As individuals become newly differentiated in the AI era, what happens to the companies, teams, and working relationships they form?
Editor's note
Welcome to Tinsight, Tabbit's column for thoughtful, in-depth reading. We previously shared two essays from offbook.press: The Real Divide in the AI Era and Rebuilding Individual Learning.
The Real Divide in the AI Era and Rebuilding Individual Learning
We now begin The Breakdown of Organizational Forms. The earlier essays explored the breakdown and reconstruction of individual cognition and learning in the AI era. This series examines what changes for the companies, teams, and relationships formed by those individuals. The original essay is about 20,000 Chinese characters long; we are publishing it in six installments. This is chapter one.
Author: Dawei Geng
Source: https://offbook.press/essays/breakdown-of-firms/
https://offbook.press/essays/breakdown-of-firms/
Follow the original article link to read the full essay on the author's website.
Preface
The previous two essays examined individuals: the new differentiation of cognitive abilities in the AI era (issue 1), and the breakdown and rebuilding of individual learning (issue 2).
This essay examines organizations: as individuals become newly differentiated in the AI era, what happens to the companies, teams, and working relationships they form?
It is a colder question: which organizational forms, management methods, and working relationships have already become structurally obsolete, even as we keep describing their failures with outdated words?
This is difficult to discuss because our vocabulary for organizations, management, and collaboration—company, employee, manager, team, collaboration, efficiency—comes from the industrial age. Those words carry industrial-age assumptions. Using them to discuss AI-era phenomena is like discussing electric cars in terms of coal-burning efficiency. The problem is not simply unclear discussion; the words themselves are wrong.
But we do not have better words. This essay therefore does one thing: use the old vocabulary as precisely as possible to describe what is happening, while acknowledging the limits of language itself.
No prescription. Just as honest a diagnosis as possible.
Chapter 1 · The question we skip: what did companies originally solve?
Before asking how companies should evolve in the AI era, we need to ask the question almost everyone skips: what problem did companies originally exist to solve?
It seems too basic to discuss. Most people assume a company is simply an organizational form for people doing things together—a natural state. But companies are not natural. They arose under particular historical conditions as one way to solve specific problems. If new technology solves or changes those problems, the need for that particular solution becomes less obvious.
Coase's question
In 1937, Ronald Coase published The Nature of the Firm and posed a central question:
If markets are so efficient at allocating resources through prices, why do we need companies? Why not conduct every economic activity through individual market transactions—buy A when I need A, and B when I need B?
Coase's answer was that market transactions themselves have costs. Searching for partners, negotiating, signing contracts, executing, enforcing accountability, and handling breaches can make market transactions too expensive. Bringing an activity inside an organization becomes cheaper because instructions and processes can replace repeated market negotiations.
A company is therefore, fundamentally, a solution to excessive market transaction costs. When the cost of a market transaction exceeds the cost of doing the same thing inside an organization, the activity is brought inside.
Oliver Williamson and others later developed Coase's insight into a foundation of new institutional economics. This is not a flashy business-school theory. It is one of the clearest answers to the fundamental question of why firms exist.
Eight problems companies solve
Expanding Coase's framework, companies address eight specific types of transaction-cost problems:
1. Search and matching costs
Complex work once required finding people with many different skills: programming, design, sales, finance, and law. Searching the market each time was expensive. You did not know whom to trust, they did not know whether to trust you, and both sides needed time to build confidence.
Companies solve this by binding a group of people together over time. Once you join, the programmers, designers, and salespeople are readily available resources; you do not need to search again for every task.
2. Information asymmetry
A fundamental problem in market transactions is knowing in advance whether someone will do a good job. A freelance programmer says they will write excellent code—but how do you know? You find out only after delivery, when the money or time has already been spent.
Companies address this through long-term employment, reputation, and observable daily performance. After three years at a company, your manager has a reasonably clear picture of your abilities. A one-off market transaction cannot provide that.
3. Asset specificity
Some investments are valuable only within a particular relationship. Spending three months learning a client's internal systems for a custom project may benefit only that client. In a one-off transaction, you bear the risk of that specialized investment.
