Is AI Changing the Minimum Viable Firm?
For most of modern business history, accomplishing something economically significant generally required building an organization capable of coordinating a large amount of capital and labor. Growth meant adding people, and adding people meant adding systems, management, infrastructure, and coordination around them. Scale therefore depended not only on whether a company could find customers, but on whether it could successfully organize enough resources to serve them.
Ronald Coase gave us one of the most useful ways to think about why companies take the shape they do. In his 1937 paper, The Nature of the Firm, Coase asked a basic question: if markets are capable of coordinating economic activity through prices and contracts, why do firms exist at all? His answer was that using the market is not free. Companies and individuals incur costs finding counterparties, negotiating contracts, monitoring performance, resolving disputes, and repeatedly coordinating transactions. Sometimes it is cheaper to bring an activity inside a firm than to continually purchase it through the market. Firms therefore emerge where internal coordination is cheaper than external coordination, and expand until organizing the next activity internally becomes more expensive than obtaining it elsewhere. Coase later described these frictions as transaction costs, work that became central to the 1991 Nobel Prize in Economic Sciences awarded to him. Coase’s Nobel lecture and the Nobel committee’s summary of his contribution lay out the argument directly.
Technology in the information age has spent decades changing that calculation. Personal computers reduced the labor required for administrative work. Enterprise software automated activities that previously required additional people. The internet radically lowered the cost of communication and distribution. Cloud computing allowed companies to access sophisticated infrastructure without owning it. Software-as-a-service moved capabilities that once required internal systems or departments to resources outside the boundaries of the firm. Remote work, digital marketplaces, and global communications made it easier to coordinate economic activity across geography. Each innovation changed, in one way or another, the relative cost of doing something inside an organization versus obtaining the same capability somewhere else.
Technology began to shift the ground under employment in a new way in the early 2000s with the growing ‘gig economy’. Fractional work became more commonplace and more popular in sectors that had previously been almost exclusively full-time. First taxi and delivery drivers, then on to higher-skill roles like nursing and technology. Now, few roles within an organization aren’t eligible for fractional implementations when that is the right fit. I discuss the impact on roles in more detail in this article: AI Is Changing the Minimum Viable Hire.
The newest generation of software is changing the calculation again by altering the amount of human effort required for certain kinds of knowledge work. There is already empirical evidence that the effect can be meaningful. Erik Brynjolfsson, Danielle Li, and Lindsey Raymond studied 5,179 customer-support workers and found that access to a generative software assistant increased productivity by 14 percent on average, with gains of 34 percent among novice and lower-skilled workers. They also found evidence that suggested that the technology helped disseminate practices associated with more capable workers. The full NBER paper, Generative AI at Work, is here.
A separate randomized experiment involving 758 consultants at the Boston Consulting Group found similarly significant effects on more complex knowledge work. On tasks inside what the researchers called the technological frontier, consultants using GPT-4 completed 12.2 percent more tasks, worked 25.1 percent faster, and produced substantially higher-quality results. But the benefits were not universal; on a task outside that frontier, workers using the system were less likely to reach the correct answer. The finding is useful precisely because it cuts against simplistic predictions of AI’s universal utility: the tools can clearly change productive capacity, but they do not improve every worker or every task in the same way. The published Organization Science paper is available here, with a Harvard Business School overview here.
One obvious conclusion from all of this is that some companies can get smaller. If individuals can produce more thanks to AI, and if organizations can obtain more of their infrastructure and specialized capabilities from outside the firm thanks to as-a-service models and fractional work, then fewer full-time staff may be required to accomplish something that once demanded a considerably larger organization. A founder can already access computing, payments, communications, accounting, analytics, distribution, specialized contractors, and a growing number of sophisticated software capabilities without recreating each one internally. In Coase’s terms, technology keeps moving the boundary between what is economical to organize inside the firm and what can be coordinated through the market.
There is a lot of conversation right now about this possibility, and much of it is probably correct. The minimum organizational size required to build something consequential appears to be decreasing in at least some parts of the economy. A small group today can possess technical capabilities, geographic reach, and productive capacity that would have been extraordinarily difficult for a comparably sized organization a generation ago.
But that is only one potential outcome of the application of this technology.
While greater efficiency through technology has been driving staff reductions since the industrial revolution, major technological transitions can also allow successful organizations to dramatically increase efficiency and output, even at a large scale.
A technology that makes one person more productive can also make ten thousand people more productive. A company with ten customers can apply a new capability across ten customer relationships; a company with hundreds of millions of customers can potentially apply it across hundreds of millions. An individual gains leverage from better software, but so does an organization that combines the same software with enormous capital resources, proprietary information, specialized expertise, existing customer relationships, and global distribution.
Technology can therefore compress some parts of the firm while magnifying others. It can reduce the number of people required to perform a particular function while making the organization capable of serving far more customers. It can eliminate the need to internalize one capability while creating enormous returns when investing in growth in another capability. It can allow a tiny competitor to challenge an incumbent while simultaneously giving an incumbent new ways to exploit the advantages it already possesses.
Coase’s framework is useful precisely because it does not predict that firms should become either large or small. The boundary of the firm depends on the relative costs of coordinating economic activity in different ways. If technology makes it cheaper to obtain expertise externally, companies may internalize less. If technology makes it cheaper to coordinate enormous numbers of activities inside one organization, companies may internalize more. If it reduces both kinds of costs at different rates, the resulting organizational structure may be very different across industries and even across functions inside the same company.
There is also an important distinction between productive capacity and economic value. Technology can make software, analysis, media, research, or other intellectual output dramatically cheaper to produce without making customers, attention, trust, capital, proprietary information, or distribution equally abundant. As one constraint becomes cheaper, another can become relatively more important. A small team may become capable of building something that once required hundreds of people while discovering that reaching a market remains extremely difficult. A company that already owns the route to that market may find the same technology disproportionately valuable.
That is why I am skeptical of confident claims about what the company of the future will look like. There is a perfectly coherent argument that organizations can become dramatically smaller, and the evidence that individuals and small groups are gaining productive leverage is real. There is also a fairly obvious historical observation that technological transitions repeatedly create new opportunities for companies to become extraordinarily large, and to get there faster than companies could before them. The current frontier technology race is displaying both forces at once.
Predictions are not really my business. Observations are. The observation, for now, is that technology is making it possible for very small groups of people to do things that once required large organizations while simultaneously allowing successful organizations to reach extraordinary scale at increasing speed. Coase gave us a way to understand why the boundaries of firms change when the costs of coordination change. Those costs are changing again. What that ultimately does to the shape of the company is a much more interesting question than pretending we already know the answer.
