have been screaming about and quietly bring real professionals back with AI as a tool not a pink slip factory.

Right now we are firmly on path one with occasional detours into clown world.

The backlash is here and it is delicious. AI sucks has gone from edgy whisper to open conversation. The magic is gone. The emperor is not just naked. He is generating PowerPoint slides about how his nudity is actually a feature.

Sam Altman the widely disliked face of OpenAI has become something of a tech pariah. Dario Amodei of Anthropic keeps dropping pronouncements about the glorious end of white collar, never mind if that came true nobody would be left to buy his tokens. The field sprinted into the gutter then complained when it got wet.

The heads of these AI companies are not going to be comfortable where this lands. The executives and believers who think their systems are fine are in for a hard lesson. Their bosses may be impressed by the headcount cuts and the shiny demos today but the market will have very different ideas soon enough. Customers notice when everything is broken. Revenue notices. Shareholders eventually notice.

Nobody knows exactly how this ends. Some forecasters are already pointing out that token costs will soon exceed hiring humans and humans shockingly are often better at not hallucinating. What is predictable is the trajectory: more slop more broken products more executives pretending this is fine.

The status quo is not sustainable. It is not good for users. It is not good for builders. It’s not good for firms. And eventually it will not be good for the balance sheets either.

Enjoy the slop era while it lasts. Just do not say nobody warned you.

Subscribe now

Share

Disruption has a dual meaning. One is the neighbor with a taste for loud heavy metal disrupting your sleep. The other is the business meaning defined by the late Prof. Clayton Christensen. Put simply, small things do the task that big things do for less cost until people switch to the small thing, which becomes the norm. Thanks to time, AI, and the end of COVID, we are seeing the end stage of that mess with calendaring and video conferencing.

The neighbor disrupts by forcing unwanted noise into your private space. SaaS marketers are now doing the same thing. They are disrupting the user experience with feature noise that nobody requested.

The setup question

Do you keep a calendar by Google, Outlook, Apple, or something else? Video calls by Zoom, Teams, Google Meet, or something else? Most readers will answer “all of them, depending on the meeting.” That is the problem in one sentence.

These systems were meant to be interoperable

The technical specs for calendars and video calls that work across platforms were settled long ago. iCalendar (RFC 5545) is open. CalDAV is open. WebRTC is open. Every system has APIs. Every system does the same thing. And yet they are all islands.

Why they are islands

Marketing dark patterns want to keep you locked in. The interop is technically trivial and deliberately broken. Subscribe to a Google calendar in Apple Calendar and the colors do not survive. Try to dial a Zoom from Teams and you cannot. Accept a Google invite in Outlook and the Meet link sometimes goes missing. Schedule a meeting from Bangalore to Austin and the timezone field will quietly disagree with the displayed time, leaving you to do the arithmetic in your head while half awake. Each system competes once at acquisition and then degrades the experience for years because switching costs are the moat. The lock-in is not a bug. It is the business model.

Subscribe now

Value down, cost up: the late SaaS game

I write a lot about value and cost and where cost can increase while value decreases. SaaS in its late stage is a clean example. Anybody who has worked with Zoom lately knows there are more options than at the Cheesecake Factory, and yet the vast majority of people simply want to create a video conference.

The product that was great in 2019 because it was simple and fast is now Zoom Mail, Zoom Calendar, Zoom Phone, Zoom Whiteboard, Zoom Notes, Zoom Docs, Zoom Clips, Zoom AI Companion, and Zoom Workplace. Each addition defends a revenue line. None of them improves the core experience of “click link, see person, talk.”

Calendaring is the same story in a different costume. Am I free for this meeting at this time in this timezone? Yes or no, and schedule. That is it. Yet thanks to disparate systems and timezone fields that lie about themselves, this has become a common cognitive tax. What was once the disruptor now looks ready to be disrupted.

The pattern, named

Successful tools that do one thing well get pulled toward becoming suites once growth slows. A focused tool has a price ceiling. A suite has higher willingness to pay per seat. The economic gravity of the public company quarterly cycle always wins against the engineering instinct to keep things simple. Slack added Lists and Canvases. Notion added databases and AI and forms. Dropbox tried to become a productivity company. The user experience erodes predictably. Software gets worse as companies get more successful at extracting value from it.

This is the predictable output of the SaaS business model meeting public market revenue expectations. The product strategy that maximizes shareholder value is identical to the product strategy that alienates the core user. The two converge until a disruptor shows up.

Why GenAI accelerates this

The underlying calendar, the underlying video stream, and the underlying email are all commodities. What matters increasingly is the layer above them. What is on my plate today? Summarize this meeting. Draft a reply.

If a neutral AI layer can answer those questions across any platform, the platform becomes the backend. The interface moves up a layer. The walled garden survives only as long as nothing can read across the walls. AI can read across the walls.

This is why every incumbent is racing to build their own AI inside their walled garden. Google Gemini in Workspace, Microsoft Copilot in Teams, Zoom AI Companion, Apple Intelligence. They are racing because a neutral assistant breaks the lock-in. Users did not ask for any of it.

What the disruptor looks like

If the assistant layer is where the value moves, the disruptor is whoever owns it. Christensen would predict it will be cheaper, simpler, and initially viewed as inadequate by the incumbent’s best customers.

It will operate above the platform layer, not at it. It will treat the underlying calendar, video, and email as commodity backends. It will solve the question the user actually has without caring which platform the data lives in. It will lose every feature comparison and win on the only thing that matters: doing the job. Plain text scheduling. A simple notification layer. No bundle. No upsell. It will charge less than the suite because it does not have to maintain the suite.

Closing

What was once disruptive becomes incumbent. What was once a sharp tool becomes a Swiss Army knife with too many blades. What was once “fast and simple” becomes “fast, simple, plus fifteen things you did not ask for, plus a price increase, plus an AI assistant trying to upsell you.” People want value typically in the form of less friction to match the expected output. Companies want lower cost, hopefully by removing cost that does not add buyer value. Aiming that lower cost/higher value offering at both customers and noncustomers of a current offering is the path towards a blue ocean. And yet, we’re seeing exactly the opposite become the norm.

Share

I have serious questions about AI’s ability to do certain things. I’ve made that clear. But there is one area where machines will be vastly better than people, and I have the anecdote to prove it.

I’m self-employed. I have an ACA plan. My wife is on that plan. I signed up over the phone on the advice of an insurance rep, despite doing almost everything online. They typo’d her birthday by one day. One digit. Their mistake, not mine.

This should take 30 seconds to fix. Instead, I got an interrogation about revenue, taxes, and whether we’re planning to move. A machine would see that the date entered differs by a single digit from all corroborating documents, fix it, and move on. No script reading. No irrelevant questions. No agent fishing to sell another policy, which I suspect is exactly what was happening.

The “Human-in-the-Loop” Fallacy

We often talk about keeping humans “in the loop” to ensure empathy and judgment. But when a process is governed by a rigid, poorly designed flowchart, the human isn’t providing judgment. They are merely a slower, more expensive, and more biased processor than a piece of Python code.

A basic validation script would catch a one-digit discrepancy against existing records and execute the change. The human agent, by contrast, has incentives that are diametrically opposed to my goal of a 30-second correction. Commissions. Call-duration metrics. Fear of deviating from a script. Their employer designed the process to extract revenue from every interaction, not to resolve problems efficiently.

The Billion-Dollar System That Still Can’t Validate a Date

We all remember the ACA Marketplace website launch. A billion-dollar disaster. That is the same system I just called into. Not a metaphor. The same platform. They spent a billion dollars building it, it failed, and their solution was to route people to a human on a phone, reading from a script, asking questions that have nothing to do with a transposed digit. The system that couldn’t validate a date in 2013 still can’t validate a date in 2025. Technology evolved; the ACA website apparently did not.

