One Person, One Company: But Not for Everyone
AI hands everyone the same toolkit. Yet the output gap is not shrinking. It is widening.
Consider the numbers. A decent software engineer, augmented with Claude and Copilot, might see productivity jump from 1x to 3x. A top-tier engineer with the same tools goes from 5x to 33x. Same tools, same models, same tokens. The gap stretches from 5x to 11x. A 2026 analysis of engineering teams found that the productivity gap between elite and average engineers widened from 4x to 5x after AI adoption, with the best accelerating faster than everyone else.
The implication is uncomfortable. "One person is a company" is true, but only for a very small number of people. For the majority, AI creates a different problem: your job functions are being absorbed by the same tools that amplify your best colleague, but you lack the cognitive capacity to use those tools independently. You used to be absorbed and protected by an organization, which assigned you tasks, provided context, and caught your mistakes. When the organization no longer needs as many people to do the same work, where do you go?
The real shock to organizational form is not "everyone becomes free." It is that a tiny minority gains the ability to operate independently at massive scale, while the majority loses even the organizational value they once contributed.
Coase, Inverted
In 1937, Ronald Coase asked a deceptively simple question: why do firms exist? If markets are so efficient at allocating resources, why doesn't everyone just contract with everyone else on the open market?
His answer was transaction costs. Hiring a freelance designer every time you need a mockup means negotiating price, aligning on requirements, iterating through revisions, chasing delivery. All of that is expensive and slow. It is cheaper to employ a designer full-time and manage them internally. The firm exists to internalize coordination that would be too costly to conduct through the market.
Coase's theory also tells us where firms stop growing: at the point where the cost of organizing one more transaction inside the firm equals the cost of doing it on the market. This boundary has been remarkably stable for decades.
AI disrupts both sides of this equation simultaneously.
Market transaction costs have collapsed. Need a design? Generate it in 20 seconds. Need code? Prompt it out. Need copy, analysis, a data visualization? Done, done, done. No negotiation, no alignment meetings, no revision cycles. When the market can deliver in 200 milliseconds what used to take two weeks of coordination, the advantage of internalizing that function disappears.
Internal coordination costs have not moved. Departments, reporting lines, approval chains, cross-functional alignment meetings: these are still here. In fact, they may have gotten worse. When every individual in the organization is producing faster and more densely, the cost of waiting for alignment goes up, not down. Your AI-accelerated output just piles up at the next bottleneck.
But there is a deeper problem with Coase's framework that AI exposes. Coase assumed roughly homogeneous labor. Each additional employee contributes approximately the same marginal output. That assumption was never perfectly true, but it was close enough to make the math work.
AI shatters it. When one person can produce at 33x and another at 3x, adding the 3x person to a team that already has a 33x person barely moves total output but adds a full communication channel. The optimal team size does not gradually shrink. It collapses at a threshold.
This is not a Coasian adjustment. It is a phase change.
Free Agents in History
The idea that individuals might operate independently of permanent organizations is not new. History offers several precedents, each with instructive parallels and critical differences.
The retainer system of ancient China. During the Warring States period (475-221 BCE), powerful lords maintained hundreds or thousands of retainers (门客, menke). These were not employees. They received no fixed salary, had no fixed duties, and were free to leave whenever they chose. They stayed because the lord provided food, shelter, and a platform for their talents. In return, they offered expertise: diplomacy, military strategy, rhetoric, administration. Lord Mengchang famously retained 3,000 menke, each valued for a specific capability.
The parallel to AI-era independents is structural: menke matched their abilities to demand, stayed for the duration of a task or engagement, and moved on when the fit was no longer right. But there is a critical difference. Menke depended on their lord for physical protection and material resources. Today, AI plays the role of resource provider: it supplies the tools, knowledge base, and execution capability. The difference is that AI will not exile you over a political disagreement. The dependency is on infrastructure, not patronage.
The Renaissance workshop. Leonardo da Vinci ran a bottega with a handful of assistants and apprentices. The master provided the creative vision and executed the most important elements personally. Assistants handled preparation, backgrounds, and repetitive work. Contracts sometimes stipulated which parts the master must paint with his own hand. The workshop took on projects ranging from murals to military engineering to urban planning, with a core team of perhaps five to eight people.
