In 1966, when Walmart had just 20 stores, Sam Walton enrolled himself in an IBM school for retailers in Poughkeepsie, New York. His own framing of the decision, recorded in his 1992 autobiography Made in America, was simpler than any management theory would later make it: “I was curious. I made up my mind I was going to learn something about IBM computers.” A 1998 TIME retrospective added the structural detail that mattered most. By the time Walton walked into that classroom, he had already decided what the trip was actually for: “His goal: to hire the smartest guy in the class to come down to Bentonville, Ark., and computerize his operations. He realized that he could not grow at the pace he desired without computerizing merchandise controls.”
The classroom anecdote inside the autobiography has a detail business writers like to skip. While at the IBM school, Walton cornered Abe Marks, a retail CPA and the president of the National Mass Retailers’ Institute, going through Marks’s briefcase and pulling out his own hand-written ledgers to ask Marks to audit them. By 1969, three years after the trip, Walmart had installed an IBM System/360 Model 20 to receive daily sales reports. The competitive moat that compounded for the next four decades, the moat that turned a 20-store regional chain into the largest retailer in the world, was built on a decision Walton made personally in a classroom before any of his deputies had been hired to run it.
This is not a story about technology. It is a story about a leadership move that recurs across paradigm shifts, almost always associated with founders who built durable companies, almost always missing in founders who didn’t. The move is this: when a new operating logic arrives in the industry, the founder shows up personally, learns enough to have firsthand judgment, and only then hires the specialists who will run it at scale. The personal understanding is the entry condition for the discriminating-hire move that follows. Skipping the first step makes the second step stochastic.
The lesson for AI is not that leaders should pay attention. The vocabulary of “pay attention” lets the executive off the hook for a much harder move. The actual move is more demanding: leaders who want to win cannot outsource the formation of their own judgment.
The pattern, named
Walton’s three-step sequence is the pattern that recurs across the founders profiled here. Stated cleanly: first, personal understanding; then, the right talent; then, organizational advantage. Each step costs something different. The first costs ego, because it requires the founder to sit in a learner’s chair when most peer CEOs are sitting in approval meetings. The second costs accuracy, because the discriminating-hire judgment depends on the first step being real, not theatrical. The third costs time, because the integration of the specialist’s work into the company’s structure is a multi-year reorganization, not a vendor selection.
What the pattern is not is a celebration of founder-as-polymath. None of the founders below personally became the technical operator. Walton did not run Walmart’s IT department. Schultz did not pull espresso shots in production stores. Nadella did not write Azure code. The pattern is narrower than that and more transferable: the founder did the personal-understanding work that most of their peers skipped, and that work let them make a hire and an integration call their peers could not.
Schultz: the body in the espresso bar
Howard Schultz was on a buying trip to Milan in 1983. He was Starbucks’ director of retail operations and marketing at the time, hired the previous year, in his early thirties. The trip was for housewares, not coffee. What changed Starbucks (and what changed Schultz) was that he walked. His description of the moment, in Pour Your Heart Into It (1997), opens with the line that compresses everything: “I love to walk and Milan is a perfect place for walking.” Over the days that followed, Schultz visited more than 500 espresso bars across Milan, watching the choreography of a culture he had never seen before. He came home and tried to convince Starbucks’ founders, Jerry Baldwin and Gordon Bowker, to add traditional espresso beverages to a coffee company that until that point sold whole bean coffee, leaf teas, and spices.
They refused. Schultz left, started his own company, then bought Starbucks back in 1987 and rebuilt the entire chain around what he had personally watched and felt during those Milan walks. He did not commission a market study. He did not hire a consultant to assess the Italian espresso bar opportunity. He went, he walked, he counted, he sat. The body did the work that no PowerPoint could.
The transferable point for AI is not “your CEO should walk through 500 espresso bars.” The transferable point is that the discriminating judgment about which AI capability matters and which is theater is built up the same way Schultz built his judgment about which espresso bar configuration mattered. By being inside it, repeatedly, with no buffer between the executive and the thing.
Nadella: from defensive to curious
Satya Nadella took over Microsoft as CEO in February 2014. The company had spent a decade defending market positions it had won in the 1990s, missing mobile, missing search, watching internal divisions fight over whose budget would absorb the next loss. Nadella’s pivot is now well-documented in Hit Refresh (2017), and the line from that book that did the most cultural work is one he repeated until the company internalized it: “For too long we’ve been the know-it-all company, and we need to become the learn-it-all company.”
