Part of his argument was strictly on the numbers. Yes, recently AMD has grown faster, by percentage increase. But Robert pointed out NVDA’s growth comes from a far larger base, and in absolute terms is still growing faster than AMD.

The clincher, however, was how he pointed out that the future of the AI economy remained NVDA’s.

“Nvidia’s competitive advantage still extends beyond accelerator performance,” Dr. Castellano wrote. “Customers are buying an integrated architecture encompassing GPUs, CPUs, networking, rack-scale design, software libraries and deployment support. The economic value lies in reducing the time and execution risk required to bring extremely large AI systems online.”

In other words, in overvaluing AMD versus NVDA, the market was neglecting NVDA’s strategic importance, its ability to shape the future of its industry, and the economy itself.

That’s the question Robert highlighted for me. Why do markets miss, or at least undervalue, strategic importance?

It’s a bit like the guy who lost his keys and looks for them under the light, though he lost them in the dark, in quite another place. Companies issue financial results. That’s the information under the light. That easy visibility prompts investors to ask, “how much money is this company making today and maybe the next few quarters?”

They are slower to ask: “How important is this company becoming to the rest of the economy?” That’s a systems question. There is no Securities and Exchange Commission (SEC) form to cover systems. Markets price performance long before they price strategic position. And strategic position cannot be understood by studying the company alone. We must understand the system around it.

The AI economy increasingly requires investors to understand systems before they can understand firms.

Share

We have held Nova (NASDAQ: NVMI) in the Gilder’s Technology Report portfolio since 2024. For much of its history, it was a nice little company with modest revenues, though improving margins. Then everything changed. Not because Nova made a hot product. But because it became clear that TSMC could not continue shrinking transistors and putting them into ever more complex 3D systems without Nova.

Crucial semiconductor features have been smaller than the wavelengths of visible light since the 1990s. But as chipmakers moved into increasingly complex 3D structures, simply seeing the features mattered less than determining their dimensions, profiles, and material properties. Nova’s systems infer those characteristics from the way light interacts with the structures.

For years, investors saw in Nova a profitable metrology company. Eventually, they realized Nova isn’t just measuring wafers. It’s providing for the continuation of Moore’s Law. Then the stock began to move.

In the old economy (okay, five years ago, but who’s counting?) analysts evaluated a company’s performance. They measured revenue, earnings, margins, growth, and cash flow.

In the AI economy, the heightened complexity of value chains increasingly requires us to evaluate companies’ positions in those chains. Investors must ask

• Who does everyone else depend on?

• Who determines the pace of progress?

• Who sits at the bottleneck?

• Whose roadmap shapes everyone else’s roadmap?

Investors are naturally obsessed with what is visible for a company, product, or revenue. But the fantastic complexities of the AI economy mean that wealth increasingly accumulates at hidden control points.

The stock market can do a good job at valuing what already exists. But strategically important companies shape the future organization of the economy. This makes the job of the investor analyst fundamentally more difficult because as the organization of the economy changes, so does the strategic importance of companies.

Nova did not suddenly become better at measuring chips. Instead, chips became dramatically harder to measure. It was at that point that Nova began to be a control point in the semiconductor manufacturing system: progress elsewhere increasingly depended on solving the measurement problems Nova was built to solve.

And then the stock began to move.

Subscribe now

This adds another task to the investor’s job: paradigm detection. It is not enough to find today’s control points. Investors must recognize when the technological architecture itself is changing, because a new paradigm rearranges dependencies and moves the bottlenecks. A company that looks peripheral under the old paradigm can become a control point under the new one.

If the paradigm of semiconductor progress is “smaller transistors,” Nova looks like a useful metrology supplier. If the paradigm becomes ever more complex, with 3-D structures, materials, chiplets, and advanced packaging, measurement and inference become much more important to continued progress. Nova hasn’t merely improved its financial performance. Its position in the semiconductor architecture has changed.

In the AI economy, the worst aspect of traditional measures of financial performance—revenue, margins, earnings, and cash flow—is accentuated. More than ever, these financial measures are backward-looking. They tell you how valuable the company has been under the existing architecture. They don’t necessarily tell you how valuable its position is becoming as the architecture changes.

Financial statements remain indispensable. But they are records of a company’s performance within the existing architecture. Paradigm detection asks a different question: As the architecture changes, is this company’s position becoming more important or less?

The analyst used to ask, “How good is this company?” The paradigm detective asks, “What is happening to the system that determines how important this company will become?”

We all need to become paradigm detectives.

Leave a comment

The first time I can remember this coming up was when George explained “low and slow.” This was back in the late 1980s when I was helping him with his great book Microcosm. At the time, the conventional “high tech” path to multiply computing power was to ramp up electrical power and switching frequency, boosting the number of calculations per second for any given circuit.

What George saw was that “high and fast” would build so much heat per circuit that transistors would have to be spread further apart, limiting how many transistors could be fit on a chip.

Far better, he argued, was to reduce power and speed, enabling a radical increase in transistor count—and computing power.

The technological race seemed to favor high and fast. The economic race favored low and slow. Low and slow looked less glamorous, but it was where the value would ultimately accumulate.

It was also a classic Peter Drucker outcome, which may have been part of its appeal to George. “Don’t solve problems, pursue opportunity,” Drucker taught us. The winning idea wasn’t where engineering effort was concentrated—it was where the economic leverage ultimately emerged.

Today, I think about this using another set of George terms: abundance and scarcity. As George also taught me, every major technological advance creates a new abundance. But every new abundance exposes a new scarcity. Every scarcity becomes a bottleneck. And the companies that remove those bottlenecks become the places where value accumulates.

Because AI accelerates innovation itself, abundances arrive faster than ever. Each abundance exposes a new bottleneck before investors have fully recognized the last one. Value migrates just as quickly, rewarding the companies that solve the newest constraint rather than yesterday’s.

This rearranges the investment search. The central question isn’t “Which company has the best technology?” Instead, the key question is: “Which company has become the solution to the economy’s newest binding constraint?”

Share

This explains many of the investment picks for our Gilder Technology Report Nova Ltd. (NASDAQ: NVMI) has become vastly more valuable as the computational power of microchips became more abundant. One source of that abundance was the continued shrinkage of circuits to levels below the visible wavelengths of light. That created a new bottleneck because now testing firms needed to assess invisible features. Like bats, Nova sends bursts of energy toward these submicroscopic features, and then (with lots of help from AI) interprets the echo.

And then there is the learning bonus. Once a company like Nova becomes the solution to a difficult bottleneck, every customer interaction makes that solution better. Each customer learns about its own products from Nova, but Nova learns from every customer’s product.

Similar logic applies to Camtek (NASDAQ: CAMT), Palantir (NASDAQ: PLTR), Nvidia (NASDAQ: NVDA), ASML (NASDAQ: ASML), and any company that specializes in powering past bottlenecks holding back an industry.

Subscribe now

Do note, however, that last point. Every economy contains thousands of bottlenecks. Most are economically unimportant. The fortunes are made by identifying the one that has become the binding constraint on an entire industry. Nova isn’t valuable because it removed a bottleneck. It’s valuable because it removed the bottleneck that has come to matter most.

To be useful, an investment theory must tell you where value accumulates. Once you are looking in the right place, opportunities become much easier to recognize.

Here is our theory. Technology creates new abundances. New abundances create or at least reveal new scarcities. Scarcities determine where value accumulates. Find the new scarcity—and its resolution—and you’ll often find the next great investment.

Leave a comment

Most recently, I have been reading “Innovation and Entrepreneurship,” published in 1985 and covering events starting in the late 1960s. So my late childhood, adolescence, and early adulthood.