Here’s what should worry us: the danger isn’t that AI will be “evil.” It’s that it will be wrapped in the same bureaucratic layer that crippled the website. Give a broken process a faster engine and you just reach the wrong destination sooner.

Why This Isn’t Just an Annoyance

I have MS. I do NOT want a different policy. My meds are pre-approved and set. My wife went for a routine checkup and discovered this problem. All I want is to fix a typo they made. That’s it.

When you are managing a condition that requires precise continuity of care, a “cascade of re-verification” isn’t an inconvenience. It is a threat to medical stability, arguably an American with Disabilities Act (ADA) violation on its own. The agent doesn’t benefit from a quick fix. They benefit from a longer call, a new enrollment, and a fresh commission. The incentives are structural, not incidental.

Can AI handle ACA enrollment? Almost certainly. The waste generated by the current system is externalized onto the people it’s supposed to serve. We don’t just need better code. We need a complete dismantling of the commission-and-script architecture that treats patients like data-entry errors.

The solution is obvious. Bring on the robots.

Subscribe now

Share

I recently hired a contractor to rebuild the front end of VSTRAT. He submitted regular progress updates and said everything was on track.

After a week, I asked to see what he’d built. There was nothing. A stub. The kind of empty scaffold that Claude Code generates when you type “build me a Next.js app” and hit enter, including fingerprints showing that’s exactly what he did. A full week of reported progress and the sum total was a default template with nothing inside it.

Scammer? Maybe. Or maybe he read the same articles you’ve been reading. AI writes code now. Anyone can build software! He likely fell for the allure of “Vibe Coding,” the practice of prompting an AI to generate code that looks right and runs once, without actually understanding the underlying logic, state, or architecture.

He took the gig believing he could prompt his way through it. He couldn’t, though after being fired did hack out something in a few hours. Anyway, instead of admitting being lost, he faked updates until pressed on specifics. I caught it because, unlike most executives thinking they’ll save some money, I’ve been doing this a long, long time.

Subscribe now

The Invisible Barrier

AI code generation produces starting points. Scaffolds. Boilerplate. Or focused changes. For an experienced developer, this is a genuine power-up. I use it daily and it makes me significantly faster. It augments the expertise I already have much like VSTRAT augments strategy work.

But there is a gap between a scaffold and a working product. That gap is where engineering actually lives. It’s where you debug the state management issue that only surfaces under load. It’s where you make the architectural decisions that determine whether the thing holds together or falls apart the first time a real user touches it.

Vibe coding is about the feel of progress. Engineering is about the fact of reliability.

My contractor sat in front of a tool that gave him an empty shell and apparently packed, unsure what to do next.

The Webflow Canary

After I let him go, I tried Webflow, the well-known visual site builder. No code required.

The most recent incarnation of Webflow itself appears to have been vibe coded. The platform is increasingly full of bugs that suggest a lack of oversight. In my case, the credit card field on the payment form simply didn’t render. It wasn’t a styling issue: the component just wasn’t there. Wanting my front-end, I found a workaround spending more time than the Webflow team did testing their site.

There is a profound irony in a multi-billion dollar company, designed to democratize web development, failing at a basic task like collecting a payment. I signed up with a temporary credit card number just in case. If their payment processing is as buggy as the rest of the product, I wanted to ensure they couldn’t keep charging me.

I chose a template and tried to export it, figuring I’d modify things in code. The export failed, with a dialog I left on for half an hour before realizing it’s clearly another bug.

I demanded a refund, answered by a person, and built it myself. It’s not my focus at all but took less time than I’d spent managing the contractor and fighting with Webflow combined.

The Real Accounting

Dario Amodei has said publicly that AI will write most code soon. He runs the company that makes Claude, and with an IPO on the horizon, the boldest possible claims about AI serve his interests. But I think he’s wrong. Not about what AI can do, but about what it replaces.

AI doesn’t replace engineering. It replaces typing. Just as spreadsheets didn’t replace financial analysts but made the good ones more efficient, AI augments the expert while leaving the novice stranded in a hallucination. The thinking, the debugging, the architecture, the hard decisions about tradeoffs under constraints: that’s the job. The typing was never the expensive part.

By freeing experts from routine tasks, AI enables them to imagine and experiment more. But that only works if you already have the mastery to know where the routine ends and the critical thinking begins. When companies cut experienced developers and replace them with junior staff armed with AI tools, they don’t save money. They move the cost.

It shows up weeks or months later as rework, as outages, as cancellations, as products that look finished but break the moment a real user touches them. The invoice from a professional developer feels expensive because you can see it. The cost of vibe-coded software is invisible until the damage is done.

The Fuse

Was my contractor a bad person? OK - the fake status reports aren’t a good sign. But even if he was innocent in his intentions, the outcome is the same. He took a contract he couldn’t deliver. He submitted updates for work that didn’t exist. When pressed, there was nothing to show. Fraud? Maybe. But it’s definitely a reputational hit and reputations, once broken, don’t get patched with a prompt.

He didn’t just lose this contract. He lost the next one, and the referral after that. The person who recommended him knows what happened. So do I. The vibe coding hype didn’t just fail him on the technical side. It cost him something much harder to rebuild.

AI is spectacular at augmenting people who know what they’re doing. It does not turn novices into experts. It turns novices into people who can’t tell the difference between working software and a convincing hallucination. I call GenAI a Dunning-Kruger machine.

If you’re out there right now taking on work because AI told you the barrier is gone, be careful. The barrier is still there. You just can’t see it until you’ve already committed to delivering something you can’t build.

Don’t vibe code your reputation.

Share

Conspicuous Compute

Take a look at the above image of a real leaderboard. They call it “tokenmaxxing.” Engineers and business users are competing to see who can burn the most AI tokens per month. Leaderboards are posted. Bragging rights claimed. Jensen Huang recently said he’d be “deeply alarmed” if a $500K engineer wasn’t burning through $250K in tokens a year.

Read that again. The CEO of the company selling the shovels wants you to dig more holes.

Subscribe now

Conspicuous consumption, updated for the cloud era.

The old version: a Russian oligarch bathing in Cristal because he could. Stupid, but at least the champagne was real. The new version: Michael Lee at the top of the leaderboard burning through 5.8 billion tokens because … ok; we don’t know why he did it or what he produced.

There was a time, not long ago, when engineers cared about efficiency. Mainframe operators rationed compute like water in a desert. Every CPU cycle had a cost and a purpose. Then cloud computing arrived and the discipline evaporated. “Scale” became the excuse for waste. And now tokens have completed the transformation. They rebranded compute cycles, gave them a friendly name, and turned profligacy into a personality trait.

Cognitive Surrender’s Cousin

Recently, I wrote about cognitive surrender, the quiet trend of outsourcing judgment to AI. Tokenmaxxing is its financial cousin. One is about surrendering your brain; the other is about surrendering your budget. Both assume that more AI is always better.

Adam Smith saw this coming in 1776. In The Wealth of Nations, buried hundreds of pages past the famous bit about pin factories (as was the Invisible Hand), he warned that when a person’s work is reduced to a few simple operations, he “naturally loses the habit of such exertion, and generally becomes as stupid and ignorant as it is possible for a human creature to become.”

Smith was writing about the division of labor in early factories. Replace “few simple operations” with “prompting an agent to do your thinking” and the sentence doesn’t need updating. When you stop thinking because the machine thinks for you, you lose the muscle. Both lead to the same place: dependency without understanding.

In the movie Idiocracy, the population waters crops with a sports drink because an algorithm told them to. Crops die. Nobody can figure out why. The machines keep running. The humans keep not thinking. The movie set this hundreds of years in the future. They got the premise right but blew the timeline. We’re building that world right now, on a sprint, and celebrating the people who get us there fastest on a token leaderboard.