The framing came from a book his wife Anu had given him before he took the role: Carol Dweck’s Mindset: The New Psychology of Success. Dweck’s research distinguishes a fixed mindset (intelligence and ability are static) from a growth mindset (both can be developed). Nadella read it, decided the company needed to operate the second way, and built every senior-leadership cadence around the shift. The Azure pivot, the OpenAI partnership, the cultural rewrite that allowed Microsoft to ship multimodal AI products faster than any peer of comparable size, all of them rest on the personal-understanding move he made before any of them.
He did not delegate the cloud and AI strategic call to a specialist who would have arrived at it through their own career incentives. He immersed himself in the technology personally, pushed his senior team to do the same, and then made the strategic bet from a position of having actually understood it. This is the same Walton move (enroll in the school, then hire the smartest person in the class), updated for a 200,000-person company.
Chesky: the designer who wouldn’t delegate the experience
Brian Chesky graduated from the Rhode Island School of Design in 2004 with a Bachelor of Fine Arts in industrial design. By the time he co-founded Airbnb four years later, the relevant credential was not the BFA itself; it was the operating discipline RISD had taught him. Designers do not delegate the experience of using a product; they prototype it, sit with it, throw it away, do it again. Chesky imported that discipline into a tech company and refused to give it up as Airbnb scaled.
The clearest example is the Snow White project. Chesky read Walt Disney’s biography in the early 2010s and discovered the storyboarding technique Disney’s team had developed to make Snow White and the Seven Dwarfs in 1937. He decided Airbnb’s product should be designed the same way: as a sequence of frames, scene by scene, drawn out before being built. Airbnb hired a Pixar animator, Nick Sung, to produce three full storyboards: the host process, the guest process, and the hiring process. The storyboards then drove engineering, marketing, and customer service decisions across the company. When Chesky needed to make a product call, the question on the table was “what does the storyboard say should happen here?” not “what does the latest A/B test suggest?”
The pattern transfers to AI cleanly. Most companies are running A/B tests on AI features bolted onto products designed for the pre-AI customer experience. Chesky’s discipline says the experience gets storyboarded first, then AI gets dropped into the storyboard where it actually matters, then the product gets built. The personal-understanding move (the founder doing the storyboarding work himself, not delegating it to a product manager) is the entry condition. Skip it, and the AI features sit on top of an experience that no one in the company has actually walked end to end.
Schreiber: the chassis from day one
Daniel Schreiber co-founded Lemonade in 2015. Unlike the other four founders here, Schreiber is the one who built a company knowing the new operating logic from inception, not retrofitting an existing organization to it. At the Insurtech Insights Europe conference on March 19, 2025, Schreiber stated his thesis directly: “Companies in the insurance industry that are built on top of an AI substrate will enjoy a structural and competitive advantage.”
The phrase “AI substrate” is doing structural work. Most insurance companies have AI projects layered on top of decades-old policy administration, claims, and underwriting systems. Lemonade has it the other way: AI is the floor, not the ceiling. Customer onboarding, claims handling, and pricing all run on AI from the first interaction, with humans as the exception layer rather than the default. The outcomes Schreiber attributed in the same talk were qualitative and direct: “Our customer satisfaction scores go up every time we move something from humans to AI, while collapsing costs and increasing response times.”
The strategic argument Schreiber offered at the same conference is the one this article is about: “It’s very, very difficult for large corporations, willing as they are, to reinvent the very chassis upon which they are built while the car is moving.” The diagnosis is structural, not motivational. Established competitors have not failed Schreiber’s challenge because they lacked AI talent or AI budget. They have struggled because the chassis itself is the obstacle, and the chassis can only be redesigned by a founder or CEO who personally understands what AI does to the underlying business model. That is precisely the move every other founder in this article made before the AI era arrived.
The Lemonade case is also the closest analog to the position most CEOs reading this are in. Most are not building from zero. They are running 80 to 200 people in a company whose chassis was designed before AI was a serious operating constraint. Schreiber’s lesson for them is not “rebuild from scratch.” It is more practical: read the chassis honestly, decide which parts of it are worth retrofitting and which parts are worth running parallel, and own the call personally.
What it costs to skip step one
The five founders above are survivor-bias evidence, and that needs to be named. We remember Walton, Schultz, Nadella, Chesky, and Schreiber because their bets paid off. The harder cohort to study is the one we don’t remember as well: founders who personally immersed in technology shifts they did not have the disposition to evaluate, made wrong calls, and cost their shareholders billions.