Having lived through this period, I had a picture of it in my head, and it turns out my picture was all wrong. The 1970s I remembered as locked in the doldrums of stagflation, with no relief in sight, until Reagan came along.

Reagan did a lot of good, but Drucker explains that the biggest economic shift of the era started at least a decade before his inauguration.

The great shift came in the sources of job creation. “In the quarter century after World War II, the Fortune 500—the largest companies in the United States—added essentially all the new jobs created in the American economy.”

But, Drucker explains, in the early 1970s, job creation and job growth in the United States “shifted to a new sector.” The old job creators began to lose jobs, at first slowly, and then very quickly. “By 1984, the Fortune 500 had lost permanently at least 4 to 6 million jobs.”

Meanwhile, small and medium-sized businesses started adding jobs at a tremendous rate. Total employment in the United States grew by some 35 million net new jobs. Since big business had lost at least 5 million jobs, that means these smaller enterprises “created 40 million or more [jobs], since we had to offset a permanent job shrinkage of at least 5 million jobs in the traditional employing institutions.”

Okay. Interesting. Even surprising. But hardly shocking, I thought. Hi-tech started creating lots of jobs in the mid-1960s, so that explains it.

Nope. Dead wrong, said Drucker. “Of the 40 million-plus jobs created since 1965 in the economy, high technology did not contribute more than 5 or 6 million. High tech thus contributed no more than ‘smokestack’ lost.”

That line made me sit up and think. How did this happen?

Share

Drucker attributes most of that growth to the emergence of a new information technology that most people never think of as a technology at all. The new technology was “management,” a technology for processing information—collecting it, organizing it, transmitting it, and turning it into coordinated action. Drucker argues that management came into existence between the two world wars but was not recognized as a discipline to be learned until the late 1940s.

Drucker himself was the pioneer of the discipline starting with The Concept of the Corporation, which grew out of his two-year study of GM. By the time he published “The Practice of Management” in 1954, management was on its way to being studied by hundreds of thousands of aspiring executives.

Organizational knowledge that had once been confined to America’s largest corporations became available to tens of thousands of firms through business schools, consultants, books like “The Practice of Management,” MBA programs, and an emerging managerial profession.

As an information technology, management reduced the cost of coordination. It made information visible throughout the enterprise, allowing managers to recognize opportunities, delegate decisions, and scale organizations that previously depended on the founder’s direct supervision and dramatically lowered the organizational barriers to entrepreneurship.

Management codified knowledge, created repeatable processes, and delegated authority without sacrificing control. Above all, it made organizations teachable.

Management did for organizations what the machine tool had done for manufacturing. It turned craftsmanship into a repeatable process. More than a way to supervise large companies, management allowed small companies to become large ones.

Drucker wisely and correctly does not attribute all the job growth of the period to entrepreneurial management. Important social realities also contributed, not the least being momentum and the popularization of the entrepreneurial ideal.

Oh, OK, I hear you asking, “why should we care about industrial era management?”

Because the lesson learned is not simply that management mattered. It is that by reducing the cost of processing information, management unleashes entrepreneurship.

Drucker watched that happen once, when management escaped the Fortune 500 and became available to ordinary businesses. We are about to watch it happen again, as AI once again dramatically accelerates the flow of information, reduces the overhead of management, slims organizations, and reduces the capital required to turn an idea into a product or service.

The result? We are about to see an explosion in job creation. Jobs will be lost, but many more will be created. To the extent that new job creation is limited by population, the likely outcome will be to shift our human resources to the enterprises that get the most out of their employees, which means employees will get even more out of their employers.

America and Americans are about to get a whole lot richer.

Leave a comment

That has not happened yet. Instead, most discussion of AI investing consists of deciding whether Nvidia is overvalued, whether hyperscaler capital spending has peaked, whether this or that software company is “using AI,” or whether another startup has become the next unicorn: reasonable questions, but strictly tactical. They assume that the underlying theory of investing remains unchanged.

A “General Purpose Technology” (GPT) is a foundational innovation that improves continuously, spreads throughout the economy, and enables countless new products, industries, and innovations, transforming the way every industry creates value. Think railroads, electricity, the computer, and the internet.

Over decades, these technologies reshape industries. They change how businesses create value. Artificial intelligence appears to be doing so faster, and perhaps more broadly, than any technology since the computer.

Because investment ultimately always comes down to valuation, we should expect this new source of value to change how investors invest.

For decades, investors have relied upon concepts developed largely during the industrial age. We ask whether a company enjoys economies of scale. Whether it possesses pricing power. Whether barriers to entry protect profits. Whether management allocates capital wisely. Whether assets generate attractive returns.

None of these questions disappear. But they are no longer sufficient.

Consider a few ideas we have explored over recent months.

Before AI, intelligence was rare and expensive. AI is making intelligence abundant and cheap .But abundant intelligence no longer commands premium returns. Instead, value migrates toward whatever remains scarce.

We argued that AI reduces friction

When, as a result, planning, design, coding, marketing, research, and coordination become dramatically easier, entrepreneurship changes. Organizations change. Competitive advantages built upon administrative complexity begin to weaken.

We argued that visibility enabled by AI processing of chaotic data has become a new source of national power.

Industrial civilization rewarded simplification. The AI economy increasingly rewards those able to manage, coordinate, and profit from growing complexity.

None of these observations, standing alone, tells us which stock to buy. Taken together, however, they say that the underlying architecture of value creation is changing.

Increasingly, the most valuable companies do not simply produce products. They remove bottlenecks. They coordinate complexity. They illuminate the invisible. They occupy critical positions inside systems that continue growing more interconnected every year.

These observations are not merely theoretical. They already help explain why certain companies have become strategically indispensable. Companies like Nova (NASDAQ: NVMI) and Camtek (NASDAQ: CAMT), for example, do not make products. They use AI to analyze defects in semiconductor manufacturing processes that are impossible to see conventionally. They are deeply embedded in manufacturing firms such as Taiwan Semiconductor (NYSE: TSM). They remove bottlenecks and constraints that would render current microchips impossible and thereby enable thousands of other firms to become more productive.

Such AI-driven companies require investors to ask new questions.

Instead of asking, “How large is this market?” ask, “What constraint does this company remove?”

Instead of asking only whether a company possesses a moat, perhaps we should ask whether it occupies an indispensable control point within a complex system.

Instead of asking merely whether earnings are growing, perhaps we should ask whether the company’s strategic importance is growing faster than investors currently recognize.

Different questions produce different portfolios.

I am not suggesting we discard the wisdom of Benjamin Graham, Philip Fisher, or Warren Buffett. For each, their great strength was to identify the source of value that dominated in their era. For the AI era, we should follow their lead.

Great investors begin by asking the question everyone else was neglecting.

The AI economy demands a new set of questions.

Leave a comment

Majestic responded with an AI processor incorporating with far more native memory than the Nvidia models, allowing for fewer GPUs per bundle and dramatically lower costs in energy, cooling, and capital.

Another top 10 startup is Qodo. One of AI’s superpowers is writing software code, cheaply and abundantly. But every new abundance creates a new scarcity, a new bottleneck. The new bottleneck in AI software is checking the proliferating pile of new algorithms for errors, bugs, and inefficiencies. Qodo figured a way to automate that process, preserving the value of AI-generated code.

This phenomenon of abundance-producing scarcity and flow generating chokepoints well-predates AI. Every wave of technological progress does this. Often the abundances and constraints follow so closely that they may be hard to distinguish from each other. The industrial era was largely the product of suddenly abundant horsepower, steel, and railroads, yet the demand for all three of th

ose generated powerful and profitable bottlenecks.

