Crosscurrents | Machines That Don’t Pay Attention, The Refinance Economy & Unintended Consequences

Sanjeev Krishnan | October 20, 2026

My Agent Doesn’t Watch the Super Bowl Ads — 100 Years of Advertising Stability and the New Economic Inversion 

In 1919, advertising accounted for ~2.5% of GDP. In 2007, it was around 2%. Today it sits between 1.4 – 2%. Despite a century that brought radio, network television, cable, the web, mobile, social media — two world wars, a depression, three asset bubbles — the number has barely moved.

Almost nothing in the economy has behaved like this. Agriculture fell from 15% of output in the early 1900s to less than 1% in 2023. Manufacturing declined from 25% in the mid-20th century to 9.5% in 2025. Financial services roughly tripled between 1950 and 2006, and healthcare spending rose from 5% of GDP in 1960 to 18% in 2024. Advertising has stayed constant. 

The medium and plumbing of advertising have obviously changed. In 1919, newspapers and magazines took 60 cents of every dollar. In 2007, they took 21 cents while broadcast and internet took 40 cents. In 2025, it is estimated that internet channels accounted for nearly 70% of ad dollars. In 2025, Google, Meta, and Amazon captured nearly 60% of ad spending and digital ads accounted for ~75% of all global ad spending.

This 2% of GDP was the patron of popular culture and broader media, from the penny press in the 1830s to the newsrooms of the 1920s, the golden age of television, and even Super Bowl ads. Prior to this, the popular culture commons were financed by the church, the state, or individual patrons. During the internet and mobile age (which we are obviously still in), the collectors of this 2% of GDP became some of the most profitable companies in the history of capital markets (i.e., Google, Meta, et al). Rather than finance more content and newsrooms, they financed engineers and data scientists. We got free internet and they amassed some of the greatest EBITDA-to-free-cash-flow machines in history.

However, AI and agentic consumption may once again change the structure of consumption. For 100 years, the structure of how we consume stayed relatively constant — a firm bought advertising, the consumer got content and services for ​“free,” and the consumer’s attention was the thing that was sold. The consumer was both the target and the product. AI and agentic consumption could herald a new era. Currently, large LLMs earn most of their consumer revenue from paid subscriptions rather than ads. I suspect this will change but it is notable that we haven’t started with an ad-supported consumer AI value proposition. Agentic consumption also threatens to compress choice, lower or eliminate brand premiums, and reward consumption based on preset criteria. This means less emotion and more metrics, with consumption becoming more agnostic to the attention economy of the last 15 – 20 years.

The economic irony is potentially profound — the companies that monetized human attention (and still do) are spending that cash flow on machines that don’t pay attention. For 100 years, every new medium made money by reaching more people. Radio reached more than print. Television reached more than radio. The internet reached nearly everyone. Advertising followed the audience each time. AI is different. It makes money by replacing the audience. An agent buys the truck, and no one needs to watch the ad. The cash flow that was produced by the 30-second Super Bowl ad doesn’t need anyone to watch it to consume the product. 

This has implications for the 2% and for many of our systems, including the food system anchored around advertising. The phone used to eat first and maybe the agent doesn’t need to eat. 

The Refinance Economy — Don’t Get Repressed, Get Real 

We have built an economy based on refinancings. One estimate that has surfaced consistently in conversations across the market is that roughly 80 – 85% of gross credit creation is the U.S. finances existing assets or rolls existing debt — only 10 – 20% funds net new real investment. By gross flows, refinancing has outweighed new asset finance in the U.S. for 30 years. From 1982 to 2021, the refinance economy ran on falling rates, facilitated by lower coupon payments, extracted equity, and higher asset prices. Refinance has been the wealth creation engine.

This is a modern phenomenon. Prior to 1980, America was a nation of builders. Historically, a much larger share of credit funded physical capital. Beginning in 1826, America built canals, railroads, interstates, industrial facilities, defense-industrial complexes, and housing. The 200-year arc of this metric is well documented:

  • 1826 – 1865 — Land and Canals (state bonds, bank notes, and frontier land mortgages). By 1840, the United States had created 3,326 miles of canals, converting public and private borrowing into new transportation capacity.
     
  • 1865 – 1914 — Railroads and industry (public bond markets developed to fund railroads and factories). U.S. railroad mileage increased more than sevenfold between 1865 and 1914 as securities markets financed the construction of a national industrial network.
     
  • 1914 – 1945 — War finance and housing mortgages (Liberty bonds, real estate, et al). Four wartime Liberty Loan drives raised more than $17 billion from 20 million individual buyers, and mortgages increased from 34% of U.S. bank lending in 1914 to 51% in 1945.
     
  • 1945 – 1980 — Suburbs, interstates, industry (banks, public bonds, et al). Occupied U.S. housing units rose from 34.9 million in 1940 to 80.4 million in 1980, while the Interstate system expanded to 41,120 miles open to traffic. But the balance-sheet shift continued as well, with mortgages rising from 51% of U.S. bank lending in 1945 to 59% in 1980. 