Nobody Left to Pay

Here’s what keeps getting ignored. The founders of the major AI labs enthusiastically predict the end of white-collar work. They describe a future where lawyers, analysts, consultants, and middle managers are replaced by agents.

Uh .. ok. Maybe. But if they’re correct, who pays for the tokens or the energy to create them?

If the white-collar workforce is gutted, who has the income to subscribe to the tools that gutted them? This is the dog catching the car. The entire business model depends on the continued existence of the people the technology claims to eliminate. Agents sell to agents, consuming tokens paid for by budgets that no longer have a revenue source.

There’s also the issue of what happens with the cranky displaced people which hasn’t ended well historically.

The Cerebellum Check

The cerebellum is the part of the brain that controls coordination. Balance. Knowing where your body is in space. The tokenmaxxing crowd is losing something similar: the coordination between spending and value, between capability and judgment, between what the machine can do and what it should do or, more accurately, how we should use it.

Before you celebrate your spot on the top ten list, ask one question: what did those tokens produce that you couldn’t have done with a tenth of them and five minutes of your own thinking?

If you can’t answer that, you might want to revisit not how many tokens you’re producing but why, what is being produced with them, and where is the value. If it’s not commensurate with the cost of the tokens rethink whether being on the leaderboard is a good idea.

Share

The first sign of my multiple sclerosis sent me to the hospital for a CT scan and an MRI. Something was clearly wrong because I couldn’t feel half of my face. The doctors found nothing on the CT scan and suggested a silent stroke based on the MRI. Looking back, the diagnosis was, well, staring me in the face. I have worked with AI most of my life and assumed a machine would see the scans and solve it. In reality, reading radiology is a desk job with a supposed absolute truth, yet AI still fails to master it.

We see this in finance too. Quant trading is how the massive banks make their money. Smart humans program machines to trade stocks. Those machines mint money for a while. Eventually, often in seconds, another machine catches up. Then the humans must return to the drawing board to build something new.

Subscribe now

This leads to the talk of a job armageddon where all desk work disappears. Dario Amodei from Anthropic brags about eliminating white collar jobs to please his investors, yet his own firm is hiring more software engineers. If radiologists and quant traders are still highly paid after years of AI development, something is wrong with that theory. Even software engineers are working more hours now, and total job listings are increasing.

Turns out that even on medical scans, an absolute truth, AI’s were making it up. They’d ignore the pictures, take the written record into account, and essentially guess based on information the patient and doctors entered, and do a pretty good job guessing. If asked simply to guess, they’d do markedly worse suggesting they were leaning on human input more than machine scans or random.

Our firm, VSTRAT, focuses on strategy. If a machine cannot find a single truth in a medical scan, it has no chance with corporate strategy. Strategic answers are subjective. A machine can produce a beautiful deck that looks like a consultant wrote it, but those slides are often nonsensical.

Take the recent hype around Manus, the system Meta just bought for 2 billion dollars. It predicts that AI will grow the fine wine market. That is idiotic. Boomers are the primary wine buyers and they are dying off. Younger generations simply do not drink wine at the same rate. The machine does not see a shifting culture. It just sees a giant average of past data and assumes the future will be more of the same. It is a giant average machine but strategy, hopefully, is not about being average. It is about being the outlier. If every company uses the same AI average to generate a roadmap, they will all arrive at the same conclusion and compete on the same thin margins.

I write these pieces by hand and use various LLM’s to edit them. I shared a draft of this piece with Claude, often considered one of if not the best. It confidently corrected me on Manus, not on the incorrect facts but insisting Manus was never a Meta products, that Meta had made no such acquisition! Umm … off by $2 billion dollars.

From Claude: “The Manus paragraph has a factual problem. Manus is not a Meta product. It’s a Chinese autonomous AI agent from a startup called Monica. You may be thinking of something else, or conflating it with another acquisition. That needs to be fixed before publishing — it’ll get flagged immediately by readers.”

The machine had no idea what it was talking about and delivered the error with complete authority. If an AI cannot accurately recall a two billion dollar acquisition that was widely covered in the press, it has no business advising a company on where to compete next year.

A lot of the layoffs we see today use AI as a convenient excuse. Jack Dorsey at Block recently cut 40 percent of his staff while blaming the machine. In reality, Block had a massive headcount compared to its larger competitor, Stripe. Dorsey has a long history of over hiring. Blaming managerial incompetence on AI is a dishonest way to escape moral culpability and the very real people involved.

My daughter, Cienna, wants to go into behavioral economics. This is the exact right path for the age of AI. The field rejects the myth of Homo economicus in favor of the reality of Homo sapiens (that’d be .. us). Traditional models and AI algorithms are built on the logic of Homo economicus where data and optimization rule. They assume that all things being equal, people will maximize financial returns. Nope. Humans are fundamentally irrational. We do entirely off the wall things based on emotion, tribalism, or simple whim.

Leveraging and understanding that irrationality is how one builds effective strategy. You do not win by assuming humans will behave rationally. You win by understanding they will not. Strategy especially is an art because it requires empathy for these biological quirks. When it comes to subjectivity and feeling, us biological blobs of protoplasm still have an edge over the giant averaging machines.

Like I said, I use the LLM’s as copy editors trying different one’s for feedback. I ran this through ChatGPT and it made some suggestions on accuracy. I told it all the facts are well sourced and it became defensive. The machines are useful as copy editors, or co-thinkers. But anybody who uses them for ground truth is Cruisin' for a Bruisin', as us humans say.

Epilogue: I write these pieces in advance then give one last review before publication. I ran this through Gemini with the instruction to comment, not rewrite. Despite being created and owned by Google, it made the exact same mistake Claude did, that Manus wasn’t acquired by Meta! I corrected the machine by pointing out that the first answer in Google to that question is a high quality news article about the acquisition and there are plenty more high quality cites. It then displayed silicon embarrassment. Save yourself from the same embarrassment in front of a boardroom: use the machines, and VSTRAT, to augment your own thinking, not replace it.

Subscribe now

We are currently drowning in a sea of crazed AI pronouncements and a lot of very expensive crap.

In 2024, Anthropic CEO Dario Amodei famously predicted AI would kill all jobs by 2025. It is now 2026. Entry level jobs are still here. Amodei’s prediction may come true or may be him telling investors he’ll be king of the world and hoping for giant payday. In any event, the idea that robots will be replacing or ordering us around brings us to a new term: Cognitive Surrender.

Subscribe now

It describes our growing habit of deferring to an LLM, even when we know it is wrong, simply because it sounds authoritative and produces beautiful slides. It is the high tech version the well dressed consultant showing a nonsensical deck. Dress nonsense in a tuxedo and people will apparently believe almost anything.

The examples are already piling up. Deloitte billed the Australian government AU$440,000 for a report that turned out to be largely AI generated, parroting attorneys who have fabricated court rulings and non existent academic citations. Air Canada’s chatbot invented a bereavement discount policy that did not exist, and when sued for the discounts, the airline actually tried to argue the bot was a separate legal entity responsible for its own actions. Academic publishing has been flooded with AI generated slop. The 2026 ICLR conference saw a 70% surge in academic paper submissions, nearly 20,000 papers, far more than human researchers can actually peer review. Humans can’t sort through that pile of silicon generated bullshit so AI’s are poised to review the work of AI’s.

In each case the output looked authoritative. In each case someone surrendered their judgment to a machine, either because they were forced to or thought the machine must be right … and paid for it.

The €38 Billion Hallucination

Last week I tested Manus, the tool Meta bought for billions in late 2025, using the fine wine market as my test case. Having worked directly with the authors of Blue Ocean Strategy and taught the bestselling case on the subject for over a decade, I know this territory.

Manus spent ten full minutes thinking before proudly projecting €38 billion in growth by 2030.

Its brilliant insight? Treat wine like a collectible market.