Share

When every product or service must pass through the same narrow point, the firm that owns that passage gains pricing power and strategic importance. The companies that resolve those constraints, allowing passage through these chokepoints, often become toll booths. They can make lots of money for investors who catch the wave in time.

But if this phenomenon has been repeated throughout history, why is it particularly relevant to investing in the AI economy?

First, AI is new and is creating a whole new set of abundances and scarcities, flows and bottlenecks. That would be reason enough for investors to focus on AI’s chokepoints. But AI is also doing this—rearranging the economy and relocating the economy’s bottlenecks—at extraordinary speed.

Historically, bottlenecks often took decades to migrate. The railroad revolution took roughly 50 years to rearrange the economy. Generative AI is rearranging it in five. Railroads transformed transportation first, then steel, then manufacturing, then retailing, then finance over the course of half a century. AI is transforming entire industries simultaneously.

So yes, every technological revolution shifts bottlenecks. But AI is driving the fastest such shift in economic history.

Two years ago, everyone was talking about GPUs and how we needed more of them faster. Now everyone is talking about memory.

But not everyone got there at the same time. My esteemed colleague Dr. Robert Castellano, PhD, was one of the first to point out memory as the new bottleneck.

Now that everyone is talking about that, Robert has already leapt ahead to the next big issue the power and heat crisis caused by connecting thousands of GPUs with copper links. He also names the company that is conquering the bottleneck, but you must subscribe to the Gilder Technology Report to get that!

Leave a comment

Henry Ford simplified manufacturing. Walmart simplified retailing. Southwest simplified air travel.

No longer. In the AI economy, complexity is unavoidable. Those rewarded are companies that make overwhelming complexity manageable.

Modern semiconductor manufacturing is almost unimaginably complex. Microchips require thousands of manufacturing steps, atomic-scale precision, and tolerances measured in billionths of a meter.

Yet companies like Taiwan Semiconductor Manufacturing Company (NYSE: TSM) continue pushing the frontier of possibility year after year.

TSMC can do this not because semiconductor manufacturing has become simpler, but because a supporting ecosystem of companies has emerged to make that complexity comprehensible.

Share

Nova (NASDAQ: NVMI) measures processes that cannot be observed directly. Camtek (NASDAQ: CAMT) inspects structures too small for the human eye. Synopsys (NASDAQ: SNPS) and Cadence Design Systems (NASDAQ: CDNS) allow engineers to design chips containing tens of billions of transistors without becoming hopelessly lost inside their own creations.

These companies do not reduce complexity. They make complexity intelligible. They turn systems so complicated that they would otherwise overwhelm human understanding into systems engineers can observe, measure, diagnose, and improve.

For decades, size created competitive advantage. Large companies enjoyed economies of scale. Bigger factories produced more efficiently. Larger organizations spread fixed costs across greater output.

Today’s greatest companies grow not because they simplify the world, but because their supporting ecosystems enable complexity.

TSMC can build the world’s most sophisticated semiconductor manufacturing system precisely because companies like Nova, Camtek, Synopsys, Cadence, Applied Materials (NASDAQ: AMAT), and KLA (NASDAQ: KLAC) continuously transform impossible complexity into something engineers can actually manage.

The same pattern appears across the modern economy.

Palantir Technologies (NASDAQ: PLTR) does not reduce the complexity of military operations or global supply chains. It allows decision-makers to operate successfully despite that complexity.

NVIDIA (NASDAQ: NVDA) does not reduce computational complexity. It makes vastly greater computational complexity economically useful.

Cloudflare (NYSE: NET) does not simplify the Internet. It helps make an increasingly complicated Internet secure and reliable enough to continue expanding.

These companies occupy what might be called the economy’s control points. They sit where information becomes action. They transform observation into decision. And as technology advances, those control points become progressively more valuable.

Every major technological advance creates new complexity. Artificial intelligence creates more data. More sensors create more information. Better software creates more interactions. Greater connectivity creates more dependencies.

Today, technology does not eliminate complexity; it drives it. Far from impeding growth, complexity has become one of the great engines of economic growth.

Every new layer of complexity creates demand for companies capable of making that complexity usable. The winners are the companies that make an increasingly complicated world intelligible.

The industrial economy rewarded those who mastered scale. The AI economy rewards those who make complexity work, although to be fair, this trend predates AI by decades.

The Israel Test is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Subscribe now

Eons ago—that is, in the 1990s—George Gilder and I visited a company called Akamai Technologies (NASDAQ: AKAM).

In the early days of the Internet, many people assumed the solution to growing traffic—and traffic jams—was simply bigger, faster servers. Akamai recognized the real problem wasn’t computing power; it was complexity. Rather than trying to centralize the Internet, Akamai distributed it, cached it, and made it manageable.

The Internet generated a tangle of complexity and produced Akamai. Storage threatened to make the world one gigantic basement of horrors. Storewidth, as George Gilder called it, was the ability to make enormous stores of data accessible with low latency, supported in those days by companies such as NetApp (NASDAQ: NTAP) and EMC (formerly NYSE: EMC; acquired by Dell in 2016).

Intelligibility became a huge source of value.

Akamai made a global Internet workable. Google—now Alphabet (NASDAQ: GOOGL)—made a global web searchable. Palantir makes massive organizations intelligible. Nova makes atomic-scale semiconductor manufacturing measurable. AI is beginning to make oceans of knowledge navigable.

Here is a guiding investment principle for our time. In the face of mounting complexity, wealth flows to the companies that make complexity legible.

Leave a comment

Simplify. Simplify. Simplify. It was the Prime Directive of the Industrial Era. It was the essence of the assembly line and carried over to business organizations of all sorts. Succeed by simplification.

Henry Ford simplified manufacturing. Walmart simplified retailing. Southwest simplified air travel.

No longer. In the AI economy, complexity is unavoidable. Those rewarded are companies that make overwhelming complexity manageable.

Modern semiconductor manufacturing is almost unimaginably complex. Microchips require thousands of manufacturing steps, atomic-scale precision, and tolerances measured in billionths of a meter.

Yet companies like Taiwan Semiconductor Manufacturing Company (NYSE: TSM) continue pushing the frontier of possibility year after year.

TSMC can do this not because semiconductor manufacturing has become simpler, but because a supporting ecosystem of companies has emerged to make that complexity comprehensible.

Share

Nova (NASDAQ: NVMI) measures processes that cannot be observed directly. Camtek (NASDAQ: CAMT) inspects structures too small for the human eye. Synopsys (NASDAQ: SNPS) and Cadence Design Systems (NASDAQ: CDNS) allow engineers to design chips containing tens of billions of transistors without becoming hopelessly lost inside their own creations.

These companies do not reduce complexity. They make complexity intelligible.

They turn systems so complicated that they would otherwise overwhelm human understanding into systems engineers can observe, measure, diagnose, and improve.

For decades, size created competitive advantage. Large companies enjoyed economies of scale. Bigger factories produced more efficiently. Larger organizations spread fixed costs across greater output.

Those advantages remain real. But today’s greatest companies grow not because they simplify the world, but because their supporting ecosystems enable complexity.

TSMC can build the world’s most sophisticated semiconductor manufacturing system precisely because companies like Nova, Camtek, Synopsys, Cadence, Applied Materials (NASDAQ: AMAT), and KLA (NASDAQ: KLAC) continuously transform impossible complexity into something engineers can actually manage.

The same pattern appears across the modern economy.

