Run Towards Chaos: Finding Career Safety in the Time of AI

Here's a riddle. If you're a company looking to cut costs with AI, which workers do you replace first? The expensive ones, obviously. A senior analyst earning $180,000 costs three times as much as an entry-level hire. Replace the senior, save three times the money. Right?

Researchers at Stanford recently analyzed payroll data covering 25 million American workers to find out.¹ They expected AI would hit experienced professionals hardest, but the data showed the opposite. Entry-level workers in AI-exposed fields are losing jobs. Older workers in the same occupations are gaining them.

This surprised the researchers, but it wouldn't have surprised a Victorian economist named William Stanley Jevons, who noticed something similar 160 years ago. His insight helps explain not just who's at risk, but also gives us clues about where to find shelter.

When Less Becomes More

In 1865, a twenty-nine-year-old statistician sat down to write a book about coal. It was a calculated move. His previous work on logic, he admitted to his brother, had "made no noise."² He needed a subject that would get attention.

Britain was then the world's industrial engine, burning coal to power everything from trains to textile mills. Then James Watt invented his version of the steam engine, and suddenly the world could generate a lot more power with a lot less coal. Logically, you would expect Britain to use less coal.

Instead, Britain started using more coal than ever. Between 1760 and 1860, consumption skyrocketed from around 10 million tons per year to over 100 million.³

Jevons figured out why. When a technology makes something cheaper, two things happen. First, people save money. That's obvious. Second, and less obviously, applications that were previously uneconomical suddenly make sense. "It is wholly a confusion of ideas," Jevons wrote, "to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth."⁴

This is now called the Jevons Paradox. Make something more efficient and you don't get less of it. You get more.

The Displacement

There's comfort in the idea that efficiency leads to abundance. There's also a catch. The people who thrive before an efficiency revolution and those who thrive after it are usually not the same people.

In 1996, Microsoft launched Expedia. Within a decade, you could comparison-shop flights, book hotels and read reviews from your couch. At the same time, commercial jets were getting cheaper to fly; fuel efficiency roughly doubled between 1960 and 2008.⁵ Two efficiency gains compounding.

The result was pure Jevons. International tourist arrivals grew from 500 million in 1995 to 1.5 billion by 2019.⁶ We didn't travel less. We traveled vastly more.

Meanwhile, travel agents were disappearing. Employment peaked at 124,000 in 2000. By 2021, it had fallen 70%.⁷

Think about what a travel agent actually did in 1995. They searched for fares, compared prices, issued tickets and processed bookings. Everyone offering that service was essentially interchangeable. When everyone is interchangeable, efficiency gains don't create opportunity. They create extinction.

Knowledge Work's Watt Moment

Which brings us back to that Stanford finding.

The researchers think they know what's happening. Large language models are trained on text: books, articles, documentation, code.⁸ They've absorbed the kind of explicit knowledge that young people acquire in universities and early careers. What the models lack is tacit knowledge, the unwritten rules and judgment calls that experienced workers accumulate through practice, knowledge that never makes it into any manual.

AI is doing to entry-level knowledge work what Expedia did to fare searching. It's commoditizing the commodity parts. The World Economic Forum projects 92 million jobs displaced by 2030, with 170 million new ones emerging.⁹ If history is any guide, we'll end up doing vastly more knowledge work than before, in forms we can barely imagine.

The question is who will make the journey, and who will get left behind.

Where Value Actually Lives

Here's where it gets interesting.

For the past several years, our data science research has focused on finding hidden patterns in how markets behave, particularly during periods of disruption. We've analyzed value chains across dozens of industries, tracking where profits pool and where they dissipate. The patterns are remarkably consistent.

A value chain is simply the sequence of activities required to deliver something useful: raw materials become components become products become services become outcomes. Michael Porter mapped this decades ago.¹⁰ Clayton Christensen spent his career studying how value migrates along these chains when technologies shift. What we've found is that the same framework that explains corporate strategy applies almost identically to careers.

Value isn't distributed evenly across a chain. It pools in some segments and drains from others. More importantly, it moves. The segment that captured value in 2015 may be hemorrhaging it by 2025. The trick is seeing where it's headed.

We look at three basic states. In the first, supply exceeds demand. Call it excess capacity. In the second, supply roughly equals demand. Call it equilibrium. These two states share a crucial feature: they're where value dissipates.

In areas of excess capacity or equilibrium, commoditization increases. Margins compress. Competition intensifies. For a business, these are segments to exit. For a career, these are capabilities to divest.