I happen to have written a bestselling case about Marvel’s plunge into bankruptcy in the 1980’s. I know exactly what happens when you try to pivot a consumer product into a collectible play: you go bankrupt. For the wine market, where the Wine Market Council 2026 data shows overall consumption is down and 35% because Millennials simply do not like the taste of wine, projecting a €38 billion in growth should taste off, like a bottle with a bad cork. Besides, wine as collectible isn’t exactly new.

Honestly, anyone predicting that kind of growth in a shrinking demographic sounds like they have been sampling too much of the product themselves. The slides looked fantastic. The analysis sounded like something from a comic book.

The Fork in the Road

When I ran the exact same prompt through ChatGPT, it took four seconds to spit out exactly 800 words of generic filler (what a coincidence on the precise word count .. not). Just a wall of bland text designed to make you feel like the work was already done.

The irony is brutal: the doom and gloom predictions might actually come true, not because the machines become sentient, but because they make humans confidently incompetent. Or simply lazy.

Type a question, receive a polished and confident nonsense; hand your thinking over to the machine. The other, our approach at VSTRAT, treats AI as infrastructure for human judgment, interactive, grounded in data, forcing you to actually think. Plenty of people take the lazy path even telling others that’s the correct way to do things.

The cost of choosing wrong is not just a bad strategy deck. Business schools, for example, that cut rigorous thinking tools to save a few hundred euros, replacing them with systems that produce confident beautiful nonsense, are training the next generation to surrender cognitively before they even start. That is the jobs story nobody is telling. Not AI replacing humans, but humans surrendering what makes us human to replace themselves.

We know that living in an opium fog may be nice (I’ve never tried it myself but who knows…) but we also know it isn’t a good idea. Yet we’re, as humanity, heading down the same path.

Look to the Light Table

Recent history already gave us the better model.

Not long ago, graphic designers worked with Exacto knives, light tables, and physical templates. When desktop publishing arrived, the experts predicted mass unemployment. Instead, we saw an explosion in the industry. The tools lowered the floor so anyone could make a flyer, without lowering the ceiling for people who actually understood what good design looked like.

The same thing is happening now. AI can augment us. I use it as an editor: I write every piece by hand, then ask the LLM to act as a ruthless thought partner on structure and clarity. It is a fine editor. It is a terrible strategist. If I asked it to think from scratch, it would still tell me the wine market is a gold mine.

When I digress, as humans do, the machines also tell me to get back to task even if it’s an evening or weekend. They’re not very empathetic task masters, which makes sense given they’re robots with no emotions.

This worry isn’t new. When I first learned to program as a teenager decades ago I created a poem generator that took words from a list and randomly put them together. I kept creating new poems until one made sense and even submitted it to a poetry journal. I did receive a response - a few choice not very poetic words - and if that journal is still in existence I’m apparently banned for life.

A final note: do not blame the AI. Blaming a machine for cognitive surrender is like blaming a car for driving off the road. The car did not let go of the wheel. You did.

Subscribe now

Adam Smith was a shy, studious man who spent twelve years writing a book that became the founding text of modern economics. He had a steady, if modest, income as a professor. He had a library. He had time.

James Watt was at the same university, at roughly the same time, not as a professor but as the university’s scientific instrument maker: embedded in the academic world, useful to it, respected by it, but not protected by it. He worked in a small shop on campus fixing instruments for scholars and helping make their work possible. He moonlighted as a surveyor. He spent years trying to interest anyone with money in a more efficient steam engine before he finally found Matthew Boulton, a Birmingham manufacturer who had both the capital and the commercial instincts to turn Watt’s invention into a business.

Subscribe now

The two men were not strangers passing in some abstract historical fog. They were at Glasgow together. They knew one another. They moved in the same intellectual world. Smith theorized the invisible hand. Watt had to shake it and negotiate terms.

History has never much cared that this was no coincidence. Smith gets the canonical status, the economics textbooks, the currency. Watt gets a unit of electrical power named after him and a footnote. The man who described how capitalism worked is remembered. The man whose engine powered it is not.

It is a recurring trick of history: we remember the observer, but we forget the observed. We honor the man who explained the machine, while the man who built it, and the people who eventually broke themselves against it, become mere data points in someone else’s thesis.

The oversight matters because it helped create a hole in Smith’s framework that Marx later drove a train through. And not metaphorically.

Smith’s model was elegant and, for its time, largely correct. Free markets, free trade, the division of labor as a source of productivity. The pin factory illustration is still taught today. But Smith theorized markets from the outside, from a lectern, with a guaranteed salary. Watt stood closer to the seam between ideas and labor, between the university and the market, between intellectual life and commercial necessity.

Smith glimpsed the cost of his own theories, warning that a person whose life is spent performing a few simple operations becomes “as stupid and ignorant as it is possible for a human creature to become.” He called it mental mutilation. But he treated it as a tragic side effect of progress, not a structural flaw that would eventually set the world on fire. He saw the efficiency of specialization without fully reckoning with what Watt’s engine would do to the people performing it.

The Industrial Revolution that followed was not the genteel marketplace of Smith’s imagination. It was factories, child labor, sixteen-hour days, bodies broken by repetitive work at a pace no human guild system had ever demanded. The division of labor that Smith celebrated as a source of wealth became, in practice, a source of profound human misery for the people doing the dividing.

Marx looked at that gap between what the observer promised and what the observed actually delivered, and found everything he needed. The exploitation was not incidental. It was structural. Capital accumulated precisely because labor could be compressed, cheapened, and discarded. The theorist’s blind spot handed the critic his entire thesis.

And then labor found a way. Not by burning the machine, but by learning to drive it, bargain with it, and eventually shortening the workday from sixteen hours to eight. Not without pain. Not without decades of genuine suffering, political upheaval, and movements that changed the relationship between capital and work permanently. But the armageddon did not arrive. The steam engine did not end work. It destroyed some kinds of work and created others. The people who adapted, who found the new capabilities and learned to use them, got better. The ones who could not or would not faced real hardship. That is not a comfortable story, but it is the true one.

We are in that cycle again.

The Darios of our moment are sincere. The concerns about labor displacement from generative AI are not imaginary. There will be real disruption, real pain, real jobs that do not come back in their current form. Anyone who waves that away is not paying attention.

But the fear today has a new shape. If Watt’s engine threatened the body with physical exhaustion, the Large Language Model threatens the mind with cognitive surrender.

This is the modern evolution of Smith’s “mental mutilation”: the slow atrophy of the muscles of critical thinking, writing, and decision-making because we have outsourced the “why” to a prompt. The risk is not just that the machine replaces us, but that we become as “stupid and ignorant” as Smith feared by letting the machine do the heavy lifting of the soul.

The steam engine did not end work. The printing press did not end scribes and create only unemployment. The word processor did not destroy writing. In each case, the technology changed who could do what, at what cost, and the people who were already good at the underlying thing, thinking, communicating, building, got considerably better at it with the new tool.

Watt was not trying to put surveyors out of work. He was trying to pump water out of mines more efficiently. Marx was not trying to predict the gig economy. He was trying to explain what he saw in front of him. And Smith, brilliant as he was, could not have imagined what was being built in the shop down the hall by a man he actually knew.

The engineer at the next desk is always building something the theorist has not accounted for yet. That has always been true. It is true now.

The question worth asking is not whether AI will end work. It is the same question it has always been: who learns to use the new engine, and who waits for someone else to tell them what it means.

Watt did not wait for a footnote. He found Boulton, shook the invisible hand, and shipped the engine that would rewrite the book Smith was still writing down the hall.

Subscribe now

I have MS. I do not talk about it constantly, but it is there, and it shapes how I think about a lot of things, including technology, medicine, and the gap between what is possible and what actually gets built.

Here is the good news about MS in 2026: thanks to modern medicine, a degenerative neural disorder is often not so degenerative anymore, as long as you take your meds. The meds, however, are not aspirin. They are sophisticated biological agents administered intravenously at something called an infusion center.

