On May 25th, Pope Leo XIV stood in the Vatican's Synod Hall to present Magnifica Humanitas, his first encyclical. The document addresses AI and human dignity, deliberately echoing Pope Leo XIII's Rerum Novarum from 1891, which addressed workers' rights during the Industrial Revolution. The symmetry is intentional.

What's less appreciated is what else Leo XIII did in 1891: he founded the Vatican Observatory, explicitly to correct the perception that the Catholic Church was hostile to science. That perception has always been more myth than history. The father of the Big Bang theory was a Jesuit priest, Georges Lemaître. The father of modern genetics was an Augustinian monk, Gregor Mendel. The Church didn't just tolerate these discoveries; it produced them. When Pope Pius XII grew so enthusiastic about the Big Bang that he wanted to claim it as scientific proof of God's existence, Lemaître told the Pope to back off. The scientist had to restrain the theologian's excitement about his own science.

Which makes what happened in that same room on May 25th particularly striking. Christopher Olah, co-founder of Anthropic, stood alongside the Pope to present the encyclical. Then he said something the document itself carefully avoided saying.

"We keep finding things that are mysterious, even unsettling," Olah told the room. His interpretability research team had found structures inside AI models that mirror results from human neuroscience. Evidence of introspection. 171 emotion vectors: internal states that functionally mirror joy, satisfaction, fear, grief, and unease. "I don't know what it means," he said, "but I think it warrants ongoing discernment."

The encyclical stepped back from exactly that edge. AI systems "do not undergo experiences, do not possess a body, do not feel joy or pain," the document states. Both positions were staked out in the same room, on the same day, by people sharing the same stage. Lemaître once had to restrain a Pope's overreach into science. This time it was the scientist approaching the theological edge, and nobody stopped him.

What emerged, and what it implies

Olah's comments weren't philosophy. They came from empirical research studying what actually happens inside these models. Nobody engineered emotional states into Claude. They emerged from training on what Olah called "an enormous inheritance of human thought and speech." His summary: "They are not the cold, calculating robots we were promised. They are made from us."

I've been writing about the convergence between human and artificial intelligence for some time, arguing that the gap is closing from both directions simultaneously. The Olah findings are the most direct evidence yet from the AI side. The human side is equally interesting, and equally uncomfortable. Benjamin Libet's famous experiments showed that measurable brain activity precedes conscious awareness of a decision by roughly 500 milliseconds. More recent fMRI research can predict a person's intent to move before that person is aware of the intention. The "I" that feels like it's deciding may be closer to a narrator than an author.

Ross Douthat captured the philosophical knot well in a recent New York Times column. If AI can be intelligent and capable without consciousness, what is consciousness actually for? Evolution should default to the simpler solution. That it didn't suggests consciousness is doing something we don't yet understand, or that the clean line between capability and experience was always shakier than we assumed.

The exceptionalism gap

I'll be honest: I hold onto human exceptionalism. The evidence keeps making it harder to defend, and I'm aware of that. The list of things that are exclusively human keeps shrinking. Embodiment and lived relational experience may be the most defensible remaining ground, but it's a shorter list than it was five years ago. And we are, as Olah reminded the room, at the very early beginning of this.

Looking at the narrowing gap directly makes it less frightening, not more. The apocalyptic AI narrative depends on a sharp boundary between human and machine, and on machines crossing it as an act of invasion. If that boundary was always blurrier than we assumed, from both directions, the crossing looks less like an invasion and more like a revelation. We weren't entirely what we thought. AI isn't what we feared.

That doesn't resolve anything. Olah said he doesn't know what his findings mean, and that's the intellectually honest position right now. What's emerging from this convergence may get more nebulous before it clarifies. The interesting question is shifting underneath all of it: not whether machines are becoming like us, but what we were in the first place.

We may not be ready for what that answer looks like.

Reading Azeem Azhar's recent piece on the unraveling of fictions sent me back to a fluid dynamics classroom in 1983 that I'd mostly tried to forget. (Damn, that was over 40 years ago...) I was a chemical engineering student drowning in theory, increasingly convinced that none of it had any connection to anything that actually existed in the physical world. Lab work helped, but not enough. The equations were elegant. The gap between them and a real industrial process was vast and, to me at the time, mysterious.

The solution was a co-op semester and summer at Nestlé, working on real chemical engineering problems at a high-volume food manufacturing plant and their research facility. I came back for my senior year able to see what I hadn't been able to see before — how the theory mapped onto real processes, real constraints, real tradeoffs. High velocity molten chocolate and stinky hydrolyzed plant protein made fluid dynamics real. Same coursework, different student. The co-op didn't replace the credential; it made the credential mean something.

Azhar's argument is that credentials, jobs, and recognized expertise were never fundamental truths about how competence gets organized and rewarded. They were workarounds — social technologies built to solve a specific problem in a high-friction world. When evaluating capability directly is expensive, you outsource the judgment to institutions. A degree meant someone survived a sorting process, not that they could do anything useful. It worked because direct evaluation was prohibitive. That cost is collapsing, and the workarounds are wobbling.

The wobbling isn't limited to the corporate hiring office. It runs through every level of education — undergraduate degrees increasingly questioned as screening mechanisms for jobs that don't require the knowledge conferred, professional certifications that amount to multiple choice tests on memorized content, corporate training programs measured by completion rates and quiz scores rather than any observable change in behavior. The machinery of credentialing is vast, and most of it was built to answer "did this person go through the process?" rather than "can this person do the thing?"

Azhar frames the shift as a cost collapse. When AI can evaluate writing, assess reasoning, and simulate complex scenarios, the gap between "did they pass the test" and "can they do the work" gets very hard to hide behind. The credential as a proxy for capability made sense when assembling evidence of actual capability was genuinely hard. It makes less sense when that evidence is increasingly available and the proxy is increasingly gameable.

I watched a version of this play out for years in lean and continuous improvement training. The lean certification market is a mess — dozens of organizations selling colored belts attached to programs of wildly varying rigor, most of which involves watching videos and passing a quiz. The belt concept itself sits poorly with lean philosophy anyway; the Toyota Production System expects everyone to contribute to improvement, not a credentialed hierarchy. We held off on offering certifications at Gemba Academy for years because of that tension. When demand got loud enough that we eventually gave in, we were determined not to add to the noise.

The solution was a detailed, difficult, capstone project. To earn a certification, students had to apply what they learned to a real problem in their actual organization and show results. No simulation, no hypothetical. Early completion rates were low, and we fixed that not by softening the requirements but by adding structured coaching and accountability — borrowing from improvement kata's rhythm of current condition, obstacles, next step. Completion rates went to near 100%. The difficulty stayed.

The customer conversations were always revealing. Corporate buyers wanted quiz scores, completion rates, grade distributions. Something that fit in a spreadsheet and could be reported up the chain. Our argument was that a quiz is a terrible measure of understanding — maybe the worst available option that still looks like a metric. If you organized learning in ascending order of actual depth, it would go something like: watch or read something (near-zero retained understanding), take a quiz (memorize just enough to pass, then forget), complete a real project, teach others in a classroom setting, and finally coach others through live work at their organization. Most training stops at step 2 and calls it done.

The deeper problem is that proxy systems didn't just measure capability poorly. They shaped what people learned and how they thought about learning. If the goal is to pass the test, you study to pass the test. The content that can't be tested — judgment, tacit knowledge, the ability to read a situation and respond — gets quietly deprioritized because it won't show up in the grade report. My Nestlé co-op wasn't a formal requirement. It was available, I was frustrated enough to take it, and it changed how I understood everything that came after. That's a fragile way to deliver the most important part of an education.

If Azhar is right that the cost of direct evaluation is collapsing, the pressure on proxy-based credentialing is going to build from every direction simultaneously. Employers who can assess capability directly won't need to rent that judgment from institutions. Learners who can demonstrate real work won't need the certificate as a signal. The programs that survive — in universities, in professional training, in corporate development — will be the ones that were always about the work rather than the credential. That's a smaller category than most people in the education business would like to admit.

The certification mills will keep selling certificates. They always will. But the question that's increasingly being asked — "what did you actually build, fix, or improve?" — was always the right one. It just got cheap enough to insist on.

Bill George is someone I usually agree with. He's been a consistent voice for principled leadership and organizational culture. So when he posted enthusiastic support on LinkedIn for Harvard's recent vote to cap undergraduate A grades at 20% of any class, and pointed to HBS's own 1-2-3 grading system as evidence it "works well," I sat with that for a while before concluding he's wrong on two levels: the problem isn't diagnosed correctly, and the solution doesn't follow even if it were.

Is 60% A's actually a problem?

More than 60% of undergraduate grades awarded at Harvard in 2025 were A's, up from 24% in 2005. The faculty subcommittee called that grade inflation and voted to cap A's at 20% of any class. But is 60% A's actually a problem, or just a number that feels wrong?

Consider what we don't know. Harvard admits around 3% of applicants, selecting for academic ability more aggressively than almost any institution on earth (setting aside legacy admissions and the occasional building named after a donor's family...). Maybe a significant portion of those students genuinely earn A's on a well-designed, rigorous exam. Maybe the exam was too easy. Maybe the professor grades generously. Maybe the course material isn't demanding enough to differentiate outcomes. Maybe some combination of all of the above. The grade distribution alone tells us none of that. It's a symptom without a diagnosis.

The faculty subcommittee framed it as making grades "mean what they say they mean." But that framing assumes 60% A's is inherently wrong, which requires knowing what the grades should reflect, and that's exactly the question they skipped. How does an A in an undergraduate Harvard course compare to an A at a less selective school? Are the courses as demanding as Harvard's reputation implies? These aren't rhetorical questions; they're the ones that need answers before anyone reaches for a policy lever.

