the 1-person company

(reading time: 8-10 mins)

At the turn of the 19th century, the East India Company was run from a building on Leadenhall Street by a few hundred clerks. Its army in India numbered some 260,000 men, twice the size of the British Army. It is tempting to see this as an astonishing feat of leverage: a few hundred people in London commanding a quarter of a million on the other side of the world.

It wasn't, quite. The Leadenhall clerks did not command this ‘army’ directly. They wrote to a handful of managers, who passed orders to sub-managers, who passed them to sub-sub-managers, who passed them to sub-sub-sub-managers, each of whom had perhaps ten men to keep in line. This human ladder drove planning through to execution, every rung existed because one person can only really oversee a few others in the organization.

In October 2026, Anthropic published a study of how employment is changing using itself as an example [1]. By August, employee-directed agents were logging roughly six hours of runtime for every estimated human working hour. Between each employee and their agents, there is no ladder at all. Agents become teams for a single employee.

If the old rungs within organizations are going away, what will companies look like in the future? Sam Altman has said his group chat of tech CEOs keeps a betting pool on the year the first one-person billion-dollar company appears [2]; Dario Amodei has put it as early as 2026 [3], and Marc Andreessen calls it the industry's ‘holy grail’ [4].

how humans learned to coordinate

Historically, human groups have been shaped by human limits. We can recognize thousands of faces [5], but the relationships we can actually sustain come in nested rings of roughly 5, 15, 50 and 150 [6]. Beyond about 150, groups tend either to split (Hutterite colonies typically divide at around 168 people, and prehistoric Andean villages at around 170 [7]) or to build a human ladder to manage the extra scale. One plausible explanation of this phenomenon is interpersonal trust: can I really rely on someone in my tribe whom I don't know, and couldn't possibly know, on top of the 150 others I'm meant to keep track of? Evolutionary psychologist R. Dunbar’s 2022 paper cites a study (Lee, 1979) that shows, among the !Kung, groups of 100-150 had 4-8x as many ‘serious disputes’ as groups of 30-50 [7], which lends well to a proposed theory of trust.

In 1933 the management theorist V.A. Graicunas proposed a model in which the number of potential relationships rises rapidly: five reports means 100 relationships, twelve means more than 24,000 [8]. That might explain why armies from Rome to the US Marine Corps are built from nested units, each with its own accountable leader. The Roman legion ran from tent groups of 8, to centuries of 80, to cohorts of 480, to a legion of over 5,000 [9]; the Marines still organize by a "rule of three," from four-man fire teams upwards [10]. It isn't hard to see why such structures served these forces so well. They work well at turning commander's intent into outcomes.

In his 2011 book Sapiens, Yuval Harari argues that writing was the next great scaling invention for human coordination [11]. The earliest texts we have aren't poetry or law but accounts: about 85% of the oldest tablets from Uruk are administrative records [12]. Writing something down lets people agree on its ‘truth’ across space and time; it becomes a kind of living memory for the organization. Writing also helped preserve and transmit the ‘shared fictions’ that, Harari argues, already had allowed strangers to cooperate. Religion, money, and law were expressed in the written form of scripture, banknotes, and statute [11].

Fast forward to the twenty-first century, walk into any modern workplace, and you'll find these two mechanisms, nested hierarchy and the written record, at the heart of how it runs. The first organization chart was drawn in the 1850s, for a railroad too long for any one superintendent to oversee [13]. Today's corporations are highly tuned versions of the same machine, built to let large numbers of people move quickly towards a shared intent, such as growing revenue.

is it really the best way?

By the turn of the century, generational entrepreneurs had begun to question whether these tried-and-true mechanisms were the best way to scale fast-growing companies. Famously, Amazon's Jeff Bezos introduced the idea of two-pizza teams: teams small enough to be fed with two pizzas, typically five to seven people [14]. These were designed as autonomous units with as few dependencies as possible [17]. Amazon initially experimented with a composite ‘fitness function’ before moving toward individual input metrics [15][16]. To stay coordinated and coherent, teams are organized into lines of business. The institutional memory lives in writing. Six-page narrative memos are read in silence at the start of meetings, and for new products there are PR/FAQs, a mock press release plus FAQs, that settle the ‘what’ and the ‘why’ [18][19]. Leadership Principles govern the ‘how’ (e.g. "Have Backbone; Disagree and Commit" [20]). The goal? Fast-moving teams that stay unencumbered, even as the overall org grows, to delight customers.

Elon Musk is known for running flat businesses that do away with human ladders. He weeds out managers who don't do hands-on technical work, and he himself acts as orchestrator to front-line engineers. He bypasses levels to check on progress and remove as many blockers as possible, as quickly as possible. Judging from the results, this seems to work. But it is an operating model that management guru Jim Collins calls ‘a genius with a thousand helpers’ in his 2001 book Good to Great. As long as the genius is there, things run well; remove the genius, and things tend to fall apart. Hence, a business that isn't built to last [21].

Mark Zuckerberg and Alexandr Wang have taken a page from Musk's playbook and organized Meta Superintelligence Labs (MSL) in a similar fashion [22]. Wang himself has more than 200 direct reports: every researcher in its core lab reports to him [22]. He hires for talent density. He doesn't provide soft-skill management oversight ("They don't need managers, they need a great environment … They just need the room to cook"). Instead, he serves only to coordinate and ensure technical progress, with the technical leads of small ‘pods’ breaking ties [22]. The careful selection of talent reduces the subsequent effort to supervise them. Again, judging by the results (Meta's Muse models are now arguably at the frontier of personal-agent work [23]), the method seems to be working, at least for now.