Palantir Technologies (NASDAQ: PLTR) does not reduce the complexity of military operations or global supply chains. It allows decision-makers to operate successfully despite that complexity.

NVIDIA (NASDAQ: NVDA) does not reduce computational complexity. It makes vastly greater computational complexity economically useful.

Cloudflare (NYSE: NET) does not simplify the Internet. It helps make an increasingly complicated Internet secure and reliable enough to continue expanding.

These companies occupy what might be called the economy’s control points. They sit where information becomes action. They transform observation into decision.

As technology advances, those control points become progressively more valuable.

Every major technological advance creates new complexity. Artificial intelligence creates more data. More sensors create more information. Better software creates more interactions. Greater connectivity creates more dependencies.

Today, technology does not eliminate complexity; it drives it. Far from impeding growth, complexity has become one of the great engines of economic growth.

Every new layer of complexity creates demand for companies capable of making that complexity usable. The winners are the companies that make an increasingly complicated world intelligible.

The industrial economy rewarded those who mastered scale. The AI economy rewards those who make complexity work, although to be fair, this trend predates AI by decades.

Eons ago—that is, in the 1990s—George Gilder and I visited a company called Akamai Technologies (NASDAQ: AKAM).

In the early days of the Internet, many people assumed the solution to growing traffic—and traffic jams—was simply bigger, faster servers. Akamai recognized the real problem wasn’t computing power; it was complexity. Rather than trying to centralize the Internet, Akamai distributed it, cached it, and made it manageable.

The Internet generated a tangle of complexity and produced Akamai. Storage threatened to make the world one gigantic basement of horrors. Storewidth, as George Gilder called it, was the ability to make enormous stores of data accessible with low latency, supported in those days by companies such as NetApp (NASDAQ: NTAP) and EMC (formerly NYSE: EMC; acquired by Dell in 2016).

Intelligibility became a huge source of value.

Akamai made a global Internet workable. Google—now Alphabet (NASDAQ: GOOGL)—made a global web searchable. Palantir makes massive organizations intelligible. Nova makes atomic-scale semiconductor manufacturing measurable. AI is beginning to make oceans of knowledge navigable.

Here is a guiding investment principle for our time. In the face of mounting complexity, wealth flows to the companies that make complexity legible.

Leave a comment

That’s a mistake. Organizations are technologies for coordinating human effort. Like every technology, organizations solve certain problems well, solve others poorly, and eventually get replaced when a better technology comes along.

The modern corporation did not simply evolve because people preferred bureaucracy. It evolved because, for more than a century, bureaucracy was the most effective technology available for organizing large numbers of people. Managers, departments, reporting chains, committees, and middle management were all engineered solutions, as was the firm itself.

Nearly 90 years ago, Nobel Prize-winning economist Ronald Coase asked, “why do firms exist at all?” His answer was that coordination is expensive.

Markets require buyers and sellers to find one another, negotiate contracts, enforce agreements, and share information. Firms exist because, beyond a certain point, it becomes cheaper to organize those activities under one roof than to conduct them continuously through the market.

Peter Drucker spent much of his career explaining how these organizations should be managed once they existed. Questions like span of control, reporting relationships, and decentralized decision-making became the central management problems of the 20th century.

But both Coase and Drucker were writing in a world where information moved at human speed, ascending and descending via a hierarchy of managers. That hierarchy was dictated by the rule that one manager could not handle more than half a dozen or so direct reports.

AI moves much faster than human speed, and doles out information not to the half-dozen people directed by a manager, but to anyone on a network.

Artificial intelligence summarizes meetings, routes information, prioritizes tasks, coordinates projects, finds expertise, explains anomalies, and answers routine questions.

In this way, AI is revealing that most of what managers did was not making decisions. It was moving information from one human being to another. Thus understood, the bureaucratic hierarchy was an obvious candidate for replacement by an automated system. AI is that replacement.

Share

Not all managers will disappear. But their ranks will thin as the question, “how many people can one manager supervise?” becomes “how much coordination can AI absorb before another layer of management is needed?”

When the technology enabling an organizational form changes, the enabled organizations change as well.

The railroad turned local markets into regional, then national markets.

The telephone enabled geographically dispersed companies. The internet enabled communication at an unprecedented volume and speed, as we have argued previously, radically reducing those very market costs that Coase said inspired the firm.

Artificial intelligence will be the next great organizational technology—not because it thinks better than people, but because it helps people think together. It will upend hierarchies that did little more than facilitate communication—and slow communication at that.

The first evidence is already appearing.

Small, AI-enabled teams routinely produce software that once required much larger engineering organizations. Developers using AI coding assistants iterate faster. Customer service systems resolve issues automatically that once required several layers of escalation. Project management increasingly happens through continuously updated software rather than periodic meetings.

None of these developments look revolutionary in isolation. Together, they are overthrowing the essential organizational principle of the last 150 years.

It turns out that the managerial hierarchy was not the endpoint of economic evolution. It was simply the best organizational technology available—until something better came along.

Leave a comment

Burnham argued that capitalism was gradually giving way to a new social order dominated not by owners but by managers. Modern industry had become so technologically and organizationally complex that neither entrepreneurs nor owners could run it directly. Professional managers increasingly exercised real power.

Factories, railroads, oil companies, automobile manufacturers, chemical companies, telecommunications: the great enterprises of the industrial age required armies of specialists. But the activities of those specialists had to be coordinated, supervised, measured, budgeted, scheduled, and integrated. Managers did that work. Industrial enterprises would have been incoherent without them.

The economic logic that made managers necessary also explained the very existence of firms.

Years before Burnham’s work, Nobel laureate Ronald Coase asked why do firms exist at all? Why doesn’t every worker simply contract independently in the marketplace?

His answer: because free markets are not free.

Finding suppliers costs money. Negotiating contracts costs money. Monitoring performance costs money. Communicating costs money. Every market transaction carries costs.

Firms exist because, beyond a certain point, replacing thousands of market transactions with managerial direction becomes cheaper. The firm substitutes organization for negotiation.

One might say firms come into existence the moment managers become necessary.

Share

Coase argued from economic theory. Peter Drucker reached a remarkably similar conclusion empirically. After spending two years studying General Motors from inside, he showed in Concept of the Corporation that what managers did was coordinate specialized knowledge across large organizations while allowing decentralized divisions to operate effectively.

But Coase’s economic assumptions weakened over time. For half a century technology has been reducing the transaction costs that once justified enormous organizations.

The Internet facilitated outsourcing functions that once had to live inside the corporation. It dramatically lowered the cost of finding information, suppliers, customers, and expertise. It allowed for cloud computing so that enterprises would not have to maintain vast data farms. Contract manufacturing, not entirely new but eased by the Internet, allowed entrepreneurs to rent factories instead of owning them.

The need to avoid market costs had motivated the large firm. The Internet shifted work back toward markets, flattening the enterprise.

Artificial intelligence goes the next step. The Internet reduced the need to own physical infrastructure. AI is beginning to automate the very coordination function that justified management in the first place.

Research, writing, software development, legal drafting, customer support, scheduling, project management, and marketing increasingly become services that can be summoned on demand rather than coordinated through multiple managerial layers.

Teams of ten will become teams of one or two with no need of intermediaries.

A manager will not need to walk down the hall to persuade customer support, or marketing to modify their message. If the message is not working the AI will flag the problem, propose a fix and swiftly test to see whether it is working.

The economics of organization are being reshaped.

Share

For most of the twentieth century, if an entrepreneur wanted to compete with General Motors, IBM, or AT&T he would have to scale up. Build a bigger organization. Hire more managers to coordinate more specialists.