AI plays the role of the assistants. You provide the judgment and taste; it handles execution. But there is one asymmetry worth noting. Renaissance apprentices eventually matured, developed their own style, and opened competing workshops. AI does not. It remains a permanent extension of your cognitive capability, never developing independent judgment. This makes the "master plus AI" arrangement more stable than the Renaissance workshop, but also more closed: there is no next generation of masters being trained.
The film crew. A director, cinematographer, editor, and actors assemble for a project, collaborate intensively, then disperse. Each is a highly skilled independent professional. They come together based on the project's needs, not organizational loyalty. This is probably the closest historical analogue to what AI-era production will look like: small, elite teams that form around specific tasks and dissolve when the work is done. No permanent org chart. No departments. No annual review cycle.
The Bottleneck Has Moved
Here is where most organizations are getting the AI transition wrong.
The dominant strategy in 2025-2026 has been to equip employees with AI tools: Copilot subscriptions, Claude access, prompt engineering training, agentic workflow platforms. This is undeniably effective at the individual level. People produce more, faster, with higher quality.
But it does not move the organization's total output in proportion.
The reason is that the bottleneck has shifted. When individuals were the constraint, the economy was in the human being. Humans could only write so much code, design so many screens, analyze so many spreadsheets. Investing in individual productivity directly increased organizational throughput.
AI has broken that bottleneck. An engineer equipped with frontier models can produce in a day what previously took a week. The constraint is no longer individual output. It is coordination: waiting for a colleague to review, a manager to approve, a cross-functional team to align. In any coupled system, throughput is determined by the tightest bottleneck. Optimizing anywhere else just creates work-in-process inventory.
The mathematical structure of coordination costs makes this worse at scale. A 10-person team has 45 pairwise communication channels (10 times 9 divided by 2). A 100-person team has 4,950. Coordination cost grows quadratically; individual output, even amplified by AI, grows linearly. At some team size, the friction tax dominates the productive output.
Every time you make individuals faster without addressing the coordination bottleneck, you are making people produce more output that will just queue up at the next waiting point. You are optimizing everywhere except the constraint. That is the definition of wasted effort.
Firms Will Not Disappear. They Will Mutate.
The "end of the company" narrative is too simple. Organizations will not vanish. But their shape, purpose, and value proposition will change fundamentally.
Trust is infrastructure; labor is a component. In any system, the foundation determines capability more than the individual parts. When labor, once a scarce and expensive component, becomes abundant and cheap (AI provides unlimited execution), the relative value of infrastructure, meaning trust, brand, regulatory compliance, and capital access, surges. The firm's core asset shifts from "we have the best people" to "we have the trust that lets our output enter the market."
Not everyone can leave. The ability to operate independently depends on whether you are a builder or a user. Builders actively reshape their tools, compose new workflows, and extend AI's capability boundaries. Users consume what the tools provide out of the box. The gap between these two groups widens with every AI advance, because each new capability gives builders more raw material to compose, while users simply get a slightly better default experience. Only builders can truly become one-person companies.
This points toward a three-layer structure emerging in the AI economy:
- Top layer: Super-individuals. A small number of cognitively amplified builders, operating independently or in micro-teams of 2-5, producing at the scale of traditional mid-size companies.
- Middle layer: Trust nodes. Small, focused organizations that exist not to aggregate labor but to aggregate trust: brand reputation, legal entity status, regulatory licenses, capital relationships. They are thin shells of corporate infrastructure wrapped around networks of super-individuals.
- Bottom layer: AI-assisted labor. A large population of workers whose tasks are increasingly defined by AI tools, whose output is increasingly commoditized by those same tools, and whose organizational value is declining. They remain attached to platforms or companies, but the attachment is more fragile than ever.
Mid-size companies are the most exposed. Large firms have brand moats, capital advantages, and regulatory relationships that take decades to build. Individuals have extreme flexibility and minimal overhead. Mid-size firms have neither the scale advantages of the large nor the agility of the small. They are the ones most vulnerable to being disintermediated by AI-enhanced individuals or absorbed into trust-node structures.