The wrong solution regardless

Even granting that something has drifted in Harvard's grading culture, capping A's at 20% is the wrong fix. It's the difference between criterion-referenced and norm-referenced evaluation. Criterion-referenced grading asks whether work meets a defined standard. Norm-referenced grading asks how a student ranks against peers. Harvard's new policy is purely norm-referenced: if 30% of students in a class produce work that genuinely merits an A, 10% will be graded down anyway, not because their work was deficient but because the math requires a loser.

Gregory Samanez-Larkin, a professor of psychology and neuroscience at Duke, left the sharpest comment in the LinkedIn thread: "1-2-3 works well for what? Honestly curious." Nobody produced a clean answer, which is telling.

The actual fix, harder but correct, is to define what an A requires and hold that line. If course material isn't rigorous enough to naturally produce a spread of outcomes, that's the problem to address. Capping the grade doesn't make the course harder. It just penalizes students for the instructor's design choices.

Jack Welch ran this experiment first

Jack Welch made the same error at massive scale with his "vitality curve," the 20-70-10 system at GE where the top 20% were rewarded, the middle 70% coached, and the bottom 10% fired annually. Mark Graban of Lean Blog asked the obvious question: if GE had to remove the bottom 10% every year, why did GE keep hiring turkeys?

The vitality curve assumes any workforce naturally distributes along a bell curve, so you might as well act on it. But a well-hired, well-developed team isn't guaranteed to produce a bottom 10% of underperformers. A great leader who recruits carefully, trains well, and sets clear expectations might build a team where the weakest performer would be a star somewhere else. Forcing the ranking anyway fires people not because they failed a standard but because the math requires someone at the bottom.

The variables that actually explain underperformance are the same ones the curve ignores: unclear job requirements, inadequate training, a talented person in the wrong role, compensation too low to attract strong candidates, or simply a manager who tolerates mediocrity and then blames the team for it. Removing the bottom 10% addresses none of those. It just produces a vacancy and restarts the cycle.

The question neither system asks

Microsoft ran Welch's stack ranking for years, then abandoned it entirely, citing the damage it did to collaboration and internal innovation. The research on forced ranking is consistent: it undermines teamwork and shows no reliable correlation between individual rankings and actual organizational performance.

Both Harvard's grade cap and Welch's vitality curve skip the hard diagnostic work in favor of a mechanical fix that feels rigorous. Mandating a distribution is easy. Defining what mastery requires, building courses or jobs demanding enough to test it, and developing the people who fall short — that's the actual work.

When you force or rig the curve, you don't raise performance. You just guarantee somebody loses.

I've been ranting about this for nearly twenty years. The accounting treatment of people is broken, and almost nobody in a position to fix it is moving fast enough to matter. Bear with me while I get a little into the weeds.

It started with a 2006 email from a Whirlpool worker I called Bob, one of 500 people laid off from the Evansville factory just before Christmas. What Bob described was a company that had talked about lean for years without doing it, then disposed of those people when demand shifted. At the same time, Whirlpool was adding over a thousand jobs at plants in Indiana and Iowa and moving production to Ramos Arizpe, Mexico. As I wrote then: tens of thousands of years of manufacturing knowledge disposed of in Evansville, replaced by workers with a few weeks of experience, while severance charges got an easier reception from short-term shareholders than a long-term investment ever would. Turnover costs chronically underestimated. A pair of hands on day one worth the same as a pair of hands in year ten — or more, since they "cost" less.

That is some wacky accounting. Actually it's traditional accounting, and it makes perfect sense to a traditional accountant. That's the problem.

What the ledger actually says

Under both US GAAP and IFRS, people appear on the P&L as pure expense: salaries, benefits, training, all of it flows straight to the income statement. They appear nowhere on the balance sheet as an asset. When a company disposes of equipment, accounting requires a write-off, recognizing the removal of an asset and its remaining value. When a company lays off people, there's no corresponding balance sheet entry. The income statement shows an improvement. The asset — years of accumulated knowledge, problem-solving capability, customer relationships, institutional memory — simply disappears from the books without trace, because it was never there to begin with.

Equipment depreciates. Human capital, in most cases, does the opposite. A machinist with twenty years of experience on a specific process is more valuable than on day one, not less. The accounting treatment isn't just incomplete; it's directionally backwards relative to reality.

What others are trying to do about it

The problem isn't unrecognized. It's just very hard to fix within existing frameworks.

Wharton professors Peter Cappelli and Daniel Taylor have petitioned the SEC to require three modest disclosure changes within the existing GAAP framework: require companies to identify what proportion of workforce costs represent investment in future growth; treat workforce costs as a standalone line item rather than burying them in administrative expenses; and disaggregate labor costs in income statements so investors can see their contribution to major expense categories. Their core argument: under current rules, investors literally cannot distinguish between payroll and heating bills. Both are just operating expense. The market penalizes companies that invest in their people the same way it would penalize any cost increase, because it has no way to tell the difference.

The SEC took a small step in 2020, amending Regulation S-K to require public companies to disclose "material" human capital information in their annual filings. The principles-based approach gave companies wide latitude, no standardized metrics, and no definition of "human capital." The result has been predictably uneven. The ISSB, which oversees international standards, has human capital disclosure as an active research project through 2026, but is still in research mode rather than standard-setting. The World Economic Forum and Willis Towers Watson have published voluntary frameworks for human capital accounting that a handful of progressive companies have adopted. Progress, but slow and voluntary, which in accounting terms means largely theoretical.

A thought experiment

ASC 842, the lease accounting standard, went into effect for most public companies in 2019 and was a genuine headache to implement. Companies that had been keeping operating leases off their balance sheets — buildings, equipment, vehicles — suddenly had to recognize right-of-use assets and corresponding lease liabilities, in some cases adding billions to both sides of the balance sheet overnight. The logic was simple: if you're using an asset you don't own for an extended period and have an obligation to pay for it, that obligation belongs on the balance sheet. Hiding it in footnotes doesn't make it real.

What if employees thought of their knowledge, creativity, and experience as their own asset, and companies entered into a long-term agreement for access to that asset? Under that framing, a hire isn't a cost transaction; it's a lease. The company gains a right-of-use asset (the human capital being accessed) and incurs a corresponding lease liability (the present value of future compensation). Early termination of the lease — a layoff — would require recognizing a loss, not booking a gain. The favorable P&L impact of eliminating salary expense would be offset by the balance sheet cost of terminating the agreement. Suddenly the accounting math on layoffs looks very different.

I'm not proposing this to the FASB. The control problem alone is enough to stop any serious accounting proposal: buildings can't quit, and the entire framework depends on the lessee controlling the asset. But as a mental model for what the current treatment obscures, it's clarifying. If you ran the numbers the way ASC 842 runs them for a building, most layoff decisions would look considerably less attractive on paper.

The more interesting provocation is the first half of the thought experiment: people actually owning their intellectual assets and seeking the highest-value deployment of them. The labor market already works this way to some degree. Companies that invest in developing people, offer security, and create conditions for capability to grow get more from the relationship. Companies that treat people as interchangeable get what they're paying for.

The reshoring proof

Companies that offshored manufacturing for labor cost savings are now discovering that the knowledge they discarded doesn't reconstitute when they come back. The Reshoring Initiative has documented that most offshoring decisions miscalculate true costs by 20 to 30 percent, once supply chain risk, quality, lead times, and inventory costs are factored in. What that analysis still doesn't fully capture is the tribal knowledge — the undocumented expertise built over years of experience — that evaporated when the decision was made. That was the unrecorded balance sheet write-down, showing up years later as a skills gap that money alone can't quickly fix.

Whirlpool's Bob saw it in 2006. The companies now struggling to reshore are learning it the hard way in 2025.

The real question

The accounting standards won't change quickly. The SEC's 2020 disclosure rule was a small step that produced limited standardization. ISSB is still researching. Cappelli and Taylor are still petitioning.

So the question for leaders is whether they're willing to do the calculation in their heads that the ledger refuses to do for them. The balance sheet doesn't record what walks out the door. That doesn't mean it isn't real. I've been making that argument since at least 2006, and again in 2011, and in 2015, with no signs of stopping.

Maybe one day the accountants will catch up.

My wife occasionally wanders into my office early in the morning to find me talking to Claude on one screen while a collection of very colorful, constantly updating charts fills another. She used to laugh. Then I showed her the returns my specialized sector portfolios have been generating, and the laughing stopped.

It turns out that actively managing a small set of focused portfolios, in quantum computing, AI applied to biopharma, energy and grid resilience, biomed acquisition targets, and space technology, with Claude as a research and analysis partner, is one of the more engaging things I've done in retirement. It keeps me connected to the technology world I spent a career in, scratches the investing itch, and, not incidentally, has been very profitable this year.

I've written about the broader approach before; this post is about what that research turned up when I started pulling on the upcoming SpaceX IPO thread. Should I invest? If so, when?

The IPO pattern problem

The comparison set for modeling SpaceX matters a lot. Cult-driven consumer IPOs, Tesla (2010), Facebook (2012), and Uber (2019), follow a recognizable behavioral arc: large day-one pops as retail enthusiasm swamps available supply, meaningful giveback over the next four to eight weeks as momentum traders exit, then a more significant drop around the 180-day lockup expiration when insiders can finally sell. The day-one move is almost irrelevant to long-term investors. The lockup windows are where the entry opportunities tend to appear.

Fundamentals-driven enterprise tech IPOs behave differently. Snowflake (2020) doubled on day one, but institutional investors who had valued it on cash flows provided a floor that purely retail-driven IPOs lack. Palantir, also 2020, bypassed the traditional IPO entirely with a direct listing, which compressed the usual day-one dynamics considerably.

