The number that started the question
In 1700, the East India Company operated with 35 permanent employees in its London office. By 1785, that number had grown to 159. The territory the Company influenced over those eighty-five years had grown by orders of magnitude more.
The 159 were administrators, not operators. They sat at the top of a sovereign-delegated apparatus that included an army that would peak near 200,000 by the early 1800s, thousands of London warehouse workers, and a network of agents operating with extraordinary autonomy in an era when a message from Calcutta to London took five months. The Company's leverage in this period combined monopoly charter, military force, and an employee structure that permitted personal trade.
The 159 has become a stock figure in modern productivity discussions, often cited without its apparatus. That compression is worth resisting. The number is interesting as a historical entry point, not as a benchmark a modern company resembles or aspires to. Small cores producing outsized output is not new. The Rothschild courier network, the Medici correspondents, the Fugger silver houses, the Lloyd's coffee-house underwriters: these are the rule across commercial history, not the exception. Each shift in the leverage mechanism also produced a class of newly redundant workers, from London warehouse hands to regional bankers outpaced by Rothschild speed to back-office clerks compressed by partnership efficiency. This essay brackets that distributional history and tracks the architectural one. What changes across the cases is the engine. What this essay argues is that the current shift in the engine has properties the earlier shifts did not.
Five cities, one nervous system
By the early 1800s, a different innovation in commercial coordination had matured in continental Europe: a private courier system fast enough to outrun the official mail of European governments, used to transmit financial intelligence between five connected agents.
The agents were five brothers. Mayer Amschel Rothschild had sent each of his sons to a different European financial capital in the early 1800s. Amschel to Frankfurt. Salomon to Vienna. Nathan to London. Karl to Naples. Jakob to Paris. The structural innovation lay in what the family bond enabled. As Niall Ferguson documents in The House of Rothschild, the network's couriers consistently outpaced government dispatches, and Nathan in London could act on intelligence from Jakob in Paris before the broader market had absorbed the same news.
Five people in five cities, linked by a faster nervous system than any competitor possessed, did not require operational scale to compete. The competitive edge was speed of coordination and quality of information inside a small high-trust core. During the post-Napoleonic reconstruction period, the network participated in sovereign debt issuances across multiple European states on timelines competing houses could not match.
The output was disproportionate to the headcount because the headcount was not the constraint that produced the output. The information system was. The five-person core did not sit atop a hidden operational apparatus the way the 159 in London did. The five were the operation, and the courier network was the leverage.
The partnership as technology
By the late 1860s, a different mechanism for concentrating authority in a small core had matured into the dominant form on Wall Street: the professional partnership.
Goldman Sachs was founded in 1869 as a commercial paper business and operated for its first several decades with fewer than three hundred employees. The defining structural feature in the firm's first century was its partnership architecture. A small number of partners shared profits, bore personal liability for the firm's obligations, and reinvested the bulk of earnings back into the business. The architecture aligned three things public-corporate structures struggled to align: economic exposure, decision authority, and time horizon. Partners with personal liability priced risk differently than executives with limited downside. Partners who reinvested most earnings extended their planning horizon to match the firm's.
When Goldman went public in May 1999, the firm had 221 partners and roughly 13,000 employees. The partners represented less than two percent of the workforce and held just under half of the firm's equity at the conversion. The ratio is the point. A small economically aligned core directed an operational body more than fifty times its size, and the structure persisted for 130 years before the firm took the public-company step.
The phrase is doing real work. It names a structural choice, not a sector or a scale. The same architecture appears in the Rothschild house in 1820 and in Goldman in 1900 because both arrived at it for the same reason: small high-trust cores aligned around long time horizons make better commercial decisions than large diffuse organizations aligned around quarterly metrics. The mechanisms that allowed each to operate at scale were different. The architecture they used to govern themselves was the same.
| Era | Entity | Core team | Leverage mechanism | Operational apparatus |
|---|---|---|---|---|
| 1810s–1820s | Rothschild network | 5 brothers | Private courier system, information speed | Correspondent agents in five European capitals |
| 1869–1999 | Goldman Sachs | 221 partners (at IPO) | Partnership liability, capital alignment | ~13,000 employees |
| 2026 | Lean AI companies | ~25–150 people | Software automation, model inference | Foundation-model labs and hyperscaler datacenters |
The leverage stack changes. The structure does not.
