In 1791, Tench Coxe drafted the Report on Manufactures, the document that argued a farm republic should bet on the machine economy, and supplied the statistics underneath Hamilton's signature. The machine revolution he predicted arrived. It made the country rich. And it did not save a single one of his over-borrowed contemporaries: Robert Morris, James Wilson, and William Duer were all ruined inside the very transformation Coxe called correctly, because being right about the technology has never exempted anyone from the physics of credit.
That's the finding the machine-question section of the crosswalk left hanging, and this article picks it up, because we are standing exactly where Coxe's readers stood. Stanford's AI Index reports that 78 percent of surveyed organizations used AI in 2024, up from 55 percent the year before; U.S. private AI investment hit $109.1 billion; and the cost of a fixed level of model capability fell more than 280-fold in under two years. The transformation is real, the way textile machinery was real. Which means the useful question isn't whether AI changes the economy. It's how you position inside a genuine revolution without becoming one of its Morrises, and the seven ledger laws distilled from fifteen founding fortunes answer it law by law.
Law 1: Liquidity before ambition, or, the 280-fold argument for patience
Bingham banked his fortune before he bought his empire; Morris bought the empire on credit and died in debtors' prison. The AI edition of this law comes with a bonus the founders never had: the asset you're tempted to over-reach for is getting cheaper at a historic rate. When inference costs fall 280-fold in two years and hardware costs decline about 30 percent annually, per the AI Index, every month you spend building cash instead of borrowing capability is a month the capability gets cheaper anyway. In 1791 the machines crossed the ocean slowly and dearly; in 2026 they race toward free. Debt-funding a rapidly deflating input is paying interest to skip a line that's shortening on its own. Build the cash buffer first; the Federal Reserve's household survey says only 63 percent of adults can cover even a $400 surprise with cash, which means a third of the country is trying to fund a technological transition from a standing start with no floor under it.
Law 2: Carrying costs are senior to your plans
James Wilson wasn't destroyed by land prices; he was destroyed by land warrants that billed him maintenance payments on a calendar he didn't control. Look at a modern AI budget and count the warrants: per-seat licenses, token metering, GPU reservations, data pipelines, the platform fees under the platform fees. Every one renews on its schedule, not your revenue's. An AI capability stack is an asset that bills its owner, exactly Wilson's instrument in modern dress, and the law prices it the same way: against your worst year, not your launch-week optimism. The teams that survive technology transitions are the ones whose fixed obligations were sized to survive the quarter when the pilot didn't convert.
Law 3: Never sign personally for the mission
The cleanest divider in fifteen founding lives: Morris and Salomon fused their causes to their family balance sheets and their families inherited the wreckage; Clymer, Willing, and Girard served through structures with walls, and died solvent. The AI translation is blunt. If your route into this transition involves a personal guarantee, on venture debt for compute, on a co-signed loan for the startup, on cross-guaranteed obligations among founders that echo Duer's circle of mutual endorsements, you have connected the technology's worst-case timeline directly to your kitchen. The revolution being genuine doesn't make the note safe; the 1790s proved that combination fatal at national scale. Serve the transition with bounded money and unlimited skill, never the reverse.
Law 4: Revenue is not retention, and productivity gains get competed away
Haym Salomon moved the Revolution's money at half a percent when the market charged two to five, and died worth minus $560. Here is his trap wearing 2026 clothes: AI hands a skilled worker a productivity multiple, and the Salomon instinct is to pass the whole gain through, same rates, more output, gratitude as the margin. Markets compete that away with brutal speed; the gain you don't price into your rates, your salary, or your equity becomes your client's, permanently. The AI Index notes the technology "helps narrow skill gaps across the workforce," which is precisely why undifferentiated output is about to be the cheapest it has ever been. Price the judgment, the accountability, and the outcome, the things the tool doesn't carry, and collect while you're alive. Salomon's heirs petitioned Congress for six decades on the strength of documented gratitude and received nothing.
Law 5: Match your money's duration to the transformation's
Duer ran a months-long corner on thirty-day loans, and the renewal date, not the thesis, destroyed him in nine days of March 1792. The AI buildout is a long transformation being traded on short conviction, and this June's tape showed what that mismatch does at scale: my markets desk logged an AI-stock selloff on June 25 amid $25 billion of weekly equity outflows, bitcoin 28 percent off its peak by June 30, and then, on July 1, the "strongest U.S. equity H1 in 5 years" printing right through the wreckage of the levered trades. The 1792 pattern, a violent unwind in the crowded asset with little spillover to the broad economy, ran live this month. Positions in a decade-long transition funded by money with a monthly clock belong to the clock.