Today the effective leader asks how he can keep making his organization smaller.

Instagram reached extraordinary scale with only a handful of employees before Facebook acquired it. WhatsApp served hundreds of millions of users with a staff that would scarcely have filled one floor of an office building.

AI is taking that trend further. AI-native companies have demonstrated astonishing leverage with teams that would once have been considered absurdly small for the markets they address. Companies with revenues in excess of $1 million per employee are no longer uncommon.

The minimum efficient size of the firm is falling because technology has resolved the problem managers were hired to solve.

Managers existed because coordination was expensive. As coordination becomes cheaper, the relative value of management declines.

Neither management not managers will disappear. Some large organizations will remain indispensable. But they will need fewer managers, fewer coordinators, and fewer organizational layers. The manager will no longer be the characterizing persona of society. People will forget what “middle management” did.

Artificial intelligence will do more than make workers more productive. It will redraw the boundary between markets and firms.

Coase taught us that firms exist because markets are costly. Artificial intelligence makes markets cheaper.

Drucker taught us that management could tame industrial giantism. AI will lighten the giants’ load.

Burnham described the organizational equilibrium created by twentieth-century technology that made so many managers necessary. Artificial intelligence is upending that equilibrium.

Managers became dominant because twentieth-century technology required them. As AI changes the economics of coordination, it changes the economics of management itself. The managerial age is ending for the same reason it began: technology has rewritten the rules of organization.

Leave a comment

Ayn Rand’s heroes and villains confronted an economy fraught with friction. Entrepreneurship required overcoming immense physical, financial, and organizational resistance. Her heroes struggled to overcome that resistance; her villains used it against them.

The entrepreneur needed not only vision but an industrial army; as John Galt peeled away that army the heroes’ struggles would become futile.

Rand could not—would not—write that book today. Artificial intelligence is a lubricant on a civilizational scale. Friction, the resistance to the entrepreneur proceeding from idea to execution, is being radically reduced.

The heroes in Atlas Shrugged are constantly fighting to acquire and control scarce physical assets—railroads, mines, mills, foundries, locomotives. Increasingly today the entrepreneur’s challenge is different. The scarce asset is less often physical capital than the ability to see an opportunity, exercise judgment, and assemble readily available capabilities into something new.

In the three quarters of a century since Atlas Shrugged was published technology has steadily lowered the capital cost of starting a new business. Contract manufacturing, cloud computing, open-source software, and the internet all made entrepreneurship progressively cheaper, long before AI.

AI, however, attacked another sort of friction. Rand’s entrepreneurs spend much of their lives overcoming organizational friction—negotiating, coordinating, persuading, waiting, routing information through bureaucracies. AI doesn’t make steel mills unnecessary. It makes many forms of coordination, communication, drafting, analysis, and execution vastly easier.

Contract manufacturing let entrepreneurs outsource physical production. Cloud computing let them outsource computing infrastructure. The Internet dissolved distance.

AI is different. It lets entrepreneurs outsource an increasing number of cognitive capabilities. AI is reducing the amount of human organization required to accomplish tasks.

Share

For most of the twentieth century, to build a significant company required programmers, marketers, lawyers, customer support, system analysts, HR and financial teams. Much of the entrepreneur’s job was to assemble and manage such specialists. Management is not innovation. Load down the entrepreneur with routine management tasks and you no longer have an entrepreneur.

Today AI “agents” can do most of that work and coordinate it more easily and faster than any human manager.

The entrepreneur keeps surrendering assets but retains one thing: judgment, which early in the 20^th century Frank Knight (among others) identified as the essential job of an entrepreneur. It is by exercising his judgment that entrepreneur creates “true profit,” that is to say, output exceeding measurable inputs: capital, labor, and management.

Innovation is always the product of a judgment made under uncertainty. It is a leap forward. It is not a leap the dark, because the entrepreneur’s first task is to reduce uncertainty as much as possible. Nevertheless, he makes the leap in the face of some residual gap in knowledge, rendering the result of the leap a surprise.

Subscribe now

Rereading last night, I came upon one of my favorite scenes. Just before Dagny’s freight train is about to start its progress over an untried bridge made of “Reardon Metal” a reporter asks her:

“Miss Taggart, what’s going to support a seven-thousand-ton train on a three-thousand-ton bridge?”

Dagny answers “my judgment.”

Maybe Atlas Shrugged is not so hopelessly outdated after all.

Leave a comment

Information Theory tells us that only surprise counts as information. If the receiver can infallibly predict the next signal from the sender, that predictable signal adds no information.

If I transmit “Mr.”, none of the letters in the group “i, s, t, e, r” count as information. We trust that those letters come next.

Trust is reduced uncertainty about future behavior. But trust is generated largely by past behavior.

When we trust a bridge, we expect it to continue standing. When we trust a bank, we expect it to honor withdrawals as it has done in the past. When we trust a person, we expect him or her to behave tomorrow roughly as they behaved yesterday.

We are trusting in the past. We are trusting in reputation.

Reputation is accumulated evidence from the past about future behavior.

Even if, as is usually the case, our trust is in an expert rather than our own experience, we trust because other persons or institutions have affirmed the expert’s reputation—his history of being right or wrong.

Now here’s a wrinkle.

Suppose we have strong reason to trust an expert within a particular field. Then one day we encounter him discoursing on dozens of subjects far outside his area of expertise. Every proposition sounds plausible. Better still, many of them are entirely new to us.

He is providing enormous amounts of information. Do we believe him?

If he offers a thousand plausible assertions, do we accept them?

This is the AI problem.

Share

Artificial intelligence (AI) floods the world with information. But because information is surprise, every new assertion is also a new source of uncertainty.

Every statement leaves us asking the same question: true or false?

One apparent response is to verify every claim. Unfortunately, that destroys much of the value of AI, which is to get answers quickly with little effort.

If an AI gives me 10 facts, I can check them.

If it gives me 100 facts, I can check some of them.

If it gives me 10,000 facts—fuhgeddaboudit!

So what is to be done?

Stop verifying individual facts and start evaluating the sources that produce them. We verify not facts, but factories.

We verify reputations. That may sound risky, but it is how human beings have always managed complexity.

No scientist verifies every result in physics. No investor reads every balance sheet. No physician reads every medical paper. No historian checks every archive.

Instead, we rely on very human networks of trust built from accumulated experience: Professional communities. Institutions that have earned confidence over time. Colleagues. Friends. Your parish church or synagogue. These days, even prediction markets.

Trust is reduced uncertainty about future behavior. Reputation is the mechanism by which that uncertainty is reduced. Reputation is a form of compressed history. Multiple observations become a single judgment that a source is reliable.

Artificial intelligence does not eliminate the need for trust. It increases it.

The more information AI creates, the more valuable reputation becomes.

The machines generate the information. Humans still decide whom to believe.

The more we rely on the machines, the more we need the humans.

Leave a comment

The world has been talking about the Information Economy since I was a teenager. Thanks to AI, it has finally arrived.

For most of industrial history, companies existed to make things or to move them. Steel companies made steel. Railroads moved freight. Manufacturers produced cars. Retailers sold goods.

Increasingly though, “heavy” industries make information, generated by massive sensor networks that see more than humans ever could.

A modern tractor is not merely a tractor. It is a rolling sensor platform measuring soil conditions, moisture levels, fertilizer application, fuel consumption, machine performance, and crop yields.

A modern automobile is not simply transportation. It is a continuous stream of information about roads, driver behavior, vehicle performance, traffic patterns, and mechanical health.