Companies address this through long-term relationships. Learning internal processes, building relationships, and acquiring company-specific know-how becomes worthwhile because the relationship lasts long enough.
4. Coordination costs
Complex work requires many people to cooperate. Coordinating each step through a market transaction is impractical—imagine signing a contract every time a programmer discusses a minor detail with a designer. Companies substitute hierarchies and processes, with clear reporting lines, routines, and shared expectations about how things are done.
5. Risk sharing
Independent market participants bear their own risks. Companies share risk by bringing people together: a project's losses are distributed, someone can cover for a sick colleague, and the organization survives the loss of a client.
6. Pooling capital
Some activities require large upfront investments: building factories, developing products, or buying equipment. Individuals and small teams cannot always raise that money. As legal entities able to obtain financing, companies pool dispersed capital for large-scale production.
7. Legal and contractual capacity
Many activities require a legal entity: signing major long-term contracts, bearing bankruptcy risk, providing employee social insurance and stock options, appearing in court, undergoing regulatory review, or obtaining government licenses.
Individuals either cannot do these things because they lack the relevant legal status, or face excessive personal risk—for example, exposing all their assets when a major contract goes wrong. A company provides legal personality. Shareholders' liability is limited to their investment, while the company can contract, borrow, be sued, and go bankrupt. This is no incidental benefit; it is a central reason certain activities must be undertaken through a company.
8. Brand and institutional trust
Some activities need a credible institution to stand behind them, especially when consequences are serious and quality is difficult to verify.
For surgery, litigation, bank custody, or strategic advice, customers cannot fully judge quality beforehand. Part of what they buy is a promise that an institution will take responsibility if things go wrong. That promise derives its weight from the institution's brand, history, scale, and incentive not to lose them.
An equally capable independent individual cannot offer the same assurance. They have less at stake and customers have less institutional recourse. In these settings, customers are buying both capability and institutional backing; individuals cannot replace the latter on equal terms.
What happens to these eight problems in the AI era?
Once we understand what companies originally solved, we can ask the real question: do these eight problems still exist in the AI era, and is the company still the best solution?
Let us take them one at a time.
1. Search and matching costs: reduced
Finding a reliable programmer, designer, or lawyer used to be extremely expensive. Now:
AI handles a substantial part of these roles, reducing the number of people you need to find.
For work that still requires people, remote collaboration platforms, Upwork, LinkedIn, Twitter, and Discord communities greatly reduce search costs.
AI's general capabilities mean every role need not be filled by a specialist. A versatile individual who uses AI for routine coding, design, and legal work can handle what once required a team.
The cost-effectiveness of binding a group together long term is therefore falling rapidly. Much can now be done through temporary collaborations or by an individual with AI.
2. Information asymmetry: partly reduced
Assessing someone's ability once required long observation. Now:
Public work—GitHub, Twitter, personal websites, and previous projects—provides signals a traditional résumé cannot.
AI makes it cheaper to assess whether a candidate can do a task: give them a trial assignment and use AI to help evaluate it.
But the deeper problem of trust remains. Complex collaborations requiring sustained commitment still need long-term relationships to build trust.
This cost has been reduced, not eliminated.
3. Asset specificity: the underlying problem remains, but its form changes
Specialized investment is still necessary for serious work. What changes is the nature of that investment:
Previously, it meant learning a company's internal processes—knowledge valuable mainly to that company.
Increasingly, it means understanding a field, mastering a toolset, and developing a distinctive perspective. These investments can be reused across relationships.
Individuals' specialized investments are becoming less tied to a single employer. What you know is increasingly your own know-how rather than one company's, giving you greater independence and bargaining power.
4. Coordination costs: sharply reduced
This is where the change is most dramatic.
Coordinating complex work once required meetings, processes, hierarchies, and production lines. Now:
Asynchronous communication removes much of the need for meetings.
AI acts as an endlessly patient collaborator, taking on documentation, information organization, and progress tracking that once needed substantial human coordination.
One person with AI can do work that previously required a team of five.
Where coordination costs have fallen most sharply, small-scale collaboration is cheaper than ever. Three people with AI can do more than ten people could before.