Tonight's snapshot makes the point with unusual precision. As of this evening's pull from my Corvus market pipeline: bitcoin at $62,573, down 23.88 percent from its 60-day peak, with a negative 30-day Sharpe ratio, while high-yield credit spreads sit at a tight 2.75 percent (the snapshot's own label: "risk-on") and the VIX reads a normal 16.59. Read those three dials together and you're looking at the duration mismatch isolated on one screen: the crowded, levered asset repricing violently while credit and volatility markets, the broad economy's nervous systems, register calm. Even inside the wreck there's dispersion (one major token in the same snapshot shows strongly positive momentum), which is the market grading positioning, not the technology. Duer's Sixes fell while the young republic's economy barely noticed. Same physics, refreshed hourly. Coxe's coal lands are the counter-example: unlevered exposure to a real revolution, held on no deadline, paying whenever the demand curve finally arrived, in his case a generation late and enormous.
Law 6: Structure beats sentiment, so own what the machine multiplies
When the machines came the first time, the durable fortunes went to the people who owned the things machines multiplied: land with coal under it, mills, banks, ports, charters. The wage for tending the machine was competed down; the title to what it amplified compounded. AI is running the same division. The model multiplies whatever it's pointed at, and if you own none of the things it points at, no equity, no product, no audience, no data, no IP, then your relationship to the revolution is Salomon's: essential, skilled, and structurally excluded from the compounding. Girard's estate is still following his orders 190 years later because his wealth lived in owned, structured assets with documents around them. Convert AI-era income into owned assets while the income is high; that conversion, not the income, is what survives, and it's the entire argument of my book The W-2 Trap.
Law 7: Custodians compound, heroes combust, and adoption is now table stakes
The 55-to-78-percent adoption jump in a single measured year carries a quiet message: using AI is no longer an edge; it's the new literacy, the way double-entry bookkeeping was for Willing's Philadelphia. The founding pattern says the durable returns from a general-purpose technology go to its custodians, the ones who adopt thoroughly, run bounded experiments, keep their obligations matched, and let the combustors make the headlines. Hillegas, Meredith, Wolcott, and Gallatin kept the books through the republic's wildest financial weather and died solvent; the fresco went to the man who didn't. Your feed will keep serving you the Morrises of the AI era, spectacular raises, spectacular collapses, because the attention economy inherited the memory bias. The estates will go, as they went then, to the people nobody makes documentaries about.
How adopted is adopted? The 18-41-78 problem
Law 7 called adoption table stakes, and the honest footnote is that "adoption" is currently the most slippery number in the economy, slippery enough that the Federal Reserve published a note in April 2026 just to reconcile it. That note's three measures, verbatim: the Census Bureau's firm-level survey finds "about 18 percent of firms have adopted AI as of year-end 2025"; an individual-level survey puts "work-related Generative AI adoption... at about 41 percent" of the workforce; and an employment-weighted business survey finds "78 percent of the labor force works at firms that have adopted AI." Eighteen, forty-one, and seventy-eight, all official, all measuring the same phenomenon. The Fed's explanation is the one a founding-era merchant would recognize instantly: denominators. Most firms are small, small firms adopt least, and surveys that weight by employment or sample large organizations (like the 78 percent figures here and in the AI Index) tilt toward the adopters; in the Fed's words, "the different sampling distributions combined with adoption heterogeneity across firm size classes likely drives a considerable share of the gap."
The Census Bureau's own current reading puts overall business AI use between 17 and 20 percent from December 2025 through May 2026, with the gradient the Fed described fully visible: 37 percent adoption at firms with 250 or more employees, 32 percent at 100 to 249, and under 20 percent at firms with fewer than 20 people, with the Information sector at 39.7 percent while Retail sits near 14. Two lessons fall out. First, the verification law applies to the revolution's own scoreboard: before you position on any adoption statistic, ask whose denominator, because 18 and 78 are both true. Second, the gap itself is the opportunity map. Four out of five small businesses haven't meaningfully adopted the technology that large firms have made table stakes, which is precisely where a custodian-tempered operator, cash first, obligations matched, no personal guarantees, gets to be early without being levered. Tench Coxe compiled the machine economy's first statistics because he understood that whoever measures the transition correctly positions for it correctly. The measurement, then as now, is half the trade.
The Arkwright postscript
One more 1791 rhyme is now on the wire, and this time the fence has docket numbers. Britain treated Arkwright's machinery as a strategic asset and fenced it by law, which is why Coxe's import attempt failed. The modern fence is under active construction in the Federal Register: the Commerce Department's Bureau of Industry and Security published a rule on January 15, 2026, "Revision to License Review Policy for Advanced Computing Commodities" (document 2026-00789), governing exports of frontier-class chips, and this very week the Energy Department filed "Implementing Voluntary Agreements Under the Defense Production Act" (document 2026-13486), whose scope explicitly spans "Artificial Intelligence & high-performance computing" alongside export controls. Pause on that second one: the Defense Production Act is the modern descendant of the mobilization authority Henry Knox's War Department improvised in the 1790s, and it is now being pointed at compute. Then on June 26, my Apprised intel desk led with "U.S. Invokes National Security to Freeze Anthropic's Top AI Models for Foreign Nationals." Spinning frames, chips, models: when the state starts fencing the machine, access stops being purely a market question, and the founding-era lesson for individuals is the humble one: policy will move faster than your plans, so position in ways that survive access changing, which is laws one through six again. The fence changes who gets the machine. It has never once changed the ledger.