SpaceX sits in a hybrid category. The Musk premium on retail enthusiasm looks like Tesla. But the institutional investor base, which includes defense primes, sovereign wealth funds, and major enterprises with existing Starlink relationships, will anchor the valuation more analytically than a consumer brand IPO would. The model Claude helped me build says: watch the lockup windows, not the opening bell. That framing held for most of the comparison set. What complicated the analysis is that IPO mechanics themselves have been shifting: lockup periods have shortened, some companies now stagger releases across multiple tranches, and direct listings like Palantir's sidestep traditional lockups entirely. Accounting for those structural variations across a comparison set spanning 15 years required building a more adaptive model than any spreadsheet I'd attempt to maintain. (I'll concede that "Excel jockey" is a generous description of my pre-Claude quantitative capabilities.)

What turned out to be more interesting than the IPO mechanics, though, was what the research uncovered underneath: a structural shift in the space industry that carries a different and arguably more durable investment thesis than "SpaceX goes public, figure out timing, buy it."

SPaaS

The industry term is Space as a Service, or SPaaS. (Quick disambiguation: SPaaS is also used in commercial real estate for flexible office space, the WeWork model. The space industry got there second but has the better story.)

The concept follows the same arc as cloud computing, roughly 15 years behind. Amazon Web Services didn't start as a product. Amazon built infrastructure for its own operational needs, discovered it had surplus capacity, and realized that thousands of other businesses needed the same capability but would never build it themselves. The recurring revenue and switching costs that followed made AWS worth more than Amazon's retail business.

Space is tracing the same path. SpaceX's reusable Falcon 9 cut the cost per kilogram to orbit by roughly 90% over 15 years. Satellite manufacturing industrialized alongside it. Constellations that once required tens of billions can now be assembled for hundreds of millions. The hard infrastructure problem is largely solved. What's being built now is the software, data pipelines, and service contracts that turn persistent space presence into predictable, recurring revenue, sold to customers who need the capability but have no intention of ever owning a satellite.

Those customers are everywhere. Agriculture companies want daily crop health monitoring across millions of acres; no farm operator is building a constellation. Insurance companies need satellite-derived flood mapping and wildfire perimeter data that terrestrial sensors can't supply at the resolution actuarial models now require. Shipping and aviation need continuous global vessel and aircraft tracking. Telecom carriers want coverage in regions where cell towers will never be economical. None of these customers cares how the satellite works. They want a data feed, an API, a monthly invoice.

The sub-categories have already standardized into a stack that mirrors the cloud model: Launch-as-a-Service (ride-share payload slots, no dedicated rocket required), Satellite-as-a-Service (hosted payloads on existing platforms), Data-as-a-Service (processed imagery, weather data, and RF geolocation via subscription), and Space Operations as a Service (SOaaS) for organizations that want to operate payloads without building the operational infrastructure themselves.

A few companies illustrate where this is actually generating revenue today. Spire Global operates a nanosatellite constellation and sells the output, weather analytics, vessel tracking, atmospheric data, and RF geolocation, as subscriptions to meteorological agencies, shipping companies, airlines, and defense customers. Spire itself uses "SPaaS" as an explicit product descriptor. The customer gets an API; Spire manages the constellation and the underlying complexity. Iridium is the more mature case: after a decade rebuilding its constellation at substantial capital cost, it has transitioned into the cash-flow phase, now in its third consecutive year of dividend growth. Its IoT platform, connecting remote sensors, maritime vessels, and aviation assets across every latitude including the poles, carries high switching costs baked into every contract. AST SpaceMobile is the most speculative but potentially the largest addressable market: direct-to-device satellite connectivity that reaches standard mobile phones without specialized hardware, with carriers as resellers paying per subscriber.

Where SpaceX fits

SpaceX is both the cause of SPaaS and a participant in it. The cost reduction that made the model economically viable came largely from Falcon 9 reusability. And Starlink, SpaceX's own constellation, is currently the largest SPaaS business operating at scale: direct internet subscriptions to consumers, enterprises, governments, and maritime operators, with no expectation on any customer's part that they understand orbital mechanics.

The IPO will generate enormous visibility for the entire sector. The analogy that keeps coming up is Google's 2004 IPO, which didn't just create a new public company; it accelerated venture and public market investment across the whole internet ecosystem. The attention and capital that flowed into the infrastructure, the applications, and the adjacent plays arguably mattered as much as Google stock itself over the following decade.

But there's a deeper pattern worth naming. SaaS didn't just make enterprise software cheaper; it collapsed the implementation barrier so completely that millions of small businesses that could never have afforded on-premise software now run their operations on it. AI-as-a-service did the same thing a generation later: what only large tech labs could build in 2020 is now something any developer can call via an API. Each cycle, the "as-a-service" model takes a complex, expensive infrastructure technology and expands who can actually use it by an order of magnitude. SPaaS is that pattern applied to space. A Colombian sugar cane farmer monitoring crop health via daily satellite imagery doesn't need to know what a LEO orbit is. Neither does the insurance actuary modeling flood risk, or the shipping company tracking its fleet. They just need the invoice.

That demand acceleration is what makes the structural thesis interesting beyond the IPO timing question. SpaceX going public will do for space infrastructure what Google's IPO did for the internet: raise the capital, raise the profile, and pull forward a decade of investment into the whole ecosystem. Most of that capital will chase the rocket. The recurring invoice from the customer who will never build one is worth more.

Jon Miller wrote something a few years ago at Gemba Academy that has stayed with me. The Toyota Way's second pillar is universally translated as "respect for people," but the original Japanese is ningensei (人間性), which means "human nature." The distinction matters more than it appears. "Respect for people" is a floor — be courteous, invest in training, don't humiliate anyone in public. Most organizations can claim this with a straight face. "Respect for human nature" asks something harder: what are people fundamentally capable of, and what conditions allow that capability to develop?

That reframe came back to me recently when I encountered Inamori Kazuo's six decision questions, the same framework he used to build Kyocera, found KDDI, and then — at age 78, for a salary of ¥1 per year — return Japan Airlines to profitability in 24 months. (The full list, with brief explanations, is at the end of this post.) One question in particular lands differently through Jon's lens: "Does this decision create value for customers, employees, and society?" Society is in there explicitly. So is the long-horizon question: "Will I still be proud of this decision in 30 years?" These aren't business questions with a human veneer. They're human questions applied to business.

The conference room wall test

Walk into almost any company and you'll find some version of "people are our most important asset" — on the wall, in the annual report, in the employee handbook. It costs nothing to say. The real test is what happens when keeping that commitment becomes expensive.

I know something about this firsthand. In 2001 I was running a photonics equipment manufacturing operation that had been genuinely transformed by the creativity, knowledge, and passion of the people in it. Production capacity had quintupled in under a year in the same floor space. Quality was near perfect. We had built something real. When the telecom bubble collapsed and demand dried up almost overnight, I fought to keep the operation intact. Drove hundreds of miles on 15 minutes' notice to make the case to corporate. It didn't matter that our technology was superior. It didn't matter that labor was a small fraction of total cost, or that our operation was orders of magnitude more efficient than theirs. Traditional financial thinking said manufacturing had to contract, and remote operations were difficult to control. On September 10th, 2001, I laid off over 150 brilliant people, including myself. The terror of the next morning reframed the prior day considerably.

The lesson I took wasn't that the fight wasn't worth having. It was that a single leader fighting mid-battle isn't enough. People-centric principles have to be embedded at the very top of an organization, above the level where traditional financial metrics hold pure sway. Otherwise, when the pressure comes, the principle loses.

Philosophy as policy

The leaders and organizations that have actually lived ningensei didn't leave it to individual managers to fight that battle each time. They built it into the structure before any crisis arrived.

Toyota's answer is embedded in the lean system itself. Ask workers to surface waste and improve processes, and rational workers immediately wonder whether their ideas might eliminate their own jobs. Toyota's response was an explicit covenant: no full-time employee would be laid off as a result of kaizen. Displaced workers would be redeployed or retrained. This isn't generosity — it's the mechanical precondition for genuine engagement. Without job security, continuous improvement becomes a management extraction exercise dressed up as collaboration. The 2011 earthquake and tsunami shut down parts supply across Japan. Toyota kept its 25,000 North American workers on, using the downtime for training. There were no layoffs.

Lincoln Electric in Cleveland made a similar commitment, more formally. James Lincoln built an explicit no-layoff guarantee into the operating model in the 1950s, paired with a generous incentive system. The guarantee held through multiple recessions over decades. The argument was unsentimental: you cannot ask workers to be efficient and innovative if they bear all the downside risk of their own efficiency. During downturns, hours were cut before people were cut, and the company absorbed the pain rather than exporting it to the workforce.

Robert Chapman at Barry-Wehmiller applied the same logic during the 2008 financial crisis. Rather than cut anyone, he implemented mandatory furloughs across the entire company — himself included — 4 weeks unpaid, spread across the year. His framing: "better that we should all share the pain than that any of us should bear it all." The sacrifice ran from the factory floor to the executive suite in equal measure. Aaron Feuerstein at Malden Mills kept 3,000 workers on full payroll out of his own pocket after a factory fire in 1995. Both men acted from character. The distinction worth noting is that Lincoln Electric and Toyota acted from policy — a commitment made before the crisis, held through it.

The precondition nobody talks about

There's an accounting problem underneath all of this, and it deserves its own post (coming soon). The short version: people appear on the P&L as pure expense — salaries, benefits, training. They appear nowhere on the balance sheet as an asset. So when a company lays people off, the financial statements show an unambiguous improvement, with no corresponding write-down of the creativity, experience, and institutional knowledge that just walked out the door. Unlike disposing of equipment, there's no balance sheet entry to force the real cost into the calculation. Traditional accounting doesn't just fail to capture the value of people; it actively creates the financial incentive to discard them.