Who Hasn't Figured This Out Yet
Governments. The entire apparatus of economic governance is built around the firm as the atomic unit. Business registration, tax collection, labor law, social insurance, industry policy: all assume that production happens inside registered organizations with employees. When the productive unit becomes an individual with an API key, the framework does not just need updating. It needs reconceiving. No government has begun this work, because acknowledging the problem means admitting that the foundational unit of economic governance is becoming obsolete.
Public companies. Capital markets reward scale and growth, not efficiency per person. "We have five people but produce what we used to with fifty" is not a compelling narrative to investors trained to value headcount as a proxy for capability. CEO compensation is tied to revenue growth and market capitalization, not to output-per-employee ratios. The incentive structure actively discourages thinking about whether the firm is still the right organizational form.
The people most at risk. Workers in the bottom layer of the three-layer structure often do not realize their organizational value is declining. They see AI as a tool that makes their current job easier. They do not see that "easier" means "more automatable." The cognitive amplification gap is invisible from the inside: you experience your own productivity improvement (from 1x to 3x feels great) without seeing that someone else went from 5x to 33x.
What Token Built
The previous article in this series, "Tokens Are the New Salt and Iron," argued that tokens are the strategic resource of the AI era: a resource that moves at the speed of light, concentrates production in two countries, and resists traditional governance. This article asks what gets built on top of that infrastructure.
Salt and iron monopolies sustained the bureaucratic apparatus of empire. The state controlled the resource and used hierarchy to distribute it to officials, who managed production on the state's behalf.
Oil capital sustained the multinational corporation. Capital concentrated the resource, and the corporate form organized labor and technology around it to produce industrial output.
Tokens will sustain something we have not named yet. It will be smaller than a company, more stable than a freelancer, less centralized than a platform. A handful of super-individuals driving production with tokens. Small trust nodes providing brand and compliance. A broad base of AI-assisted labor attached to platforms. Three layers, no middle management.
It does not have a name yet. But it is already forming.
FAQ
Doesn't this only apply to knowledge work? What about physical industries? The argument applies most directly to knowledge production: software, design, writing, analysis, consulting. Physical industries (manufacturing, construction, logistics) still require coordination of physical assets and are less susceptible to the one-person-company dynamic. But even there, the management and planning layers are knowledge work, and those layers are being compressed.
If only a small elite can become one-person companies, isn't this just more inequality? That is a fair reading. The three-layer structure described here is not a utopia. It is a description of what the economics imply. The cognitive amplification gap means that AI increases inequality of output, which translates into inequality of organizational power. Whether policy can mitigate this is an open question.
What about the Hollywood crew model? Doesn't that require a shared physical workspace? Film crews do often share physical space, but the organizational logic does not depend on it. Distributed open-source teams, freelance consulting networks, and gig-economy platforms all use the same pattern: assemble for a project, collaborate, disperse. The physical workspace is incidental; the organizational pattern is the point.
Won't companies just adapt and become more efficient? Some will. The most adaptable firms will flatten hierarchies, reduce middle management, and restructure around small autonomous teams. But this is precisely the mutation described in section 5: they are becoming trust nodes, not labor aggregators. The ones that try to maintain their current form while adding AI tools to individual employees will find that the bottleneck has moved without them.
References
- Coase, R.H. "The Nature of the Firm." Economica, 1937. Wiley
- Litowitz, A. "AI and the Nature of the Firm." SSRN, 2026. SSRN
- Yu, Howard. "Coase vs. Claude and The Future of the Firm." Substack, 2026. Link
- Coverdale, Charles. "Coase vs AI." Tangentially Economics, 2026. Link
- "From Coase to AI Agents: Why the Economics of the Firm Still Matters." California Management Review, 2025. CMR
- "AI Productivity's $4 Trillion Question." Forbes, January 2026. Forbes
- "Why AI Will Widen the Gap Between Superstars and Everybody Else." WSJ, 2026. WSJ
- "Life in a Renaissance Artist's Workshop." World History Encyclopedia. Link
- "How Artists Worked in the Italian Renaissance Workshops." DailyArt Magazine. Link