Subscribe now

Despite sounding like something you would find at a boutique hotel bar, an infusion center is actually a room full of comfortable chairs where nurses hook people up to IV bags of expensive drugs. My first visit was last week. Most of the other patients were regulars, showing up at carefully scheduled intervals for treatments targeting MS, rheumatoid arthritis, Crohn's, rare genetic disorders, and other conditions that require ongoing, precise medical management. These are not casual patients. These are people whose lives, in some meaningful sense, depend on showing up. Several were already using walkers. One arrived in a wheelchair. These appointments exist to prevent exactly that.

So you can imagine my reaction when I got home and found this text waiting for me:

“Hi Michael, Thank you for your recent service with Paragon at Austin 720 West 34th St.! Your opinion counts! Please take the time to complete a one question survey at the following link...”

My recent service.

I did not get my oil changed.

Neither did any of those other people in those chairs.

A phrase that generic is not just bad marketing copy. It reflects something bigger and worse: institutional decontextualization. The system knew perfectly well where I had been, why I was there, what drug I had received, and when I needed to come back. Yet somehow it spoke as if I had stopped in for a tire rotation.

A long time ago, back when “AI” meant expert systems and the iPhone did not exist, I built something for exactly this problem and filed it as a patent. Deliberately. I wanted it in the public record as prior art so nobody could lock it up and so anyone who wanted to build something like it could do so freely.

The idea was simple: healthcare providers already know who you are, what condition you have, what drug you were just given, and when you need to come back. Why should the follow-up communication reflect none of that?

The technology to do this crudely existed in 2002. It exists with remarkable sophistication today. So I did what any curious person with an LLM handy would do. I asked Claude to rewrite that text message given the actual clinical context of an MS infusion. Here is what it came up with:

“Hi Michael, we hope your first Briumvi infusion went smoothly. For the next 24 to 48 hours, watch for headaches, fatigue, or mild fever. These are common after a first infusion and usually resolve quickly. Anything more serious, like difficulty breathing or chest tightness, call us immediately at [number]. Your next infusion is typically scheduled two weeks from today. We’ll reach out to confirm, or click here to add it to your calendar now. How are you feeling?”

That took seconds to generate. It is personalized to the actual drug administered. It provides clinically relevant information. It is warm without being saccharine. It offers a concrete next step. And it asks a human question at the end.

No new data was required. The infusion center already had everything it needed: my name, my diagnosis, the exact drug I had just received, and my next appointment date. I knew where it was; I don’t go infusion-center hopping on the same day. The LLM simply connected the dots the way any thoughtful nurse would, if every nurse had infinite time and a perfect memory.

I am not here to predict jobs armageddon. I find those pronouncements mostly theatrical.

But this is different. This is obvious.

What I just described does not merely improve a patient message. It eliminates the need for communications that strip away context, flatten the patient into a transaction, and then wonder why healthcare feels alienating. It closes part of the gap between clinical excellence and patient experience, a gap most healthcare systems seem to treat as a routine cost of doing business.

And here is the thing: it is better for everyone. Better for patients, who get information that is actually useful. Better for providers, who build trust and reduce anxiety calls. Better for the business, because a patient who knows what to expect after an infusion is less likely to panic, less likely to miss the next appointment, and more likely to stay in the system.

Like it or not, that is the future. Not because it is inevitable in some grand technological sense. Because it is simply better, cheaper, and, though generated by a machine, more human than what is being sent now.

Your recent service.

Honestly.

Subscribe now

“Vibe coding” is the latest fad. Programming is dead, they tell us. All those people creating software will have to find other jobs because it’s all so easy now.

Let’s slow down.

There have always been two parts to coding, at least for decades: deciding how a program should work and actually creating it. LLMs are fantastic at the second part. They stink at the first.

Coding, the act of typing syntax, wiring APIs, handling edge cases in loops, has never been the magical part of programming. It’s robotic, mechanical, almost industrial. There’s no surprise that a robot can do it well, and often faster and more consistently. The real sorcery has always been upstream: deciding what to build, why it matters, how the pieces should fit together conceptually, when to break convention for a better outcome, and, crucially, when the elegant solution on paper will become a nightmare in production. That domain remains stubbornly human.

As I’ve outlined in prior pieces, LLMs just aren’t very good at reasoning. That doesn’t mean people won’t defer to them, a phenomenon we call cognitive surrender, but like Catherine the Great floating down the Dnieper admiring Potemkin’s entirely fake villages, they’re awed by the spectacle without questioning what’s behind it.

But this piece isn’t really about vibe coding.

Subscribe now

Vibe Teaching

LLMs are infinitely more patient than any teacher or professor. Some studies suggest they’re better at conveying information. They’re much cheaper. We’ve even heard about people spending hours teaching others how to “prompt” an LLM, an exercise that costs about $20 per month for anyone to access on their own (hint: just ask the LLM).

But teaching isn’t information delivery. Teaching is watching a student’s face when they hit a wall. It’s knowing when to push and when to back off. It’s the professor who pulls you aside after class and says “you’re better than this paper.” An LLM will patiently explain long division forever. It will never notice that a student hasn’t eaten breakfast or that their parents are getting divorced.

Vibe Parenting

Just feed in your kid’s age, weight, schedule, and preferences, then follow the LLM’s instructions about what to do. You become the robot, at least until real robots are produced.

The absurdity is obvious here, which should make you wonder why it isn’t equally obvious with coding.

Vibe Lawyering

This one seems to be gaining traction. LLMs are fantastic at pretending to be $1,200 per hour white-shoe lawyers. They write beautiful briefs. They cite relevant precedent (sometimes real, sometimes hallucinated, but always confident).

Except the real purpose of most lawyers isn’t legal scholarship. It’s preventing conflict or working through it. Most people don’t need a brilliant appellate litigator. They need the rumpled-suit, Maalox-drinking-from-the-bottle kind of attorney who calls the person who calls the ex’s lawyer and says “how do we move forward?”

Try getting an LLM to do that.

Vibe Doctoring

I won’t dwell on this one long because it’s personal. I have MS. Usually it’s under control. Usually. But when it first reared its ugly head, it looked like a stroke, and that’s what the doctors treated it as. After a battery of tests, they figured out what was actually happening. Scary but not fatal.

Now reverse that. Imagine a person plugs their symptoms into an LLM, gets told it’s probably an MS flare, and heads home to rest instead of going to the emergency room. And it was actually a stroke. A fatal or severely debilitating stroke.

The LLM was confident in both scenarios. That’s the problem.

Vibe Legislating

OK, this one might actually work, in the sense that the LLMs would probably get something done. What that something would be might not be exactly palatable, but it would at least make sense to a robot. Which, given the current state of Congress, could be a real improvement.

The Uncomfortable Truth About Averages

Studies that say LLMs can outperform humans tend to say they perform a little better than average. Average is the definition of a C in school, a grade most people aren’t going to crow about.

Professionals aren’t average. That’s the whole point of being a professional. A surgeon who performs “a little better than average” isn’t someone you want operating on you. You want the one who’s done this specific procedure a thousand times and knows what to do when something goes wrong that isn’t in any textbook.

Businesses laying off front line workers are pleased with the cost savings .. until the complaints and the cancellations come in. Get rid of talented techies only to find them popping up with a competitive product or for a competitor. Replace a system with a vibe coded one by an amateur (“I can prompt ChatGPT!”) only to watch your customers flee and your brand value reduced to $20/mo.

What LLMs Are Actually Good For

We love LLMs. We use them extensively. We built VSTRAT, a platform that leans heavily on LLMs as structure. But the key word is structure. The frameworks and methodologies that channel the AI are what make it useful. And it’s still up to people to work through where that structure leads.