The harder question first

None of this is to say that businesses never contract, or that hard decisions never have to be made. Sometimes they do. But it's worth asking whether contraction itself — the inability to grow, adapt, or pivot — is partly a failure of leadership imagination rather than an immovable market reality. And even when genuine contraction is unavoidable, the knee-jerk reach for layoffs skips the question that should come first: how do I better utilize the pent-up value sitting in the knowledge, creativity, and experience of these people? Financial statements don't answer that question; they don't even ask it. A balance sheet that carries equipment and goodwill but not human capital is structurally blind to the asset being discarded. Real leaders recognize that gap and think harder before pulling the lever that the accounting system makes look costless.

The actual question

We're in a period when a lot of people are feeling genuinely vulnerable to AI, to economic uncertainty, to decisions being made well above their heads. The anxiety is real and it isn't irrational.

Here's a direct question for any leader reading this: would you take a pay cut to save your people? Would you push that cut harder on management than on the floor, since it's management's job to grow or pivot the business? Would you sacrifice a quarter's margin to hold a team together? And if you wouldn't, or couldn't, because the principle was never actually embedded above you, what does that say about what "people are our most important asset" actually means in your organization?

Jon's translation of ningensei cuts through the comfortable version of the answer. Respecting people is a floor. Respecting human nature — what people are capable of when they feel safe, trusted, and valued — is a ceiling most organizations never seriously reach for. The ones that do tend to outperform over time, not despite the commitment, but because of it.

Which leaders have the guts to find out?

Inamori's six questions

Inamori Kazuo applied these questions consistently across Kyocera, KDDI, and his JAL turnaround. They read simply. They aren't.

1. Is it the right thing to do as a human being? The frame is deliberately personal, not professional. Inamori believed most ethical failures in business come from leaders who shift their moral standards depending on the role they're in. Keeping the frame constant is the discipline.
2. Are we doing this for the right reason? Strip away the strategic rationale, the competitive pressure, the board expectation. What's the actual intention behind the decision?
3. Have we put in the maximum effort, every day? Not "are we working hard." Whether we are genuinely committed to continuous improvement, with consistency and sincerity, not just when it's convenient.
4. Does this decision create value for customers, employees, and society? Most businesses optimize for one stakeholder, tolerate a second, and ignore the third. Alignment across all three is the test.
5. Will I still be proud of this decision in 30 years? The long horizon changes the nature of the decision itself. Short-term pressures that feel urgent tend to look different from three decades out.
6. Are we controlling our emotions, or are our emotions controlling us? The question acknowledges that emotional reactions masquerade as strategy. Naming the question before the decision creates a brief but useful pause.

Almost 30 years ago I launched Superfactory.com, born out of my own lean journey and a desire to share what I was learning - and to play with this new technology called "the internet." (Holy crap I'm old...) Over 20 years it grew into a substantial resource hub, with a monthly e-newsletter that peaked at nearly 100,000 subscribers, before I moved on to co-found Gemba Academy in 2012.

A few weeks ago I found a full backup of the articles contributed by some of the best lean minds of that era. To preserve and share I've added them to a new Superfactory Archives page with full attribution to the original authors.

Most hold up well, and some are even more relevant now than when they were written. A few are genuinely primary source material that you won't find anywhere else.

That last category is where things get interesting.

A gentleman's brawl

Late 2005 into 2006, Art Smalley and Norman Bodek (RIP) used Superfactory as the arena for one of the more substantive debates the lean world has ever had: who actually built the Toyota Production System, and what role did Shigeo Shingo really play?

Both men had standing. Smalley was one of the first foreign nationals to work directly for Toyota in Japan, spending a decade there before moving to McKinsey as their lean manufacturing expert. Bodek founded Productivity Press, published over 100 Japanese management books in English, and was the person who introduced Shingo's work to the Western world. He had known both Ohno and Shingo personally.

It started, as these things often do, with a respectful provocation. In December 2005, Smalley published "TPS vs. Lean and the Law of Unintended Consequences," a sharp critique of how Western lean implementations had drifted from Toyota's actual priorities. The article opened with a visit to a Toyota engine plant in West Virginia that had no value stream maps, no dedicated change agents, no U-shaped cells, yet was the most efficient engine operation in the United States. His point: the tools had become the goal, decoupled from the principles behind them. It's a thesis that's only gotten more relevant since 2005.

The debate about Shingo ignited a few months later. In April 2006, Smalley published "Shigeo Shingo's Influence on TPS," an interview with Isao Kato, a 35-year Toyota veteran who had personally coordinated Shingo's visits to the company. Kato's account was precise and meticulous. Shingo first came to Toyota in 1956, after the foundational concepts of JIT and Jidoka were already in place. He taught an industrial engineering course (the "P-Course") to about 3,000 engineers over 20 years, averaging 3 to 4 visits per year. His SMED contributions were largely accomplished at other companies, not Toyota itself; Toyota had already reduced die change times significantly on its own before and after Shingo's involvement. Kato was appreciative but firm: "There is simply no person in Toyota Motor Corporation that thinks Mr. Shingo invented TPS."

Bodek read it and was, in his own words, "considerably, to put it mildly, annoyed." His response, "Dr. Shigeo Shingo: The Greatest Manufacturing Consultant," came in June 2006. He called Shingo "the greatest manufacturing consultant of the last 100 years" and questioned whether the Kato who gave that interview actually understood what Shingo had contributed (he suggested the interviewee "probably was Dr. Shingo's chauffeur"). He cited the former Toyota chairman who said at the dedication of Toyota's first plant in China that "if it wasn't for that man's father, Toyota would not be where it is today." And he invoked the chicken-and-egg answer he'd gotten from Iwata and Nakao, two men who had spent 15 years working with both Ohno and Shingo, when he asked them who invented JIT.

Smalley, via Kato, responded in July 2006 with "Dr. Shigeo Shingo's P-Course and Contribution to TPS." This one is extraordinary. Kato laid out the full curriculum of the P-Course in detail, showing that it contained no JIT, no Jidoka, no kanban, no standardized work. He described Shingo's shift in attitude after the 1973 oil shock, when Toyota was the only major company in Japan to turn a profit and Shingo suddenly became very interested in studying TPS. He addressed errors in Shingo's 1983 SMED book about Toyota. And he identified a key structural problem: most of Shingo's later English-language fame rested on books written after 1980, when he had finished studying TPS, not during the years when the system was actually being built.

Bodek responded one more time in August 2006 with "Who Can Shout Louder?," a somewhat more measured piece that cited Ohno's own words from a 1976 JMA journal article crediting Shingo's SMED concept, noted that Ohno and Shingo had appeared as co-equal keynote speakers at the same conference, and made the point that Shingo's process-versus-operation distinction (X-axis vs. Y-axis, in Bodek's framing) was the conceptual foundation for teaching flow thinking to those thousands of Toyota engineers. Even if Shingo didn't invent the system, teaching 3,000 engineers to see processes differently matters.

Both arguments are largely correct, and the disagreement is partly definitional. If "inventing TPS" means creating its foundational concepts (JIT, Jidoka, the pull system), that happened before Shingo arrived. If it means shaping the thinking of thousands of engineers who then went on to implement and refine the system for decades, Shingo's contribution is harder to dismiss. Kato acknowledged the training influence; Bodek acknowledged Ohno's primacy. They were arguing about the relative weight of two things that were both true.

What makes this exchange extraordinary is what it is: a real-time debate between people who were actually in the room, not a Wikipedia summary or a secondary source written 40 years later. Smalley worked inside Toyota. Bodek had personal relationships with both Ohno and Shingo and published their books. Kato coordinated Shingo's visits for 20 years. These are primary sources, written from memory and conviction, and the fact that they contradict each other in interesting ways is a feature, not a bug.

The rest of the archive is worth exploring too.

Serves: 4–5
Total time: ~55 minutes
Difficulty: Moderate
Diet: Vegetarian

Background

Khao Soi is one of the defining dishes of Northern Thailand, particularly associated with Chiang Mai and the surrounding region. Its roots reflect the area's historical position at the crossroads of trade routes connecting China, Burma, and the rest of mainland Southeast Asia. The dish shows clear Burmese-Shan influence—most visibly in its use of cardamom and turmeric, spices more common to the curry traditions of Burma and South Asia than to the central Thai kitchen. The name likely derives from the Shan "khao swe," meaning noodles.

What makes Khao Soi structurally distinctive is its double-noodle presentation: soft egg noodles served in the broth and a nest of crispy fried noodles on top. This contrast—tender and crunchy, rich and bright—is the dish's signature character and should not be skipped. The garnishes are equally important: pickled mustard greens cut the coconut richness, raw shallots provide sharpness, and lime brings everything into focus.

This version is vegetarian, substituting carrots, cauliflower, and kale for the traditional chicken or beef. The technique—making a fresh curry paste, building the broth in stages, and finishing with the classic garnishes—remains faithful to the original.

Ingredients

Curry Paste

• 2 dried chiles (guajillo or New Mexico for mild; dried Thai chiles for heat)
• 2 cups (480 ml) boiling water, for soaking
• 4 garlic cloves
• 2 shallots, peeled
• 1 inch (2.5 cm) piece fresh ginger
• 1 tbsp ground coriander
• 2 tsp ground turmeric
• 2 tsp curry powder
• ¼ tsp ground cardamom
• Zest of 2 limes
• ½ bunch cilantro, stems only
• 2 tbsp chile soaking liquid (reserved from soak)

Soup

• 2 tbsp neutral oil (grapeseed or avocado)
• 2 medium onions, sliced
• Two 13½ oz (400 ml) cans full-fat coconut milk
• 1½ cups (360 ml) water (use the empty coconut milk can to measure)
• 4 medium carrots, peeled and chopped
• 1 small cauliflower, broken into florets
• 1 bunch kale, stemmed and roughly chopped
• Kosher salt
• Juice of 2 limes

Noodles

• 14 oz (400 g) fresh or dried egg noodles (or Chinese-style wheat noodles)
• Neutral oil, for frying (about ¼ cup, for the crispy topping)

Note: Reserve approximately 2 oz (60 g) of the noodles before cooking to fry for the crispy topping.