5. Risk sharing: still necessary, but changing form
People still get sick, burn out, and make mistakes, so the need to share risk remains. But:
Long-term employment and benefits, the traditional approach, are becoming increasingly rigid for individuals and companies alike.
New forms include flexible working relationships, mutual-support networks among independent people, communities, and AI as a tireless backup.
In some countries, public social security, insurance, and healthcare reduce individuals' dependence on companies for risk protection.
The underlying need remains, but companies are no longer the only or necessarily the best solution.
6. Pooling capital: no longer essential for much software, content, and service work
Many activities once required substantial upfront investment. Now:
Software often does not: one person, a computer, and AI can build what once required a team.
Content often does not: an independent creator using AI can produce what once took a content team.
Many services do not: consulting, design, writing, and translation that once needed an organization can now be delivered by an individual.
Capital pooling remains necessary for chips, rockets, medicines, and other activities requiring large hardware investments. The need for companies in those areas has not changed.
7. Legal and contractual capacity: unchanged or stronger
AI has not changed the legal system. Contract law, company law, bankruptcy protection, regulatory compliance, and tax rules still require legal entities to bear obligations.
AI also introduces new legal complexity. Who is liable for losses caused by AI output? Who pays for infringement involving training data? Whose decision is an AI-generated decision? These questions increase the need for a legal entity to stand behind the work.
This function has not weakened in the AI era and is stronger in some respects.
8. Brand and institutional trust: weaker on the surface, stronger underneath
AI affects this in two directions.
Individuals can now produce professional-looking work cheaply. The surface-level brand gap narrows: one person's deliverable can look as polished as a large company's.
More significantly, the explosion of AI-generated content makes trust scarcer. Customers find it harder to distinguish deep human thought from mass-generated output. In high-consequence settings, they increasingly need an institution to stand behind the assurance that the work was not casually assembled by AI.
AI weakens the surface layer of brand trust—polished deliverables are no longer scarce—but strengthens its deeper layer. What is scarce is the promise of real responsibility for results. Individuals still struggle to replace institutions on that dimension.
A structural inference
Put these observations together.
Of the eight problems companies originally addressed:
Two are substantially reduced: search and matching, and coordination.
Two are partly reduced: information asymmetry and asset specificity.
Two remain, but companies are no longer necessarily the best solution: risk sharing and capital pooling for software-like work.
Two are unchanged or stronger: legal and contractual capacity, and institutional trust.
This is more cautious than a claim that most foundations have been destroyed. The usefulness of companies has not collapsed across the board. Its boundary is shrinking, but unevenly.
The sharpest contraction is in knowledge-intensive fields with low capital requirements, low contractual risk, light regulation, and limited need for institutional endorsement: software, content, consulting, design, and some services. One person with AI can do what once required a team; three to five people can do what once took a mid-sized company.
The least contraction—and sometimes an increase—is in capital-intensive, heavily regulated, high-risk, high-consequence fields: finance, healthcare, infrastructure, manufacturing, energy, aerospace, and government contracting. Companies remain necessary and may become more so because of AI-related legal risks and trust concerns.
The accurate conclusion is therefore that the boundary of where companies are useful is being redrawn. They were once the default form for nearly all collaboration. They remain essential, or even more essential, in some fields, while becoming too heavy, slow, and expensive in others.
The rest of this essay focuses on the area of greatest contraction: organizational change in knowledge-intensive work. How large companies retain their position through capital, regulation, and brand lies outside its scope. If you work in those fields, the analysis applies only to the knowledge-intensive part of your work.
Yet our entire vocabulary for discussing organizations assumes that the company is the default form. On the shrinking side of that boundary, this vocabulary is about to encounter its limits.
To be continued…
Chapter 1 · The question we skip: what did companies originally solve?
Chapter 2 · Collaboration no longer happens on the same plane
Chapter 3 · Why AI productivity gains differ across organizational forms
Chapter 4 · Old management fails; new management remains unknown
Chapter 5 · The distillation curve of roles and knowledge
Chapter 6 · Closing thoughts