A senior developer using an LLM to code is a professional using a power tool. A non-developer using an LLM to “vibe code” is a person with a nail gun and no blueprints. One of them builds a house. The other stars in a horror film.

Look, despite decades of kitchen gadgets, meal kits, cooking shows, and now AI-generated recipes, people still go to restaurants. More than ever, actually. The tools to cook at home have never been better or cheaper. Yet somehow restaurants aren’t closing. Because the value was never just the food. It was the judgment, the experience, the craft, the human who knows when to break the recipe.

The same will be true of coding and every other profession the “vibe” crowd claims to have disrupted. Some people will tinker at home until they realize the true cost of maintaining what they’ve built. The rest will call a professional.

Meanwhile, the loudest voices proclaiming the end of professional work tend to be LLM company executives working hard to secure investment dollars for their money-losing businesses. The FUD is irresponsible and it’s hurting real professionals. When the people telling you the revolution is here are the ones selling the revolution, apply some of that stubbornly human intuition.

Intuition

Intuition is uniquely human and a human gift. Cherish it. It’s not the most efficient. It can be entirely irrational. And at its best, it’s beautiful.

We love our robotic friends. But LLMs are not going to substitute for people in any meaningful way, no matter what the cheerleaders say. They are spectacular assistants. They are terrible replacements.

The person who vibes their way through coding will produce something that looks right until it breaks. The person who vibes their way through medicine will be fine until they’re not. And the person who vibes their way through parenting was never really parenting at all.

Professionals guide LLMs. LLMs help professionals. Get the direction of that relationship wrong and the consequences aren’t theoretical. They’re real, and they’re coming.

Share

AI makes knowledge *look* less expensive.

As knowledge becomes less expensive, those selling knowledge have fewer buyers. As they have fewer buyers, they’re more likely to turn to LLMs to fill the gap. As they turn to LLMs with their low price points, buyers notice. As buyers notice, there are fewer of them. The cycle continues with fewer buyers, less specialty, until a crater.

There has been a lot of talk about the death of programming. But programming has always been more than syntax generation. Programmers internally joke about people thinking HTML is a language. It stands for Hypertext Markup Language, sure, but calling it a programming language in a room full of developers is a reliable way to get laughed out of that room. The joke exists because there’s a fundamental confusion between describing something and making something do something. That confusion is everywhere right now.

Subscribe now

Word processing made typing easier. It didn’t turn people into authors. The author was never the person who could operate the typewriter the fastest.

When I lived in France, I lived near Barbizon. In the 1840s, new paint technology, specifically oil paints in portable tubes, enabled artists to suddenly leave the studio and paint outside. The Barbizon school and later the Impressionists were born from that moment. But the technology didn’t turn ordinary people into painters. It gave talented painters a new capability. The ones who weren’t painters before the tubes showed up weren’t painters after, either.

A race to the bottom on cost is a classic commoditization move, and it almost always ends in tragedy unless you can afford enormous losses. Best of luck competing against Google, Apple, or Harvard with its brand and $57 billion endowment. They can lose money on knowledge delivery longer than you can stay solvent.

When I was a child, I ran an errand with my father. Those electronic organs where you press a key and they play a melody were new. My father walked out of the store and saw me, a virtuoso, hammering away. He bought the organ.

That led to two immediate discussions: a) who was going to tell my mother, and b) where to put the organ when it was delivered.

It gathered dust. I never learned to do more than press a key and play Chopsticks.

Modern vibe coding is like that organ. Someone with no programming background asks an LLM to build something, it appears on screen, and they feel like a virtuoso. But the feeling fades, the thing gathers dust, and the distance between pressing a key that plays a melody and actually being a musician turns out to be the same distance it always was.

But I got pretty good at other things though piano wasn’t one of them. And that’s the part of this story nobody is telling. The death of programming, the death of expertise, the idea that we’ll all be virtuosos because the tools got easier. None of that is happening. What’s happening is that the tools got easier and the people who were already good got better. There will be room for everybody else. Doom, gloom, and the breathless predictions aside, there always has been.

Share

Two recent papers addressing LLM’s and reasoning are worth attention.

The first, from researchers at Caltech and Stanford, is bluntly titled “Large Language Model Reasoning Failures.” The second, from Wharton, is called “Thinking: Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender.” There are many more like them but the conclusion is the same: large language models don’t reason. They produce a convincing semblance of reasoning, but it is not the same thing.

However, LLMs create authoritative sounding output. Like a snake oil salesman, the pitch is polished and the product is empty. Even purposefully wrong. People believe it anyway to their detriment.

This matters enormously for strategy.

At VSTRAT we’ve been building AI-augmented strategy tools since 2003, over a decade before generative AI existed. And from the beginning our approach has been deliberate: we use structured, proven frameworks to help humans reason better. We do not build systems that churn out a finished recommended strategy. It’s not that we can’t. It’s that we purposefully don’t.

Subscribe now

Here’s why.

The field of behavioral economics, pioneered by Tversky and Kahneman and extended by Shiller, Thaler, and others, draws a sharp line between homo economicus (the perfectly rational decision maker that exists in textbooks) and homo sapiens (the actual human being). Real people think with their heads and their guts. They have insight and intuition that no model can replicate.

And in strategy, rejecting the pattern is often the whole point.

Was it predictable, following patterns of successful businesses, for FedEx to launch an overnight delivery service? No. How about for Netflix to pivot to streaming when broadband penetration was young? For a distant third-place Nintendo to bet the company on the Wii when analysts said they were wasting money? For a recently bankrupt Marvel to risk everything on building their own movie studio?

Every one of those moves was against the patterns LLM’s love. And every one of them redefined an industry. A purely pattern matching machine would not have made any of them.

This is the danger of what the Wharton researchers call “cognitive surrender”: letting an LLM do your strategic thinking because its output sounds confident and polished. The words are beautiful. The decks shine. And the reasoning isn’t there. If the goal is to rubber-stamp a decision already made, the insta-deck works fine; the robot will parrot what you want it to. But that was never strategy.

We love AI. We’ve been working at the intersection of AI and strategy for decades, long before ChatGPT, long before generative AI itself. But we draw a hard line between augmentation and replacement. The role of AI in strategy is to help people think, not to think for them.

The market is starting to figure this out. The infamous MIT study showing that 95% of AI projects fail? Those are the projects built on hype and self-proclaimed expertise. Meanwhile, the Bureau of Labor Statistics reports a 4.9% productivity gain compared to the long-running 2.2% average. That’s applied AI doing what it does best: augmenting human work, not replacing human judgment.

One last thought. We’ve started noticing something in our Google Analytics: referral traffic from ChatGPT. When we investigated, the pattern was clear. If people ask ChatGPT for a strategy, it points them somewhere else. If they ask for help thinking through strategy, it points them to us.

Even ChatGPT knows it’s not a strategy tool.

Share

I keep coming back to pattern matching. Once you see it, you can’t unsee it.

Law is pattern matching. Playbooks, checklists, best practices. All of them are ways of saying: *here is a known structure, now recognize it and reproduce the expected response.* We’ve been training people to do this for centuries, and we’ve been calling it education. A college exam rarely asks you to create something genuinely new. It asks how closely you can approximate the pattern the instructor has in mind. The “right” answer is the most recognizable one, not the most interesting one.

We have, in other words, been training students to think like machines. Now the machines are here, and they’re better at it.

Subscribe now

Of course GenAI is phenomenal at pattern matching. That’s not dystopian. That’s literally what these systems are built to do. Saying “AI is good at pattern matching” is about as alarming as saying a food processor can slice 100 kilograms of onions faster than a human, and without crying. Machines have always been better than people at certain narrow things: speed, consistency, recall. This is not Armageddon. It’s kitchen equipment.

The confusion starts when people conflate *being good at pattern matching* with *being good at creation*.