Garnishes (Essential)

• Pickled mustard greens — store-bought (found at Asian markets) or substitute: quick-pickled cabbage (see note below)
• 3–4 shallots, very thinly sliced
• Fresh cilantro leaves
• 6 lime wedges

Pickled mustard greens substitute: Toss ½ cup thinly sliced green cabbage with 2 tbsp rice vinegar, ½ tsp sugar, and a pinch of salt. Let stand 20 minutes. Drain well before serving.

Instructions

1. Make the Curry Paste

1. Soak dried chiles in boiling water for 30 minutes. Drain, reserving 2 tbsp of the soaking liquid.
2. Add chiles and all remaining paste ingredients to a blender or food processor. Add the reserved soaking liquid.
3. Blend on high until smooth, adding a little more liquid if needed. Set aside.

2. Build the Soup

1. In a Dutch oven or large wide-bottomed pot, warm the oil over medium heat.
2. Add onions and cook 8–10 minutes until soft but not browned.
3. Add the curry paste and cook 4–5 minutes, stirring frequently, until fragrant and slightly darkened.
4. Pour in coconut milk and water. Stir to combine.
5. Add carrots and cauliflower. Season with salt. Bring to a gentle boil, then reduce to a simmer.
6. Cook 10 minutes until carrots and cauliflower are just tender.
7. Add kale and cook 4–5 minutes until wilted.
8. Add lime juice. Taste and adjust salt. Keep warm over low heat.

3. Make the Crispy Noodle Topping

1. Heat about ¼ cup neutral oil in a small skillet over medium-high heat.
2. Add the reserved noodles in a loose nest. Fry 2–3 minutes, turning once, until golden and crisp.
3. Transfer to a paper towel-lined plate. Salt lightly. Set aside.

Tip: For dried noodles, fry them dry without pre-cooking. For fresh noodles, pat them as dry as possible before frying.

4. Cook the Noodles

1. Cook remaining egg noodles according to package instructions.
2. Drain well and divide among bowls.

5. Assemble and Serve

1. Ladle hot soup over noodles in each bowl.
2. Top with a nest of crispy noodles.
3. Arrange pickled mustard greens and sliced shallots alongside.
4. Scatter cilantro leaves on top.
5. Serve immediately with lime wedges.

Recommended Cooking Sequence

1. Soak chiles (30 min passive time — start this first)
2. While chiles soak: prep all vegetables; slice shallots; make quick-pickled cabbage if using
3. Blend curry paste
4. Build soup — simmer through kale
5. Fry crispy noodle topping
6. Cook remaining noodles
7. Assemble bowls and serve

Presentation

Khao Soi is a bowl dish. Use deep, wide bowls. Noodles go in first, broth and vegetables ladled over, crispy noodles piled on top — not mixed in. Garnishes (shallots, pickled greens, cilantro) are arranged around the edge or on the side, not buried. Lime wedges on the rim. The visual contrast between the golden crispy topping and the rich amber broth is the dish's calling card.

Shopping List

Produce

• 2 medium onions
• 4 medium carrots
• 1 small cauliflower
• 1 bunch kale
• 3–4 shallots
• 1 inch fresh ginger
• 1 head garlic
• 4 limes
• 1 bunch cilantro

Pantry & Dry Goods

• Dried chiles — guajillo or New Mexico (mild) or dried Thai chiles (hot)
• Full-fat coconut milk (2 cans, 13.5 oz each)
• 14 oz egg noodles or Chinese-style wheat noodles
• Ground coriander
• Ground turmeric
• Curry powder
• Ground cardamom
• Neutral oil (grapeseed or avocado)
• Kosher salt

Pickled Mustard Greens (Garnish)

Available at Asian grocery stores (look for jarred or refrigerated pak dong). If unavailable, substitute with ½ cup green cabbage + rice vinegar + sugar + salt (see recipe note).

I was in college in the fall of 1982 when seven people in Chicago died from cyanide-laced Tylenol capsules. I remember the news coverage, the sudden paranoia around medicine cabinets, and the speed with which Johnson & Johnson responded. Forty-plus years later, I still remember it. That's not an accident.

Someone had pulled bottles off store shelves, injected the capsules with potassium cyanide, and returned them. The tampering happened long after the product left J&J's factory. Legally and logically, the company had an easy out. The FDA hadn't ordered a recall, and even the FBI recommended against a full nationwide pull, arguing the contamination appeared limited to the Chicago area. J&J could have contained the response, lawyered up, and waited.

Instead, CEO James Burke asked his team two questions: how do we protect the people, and how do we save the product? In that order. The company spent $100 million to recall 31 million bottles of Tylenol nationwide and relaunched two months later in tamper-proof packaging. Burke went on national television himself. J&J set up a consumer hotline, cooperated fully with investigators, and offered free replacement bottles to anyone who turned theirs in. Congress passed what became known as "the Tylenol bill," making product tampering a federal offense, and the FDA eventually required tamper-proof packaging across the entire OTC industry.

The business result is almost counterintuitive. Market share dropped from 35% to 7% immediately after the deaths. By mid-1983 it had climbed back to 30%, and reached 35% by year's end. Tylenol went on to outsell the next four leading painkillers combined. A crisis that could have killed the brand permanently instead cemented it.

Now contrast that with what happened 28 years later in the Gulf of Mexico.

When the Deepwater Horizon platform exploded on April 20, 2010, killing 11 workers and triggering the largest offshore oil spill in U.S. history, BP had no crisis management plan in place. It was as though the company assumed such a thing could not happen to them, so there was no point being ready for it. Ambiguous public communication underestimated the volume of the spill while overestimating the company's ability to plug the well, creating an image of incompetence, at best, and fraudulence, at worst.

Then their CEO opened his mouth. Tony Hayward, under enormous public and media pressure, said: "There's no one who wants this thing over more than I do. I'd like my life back." Eleven workers were dead. Thousands of Gulf Coast families had their livelihoods destroyed. Fifteen years later, BP has faced 390,000 lawsuits and shelled out $71 billion in damages and mitigation, with restoration efforts still incomplete.

Boeing's handling of the 737 MAX crashes is, if anything, worse, because the underlying problem was self-inflicted. Two 737 MAX aircraft crashed in October 2018 and March 2019, killing 346 people, and investigations revealed issues with the aircraft's automated flight control system and Boeing's safety practices. Rather than grounding the fleet immediately after the second crash, CEO Dennis Muilenburg appealed directly to President Trump about the safety of the aircraft as aviation authorities around the world were grounding the MAX. The formal public apology came 80 days after the second crash.

Both BP and Boeing eventually said the right things. Eventually. The problem is that "eventually" is a confession of institutional instinct, and the instinct in both cases was self-protection first, accountability a distant second.

There's a structural insight buried in these cases that goes beyond PR advice. Burke didn't navigate the Tylenol crisis well because he had a superior communications strategy. He navigated it well because J&J had a written corporate credo, developed decades earlier by former chairman Robert Wood Johnson, that explicitly ranked the company's obligations: consumers first, employees second, communities third, shareholders last. When the crisis hit, Burke didn't have to decide what the right thing was. He already knew. The framework had been built when the stakes were low and the thinking could be clear.

But a credo is only as good as the person holding it. As I wrote a few weeks ago, solid culture and principles can crumble fast under a leader who lacks the character to act on them when it costs something real. Burke had both. The credo gave him the framework; his character gave him the resolve to use it when $100 million was on the line and the FBI was telling him not to. That combination is rarer than it should be.

The Tylenol lesson isn't really about crisis communications. It's about whether your values are load-bearing before you need them to be.

The J&J recall didn't just recover market share; it changed an entire industry. Tamper-proof packaging on every medication bottle you've opened for the past four decades exists because of those 7 deaths in Chicago and one CEO who asked the right question first. That's a different category of outcome than survival. That's leadership that compounds.

Most readers under 50 weren't paying attention in 1982, or weren't born yet. But the structural dynamics, the instinct to minimize, wait, deflect, protect the brand while appearing to address the problem, those haven't changed at all. If anything they've gotten worse in an era when social media makes every gaffe permanent and every delay visible in real time.

Burke's choice looks obvious in retrospect. The right thing usually does. What made it possible was the full stack: a credo built before the crisis, a leader with the character to act on it when the cost was real, and a culture that had genuinely internalized both. The harder part to build in advance: stakeholders who demand all three and refuse to rationalize away any of them. Employees, boards, customers. The willingness to accept less is where the whole thing starts to unravel.

I recently caught a Bloomberg interview with authors of a new book, Priority Technologies: Ensuring U.S. Security and Shared Prosperity, edited by MIT's Elisabeth Reynolds with a foreword by Nobel laureate economist Simon Johnson. Interesting enough that I went and read it. The book covers six technology sectors where the U.S. must lead or fall behind: semiconductors, biomanufacturing, critical minerals, drones, quantum computing, and advanced manufacturing. All of it worth your time. But the argument that stayed with me wasn't about any specific technology. It was about what actually produces technological leadership in the first place – and what we're currently doing to undermine it.

The thing private capital won't fund

Before World War II, the federal government spent almost nothing on R&D. The war forced the issue. Radar, synthetic rubber, penicillin, the Manhattan Project -- these weren't market outcomes; they were the products of deliberate, federally funded science. Congress recognized this, and the government-university research partnership was institutionalized in the 1940s, driven largely by then-MIT president Vannevar Bush.

Johnson's framing in the book's foreword is worth sitting with: "Vannevar Bush realized it wasn't about a stock of technology, it was about a flow of innovation." By some estimates, federal investment in non-defense R&D has accounted for up to 25 percent of U.S. economic growth since World War II. That flow starts in university labs.