Here's the counterintuitive part. These states look orderly. They're where things work. The technology functions. The processes hum. Everyone knows what to do. That apparent smoothness is precisely the warning sign. When something works well enough that anyone can do it, anyone will.

The third state is different. Demand exceeds supply. Call it a bottleneck. This is where value pools, where profits concentrate, where careers become defensible. Bottlenecks don't look orderly. They look frustrating. People have problems they can't solve. Systems break down. Everyone complains.

Those complaints are a map.

Reading the Map

Apply this to knowledge work in the age of AI.

If AI handles some part of your job flawlessly, that's not safety. That's a signal you're in equilibrium or excess capacity. The smooth-running parts of any system are the parts sliding toward commoditization.

If AI fails in your domain, and the consequences matter, that's a bottleneck. The ability to catch hallucinations, to build verification processes, to know which outputs to trust, these capabilities are scarce. If AI produces too much output and decision-makers drown in plausible analysis, the ability to find the three sentences that matter is scarce. The bottleneck isn't production. It's comprehension.

The data confirms this. According to Precisely and Drexel University, only 12% of organizations report having data of sufficient quality and accessibility for effective AI implementation.¹¹ Companies are buying tools they can't use because their data is a mess. That mess is a bottleneck.

Nash Squared surveyed over 2,000 technology leaders and found that AI skills have become the world's largest tech shortage in 15 years, jumping from sixth place to first in just 18 months.¹² The technology exists. The people who can implement it responsibly are desperately scarce.

These specific bottlenecks won't last forever. Data governance may be routine in five years, and value will have migrated elsewhere. The methodology doesn't change. Find where the system is stuck. Position yourself there. When the bottleneck moves, follow it.

Value migrates. Sometimes it stays in one place for months, sometimes for decades. It's case by case. The point isn't to predict exactly where it goes. The point is to recognize the states: Which parts of your work are in equilibrium, vulnerable to commoditization? Which parts are bottlenecks, where demand outstrips supply?

Run from order. Race towards chaos.

What Jevons Got Wrong

Jevons wrote his book because he was terrified. He thought Britain would exhaust its coal and lose its industrial supremacy. John Stuart Mill cited it in Parliament. Gladstone used it to push through legislation.¹³ A twenty-nine-year-old who couldn't get anyone to read his logic papers had suddenly caught the ear of the people running the British Empire.

He was right that efficiency would accelerate consumption. He was wrong about what that meant.

Britain didn't run out of coal. The economy evolved. New energy sources emerged. Coal became a smaller part of a much larger whole.

AI won't end knowledge work any more than the steam engine ended physical work. If Jevons is right, there will be vastly more of it on the other side of this transition. The question isn't whether the work will exist. It's whether you'll be positioned to capture it.

The bottlenecks form the map. Order is danger. Disorder is opportunity. Safety hides in chaos.

Endnotes

1. Erik Brynjolfsson et al., research conducted at Stanford's Digital Economy Lab using ADP payroll data covering 25 million workers. Results reported in "New Study Sheds Light on What Kinds of Workers Are Losing Jobs to AI," CBS News, 2025.

2. Letter from W.S. Jevons to Herbert Jevons, February 1864, in R.D.C. Black and R. Könekamp, eds., Papers and Correspondence of William Stanley Jevons, vol. 1 (London: Macmillan, 1972).

3. William Stanley Jevons, The Coal Question (London: Macmillan, 1865). Coal data from B. Alcott, "Jevons' Paradox," Ecological Economics 54, no. 1 (2005): 9-21.

4. Jevons, The Coal Question, Chapter VII.

5. International Council on Clean Transportation, "Fuel Efficiency Trends for New Commercial Jet Aircraft: 1960 to 2014" (2015).

6. World Tourism Organization (UNWTO), Tourism Highlights (various years).

7. U.S. Bureau of Labor Statistics, Occupational Employment Statistics.

8. Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock, "GPTs Are GPTs: Labor Market Impact Potential of LLMs," Science, June 2024.

9. World Economic Forum, Future of Jobs Report 2025.

10. Michael E. Porter, Competitive Advantage: Creating and Sustaining Superior Performance (New York: Free Press, 1985).

11. Precisely and Drexel University LeBow College of Business, "2025 Outlook: Data Integrity Trends and Insights," 2024.

12. Nash Squared/Harvey Nash, Digital Leadership Report 2025.

13. Gladstone's letter to Macmillan, 1866, in Black and Könekamp, eds., Papers and Correspondence of William Stanley Jevons, vol. 1, 203.