A semiconductor fabrication plant does not merely produce chips. It generates an ocean of information about every stage of production. Companies such as Israel’s Nova (NASDAQ: NVMI) increasingly win, not because they manufacture chips, but because they reveal what is happening inside manufacturing processes that would otherwise remain invisible.

The internet connected computers. The global sensor networks connect reality.

Sensors are becoming cheaper. Bandwidth is becoming cheaper. Storage is becoming cheaper. These alone would generate data fit to drown in. Artificial intelligence keeps us from drowning.

Military examples are often the easiest to understand. Israel’s battlefield advantage, like America’s, increasingly comes not from possessing more powerful weapons than its adversaries, but from seeing more of what is happening. Drones, sensors, intelligence fusion systems, missile-defense radars, and artificial intelligence all contribute to a single goal: making hidden things visible.

Share

The same principle applies in business.

Amazon’s warehouses are sensor networks. Cameras, scanners, robotic systems, inventory trackers, and software continuously monitor the location and movement of millions of items. The warehouse does not merely store goods. It produces visibility.

Modern logistics systems are sensor networks. Every package scanned by United Parcel Service (NYSE: UPS), every container tracked by Maersk, every truck reporting its position generates information that allows companies to reroute shipments, identify bottlenecks, and respond to disruptions before customers notice them.

Electrical grids are becoming sensor networks. Utilities increasingly deploy smart meters, line monitors, transformer sensors, and AI-driven control systems that can detect outages, equipment failures, and demand surges in real time, rather than waiting for customers to call and report problems.

Factories are becoming sensor networks. Thousands of sensors monitor temperatures, vibrations, pressures, and machine performance. Companies can often identify failing equipment days or weeks before a breakdown occurs, reducing downtime and increasing output.

Amazon’s warehouses do not merely store inventory, they make it visible.

Modern logistics networks do not merely transport packages. They make supply chains visible.

Smart electrical grids do not merely distribute power. They make the state of the network visible.

Modern factories do not merely manufacture products. They make production processes visible.

Yes, but what do we mean by visible? We mean not an event that happens in our eyes, but in our brain. We mean not only that we see, but that we see the point.

All this data from sensors would be pointless without understanding, for which AI is indispensable. AI’s most important contribution is not generating text, images, or software code. Its most important contribution is interpretation. Indeed, without AI to interpret the data, the sensors never would have been invented.

My friend and colleague John Schroeter has long been engaged by what some technologists call the “Spatial Web.” The idea is that sensors, networks, software, and digital identities, all connected to the net, increasingly make the physical world machine-readable.

That is happening. Yet the most important question is not whether the world becomes observable. It is how to make sense of what is observed. Visibility creates value only when it leads to understanding and action.

That is to say, when it leads to learning, which, as we have learned via the information theory of the economy, is the true source of all economic growth.

AI is our learning machine.

In “The Age of the Seeker,” we argued that modern technology is shifting power from hiders to seekers. In “The Systems That See Win,” we examined how visibility itself has become a source of power.

A recent report suggests the next step. Visibility can overcome constraints that once appeared to be imposed by physics itself.

Breaking Defense recently reported on an Israeli startup called Esh-Tech and its new DroneLite anti-drone laser system. As always with early-stage defense technologies, some caution is warranted. The claims below come from the company itself.

Still, the headline numbers are impressive. According to the company, DroneLite uses only four kilowatts of power, roughly what a household clothes dryer consumes. Traditional high-energy laser weapons often require tens or even hundreds of kilowatts, making power supply and cooling major operational constraints.

DroneLite can be mounted on a vehicle. It delivers comparable capability at only about one-quarter to one-third of the cost of existing laser systems. Unlike many continuous-wave lasers that must remain on target for 10 or 15 seconds to burn through a drone, DroneLite reportedly uses rapid pulses that punch multiple holes in the target.

But here is the most interesting part.

For decades, laser weapons have faced a stubborn limitation. The atmosphere gets in the way. Dust, humidity, turbulence, smoke, heat shimmer, and other effects scatter and distort laser beams, rendering laser weapons ineffective at crucial moments. This problem has long been viewed as one of the principal obstacles to practical laser weapons.

Share

The traditional engineering response would be to add power to overwhelm the constraint. But more power means more heat and weight. Soon you have a weapon that can’t be very mobile, much less airborne.

Esh-Tech is substituting vision for brawn.

It turns out that atmospheric conditions are variable across fractions of a second. DroneLite continuously analyzes these conditions and identifies tiny “windows” during which transmission conditions improve. The company claims these windows can be detected within hundredths of a second and that exploiting them can increase effective range by 50% to 100%.

The system sees something about its environment and uses that knowledge to act at precisely the right moment.

The constraint that appeared to be power, or nature, or the laws of physics was revealed to be information. More information turned out to be more valuable than more power.

For much of industrial history, progress meant overpowering obstacles. Imagine bigger engines, bigger factories, bigger power plants, and bigger computers.

Increasingly, however, the winning systems are those that see better and learn more. There is hardly a more complex natural phenomenon than the weather. DroneLite discovers opportunities hidden inside complexity.

This pathway—preferring learning to power—is not new. Learning is ultimately the source of all economic growth. In our era, but before AI, we have seen versions of this. In telecommunications, simply increasing transmission power often created additional noise and interference. Qualcomm’s innovation was to substitute processing and coding for brute force, dramatically increasing network capacity.

Artificial Intelligence, however, is making this pattern—learning trumps power—far more common and yes, powerful.

With AI, progress comes not from defeating constraints but from understanding them in previously unimaginable detail. Without AI, the weather is “cloudy.” With AI, the weather is “cloudy with intermittent microseconds of sunshine.”

Increasingly, the greatest value of a technology accrues not at the point where force is applied, but at the point where information is gathered and interpreted.

AI converts apparent physical constraints into informational constraints, then supplies the information that defeats them

Thanks to AI, the clouds are not disappearing, we are learning to see through them.

Here comes the Sun.

Leave a comment

A piece that defends itself against all error, that backstops every assertion with equivocation, is worse than tedious. It is empty.

Yet for much of the 20th century, perfection was an industrial ideal. Factories pursued zero defects. Engineers sought flawless execution. Managers measured success by the absence of mistakes. “Six Sigma” was the Holy Grail.

This was not foolish. When products are expensive, production runs are long, and mistakes are difficult to correct, perfection matters. Under those conditions, “we learned from our mistakes” was a confession of catastrophe.

Today, the most successful systems are not the ones that make the fewest mistakes, but the ones that discover mistakes fastest and improve most rapidly.

As we have noted elsewhere, SpaceX provides a vivid example. Rockets exploded. Landings failed. Hardware was lost. Critics mocked the company’s repeated setbacks. Yet every failure generated information. Every mistake became learning. Learning ultimately produced the world’s most successful launch company. Beats NASA’s massive commitment to ‘no mistakes’ because it means NASA launches 99 percent less often.

Share

AI only underlines the point.

One of the most important military technologies of recent years is Project Maven, the Pentagon’s AI-assisted targeting system. As the Economist’s review of Project Maven by Katrina Manson put it, Maven “has spotted Iranian missiles heading for Israel and rocket launchers in Yemen. It has detected migrants crossing America’s southern border and drug boats in the Caribbean. On one day in 2022, it found more than 260 potential targets for Ukraine.”

Predictably, early versions struggled. Algorithms often misidentified objects and performed inconsistently across different environments. “Early algorithms, tested during counter-terrorism operations in Somalia in 2017, were erratic. Algorithms would label clouds as flying school buses. The next year, in Afghanistan, the software identified trees as people and rocks as buildings.” It was confused by snow.