Steve Ballmer’s personal stewardship of Microsoft Mobile produced Windows Phone, Nokia, and the eventual write-down. Travis Kalanick personally drove Uber’s autonomous-vehicle program for years, with expensive failures the post-IPO restructuring eventually unwound. Jeff Immelt’s personal sponsorship of GE’s industrial internet strategy was a multi-billion-dollar bet that did not survive his successor. Each of these founders refused to delegate the understanding. Each was wrong about what they had understood.
The honest read is narrower than “personal understanding always wins.” Personal understanding is the precondition for the discriminating-hire move that wins. Without it, the founder cannot tell a great specialist from a mediocre one, and the hire becomes a roll of the dice. With it, the founder has at least the means to evaluate, even if the evaluation itself is hard. The pattern is necessary, not sufficient. Skipping it makes the next step stochastic. Doing it does not guarantee the next step is right.
The corollary, which most CEO advice elides, is that the personal-understanding work has to be done at the right altitude. Walton did not need to learn how to write COBOL. He needed to learn enough about what computers did, how they were programmed, and what kinds of people built them, to hire the right operator and integrate the work into Walmart’s structure. Schultz did not need to learn how to roast Italian beans. He needed to learn what an espresso bar actually was as a customer experience, so he could rebuild Starbucks around something other than the existing whole-bean retail model. The altitude varies by domain. The discipline does not.
The five moves a CEO can make this quarter
The diagnostic the post implied, in the form a CEO running 80 to 200 people can run on Monday morning. None of these requires a consultant. All five can be done in a single quarter without restructuring a single department.
-
Sit in the learner’s seat. Pick the AI capability your company is most likely to deploy in the next twelve months (a customer-support agent, a research agent, a marketing agent, or the multi-agent orchestration layer that ties several of them together for your operation). Spend two hours a week using it personally for your own work, not your team’s work. Two hours times twelve weeks is twenty-four hours, which is roughly the hour count of an executive education program. The difference is that you are evaluating the tool against your actual judgment, not a curriculum.
-
Read the primary sources, not the summaries. When the next McKinsey, Stanford, or Anthropic report drops, read the original PDF, not the LinkedIn summary of the LinkedIn summary of the McKinsey summary. The summaries are usually accurate at the headline level and almost always wrong on the specifics that matter for your decisions. Walton went to the IBM school. He did not read a summary of the IBM school.
-
Hire your specialist after personal evaluation, not before. The Walton move is not “hire the AI executive first.” It is “learn enough to know what kind of AI executive you actually need, then hire that one.” Most CEOs do this in reverse, which is why most CEOs hire the AI executive whose résumé reads best, not the one whose judgment matches the strategic bet the company actually needs to make. The same discipline applies to the consulting firm, the integration partner, and the build-versus-buy call.
-
Tie AI to OKRs you personally own. Skills Are SOPs showed that strategic-integration sponsorship (AI tied to a measurable target with a tied incentive) was the activity that separated the 29% of Stanford’s cases that hit organization-wide transformation from the 71% that did not. The same logic applies to the founder’s calendar. If the AI bet is real, the founder’s calendar should show it: weekly check-ins, named obstacles, removed obstacles. Quarterly steering committees do not count.
-
Audit your delegation chain for understanding gaps. The first AI delegation question is usually framed as “what should I delegate to AI.” The harder question, the one this article is actually about, is “where in my current organization is my judgment running on someone else’s understanding?” Start with the AI program. Walk it from the executive sponsor down through the team running it, and ask at each layer whether the next person up the chain understands enough to make the call they are making. Where the answer is no, the chain is broken, and the AI bet is sitting on a fault line.
The lesson, sharpened
Six decades, four industries, one pattern. The lesson is not that founders should learn the new technology. It is that founders who win on paradigm shifts do not delegate the formation of their own judgment about that shift. The judgment is what enables every downstream move. Without it, the discriminating hire becomes a coin flip, the strategic bet becomes a bet on the consultants who briefed it, and the integration becomes a bet that the org chart will somehow learn what the founder hasn’t.
AI is the current instance of an old pattern, sharper than usual because the technology compresses the timeline. The companies that get the next two years right will be the ones whose founders treat AI the way Walton treated computers, the way Schultz treated espresso bars, the way Nadella treated cloud, the way Chesky treated design, the way Schreiber treats the chassis. Personally. First. Before the team.
The seat in the learner’s classroom is empty in most companies right now. The founders who fill it are the ones who will be writing the autobiographies in 2056.