Doing very well on a test does not mean you can invent a field. Matching the expected answer does not mean you can design the question. And here’s the uncomfortable part: patterns are, by definition, averages. They are the center of mass of prior behavior. Even when we celebrate someone as “exceptional,” we’re usually celebrating a tight fit to a known template with a bit of polish.

And despite the soothing voice of public radio hosts, everybody cannot be above average. Statistics doesn’t care about your feelings.

Why This Matters for Strategy

You generally do not want an average strategy. The incumbents already have that one, plus scale, plus distribution, plus political capital. An AI that gives you a perfectly competent, generic SWOT is not threatening consultants so much as it is threatening bad consulting.

I saw a LinkedIn post recently where someone crowed that Claude can generate a SWOT “for free,” unlike a human analyst. Shortly after, the same person admitted they hadn’t actually read the details of the output. When they did, the realization dawned: it was fine. Perfectly fine. Smooth. Plausible. And utterly average.

That’s the trap. Pattern-matched outputs *feel* authoritative because they resemble what authority has looked like in the past. Bullet points. Clean phrasing. Familiar structure. But if everyone can generate the same competent average instantly, the value shifts elsewhere.

The Line Between Tool and Crutch

This is where we’ve been very deliberate with VSTRAT. We blend interactivity with GenAI. We use the model as a push, not a replacement for thought. It helps you explore, surface options, stress-test ideas. It does not claim to carry you across the finish line while you nap.

We could have sold more licenses by slapping a big button on the homepage that says “Generate Your Strategy.” Plenty of people do. We don’t, because we don’t believe it. Not philosophically, not practically, and not ethically. Strategy worth having requires judgment, constraint, tradeoffs, and sometimes the courage to reject the most obvious pattern.

The future isn’t pattern matching versus people. It’s pattern matching in service of people who know when average is the enemy.

And that distinction is going to matter a lot more than whether a SWOT costs zero dollars or zero seconds.

Share

The Smartest Bouncer in Long Island

LLMs are becoming measurably more intelligent by the month. Benchmarks keep falling. Context windows keep growing. The demos keep getting more impressive. But intelligence alone isn’t quite the road to nirvana it’s been made out to be.

Consider the case of Chris Langan, a man repeatedly tested with a consistent IQ of almost 200. By most measures, one of the smartest human beings alive. He worked as a bar bouncer on Long Island for years before eventually moving to a farm in Missouri. Chris wasn’t dumb. He had staggering raw intelligence. What he lacked was structure: the frameworks, the mentorship, the institutional scaffolding that channels raw cognitive horsepower into something the world recognizes as achievement.

This is the dirty secret nobody in the AI industry wants to talk about.

Subscribe now

The Consulting Paradox

Multiple studies, including a well-publicized experiment at BCG with Harvard researchers, have shown that LLMs can amplify the output of high-end strategy consultants. The results are real. But we conveniently ignore a critical detail: to become a high-end strategy consultant in the first place, that person likely endured years of rigorous training. Case method. Structured frameworks. Mentorship. Thousands of hours learning how to think about problems before they ever touched a keyboard.

The AI amplified their existing structure. It didn’t create it.

Hand that same LLM to someone without that training and you get something that sounds brilliant but is ultimately mediocre. Confident, articulate mediocrity. Which, if you think about it, is far more dangerous than obvious incompetence.

Computers Were Supposed to Teach

The idea that computers should provide structure, not just answers, is not new. It’s older than most of the people building today’s AI products. In the 1970s, the PLATO system at the University of Illinois had interactive courseware, forums, and messaging decades before Netscape existed. Alan Kay at Xerox PARC invented overlapping windows, mouse-driven interaction, and graphical displays primarily to teach kids. Not doctors. Not consultants. Kids. The premise was that young minds needed structure to unleash their inherent creativity and intelligence.

That vision, computers as structured thinking systems, has largely been forgotten. We got distracted by the internet, then social media, then mobile, then crypto, and now generative AI. Each wave promising transformation. Each wave mostly delivering new ways to consume content rather than new ways to think.

What I Saw in Missouri

I live in Austin, TX, and I go to a lot of pitch events. I’ve seen hundreds of presentations from founders, consultants, and MBAs backed by serious venture money. I was recently privileged to attend the Center for Transformative Technology at the University of Missouri, where students had used a structured strategy system (full disclosure: one I built, called VSTRAT) to prepare presentations.

What I saw stopped me cold.

These students articulated specific strategies and the marketing focus to enable them. They identified precise segments and explained which archetypes would appeal to those segments, and why. They mapped pain points their systems addressed to concrete opportunities. The full chain, from strategy through segmentation through execution, presented coherently and defended under questioning.

I have seldom heard this expressed so well by professional, full-time funded entrepreneurs. Putting it mildly, maybe never. And I see those presentations regularly.

Think about that for a moment. Who are you going to sell your thing to, and why do they care? That’s not an advanced question. It’s the question. And yet most pitches I see treat it as a box to check rather than a problem to solve. You get a slide that says “target market” with a giant TAM number, maybe a vague persona, and then the founder races ahead to the product demo because that’s the part they actually enjoy talking about. The specific chain of reasoning, from segment to archetype to pain point to opportunity, gets skipped. Not because these people are stupid. They’re often brilliant. But nobody forced them to do the hard, structured thinking that connects those dots before they opened PowerPoint.

These weren’t MBA candidates with five years at McKinsey. They were students who had spent hours inside a structured system that guided them through the process step by step. Naturally, a few asked the obvious question: couldn’t you do the same thing with ChatGPT?

Why the Answer Is No (and Why It Matters)

VSTRAT is not a chatbot in the normal sense. It doesn’t take a prompt and hand you a deck. We studiously avoid the “type a sentence and get a strategy” approach, and frankly, we have some disdain for the people selling that as a serious product.

Instead, the system guides people to create their own strategies using structured interactivity. You work through frameworks. You make choices. You confront tradeoffs. The AI embedded in the system is substantial, but its role is to augment your thinking, not replace it. Similarly, we avoid the unintelligent whiteboard approach where collaboration just means moving sticky notes around. People create individually inside the structure, then use collaborative whiteboards to synthesize and finish a strategy together.

This distinction matters enormously, and it’s the one the current AI discourse almost entirely ignores.

An LLM without structure can make a person feel brilliant. That feeling is a trap. It’s the Dunning-Kruger effect running on GPU clusters: the tendency to not know what you don’t know and to believe everything is easy because the output looks polished. The prose is clean. The bullet points are crisp. The confidence is absolute. And the strategy is hollow.

Guided Intelligence

We call this approach “guided intelligence,” and we think it represents the likely future of artificial intelligence. Or at least the version of it that actually helps people rather than flattering them.

The concept is simple but apparently difficult to execute: build systems that make people genuinely smarter, not just faster at producing things that look smart. Systems where the AI’s role is to structure your thinking, surface the tradeoffs you’d otherwise miss, and force you to confront the hard questions before you’ve committed to an answer. Not systems that skip straight to a polished deliverable and let you reverse-engineer a rationale.

We’ve been working on this problem for decades, long before the current hype cycle made everyone an AI expert overnight. And after watching group after group of students produce work that outclasses what I see from funded professionals, the conviction has only deepened.

Chris Langan didn’t need more intelligence. He needed structure. The same is true for most of us. And it’s certainly true for the organizations currently spending millions on AI tools that produce beautiful, confident, empty output.

The smartest bouncer in Long Island is a cautionary tale. Not about the limits of human intelligence, but about what we waste when we treat raw capability as a substitute for disciplined thinking. That waste is about to get industrialized unless we change course.

Subscribe now

Share

During my undergraduate years, we had something called co-ops: work quarters where students ran actual businesses instead of sitting in classrooms. In one of them, the university would hand you a small budget, assign you a venture, and let you sink or swim with real money and real consequences.