Today Maven is so good, it not only sees, it also turns vision into action. “Maven does not just sense such things: it can also coordinate the response. It fuses together different sorts of intelligence—photos, text, radio, and electromagnetic pulses—and works out which plane, carrying which munitions, is closest to which target. With a single click by a human, Maven can turn data into ash.”

Large organizations tend to hide errors. Bureaucracies punish mistakes. Employees learn to conceal problems. Peter Drucker argued that, for large corporations, free markets operated only at the transactional borders between them. Deep inland, the modern corporation more resembled the Soviet Union than anything in the imagination of Adam Smith. They generate the appearance of competence but impede learning.

AI, which we have called a seeker technology, is seminal in enabling learning organizations because it creates visibility. A hidden defect remains a defect. A visible defect becomes data.

In leading semiconductor manufacturing, the principle is clear. Cerebras (NASDAQ:CBRS) can mount four trillion circuits on a chip because it can “see” defects and route around them. It turns error into information.

Companies such as Nova (NASDAQ: NVMI) and Camtek (NASDAQ: CAMT) help chipmakers identify increasingly subtle manufacturing defects. They don’t make errors disappear. They make errors visible before they become catastrophic.

Visibility enables learning. Learning enables adaptation. Adaptation enables survival.

Leave a comment

Rather than cutting silicon wafers into hundreds of individual chips, Cerebras builds a processor from an entire wafer. The result is one of the largest and most ambitious computing architectures ever commercialized.

Yet the most important lesson of Cerebras is more about “visibility” than raw computing power. For most of the industrial era, progress came from eliminating defects. Factories improved quality control. Machine tools improved precision. Semiconductor manufacturers reduced contamination.

Every generation pushed toward greater perfection. Perfection even got a name: “Six Sigma,” a quality-control discipline aimed at near-zero defects.

But as systems become larger and more complex, perfection becomes increasingly unattainable. No semiconductor manufacturer produces perfect wafers.

Defects are unavoidable. Rather than requiring perfection, Cerebras designed its architecture to identify defective regions and route around them. The system works not because defects disappear, but because the system knows where they are.

A modern semiconductor wafer may contain trillions of transistors. At that scale, perfection is impossible. The challenge is not eliminating defects, but designing systems that can detect, isolate, and work around them.

A hyperscale data center contains hundreds of thousands of processors. Global communications networks connect billions of devices. Modern battlefields generate streams of information that no human staff can fully absorb.

In none of these systems is eliminating every error plausible. Instead, success comes from identifying errors quickly enough to respond.

Share

This has been going on for a long time. The Internet was designed to route around failures. But AI intensifies the shift away from perfection and toward visibility and compensation.

Value increasingly migrates toward systems that can “see” or render systems intelligible.

Modern AI systems identify patterns hidden within oceans of data. They see the pattern and propose action. Advanced missile defense systems discriminate between threats that matter and those that do not.

One result of this shift is that many of today’s successful technology companies occupy what might be called “control points.”

A control point is a place where visibility can be translated into action.

Nova Ltd. (NASDAQ: NVMI) does not manufacture semiconductors. It measures semiconductor processes. Its tools allow manufacturers to see what is happening inside increasingly complex fabrication steps.

Camtek Ltd. (NASDAQ: CAMT) does not package advanced chips. It inspects them. Its systems identify defects that would otherwise remain invisible until costly failures occur.

Palantir Technologies (NASDAQ: PLTR) does not manufacture weapons. It integrates information. Its value comes from helping organizations see relationships hidden across multiple data sources.

In each case, the economic value comes less from ownership than observation. The company that sees gains leverage over the company that merely produces.

Share

This principle is tending toward a general rule for our time.

For the military, visibility is increasingly more important than raw scale. Armies that can identify targets, integrate intelligence, and respond rapidly may outperform larger forces with less awareness. That’s largely the story of Ukraine vs. Russia.

Israel’s military success reflects the same trend. That country’s advantage is not simply superior firepower, but superior visibility. The ability to identify threats, discriminate among them, and act quickly transforms information into operational advantage.

The same pattern appears in business. As complexity rises, learning becomes more valuable than scale. For a large organization that cannot render reality intelligible, its scale is a bug, not a feature.

A large network that cannot identify failures becomes fragile. A large model that cannot distinguish signal from noise proliferates error. It “hallucinates.”

For two centuries, industrial civilization rewarded those who could build larger systems. The AI era will reward those who can see what those systems are doing.

The future belongs not to the biggest systems, but the systems that learn. And learning begins with seeing.

Subscribe now

Leave a comment

How many workers were impoverished? How many families were ruined?

Seems like a silly question. But “reduction in work hours” can always be restated as job loss or vice versa. In that sense, “jobs” were being lost. But happiness was being gained.

The century after 1870 witnessed one of the greatest productivity explosions in human history. Americans became vastly more productive while working fewer hours. Over those 150 years, output per hour worked increased roughly 15-fold while annual working hours fell by about 40%. If productivity destroys jobs, that should have been an economic catastrophe. Instead, it became the richest period of broad-based prosperity the world had ever seen.

The reduction in working hours was one of the great achievements of modern civilization. Americans produced vastly more wealth while spending far less of their lives at work.

Workers became richer. Living standards rose. Leisure expanded. Entirely new industries emerged. Most Americans today enjoy comforts that would have seemed an impossible dream to our grandparents.

If productivity destroyed employment, most of us should have been unemployed decades ago. Instead, we work less and have more.

Share

So why, when artificial intelligence enters the discussion, suddenly assume that if AI increases productivity, then workers will be impoverished?

The mistake is assuming that society has a fixed amount of work to do. Economists sometimes call this the “lump of labor” fallacy. The idea is that there are only so many tasks to perform and that technology merely divides those tasks among fewer workers.

But human wants are not fixed. As productivity rises, possibilities multiply and people demand more—and better: housing, health care, entertainment, education, travel, food, and these days, better “experiences.”

The great French economist Jean-Baptiste Say understood this more than two centuries ago. Say’s Law can be stated as ”supply creates its own demand.” Or alternatively: ”production creates purchasing power.”

We can demand only because we first supply; consume only because we first produce.

Wealth increases with productivity; as wealth increases, new forms of demand emerge. Or, since demand is potentially infinite, one could say new demands become relevant.

A frequently cited study of AI’s economic impact comes from Stanford economist Erik Brynjolfsson and his colleagues. They found that customer-service representatives using AI increased productivity by as much as 34%. The largest gains often came from less experienced workers.

Should that 34% productivity increase translate into layoffs?

That would assume demand for customer service is fixed. Is yours? Mine sure isn’t. Customer service is horribly scarce. We spend countless hours on hold, repeating information to multiple representatives, navigating mis-automated menus, and struggling to get answers from companies who claim they put customers first.

If AI allows representatives to solve problems faster and more effectively, why assume the result must be fewer representatives? Why not better service? Why not longer service hours? Why not proactive support? Why not entirely new categories of customer engagement that were previously uneconomic?

Why not AI “accent translators” that make residents of the sub-continent intelligible to Americans?

Share

Expanding productivity expands possibilities.

Economists get excited when productivity growth rises by fractions of a percent. After averaging roughly 1% to 1.5% annually for much of the past two decades, U.S. productivity growth recently accelerated to 2.7%. That’s enough to make economists talk about a possible productivity boom, readily translated into better products, lower prices, shorter workweeks, and entirely new industries.

Transitions can cause pain. Recent research suggests that some entry-level workers in highly AI-exposed occupations may be less necessary. Software development “tasks,” and other instances of what we once thought of as knowledge work, will be revealed as rote and be automated. But wherever human judgment is relevant, workers will be more productive and better paid.