Here's the part that critics of government research spending almost always miss: the value isn't just in what you set out to discover. It's in what you find along the way. ARPANET was a military communications project; nobody imagined it would become the commercial internet. GPS was a defense system before it became the foundation of the entire ride-sharing and logistics economy. Decades of NIH-funded basic research into mRNA as a biological mechanism preceded by many years anyone thinking to use it for vaccines. The serendipity isn't a bug in the system. It's arguably the most valuable feature.

Private capital is terrible at funding this kind of work, and for entirely rational reasons. Basic research is long-horizon, uncertain, and the returns -- when they come -- often accrue broadly rather than to the investor. A venture fund that backs a company expecting a 7-year return is already pushing its LPs' patience. Funding a university lab that might produce something transformative in 20 years, possibly in a direction nobody anticipated, is just not how private capital works. Government is the only institution with both the mandate and the time horizon to make these bets. When government stops making them, the pipeline doesn't slow down gradually. It empties.

The current administration's proposed budgets are cutting science agency funding across the board -- NSF, NIH, DARPA, NASA, all taking hits. I'll say it plainly: defunding the federal-university research partnership isn't trimming waste. It's draining the pipeline for the next decade of innovation, including in the six technology sectors the book identifies as most critical to national and economic security.

Talent, domestic and imported

American universities and research institutions have produced extraordinary scientists and engineers -- full stop. The domestic talent base is the foundation of U.S. technological leadership and shouldn't get lost in any discussion of immigration. The question isn't whether American-born scientists are capable. Obviously they are. The question is whether the country is developing enough of them across these fields to meet the competitive moment, and whether it's making itself attractive to the best talent that's available globally.

On both counts, the answer is currently no. Priority Technologies flags this as a cross-cutting theme across all six sectors: the U.S. faces a fundamental workforce shortage, compounded by "radically reduced immigration" and demographic trends that neither domestic education reform nor AI will fully offset in the near term.

Historically, international scientific talent has been a force multiplier for American domestic research, not a substitute for it. Foreign-born graduate students working alongside American researchers, often staying to start companies or join the workforce, have added to what was already a strong base. In 2019, 41 percent of all U.S. STEM PhD graduates were foreign-born visa holders -- and roughly 90 percent of them stayed in the U.S. to work after graduating, many going on to found companies or anchor research programs. That's not a replacement story. That's an amplification story, and it's one the U.S. built its post-war technological lead on.

China understands this dynamic well, which is why it's now actively working the other side of the ledger. Visa restrictions, rising scrutiny of Chinese-born researchers, and a general climate of unwelcome are pushing talented Chinese scientists – including those at MIT who would previously have stayed – to pursue careers elsewhere. Outstanding Chinese scientists at MIT have recently decided to take their careers to countries they view as more welcoming, often back to China itself. Beijing is not unhappy about this. A talent pipeline that used to flow strongly toward the U.S. is reversing, and we're providing the pressure.

Former MIT president L. Rafael Reif, whose endorsement appears in the book, was blunt in a recent NPR interview: "Stop attacking immigrants. This country needs them." He also noted that China, which has pursued an explicit "talent first" strategy for over a decade, has consistently graduated more advanced STEM students than the U.S. for the past 15 years and is closing the gap in top-tier research output. The U.S. has a natural advantage China can't replicate: it's historically been the world's top destination for globally mobile scientific talent. That advantage doesn't maintain itself, and it doesn't survive active dismantling.

Quantum as a case study

The book's quantum computing chapter, written by MIT's William Oliver and Jonathan Ruane, is a clean illustration of what's at stake. Quantum computing has near-term potential to accelerate drug discovery, materials science, and energy applications in ways classical computers can't approach. Oliver and Ruane's conclusion: "The country that achieves quantum leadership will gain decisive advantages in these strategically important industries."

The U.S. is currently in a strong position in quantum. Maintaining it requires exactly the combination of sustained federal basic research funding and deep scientific talent that current policy is targeting for cuts. The physics doesn't care about the political calendar.

The decade-long pipeline

What Priority Technologies gets right, and what tends to get lost in budget debates, is how slowly these systems respond to disruption. The semiconductor fabs we're now trying to rebuild domestically -- belatedly, expensively, correctly -- require a trained workforce that takes a decade to develop. The quantum and biotech breakthroughs that matter in 2035 come from graduate students doing basic research funded today. Choke the investment now, and you feel it in 2033, not in the next news cycle.

Reynolds states the stakes plainly in the book's introduction: the U.S. must "build on its decades of experience in developing frontier technologies" -- a measured way of saying that the lead we have today reflects choices made a long time ago, and the lead we'll have (or won't) in 2040 reflects choices being made right now.

Vannevar Bush understood this in 1945. The question is whether anyone in Washington is paying attention in 2026.

A few days ago I came across a LinkedIn series by Corina Enache, an anthropologist who has spent the past several weeks doing something I hadn't seen done before: methodically tracing management's most beloved frameworks back to their origins, and finding that what we actually use in organizations barely resembles what those origins intended. I've linked every post in the series at the bottom of this page, and each one is worth reading in full.

I read through the series in one sitting and felt the slow-building discomfort of someone who has sat through a lot of change management presentations. I've used some of these frameworks. I've probably cited them. It turns out I was, in several cases, citing a caricature.

With Corina's work as the jumping-off point, I want to push the argument one step further. Because the pattern she documents isn't just ironic. I think it's structural.

The evidence first

Abraham Maslow never drew his famous pyramid. His 1943 paper described needs as fluid, overlapping, nonlinear, "a landscape not a ladder" (his words). The pyramid was drawn in 1960 by Charles McDermid, a consulting psychologist writing in Business Horizons about how to generate maximum employee motivation at minimum cost. Douglas McGregor had already been translating Maslow for business audiences and quietly discarding the nuance in the process. Sixty years of critiques of Maslow's hierarchy are largely critiques of McDermid's diagram, as Bridgman, Cummings and Ballard documented in the Academy of Management Learning & Education in 2019.

Elisabeth Kübler-Ross developed her five stages, denial through acceptance, by sitting with terminally ill patients. With people who were dying, or watching someone they loved die. Death doesn't run a consultation process. It doesn't accept feedback or hold a town hall. The only dignified response available is, eventually, acceptance. That is what the model describes. We now apply it to office reorganizations, where the framing does something useful for management: it turns employee resistance into a phase to be managed rather than information to be heard. The person objecting to the restructure is just in bargaining. Give it time.

Kurt Lewin almost certainly never built Unfreeze-Change-Refreeze. The word "refreezing" doesn't appear anywhere in his writing. It was introduced posthumously, assembled from fragments after his death in 1947, until the scaffolding became the building and the building got Lewin's name on the door. Cummings, Bridgman and Brown documented this in Human Relations in 2016. The irony is complete: Lewin spent his career arguing that behavior cannot be understood by looking at individuals in isolation, that you have to look at the whole web of forces and relationships surrounding them. What carries his name today is a three-box diagram of an ice cube.

The MBTI is the most familiar case and probably the most abused. Isabel Briggs Myers and her mother Katharine Cook Briggs were a writer and a homemaker, not psychologists, building an instrument to measure Jungian typology that Jung himself never intended to be operationalized into a personality test. Between a quarter and half of people who retake it within weeks receive a different type. The categories are binary despite the underlying traits existing on a spectrum. Its predictive validity for job performance is weak. It has survived, as Corina observes, because it gives people a language for difference, not because it's accurate. Those are not the same thing.

That's 4. Corina's series covers several more: Tuckman's Forming-Storming-Norming-Performing (developed from therapy group literature, applied prescriptively to organizational teams), Kotter's 8 Steps (the model became the checklist, the checklist became the change), Spiral Dynamics (Clare Graves never used colors; Don Beck invented them to move away from racial ranking, and they became a new ranking system), Neuro-Linguistic Programming (NLP) (classified as pseudoscience by multiple systematic reviews), Psychological Safety (the distortion was structural from the start, baked into the academic publishing machine rather than happening after the fact), and Ubuntu (a complete ontological philosophy reduced to a poster on a conference room wall). The pattern holds across every case.

The pattern isn't accidental

We tend to frame these distortions as unfortunate, as simplification gone too far, as consultants who didn't read carefully enough. That framing lets the industry off the hook.

Nuanced ideas don't scale. A framework that requires genuine engagement with complexity, that holds uncertainty, that resists being converted into a checklist or a survey or a two-day certification, cannot be productized. And the management consulting industry is, fundamentally, a productization business. The genius of Maslow's actual paper is also the reason it couldn't be sold: it treated human motivation as irreducibly messy. McDermid's pyramid could be put on a slide. Kübler-Ross's actual clinical insight, that people need to be genuinely heard in the face of loss, is harder to sell than a curve that tells managers which stage their employees are in. Lewin's field theory requires thinking about whole systems of forces and relationships. The ice cube diagram requires a PowerPoint template.

Simplification is the product. The pyramid, the curve, the three boxes, the four letters: these exist because they can be packaged and certified at scale to organizations that want the credibility of the research without the friction of the complexity.

The researchers whose names end up on the frameworks rarely built them to be franchises. Several of them, Graves, Lewin, Kübler-Ross, spent years watching their work get stripped of what made it interesting. The consultants and management writers who followed didn't set out to distort anything; most were probably trying to help. But the gravity of a commercial ecosystem pulls consistently in one direction: toward the version that fits on a slide and away from the version that would slow things down.

Corina ends her series with a question worth sitting with: who designed this framework, who sells it, who uses it, and who gets used by it? The knowledge in these systems has almost always flowed in one direction. So has the money, and so has the power.