My friend Jane drew what seemed like an easy assignment: managing a student lounge that sold grilled cheese sandwiches. Previous operator had done fine with basic grilled cheese at reasonable prices: nothing fancy, decent volume, steady profits, hungry students eating grilled cheese sandwiches at 10M. Jane looked at this setup and saw opportunity for improvement.

Subscribe now

The Death Spiral Begins

She started making gourmet sandwiches. Artisanal bread, premium cheeses, organic tomatoes … the works. She raised prices to reflect the elevated quality. Fewer people bought sandwiches, but the margins were better, so initially it seemed smart.

When sales continued dropping, Jane’s response was to raise prices again. After all, if fewer people were buying, she needed more profit per sandwich to hit her targets. She added more expensive ingredients to justify the premium pricing. The customer base shrank further.

By week six, Jane was caught in a death spiral. To maintain revenue with her tiny remaining customer base, she had to keep prices high. But to maintain margins with those high prices, she couldn’t lower ingredient costs. So she actually cut back on ingredients while keeping prices at their peak levels. Even the stoners eventually gave up.

Jane’s grilled cheese empire was bankrupt.

The Safety Net That Made Failure Valuable

The brilliant thing about the co-op program was its built-in safety net. The school knew this might happen. They’d given each student a budget at the beginning, and at the end, the school kept half for future co-ops while the student kept the other half as their “salary.” So when Jane’s sandwich empire collapsed, the next quarter’s student could start fresh with working capital. Jane, and those of us who watched and winced, learned invaluable lessons about pricing, customer behavior, and the dangers of optimization without market feedback.

This ten-week undergraduate exercise taught more about fundamental business principles than most MBA programs manage in two years. The lesson stuck because Jane felt the panic of declining sales, lived with her bad decisions, and had to face the consequences in real time with real money.

AI as Business Theater

Here’s what makes this story relevant today: many business schools are responding to AI exactly like Jane responded to declining sandwich sales. Business schools teach strategy while executing terrible strategy. They teach operational efficiency while drowning in operational dysfunction. They lecture about innovation while clinging to business models that haven’t changed since 1970.

A well-designed AI tool that actually enhances learning looks boring in a board presentation. It doesn’t have a slick demo. It requires faculty to change how and what they teach. It takes time to integrate into curriculum. It’s infrastructure, not spectacle.

But infrastructure is what moves people. Trains aren’t sexy, but they transform cities. The schools falling for vendor dog and pony shows are buying Maseratis to park in the lot while their students are still waiting at the bus stop.

What AI Actually Can’t Do

AI can simulate business scenarios, but it can’t replicate the visceral experience of real failure with real consequences. It can analyze market data, but it can’t teach you how to navigate organizational politics when your analysis threatens someone’s territory. It can generate strategy recommendations, but it can’t prepare you for how to collaborate and sell the recommendations to management.

Jane’s grilled cheese bankruptcy worked as a learning experience because she felt genuine panic watching her customer base evaporate. She had to sit with the discomfort of knowing her decisions were failing. She experienced the emotional reality of a death spiral. No simulation replicates that.

The schools that will survive aren’t the ones replacing faculty with AI. They’re the ones using AI to create more opportunities for students to experience that kind of high-stakes learning. They’re enhancing judgment, not replacing it. They’re using AI to create richer case studies, more dynamic simulations, practitioner tools, and better feedback loops.

Making the Same Mistakes

Like Jane doubling down on gourmet ingredients while her customers disappeared, many institutions are responding to market pressure by adding more bureaucracy, more complexity, more administrative friction. They’re launching “AI centers of excellence” that function primarily as PR vehicles. They’re creating new administrative positions with titles like “Chief AI Officer” without changing what happens in classrooms.

Some schools had working relationships with AI tools that genuinely enhanced learning. And then we can guess what happened. Pricing became political. Implementation became bureaucratic. Faculty lost interest. The tools sat unused while the same schools signed contracts for flashier alternatives that generated better press releases.

This is the institutional equivalent of Jane cutting back on cheese while keeping prices high. Same likely ending.

Who Will Actually Survive

The institutions that will thrive aren’t the ones with the most impressive AI demos. They’re the ones asking harder questions:

• Where does AI actually enhance human judgment rather than replace it?

• What would our curriculum look like if we designed it from scratch today?

• Are we buying tools that transform learning, or tools that look good in presentations?

They’re the ones willing to admit that a well-designed AI system might teach certain business principles better than a lecture hall, while also being clear about what only human instructors can provide.

They’re the ones treating AI as infrastructure, not decoration. They’re buying trains, not Maseratis.

The Simple Truth

The co-op program worked because it was elegantly simple: here’s some money, here’s a real business challenge, go learn. The failure mattered because the consequences were real. The lesson stuck because Jane lived it.

The best use of AI in education follows the same principle: give students powerful tools, create genuine stakes, let them experience the full arc of success and failure. Use AI to create more opportunities for that kind of learning, not fewer.

Jane’s grilled cheese bankruptcy taught her lessons that stuck for life. She learned about death spirals, margin optimization, and the dangers of losing touch with your customers.

The question isn’t whether business schools will learn those same lessons but whether they’ll learn them before it’s too late. Physician, heal thyself, as my co-founder put it in an article. Apply their own medicine to their own institutions before their customers stop showing up or stop getting jobs.

Subscribe now

People keep predicting that vertical AI systems like VSTRAT will be replaced any day now by frontier models. The irony is that the better the models get, the more necessary systems like VSTRAT become.

I’ve been hearing this claim for years. It still hasn’t happened.

That is not because frontier models aren’t powerful. They are. They offer extraordinary general intelligence and increasingly fluent guidance. But raw intelligence has never been the bottleneck in professional decision making. Structure is.

We have seen this movie before.

Subscribe now

MOOCs will replace universities. Elite lectures are now free or cheap from Harvard, MIT, and Stanford. Yet universities are thriving, charging $80,000 per year, with acceptance rates at historic lows. Content was never the scarce resource.

Blockchain will replace banks. It didn’t. At best, it may become infrastructure inside banks. That is the ceiling.

The death of programming. Predicted repeatedly since FORTRAN and COBOL. Each new abstraction promised fewer programmers. What actually happened was more software, more complexity, and more programmers.

AI will replace lawyers and doctors. I build legal expert systems and earned a law degree believing some version of this myself. After enough time in court, and enough exposure to LLMs making confidently wrong legal judgments, the reality is obvious. They are tremendous support tools, but they are not a substitute for judgment, responsibility, and accountability.

None of this denies real disruption. We are already seeing it in front line customer service, basic content creation, and low level analysis. AI is already replacing low level tasks because it is good enough for work where the cost of a minor error is low.

But strategy is not a low level task. In strategy, being 90% right is often 100% wrong because the remaining 10% is where the competitive ruin or breakthrough lies. Extrapolating the success of basic chatbots to higher order work like strategic planning is a category error. Generating fluent output is not the same thing as producing sound judgment. A model can create a beautiful, confident deck. That does not make it equivalent to years of domain experience, constraint awareness, and responsibility for outcomes.

Here is the part most people miss: as models improve, unstructured use becomes more dangerous, not less. The better the output sounds, the easier it is to mistake confidence for correctness.

When you use a raw prompt for strategy, you aren’t actually strategizing. You are accessing an average of the internet’s opinions.

That is why systems like VSTRAT don’t disappear as models get better. They become more valuable. If the frontier models are the engines, VSTRAT is the cognitive scaffold. It provides the necessary guardrails: intake, sequencing, framing, and the rigorous domain discipline required to find a Blue Ocean.

VSTRAT is not for teams looking to outsource their thinking to a chatbot. It is for leaders who want a better system to do their own thinking.

Tools are only as good as the structure governing their use. That truth hasn’t changed. And it isn’t changing anytime soon.

Subscribe now