Former Zappos CEO Tony Hsieh was even more radical: he experimented with holacracy, an extreme form of self-government with no people managers. Teams become ‘circles’, and bosses become ‘lead links’ with limited managerial authority [24]. Did it work? Not obviously. In 2015 Hsieh told his roughly 1,500 staff to commit or take severance. Eighteen percent took the buyout and another 11% left without one [24][25]. (Hsieh countered that normal turnover was about 20% a year anyway [25].) In 2016 Zappos fell off Fortune's 100 Best Companies to Work For list for the first time in eight years [24]. By 2017 it was shifting to a ‘market-based’ model in which teams run their own P&Ls, and it later quietly brought managers back [26]. Hsieh retired in August 2020 [27].

Some see Hsieh's attempt as a failure. Others think such experiments deserve celebration even if most fail. Stripe founder Patrick Collison has said he ‘love[s] the fact’ that Zappos and Medium tried holacracy, while putting the odds that any radical departure from best practice works at best one in ten [28]. The underlying instinct here is that, in some way, orgs can be designed for the better, and more experimentation is needed. To what end? Presumably, faster global economic growth.

Perhaps the thing to test directly is whether nested hierarchy is more durable and resilient in the absence of genius; and what is it about the genius with a thousand helpers model that works so well, at least in the short-run? Is it the genius or is it the genius of selecting a highly autonomous and able team? Let’s suppose for a moment that a genius planner or talent selection is the active ingredient – some form of skill and judgement seem to be the driving factors of success. If spotting and sourcing this scarce talent has been hard and expensive in the past, then how does this change with the rise of superintelligent agents that are available 24-7 in any part of the world?

a theory of firms

In the theory of firms, British economist Ronald Coase asked: if markets coordinate economic activity through prices, then why do firms exist at all? His answer was transaction costs. Transacting with the market requires finding counterparts, negotiating and enforcing terms, and collecting payment. Organizing as a firm reduces the transaction cost a single person would have to bear when transacting with the market. But internal coordination within the firm also incurs overhead costs which grow as the firm grows in size and as the firm pursues new markets. There exists a cross-over point where internal costs for the next transaction are equal to transaction costs, at which it doesn’t make sense for the firm to expand. A Coasian scaling limit [29].

It’s worth asking whether a hierarchical team of humans has higher internal cost compared to a human ‘genius’ and a thousand helper agents. Supposing this is true, it isn’t hard to see how new types of firms will form to compete in the market at lower transaction cost, at more competitive pricing than its hierarchical competition. Of course, working with agents isn’t cost free. Anyone using chatbots or personal assistants are feeling the load of the constant back and forth, and Anthropic even cites this type of human bottleneck in its study. But this also reflects the lack of a living memory, a shared written history, to seamlessly coordinate on. Once that’s in place, we may see scaling in a similar way to what ancient civilizations experienced when inventory and norms were scribed and used as ground truth. Importantly, there may be no Coasian limit in this case. In theory, you can simply add more agents to your workforce at no additional coordination cost.

So why does Anthropic still need more humans at all? Part of the answer is a wider trend: firms are hiring fewer entry-level workers and more skilled people with judgement. The rest is hidden in how those people spend their day.

cheap talk?

There is a small detail in Anthropic's report that is easy to miss. In the most agent-heavy company yet studied, machines work six hours for every human hour. Yet the median employee now spends 16-20% of their time in meetings [1]. That is more than before.

Why? One answer is accountability: a person must answer for the work. But that does not explain why meetings grow as agents do more. A better answer is that people meet to agree on what to build, and why. Agents can do the work. Choosing the right work is still a human job. If people thinking together are a kind of superintelligence that only humans have, agents will not end the meeting. They will make it the main job left for humans: making good decisions. Jeff Bezos once said: “If I make, like, three good decisions a day, that’s enough” [30].

For ninety years, firms were designed to cut talk and focus on getting work done. Amazon went further. It let front-line staff make decisions that are easy to undo, which it calls “two-way doors” [15]. Firms will probably hand the same kind of decisions to agents. The doing moves to machines, and the deciding stays with people. It sounds strange, but the firm of the future may be almost all about talk.

Note: this essay was edited and fact-checked, not written, by Grok Bot and Claude.

[1] https://www.anthropic.com/research/employment-at-anthropic

[2] https://fortune.com/2024/02/04/sam-altman-one-person-unicorn-silicon-valley-founder-myth/

[3] https://www.inc.com/ben-sherry/anthropic-ceo-dario-amodei-predicts-the-first-billion-dollar-solopreneur-by-2026/91193609

[4] https://podscripts.co/podcasts/lennys-podcast-product-career-growth/marc-andreessen-the-real-ai-boom-hasnt-even-started-yet

[5] https://pmc.ncbi.nlm.nih.gov/articles/PMC6191703/

[6] https://pmc.ncbi.nlm.nih.gov/articles/PMC1634986/

[7] https://www.cambridge.org/core/journals/evolutionary-human-sciences/article/managing-the-stresses-of-groupliving-in-the-transition-to-village-life/11DF7BD863F253F058E9BADA1F6DB075

[8] https://nickols.us/~nickols1/graicunas.htm

[9] https://www.vindolanda.com/Blog/fact-file-roman-soldiers

[10] https://www.defense.gov/Multimedia/Experience/Military-Units/marine-corps/

[11] https://en.wikipedia.org/wiki/Sapiens:_A_Brief_History_of_Humankind