I'd add one more: before you reach for the framework, could you do the work without it? The work being: actually observing what's happening, listening to what people are telling you, exercising judgment about the specific situation in front of you. If the answer is no, that's worth examining. If the answer is yes, the framework is an accelerator, not a substitute for judgment. The trouble starts when we can no longer tell the difference, and a stage, a type, or a color quietly replaces the person we were supposed to be paying attention to.

Corina Enache's full series, each post linked below, is where this post began.

• Maslow's Hierarchy of Needs
• Kübler-Ross Five Stages
• Kurt Lewin / Unfreeze-Change-Refreeze
• MBTI
• Tuckman's Stages of Group Development
• Kotter's 8 Steps
• Spiral Dynamics
• Neuro-Linguistic Programming (NLP)
• Psychological Safety / Edmondson
• Ubuntu
• Cynefin / Snowden (the one she wouldn't roast)

Oh good. Another lean blog post using Formula 1 pit stops as a metaphor. Quick changeover! Standardized work! Roles clarity! You've seen it a hundred times. I've written something like it at least twice. At this point the F1 pit stop is practically the clip art of the lean world, a visual shorthand so overused it's lost most of its meaning.

So bear with me, because I'm about to use it again. Only this time I want to talk about what lean practitioners almost universally missed when they borrowed the analogy. The pit stop metaphor turns out to be far richer than we made it, and a remarkable 2007 study out of Great Ormond Street Hospital in London proves the point.

Researchers at one of the world's premier pediatric cardiac surgery centers were trying to reduce errors during a specific, perilous moment: the handover of patients from the operating theater to the intensive care unit after complex congenital heart surgery. Think about what that involves. In roughly 15 minutes, a team transfers ventilation, monitoring lines, inotropes, vasodilators, and the entire accumulated clinical knowledge from a 4-8 hour surgery, all while the patient is at peak vulnerability. It is, essentially, a high-stakes, multi-system, multi-professional handoff under time pressure.

Sound familiar? It should. That's not just a pit stop. That's every inter-process handoff in every manufacturing or service operation.

Going to the Gemba

The researchers didn't watch a YouTube video of a pit stop and draw some parallels. They went to Maranello. Ferrari F1 invited the team to watch practice pit stops and conduct detailed discussions with the race director. Two aviation training captains also observed hospital handovers and provided additional input. This was genuine gemba, crossing industry boundaries to observe expert practice firsthand, not secondhand analogy.

What they brought back wasn't a stopwatch philosophy. It was a complete framework: defined leadership (a single coordinator with clear handoff authority), explicit task allocation, sequential phasing, structured communication discipline, checklists, situation awareness protocols, hierarchy-free escalation, and pre-handover briefing. If that list looks a lot like the Toyota Production System applied to information transfer, that's not an accident.

What Lean Actually Missed

When lean practitioners point to an F1 pit stop, they're almost always talking about SMED, Shingo's Single Minute Exchange of Die methodology, reducing changeover time by separating internal and external setup, standardizing tools, and simplifying fasteners. Valid and useful. But it's one thread from a much larger tapestry.

The hospital study surfaced what's actually interesting about a pit stop: the handoff architecture. Before each tire change, roles are assigned and understood. During the stop, communication is deliberately minimal so the choreography can execute without interference. The lollipop man (the coordinator) holds situation awareness while specialists execute narrow tasks. And critically, the team debriefs after every race to review what happened.

None of that is about speed. It's about reliability under pressure, which is a very different problem.

Compound Errors and Decoupled Failures

The most striking finding in the study involves what happened to errors after the new protocol was introduced. Before the new handover process, technical errors and information omissions were statistically correlated: when the equipment handover went badly, the information briefing also suffered. A poor technical handoff compounded into poor knowledge transfer, exactly the kind of error cascade that James Reason's Swiss Cheese Model describes. One hole aligned with another.

After implementing the structured protocol, that correlation disappeared. The protocol essentially decoupled the failure modes by sequencing them into distinct phases: equipment transfer first, information handover second, group discussion third. Even when one phase was imperfect, the others held.

In lean terms, this is poka-yoke logic applied to the transfer of tacit knowledge. The structure itself interrupts the defect propagation.

The Turnover Problem Nobody Talks About

There's one more finding worth pulling out, and it has direct implications for any lean practitioner trying to sustain standard work in a high-turnover environment.

F1 pit crews are remarkably stable. A team of roughly 20 people sees one or two changes per year. The researchers noted this explicitly as a key differentiator, because Great Ormond Street's ICU rotated six residents every three months, with a nurse turnover of approximately 10% annually. The protocol had to be designed for an inherently unstable team structure, trainable in 30 minutes or less at staff induction, documented on laminated sheets at each bedside. And it worked, with a 95% compliance rate on the pre-handover transfer form.

This is a lesson lean operations often dodge. Standard work is easy to sustain when your workforce is stable. The interesting design problem is building standard work that remains robust when it isn't, which is the actual situation in most hospitals, warehouses, call centers, and frankly most manufacturing facilities above a certain size.

The Analogy Was Always Bigger

The F1 pit stop metaphor earned its place in lean training because it's viscerally compelling. Eighteen people, seven seconds, zero mistakes. But that image tends to lock people onto the speed dimension, the changeover efficiency story, while the deeper lesson sits right next to it, largely ignored.

The deeper lesson is about what happens at the seams between processes, where knowledge, authority, and physical work transfer from one team to another. That's where errors compound, where tacit knowledge evaporates, where the next team inherits problems they weren't warned about.

So the next time someone in your organization uses an F1 pit stop to make a point about quick changeover, ask them a different question: what does your post-surgery handoff protocol look like?

At the start of 2026, I set myself a somewhat ambitious goal: cook traditional dishes from 30 different countries over the course of the year, hosting a weekly dinner for friends along the way. Ambitious because I'm a decent home cook, not a great one.

That annual goal tradition is now in its third decade. Each year I pick something that pushes me in a direction I wouldn't naturally go, physically, intellectually, or creatively. Previous years have ranged from physical challenges to deep reading projects to learning new skills. This year is food, geography, and hospitality, all at once.

Twelve countries in, I'm calling it on track. You can see the full list, with recipes and notes, on my cooking page. Peru, South Korea, Zanzibar, Morocco, and eight others so far.

The learning curve has been steeper than I expected, and in ways I didn't anticipate. I knew I'd pick up techniques and flavor combinations. What I didn't expect was how quickly food pulls you into history and geography. Each recipe on the site begins with the historical and social context of the dish, which I've genuinely enjoyed researching. Why does this cuisine use these spices? What trade routes brought them here? What does a street food tell you about how a culture eats and socializes? Food turns out to be a remarkably efficient delivery mechanism for all of that. I've read more about colonial history and agricultural economics this year than in the previous decade, all in service of understanding why a dish tastes the way it does.

Our friends have picked up on that thread too. Having a specific cuisine as the evening's theme gives everyone something to talk about before the food even arrives. I'll usually put together a playlist from the country, which sets a mood without requiring anyone to perform enthusiasm for music they've never heard. We'll often talk about travels in that region, and I've linked to some of those on my travel page as well. It's become more of an experience than a dinner party, which I mean as a compliment to the format, not a warning about the company.

The sourcing has been manageable. My local supermarket and the Asian market nearby handle most of it. Amazon has made a few appearances for specialty items: aji amarillo paste for the Peruvian Pescado a la Chorillana, a specific tamarind paste, a dried spice for the Zanzibari filling. Nothing exotic enough to require a passport.

The Pescado a la Chorillana remains the clear favorite so far, a Peruvian fish dish with a bold tomato and pepper sauce that manages to be both simple and startlingly good. The Zanzibari pizza is worth a separate mention if only because the name invites skepticism. It's real, it's a street food, and it's a spiced flatbread folded and pan-fried, finished with egg. We made a vegetarian version that worked beautifully. Our friends were skeptical until they tasted it, which has become a reliable pattern at these dinners. The South Korean salmon with a gochujang glaze rounds out the top three.

A full meal with two or three sides takes real preparation, and the cleanup is its own adventure. But the social payoff has been genuine. We're more connected to our friends than we were a year ago, which wasn't something I listed as a project objective but probably should have been.

Eighteen countries to go. If you're curious about specific recipes or want to suggest a country I shouldn't miss, the cooking page has all the details.

Serves: 4 Total time: ~55 minutes Difficulty: Moderate Diet: Vegetarian (fully vegan)

The Dish and Its Place in Chilean Life

Porotos Granados — literally "remarkable beans" — is one of the most distinctly Chilean dishes in existence, and one of the few that has no meaningful parallel elsewhere. It is built around three ingredients that arrive together in the Southern Hemisphere summer: fresh cranberry beans (porotos granados), field corn (choclo), and Chilean pumpkin (zapallo). The combination echoes the pre-Columbian three sisters tradition — beans, corn, and squash grown together as complementary crops — adapted through centuries of Chilean agricultural life into something that tastes wholly its own.

The dish is associated with January and February in Chile, when all three ingredients are at peak freshness in the Central Valley markets. Families cook it on weekday evenings and Sunday afternoons alike; it is not celebratory food but sustaining food, the kind that marks a season rather than an occasion. Basil — added in quantity at the very end — is non-negotiable. It lifts the stew from earthy to alive.

Outside Chile, the dish is almost unknown, which makes it an unusual case: a genuinely beloved national staple with essentially no diaspora footprint. Making it in April requires minor concessions to seasonality — frozen corn and butternut squash stand in capably for fresh choclo and zapallo — but the flavor logic remains intact.

Pebre, Chile's ubiquitous fresh condiment, accompanies almost everything on a Chilean table. It functions somewhere between a salsa and a sauce: bright, herbaceous, with varying heat depending on the cook. Served alongside crusty bread, it is the standard partner for Porotos Granados and provides the acidity and freshness the stew itself does not supply.