The three cases above are separated by more than a century each. They share an architectural feature: a small high-trust core directs an operational footprint many times its size. What differs across them is the bridge that spans the gap.
Read the leverage column from the top down. A private courier network required generational kinship and the capital to maintain it across borders. A partnership architecture required decades of cultural and reputational investment before it produced durable returns. Software automation requires a laptop and an API key. The leverage mechanism has gotten more efficient with each iteration, and the access barrier has fallen alongside it.
The honest version of the claim is about where the apparatus went, not whether it shrank. The current AI-native company's leverage runs through a stack it does not own: foundation-model labs employing thousands of researchers, hyperscaler datacenters built with tens of billions of capital, GPU output from a small number of fabs, training data labeled by globally distributed contractors. The apparatus moved a layer down and became a metered utility. The architecture has always been about the relationship between the small core and the operational footprint it directs, not about who owns that footprint. The Rothschild brothers did not own the European post roads their couriers raced. Goldman did not own the New York Stock Exchange. The ownership boundary moves across the cases. The architectural boundary holds.
Each leverage shift also produced new categories of work that did not exist before it. The chartered-trading era is the cleanest case. Edward Lloyd's coffee house on Tower Street began as a place where ship captains and merchants traded news in the 1680s; by the time the Society of Lloyd's formalized in 1771 with seventy-nine underwriters subscribing to a single trust deed, marine insurance had become a profession with its own apprenticeship, its own pricing literature, and its own clerical economy. Customs administration, average adjusters, and double-entry bookkeepers followed. A class of risk-bearers existed at scale that had not existed a century before. The Wall Street partnership era did the same on a different axis. The Securities Act of 1933 and the Exchange Act of 1934 made disclosure a federal regime, and in the decades that followed the certified public accountant, the sell-side analyst, and the back-office reconciliation clerk became durable middle-class livelihoods; ERISA in 1974 and the 401(k) provision in the Revenue Act of 1978 then routed equity ownership through the retirement accounts of tens of millions of workers who would never sit on a trading floor. The AI-native version is forming rather than mature, and the categories are early enough that some will not survive the decade. Anthropic's own job board in 2024 listed full-time roles for prompt engineers and red-teamers at salaries above two hundred thousand dollars, categories that did not exist as titles three years earlier; the RLHF labeling economy and the agent-operations function are growing on similar curves, and the rise of one-person companies clearing seven-figure revenue is the consumer-surface version of the same shift. The architecture is extraction-capable and coordination-capable in equal measure. Which it becomes depends on how the financial infrastructure around it is built.
The lean AI company is what the curve looks like when the apparatus is rented software running on someone else's silicon. A team that produces millions in revenue per employee sits at the top of the AI-native distribution rather than at its median. It is the current point on a curve that has been moving in one direction since five brothers ran sovereign debt syndicates from five capitals.
What the architecture predicts
If the architectural feature persists, as the cases above suggest it does in non-state-owned commercial settings, the implication is not that the lean AI company is the end state of the curve. It is a current position on it. The current generation of AI-native companies, where the top of the distribution shows teams of fifty to one hundred and fifty producing in the millions of revenue per employee, occupies the same architectural position the partnership houses occupied a century earlier. Lovable at one hundred and forty-six people producing four hundred million in annual recurring revenue. Midjourney at roughly one hundred people generating five hundred million. Gamma at fifty employees crossing one hundred million. Bolt.new built inside a thirty-five-person StackBlitz team. The names change month to month. The architecture is the same, and it is showing up across every category AI touches, at every funding stage.
The historical record also suggests which instances last. The Rothschild network endured across the nineteenth century because the partnership concentrated wealth and decision-making over multi-generational time horizons. The Wall Street partnerships that lasted longest were the ones that resisted quarterly-earnings pressure as long as they could. Structures that did not align time horizon with enterprise interest are not in this essay because they did not last long enough to enter the historical record. The technology determines how small the team can be. The structure determines whether the small team builds something durable or extracts value from a window and disappears. The companies are already small; the financial infrastructure they need has not caught up, and the next volume of this almanac argues that the gap is closing faster than most institutional capital allocators have priced in.