The biggest refinancing of all, when measured by gross rollover volume, is the U.S. Treasury (i.e., rolling the U.S. debt and funding deficits). In 2024, gross Treasury issuance reached $29.3 trillion, but privately held net marketable borrowing was only $2.4 trillion. The $27 trillion difference principally reflects securities that matured and were replaced rather than net new borrowing. 

Aggregate annual T‑bill issuance represents approximately 80% of this gross volume, which reflects an outstanding bill stock that turns over 3 – 4x a year. The U.S. Treasury has shifted from borrowing mostly long-dated bonds to borrowing short-term T‑bills. Money market funds, which held $7.24 trillion in financial assets at the end of 2024, and the overnight lending system must absorb much of this continuing supply. 

This shift has important consequences. Issuing bills instead of bonds starves the markets of duration and holds the 10-year yield — the rate that prices mortgages, auto loans, and most of the refinance economy — below where it would otherwise clear. The sovereign has become the largest participant in the refinance economy and has every incentive to keep rates repressed, below what the market would likely want. 

Can this last? It can if inflation is stable and the Federal Reserve can absorb it. The politically easy path is to monetize the debt away. If inflation remains high and our ​“refinance economy” can’t handle 14% rates (like the Volcker era), the politically easy path will be to keep rates below inflation and monetize the debt. This is the definition of financial repression. 

Assuming this repressive era, assets that rise with prices will become scarce commodities, especially real assets that do not need the refinance economy to extract value. Beware of the rest.

Predictable Patterns — ​“We’ll See” — An Anthology of Unintended Consequences 

I hear ​“the unintended consequence of X” so often that I have assumed this is a feature (not a bug) of human endeavors. Shakespeare said it better in Much Ado About Nothing: ​“O, what men may dare do! What men may do! What men daily do, not knowing what they do!”

Our siloed systems don’t help model the cocktail of variables that produce order and disorder. So I became curious about the greatest hits of unintended consequences in markets and innovation. See below: 

  • Watt’s efficient steam engine (1770s) — meant to save coal but made coal so cheap (and useful) that Britain ended up burning more coal.
  • Cotton gin (1793) — meant to save on labor but made cotton so profitable that it led to the expansion of slavery. 
  • Air conditioning (1902) — built for factories and printing presses, but rewired migration away from the industrial belt to the Sun Belt.
  • Haber-Bosch (1909) — meant to break the world’s dependence on Chile’s finite nitrogen supply but led to an addiction of nitrogen fertilizer to keep up with food production. 
  • Thomas Midgley’s work on leaded gasoline and Freon (1920s) — both meant to make engines more efficient and refrigeration safer, but ultimately contributed to widespread lead exposure and depletion of the ozone layer.

These unintended consequences follow some predictability: the benefits are often concentrated, immediate, and owned, while the harms are diffuse, delayed, and unowned. Once these ​“benefits” build balance sheets and produce jobs, they naturally acquire defenders and it can take decades or longer to unwind them. 

Unwinding these consequences also follows some patterns: 

  • A substitute was ready: Kerosene largely replaced whale oil, DuPont had alternatives for ozone-depleting CFCs, and the car replaced urban horses. 
  • Measurement made the invisible more visible: Health evidence and blood-lead data helped drive the phase-down of leaded fuel.
  • Liabilities over regulatory timelines: Asbestos-related bankruptcies preceded the EPA’s 2024 ban on ongoing uses of chrysotile asbestos, while tobacco litigation and regulation helped cut adult smoking from about 42% to 12%.
  • Catastrophes and crashes: The Civil War and the 13th Amendment ended slavery; thalidomide birth defects helped spur the 1962 reforms that strengthened modern drug approval.

These patterns of unwinding probably cause another cascade of consequences. A less Shakespearean avatar of this meta quagmire is Gust Avrakotos’s character in Charlie Wilson’s War:

A boy gets a horse for his birthday, and the village calls it wonderful; the master says ​“we’ll see.” The boy falls off and breaks his leg, and the village calls it terrible; ​“we’ll see.” Then war comes, and the boy is spared from fighting because of his injury; ​“we’ll see.”

The lesson is not that consequences are unknowable. It is that the events are unknowable while the patterns repeat. What are the unintended consequences of AI or multipolar geopolitics or an unknown price of money or climate warming or even aging — ​“We’ll see.”

—

The views and opinions expressed are those of the S2G team members identified and do not necessarily reflect the views of S2G. This content is for informational purposes only, should not be taken as legal, business, tax or investment advice, or be used to evaluate any investment or security, and is not directed at any investor or potential investor in any investment vehicle sponsored by S2G. Investing involves risk, including the risk of loss. Specific companies mentioned in this podcast are for educational purposes and should not be construed as an endorsement of any kind. Please note that S2G may maintain investments in some of the companies discussed on this podcast. Any past performance discussed is not indicative of future results. For more important information, please see s2gin​vest​ments​.com/​d​i​s​c​l​o​sures.