And a closing habit, because the machine makes the oldest scam cheaper: AI can now produce confident, well-formatted, uncited financial claims at zero marginal cost, which means the verification discipline this series practiced, chain-of-title on every number before you build on it, has gone from good hygiene for historians to survival equipment for everyone. The Scioto Company needed a Paris office and a famous poet to sell deeds to land it didn't own. The modern version needs a prompt.
Related reading
- 1792 Is Trending: The Modern Headlines Hiding in the Founders' Ledgers: the crosswalk this article extends, section 8 in particular.
- The Ledger Lessons: 15 Founding Fortunes and What They Teach About Money in 2026: where the seven laws come from, with the full evidence.
- Tench Coxe Drafted Hamilton's Masterpiece. His Land Paid Off a Century Late.: the 1791 half of the machine question.
- William Duer: The Founding Father Who Crashed an American Market First: what mismatched duration does inside a genuine boom.
Fact-check notes and sources
- AI statistics (78 percent of organizations using AI in 2024, up from 55 percent; $109.1 billion U.S. private AI investment in 2024; the more-than-280-fold inference cost decline between November 2022 and October 2024; roughly 30 percent annual hardware cost declines; the workforce skill-gap note): Stanford University Human-Centered AI Institute, AI Index Report 2025.
- Household statistics (63 percent able to cover a $400 emergency expense with cash or its equivalent, 2024): Federal Reserve, Economic Well-Being of U.S. Households (SHED), report on 2024, executive summary.
- June 2026 market and policy items (the June 25 "AI selloff" brief with $25 billion of weekly equity outflows, bitcoin 28 percent off peak June 30, the July 1 "Strongest U.S. equity H1 in 5 years" headline, and the June 26 "U.S. Invokes National Security to Freeze Anthropic's Top AI Models for Foreign Nationals" headline): Apprised daily brief archive, markets and intel desks, headlines reproduced verbatim; disclosure: Apprised is my project. Interpretations of these items are this article's own.
- The adoption measurement trifecta ("about 18 percent of firms have adopted AI as of year-end 2025," work-related generative AI adoption "at about 41 percent" of the workforce, "78 percent of the labor force works at firms that have adopted AI," and the quoted sampling explanation): Federal Reserve, FEDS Notes, "Monitoring AI Adoption in the U.S. Economy," April 3, 2026.
- Census adoption detail (overall business AI use between 17 and 20 percent December 2025 through May 2026, 20 to 23 percent expected, 37 percent at firms of 250 or more employees, 32 percent at 100-249, under 20 percent below 20 employees, Information at 39.7 percent, Retail near 14, and the November 2025 question rewording): U.S. Census Bureau, "AI Use at U.S. Businesses," May 2026, from the Business Trends and Outlook Survey.
- Tonight's market snapshot (bitcoin $62,573 and 23.88 percent below its 60-day peak with a negative 30-day Sharpe, high-yield option-adjusted spreads at 2.75 percent labeled "tight (risk-on)," VIX 16.59): official market series snapshotted July 3, 2026 by my Corvus pipeline's public market-context endpoint; disclosure: Corvus is my project.
- The export-control docket (Bureau of Industry and Security, "Revision to License Review Policy for Advanced Computing Commodities," Rule, published January 15, 2026, document 2026-00789; Department of Energy, "Implementing Voluntary Agreements Under the Defense Production Act," Notice, publication date July 6, 2026, document 2026-13486, with "Artificial Intelligence & high-performance computing" and export controls in scope): retrieved from the Federal Register's public API; documents resolve at federalregister.gov by number.
- A note on the 1791 primary text: the Report on Manufactures itself is preserved at the National Archives' Founders Online, which does not serve automated readers even via proxy; no verbatim 1791 quotation appears in this article for that reason, and its facts are carried by the Coxe article's cited sources instead.
- The 1792 mechanics and the "little or no long-term spillover" characterization: Federal Reserve Bank of New York, Liberty Street Economics, "Crisis Chronicles: Central Bank Crisis Management during Wall Street's First Crash (1792)", as applied in the crosswalk article.
- All founding-era claims (Coxe's draft and statistics, the Arkwright attempt, the coal lands, Morris, Wilson, Duer, Salomon, Bingham, Clymer, Willing, Girard, Hillegas, Meredith, Wolcott, and Gallatin): documented with per-claim citations in their linked series posts, indexed in the capstone.
This post is informational, not financial, legal, or investment advice. Institutions and publications are mentioned as nominative fair use; no affiliation is implied.