That’s one reason that despite years of warnings about an imminent employment apocalypse, economists continue to find limited evidence of broad labor-market disruption. Productivity gains are appearing, but large-scale unemployment has not.

Programmers’ tasks are being automated, but programming employment is increasing.

This is not a surprise. Going back as far as we have data, productivity gains have repeatedly been converted into some combination of higher wages, lower prices, new industries, better products, and more leisure.

Which brings us back to the original question. How many jobs were destroyed when the American workweek fell from 60 hours to 40?

It was a dumb question. It misunderstands what productivity does. The purpose of technology is not to create jobs. It is to create abundance. And the most meaningful of all abundances is the one that addresses the most tragic of all scarcities:

Time itself.

AI makes time.

Leave a comment

I barely mentioned artificial intelligence—because I had mischaracterized the change. What was happening was not so much the triumph of defenders over attackers, but the triumph of seekers over hiders.

Artificial intelligence is the technology of the seeker. AI turns the industrial era scarcity of information into an abundance by radically lowering the cost of seeking: patterns, correlations, anomalies, inefficiencies, vulnerabilities, opportunities, threats, and targets.

That’s brutal for the hiders.

The dominant military systems of the twentieth century reflected the logic of industrial civilization. Mass armies. Carrier battle groups. Tank formations. Strategic bombers. Hierarchical command structures. Long supply chains. Giant centralized intelligence bureaucracies.

All these were driven by the high cost of coordination and communication. Industrial warfare depended on the scarcity not only of intelligence but also of visibility. Information traveled imperfectly and often arrived too late. When reconnaissance is difficult, large systems can hide. If information moves slowly, opportunities for dynamic response become rare.

If, in December 1941, the Japanese fleet had been visible, Pearl Harbor would have been the first and last day of a war that we won! The entire Imperial fleet would reside on the Pacific floor.

AI is the great leveler because it is the great revealer. It compresses the time and effort between sensing, understanding, decision, and action.

Share

That compression is the principal reason Ukraine has fought Russia to a standoff. Ukraine’s adaptation of tactics and weapons is paced more like software development than industrial retooling. Small teams improvise new tactics, modify weapons to suit, and distribute both rapidly at a speed to render bureaucrats speechless.

The media commentary on Ukraine focused on drones. The real issue was visibility. Drones, satellites, sensors, communications intercepts, thermal imaging, and networked targeting systems make battlefield concealment extraordinarily difficult. The battlefield increasingly belongs to the seekers.

That is an enormous change. Throughout military history, offensive warfare depended heavily on maneuver and surprise. Alexander, Caesar, Napoleon, Patton, and nearly every successful offensive commander in history relied fundamentally on concentrated force appearing unexpectedly where opponents were weak.

Speed mattered primarily because speed enabled surprise. But speed without concealment enables only the slaughter of the attacker.

Russia has scale. Ukraine has speed of learning. Scale erodes. Learning accelerates.

So, who is ahead in the new arms race, the race to deploy cognition?

The Israel Defense Force may be the world’s first AI native military.

Over the past several years, Israel has woven AI directly into operational military systems, fusing diverse pieces of intelligence and fragmentary data, to enable targeting for drone operations and missile defense, and support real-time battlefield adaptation.

Every major military now claims some form of AI integration. But the integration of AI depends on the integration of the deploying organization.

Israel’s military structure was already ripe for the AI era. Israel’s strategic circumstances have always rewarded visibility, speed, and adaptation. A small country surrounded by larger enemies cannot rely on mass, cannot absorb losses the way continental powers can, cannot fight long wars of attrition. It must identify threats rapidly, compress decision cycles, and act before adversaries fully organize.

The shifting strategic reality favoring seekers over concealers increasingly shows in the broader direction of technological civilization.

In the industrial era, concentration conferred safety. Large factories were efficient. Large fleets projected power. Large bureaucracies coordinated complexity. Large corporations dominated markets.

But size was a dependent variable. Information was the controlling factor.

Industrial civilization depended upon the ability of large organizations to manage information asymmetries: corporations controlled distribution; governments controlled intelligence; universities controlled expertise or at least the impression thereof; bureaucracies controlled coordination; media institutions controlled visibility.

AI, the equalizer, undermines all those asymmetries.

The industrial era concentrated intelligence because intelligence was expensive. AI distributes intelligence as cognition itself becomes abundant. And that will change the world. The future belongs not to the machines at the center, but to the humans at the edge.

Leave a comment

Industrial civilization rewarded scale. Bigger countries, bigger factories, bigger corporations, bigger bureaucracies, and larger standing military systems enjoyed overwhelming advantages. Organizing the large numbers of people and machines those systems needed required hierarchy, management, and centralized control.

Both Ukraine and Israel are highly adaptive local systems outperforming larger but slower foes.

Ukraine’s battlefield evolution looks more like software development than industrial era armament. Drone designs evolve continuously. Small teams improvise new tactics, distribute them rapidly, and feed battlefield lessons back into production almost immediately. Cheap commercial components are substituted dynamically as conditions change.

The important point is not that Ukraine uses drones. Everyone uses drones. Ukraine’s advantage comes from the acceleration of the learning cycle. Smaller distributed teams iterate faster than larger bureaucratic structures can respond.

Trump did Ukraine a huge favor by ending U.S. military aid, freeing Ukraine from the inevitable restrictions in its use. It turns out that Ukraine is better off making its own weapons and choosing its own targets.

Israel’s technology and defense ecosystem operates through unusually dense local feedback loops connecting military units, startups, universities, engineers, and entrepreneurs. Small teams receive considerable autonomy. Communication paths are short. Constraint-driven improvisation—doing more with less—is a deep part of the culture.

Israel’s defensive prowess long predates AI, and Ukraine adapted to unique local circumstances. So, what does any of this have to do with AI?

Share

AI will make what seems surprising and exceptional in Ukraine and Israel into a global, historical trend. The critical advantage wielded by both those nations, tiny in comparison to their foes, was adaptability. AI will extend that advantage to the rest of the world.

Artificial intelligence increasingly lowers the coordination costs that once favored giant centralized systems. Small groups—and small nations—can now perform functions that once required large managerial structures: software development, research, design, logistics.

The onset of “Additive Manufacturing,” (which used to be called 3D printing), further reduces the scale and capital requirements for industrial production. Indeed, “additive” and “iterative” go hand in hand. As we have previously argued, “as non-recurring engineering friction declines, small-batch production becomes economically viable. Early-stage systems can evolve without committing to massive tooling investment… Additive manufacturing, meanwhile, will increasingly make iterative design the favored option.”

This evolution is not only about size, but about locality. Neither Israel nor Ukraine, fortunately for them, had to fight a global war. Yet Ukraine has fought to a draw (at least) an enemy with global reach, and Israel humiliated what appeared to be a “supreme” regional power. On the whole, the compact size of both countries has been more advantageous than not, supporting rapid, localized learning.

AI will enable a world organized less around giant uniform systems and more around highly capable local ecosystems: regional manufacturing clusters, distributed innovation networks, specialized supply chains, and dense high-trust technical communities.

I have written elsewhere on the increasing difficulty, even futility, of “force projection” far beyond any nation’s borders. Several years ago, these difficulties persuaded me that the United States would lose a war waged against China for the defense of Taiwan, with our vaunted carrier fleet sent to the bottom of the sea. A few years hence, and what will become more obvious is the futility of China trying to take a Taiwan protected by AI enabled defenses.

Leave a comment