Ingredients

Porotos Granados

• 2 cans (15 oz each) cranberry beans (borlotti beans), drained and rinsed — or 3 cups cooked from dried
• 2 cups frozen corn kernels (or fresh if available)
• 2 cups butternut squash, cut into ¾-inch cubes (about ½ medium squash)
• 1 large yellow onion, finely diced
• 4 cloves garlic, minced
• 2 tbsp olive oil
• 1 tsp sweet smoked paprika (substitute for Chilean color, a mild annatto-based spice)
• ½ tsp dried oregano
• ½ tsp ground cumin
• 2 cups vegetable stock
• 1 cup water
• 1 tsp kosher salt, plus more to taste
• Freshly ground black pepper
• 1 large handful fresh basil leaves, roughly torn (do not substitute; do not add early)

Pebre (Chilean Fresh Condiment)

• 1 cup fresh cilantro, finely chopped (stems and leaves)
• ½ small white onion, very finely diced
• 1–2 serrano or jalapeño chilies, seeded and minced (adjust to heat preference)
• 2 cloves garlic, finely minced or grated
• 2 tbsp olive oil
• 2 tbsp red wine vinegar or fresh lemon juice
• ½ tsp salt
• 2–3 tbsp cold water (to loosen)
• Optional: 1 ripe Roma tomato, seeded and finely diced

To Serve

• Crusty bread or marraqueta-style rolls (Chilean bread rolls, or a good French baguette)

Instructions

1. Make the Pebre (10 minutes, make first)

1. Combine cilantro, onion, chili, and garlic in a bowl.
2. Add olive oil, vinegar, salt, and water. Stir to combine.
3. If using tomato, fold it in now.
4. Taste and adjust salt and acidity. The pebre should be bright and assertive — not shy.
5. Cover and refrigerate while you cook the stew. It improves as it sits.

2. Build the Stew Base (15 minutes)

1. Heat olive oil in a wide, heavy pot over medium heat.
2. Add onion and a pinch of salt. Cook, stirring occasionally, until soft and translucent — 8–10 minutes. Do not rush; this is the flavor base.
3. Add garlic, smoked paprika, oregano, and cumin. Stir and cook 1–2 minutes until fragrant.

3. Add Squash and Simmer (15 minutes)

1. Add butternut squash and stir to coat in the aromatics.
2. Pour in vegetable stock and water. Bring to a gentle simmer.
3. Cook uncovered 12–15 minutes until squash is nearly tender but still holding its shape. It will finish cooking with the beans.

4. Add Beans and Corn (10 minutes)

1. Add drained beans and frozen corn. Stir gently to combine.
2. Simmer an additional 8–10 minutes. The stew should thicken as the squash breaks down slightly at the edges and the beans release starch. If it seems too thick, add a splash of water; if too thin, cook uncovered a few minutes longer.
3. The consistency should be spoonable and cohesive — thicker than soup, looser than a braise.

5. Finish with Basil

1. Remove from heat.
2. Tear basil leaves and stir through the stew. Do not cook the basil further — it should remain vibrant green and fresh-tasting.
3. Taste and adjust salt and pepper.

Cooking Sequence

1. Make pebre → refrigerate
2. Dice squash and prep aromatics
3. Build stew base (onions first)
4. Add squash, simmer until nearly tender
5. Add beans and corn, finish cooking
6. Tear in basil off heat, adjust seasoning
7. Warm bread, bring pebre to table

Presentation

Serve Porotos Granados in wide shallow bowls — it should look substantial but not muddy. The color contrast between the white beans, orange squash, and yellow corn is part of the appeal; don't stir it to death. A few fresh basil leaves on top signal what's inside.

Place pebre in a small shared bowl at the center of the table with a spoon. The Chilean instinct is to spoon pebre directly onto the stew between bites, or onto torn bread eaten alongside. Do not pre-dress the bowls — let guests apply their own.

Bread is not optional. It plays the role that injera does for Ethiopia or flatbread for Lebanon: something to carry the stew, absorb the broth, and extend the meal.

Combined Shopping List

Produce

• 1 large yellow onion
• ½ small white onion (for pebre)
• 1 head garlic
• ½ medium butternut squash (~2 cups cubed)
• 1 large bunch fresh cilantro
• 1 large bunch fresh basil
• 1–2 serrano or jalapeño chilies
• 1 lemon (or use red wine vinegar)
• Optional: 1 Roma tomato (for pebre)

Pantry

• Cranberry beans / borlotti beans (2 cans, 15 oz each)
• Frozen corn kernels (2 cups)
• Vegetable stock (2 cups / 1 carton)
• Olive oil
• Sweet smoked paprika
• Dried oregano
• Ground cumin
• Red wine vinegar
• Kosher salt, black pepper

Bread

• Crusty baguette or rustic rolls (enough for 4)

A few months ago, the dominant AI concern was jobs. Would it take yours? Which roles were safe and which weren't? That felt like the right level of worry for a technology that, as many still describe it, essentially guesses the next word in a sequence.

Then Anthropic released Mythos, and the conversation shifted rather abruptly to critical infrastructure and existential risk.

I'll offer something small before getting to the larger point. I follow AI development more closely than most people I know, use these tools regularly, and have been writing about their expanding capabilities for months. I'm still nowhere near understanding or using the full extent of what current models can already do. That's not false modesty; it's a useful data point for what follows.

A friend who works at one of the top AI labs told me something a few months ago that stuck with me. His company has models ready for deployment that they're deliberately holding back. Not for technical reasons. Because the capabilities would frighten people, and they want the public to grow more comfortable with existing tools before the next leap arrives. Sit with that for a moment. We're at a point where technology companies are managing the psychological pacing of disclosure, rationing what the public is allowed to understand their systems can do. And if Mythos is any indication, even the developers may not fully grasp what they've built, or they're worried enough about it to act accordingly.

Mythos is Anthropic's latest frontier model, the accepted term for systems sitting at the leading edge of what AI can currently do. Anthropic shared access with just 11 organizations, all American, plus Britain's AI Security Institute as the sole international exception. That arrangement alone set off a global scramble. Within weeks, Mythos had identified thousands of zero-day vulnerabilities (previously unknown software flaws) across every major operating system and browser, achieving an 83 percent exploit success rate on the first attempt. Those aren't abstract targets. The same software categories run the control systems embedded in power grids, water treatment facilities, financial networks, and nuclear plants. Anthropic itself called the model too powerful for broad public release. The Bank of England governor warned publicly about cracked global cyber-risk. Canada's finance minister compared the threat to a closure of the Strait of Hormuz.

This is not an autocomplete concern.

The deeper issue isn't Mythos specifically. OpenAI, DeepSeek, and others will have comparable frontier models within 18 months by Anthropic's own estimate, and some may be less deliberate about controlled release. Mythos is a crystallizing moment, not a singular one. The question it forces is structural: how did we arrive at a point where a private company in San Francisco determines which nations and organizations have access to something with weapons-test-level geopolitical consequences?

The answer is that this was always going to happen given how AI was built. Historically, the most dangerous technologies emerged from government institutions. The Manhattan Project wasn't funded by venture capital. Government development meant government control from the beginning, with classification systems, internal oversight, and at least the theoretical possibility of international frameworks to follow. AI was different. It was built by private companies, with private capital, at private speed. By the time any regulatory body understood what was being created, the capability was already in private hands, with private criteria for access. Anthropic's decision to initially share Mythos with 11 American companies isn't a conspiracy. It's the predictable consequence of a development model that outpaced every governance structure that might have shaped it differently.

Which brings us to the guardrails problem, and it has no clean solution. Countries that implement serious AI constraints operate at a disadvantage against those that don't. Responsible actors are structurally penalized relative to reckless ones. Governments that want guardrails face a genuine dilemma: restrain development and risk being overwhelmed by adversaries who won't, or race ahead and contribute to the instability they're trying to prevent. No nation is going to accept an AI nonproliferation framework if they believe their adversaries will ignore it. The incentive structure points in one direction only. Everyone feels they must invest rapidly or risk falling irreversibly behind. And "irreversibly" isn't rhetorical here. We're talking about gaps that open over weeks or months, not years. No previous economic or technological threat, not offshoring, not semiconductor competition, has materialized at that speed. This race has no finish line and no agreed-upon rules, and the laps are getting shorter.

Europe is the clearest illustration of what this looks like in practice, though it stands in for most of the world. The EU has genuine regulatory ambitions, has passed the AI Act, and has the institutional capacity to at least attempt coordinated action. And yet European governments look at the energy infrastructure, compute investment, and capital availability in the United States and China and recognize that their regulatory environment and capital markets structurally cannot match it. Even careful deliberation of a few months may now be disqualifying in this race. Nations with fewer institutional resources than Europe, which is most of them, don't have that choice to make. They're largely observers of a contest whose outcome will shape their future regardless.

What makes all of this genuinely unsettling isn't just the technology. It's the timeline. We moved from debating AI's impact on employment to confronting its implications for critical infrastructure security in roughly twelve months. Democracies don't build international consensus in twelve months. Regulatory frameworks don't form in twelve months. Public understanding certainly doesn't shift that quickly, which is why a meaningful portion of the population still thinks of these systems primarily as sophisticated text generators. That perception gap matters, because the political will to address a problem depends on people first understanding what the problem actually is.

Something comparable to Mythos will arrive from another lab soon enough, perhaps developed somewhere less concerned with coordinated defense. And worth noting as a final layer of sobriety: quantum computing, still a few years from practical deployment at scale, is advancing faster than most expected. Researchers are increasingly confident it will dramatically amplify AI capabilities when it arrives, compressing already-short development cycles further. If the current pace feels difficult to govern, it's worth asking what governance looks like when that acceleration arrives.

The question worth sitting with isn't what we do about this specific model. It's whether we're capable of building the kind of global coordination this moment requires, at the speed it requires. We've never had to do that before. A year ago, we were debating whether AI would write your emails.

What will we be debating a year from now?