In the coming age of artificial general intelligence, will a firm have one AI or many?
We should first be clear about what we mean by one AI or many. The question lies outside the usual concerns of technical architecture.
An AI can run in the cloud, locally, or across both. A system can serve one tenant or many. A single model can do the work, or many models and agents can do it together. We can keep an open mind about all of these choices.
What concerns us here is the relationship between AI and people.
By “many AIs,” we mean that different individuals have distinct AI agents acting on their behalf. Each agent represents a particular person’s intentions, acts under that person’s identity, and stays within the limits of the authority and resources entrusted to it. A person can delegate just some of what they wish to do and some of their authority to an agent.
At its most abstract, then, this is a question of agency. Does the system contain independent agents that act for different individuals, perhaps in only one part of each person’s affairs, with their own intentions to represent, authority to exercise, and resources to draw on?
If it does, and if those relationships remain substantially independent, we are looking at a world of many AIs. Conversely, any number of AI programs might operate as modules within a single system of execution. If they no longer need separate mandates from people with independent intentions, they count as one AI in the sense we are using here.
An AI can act on a person’s behalf. How many such relationships need to exist independently of one another? That is the number that matters here, however many programs are running.
At the level of society as a whole, the case for many AIs seems self-evident. Our society rests on the legal standing of natural persons. Each person has intentions, interests, rights, property, and responsibilities of their own. The arrival of AI does nothing in itself to dissolve those boundaries.
Inside a firm, though, we face a different question. The firm itself uses contracts to reorganize economic relationships between people. In the process, all the legal relationships involved are being reconstituted.
When someone joins a firm, their labor, resources, authority, and responsibilities are rearranged within its contractual and organizational framework. Economic activities that different people once pursued independently in the market can now take place inside a single firm and serve a common economic purpose.
So, inside the firm, should there be one AI or many?
I find that hard to answer today. We are in the middle of an AI revolution moving at extraordinary speed. How capable AI will become, and what firms might look like as a result, are still difficult questions to settle.
But we can begin with the organizing logic of the traditional firm and see where it takes us.
Suppose we think of the firm as we know it today primarily as an organization built for execution. On that view, most of what a firm does through its contracts, its organizational structure, and its culture is arranged around execution.
The firm brings different people together and coordinates what they want, the resources they command, and what they can do. Out of these it builds an economic machine. For most firms, the greater part of the work consists in using that machine to carry out a particular intention or put an established economic model into practice.
There is a familiar story about how such a firm comes into being. An innovator creates a new form of economic activity. It needs a machine to keep it running. So the innovator builds one. That machine is the firm.
Once established, the firm’s main task is to keep the activity going, replicate it, and expand it.
Of course, the people who work there still have minds of their own, and they still come up with innovations. The original organization, however, was built with another purpose foremost. Through coordination, competition, and sometimes internal politics, someone’s innovation might become a new product or line of business. Or that person might leave and start a firm of their own to pursue it.
In an organization built for execution, people’s independent intentions are therefore, to a considerable extent, something to coordinate and manage. The firm already has an economic purpose to fulfill. What other people want must somehow be reconciled with it, and the effort to do so can create friction.
If we accept this account of the traditional firm, the part organized around execution could, in principle, become a single AI.
Such a firm could still use any number of agents. Some might handle R&D, others finance or sales. How to divide the work among them and coordinate it could, in principle, become a technical question internal to one AI system. Each function would no longer need to correspond to a different person’s intentions and authority.
In an organization of this kind, some of the roles people play today may be transitional. In some cases, AI still lacks the intelligence or capability to do the work. In others, our institutions and working practices still assume that a person will be there to carry out a task or coordinate it with someone else.
As AI becomes more capable and we reorganize the relations of production and the ways we work together, forms of human participation that execution itself does not require could gradually fall away.
This is one principle behind the idea of the one-person company. The person supplies the intent. The AI, taken as a whole, serves as the machine for carrying it out and fulfilling its economic purpose.
Follow that reasoning a little further and we can imagine a future in which most firms have a single AI, while society as a whole runs through market exchange among one-person companies.
Is that how it will turn out?
I don’t know. But I want to raise an objection, and it has to do with the value of intent itself.
A traditional organization built for execution often treats people’s independent intentions as a source of friction to be managed. Could those intentions themselves be an important asset?
I start from a theoretical premise. People are the primary factor in innovation, and their role is irreducible. I will take that premise as given here.
If we accept it, another kind of firm becomes conceivable.
Could an economic organization in the age of AI evolve from a machine built around execution into one whose central purpose is to generate valuable intentions and turn them into economic value?
In more familiar terms, could it become a machine for innovation?
Here I want to draw a distinction between knowledge and intent.
Knowledge has more to do with the world as it is and with how we can go about producing things in that world. Intent points toward a world that has yet to exist. What do we want to create? What do we want to see happen, and what are we willing to act on?
Only some intentions lead to innovation. The ones that concern us here are those capable of giving rise to new economic activity and creating new economic value. These are what I mean by innovative intentions.
Why would we need an organization to generate them?
How intentions form depends a great deal on a person’s social relationships, economic circumstances, and experience of the world they live in.
The setting people find themselves in, the people they meet, the economic activities they take part in, the problems they face, and the resources and authority at their disposal all shape what they can imagine doing and come to want to do.
It is quite conceivable, then, that certain people, working in a particular setting, would be more likely to form valuable intentions of a particular kind. They might also be more likely to turn those intentions into something of real economic value.
Gathering people who already know what innovations they want to pursue is only part of the story. The very way people are brought together and organized shapes the intentions they are able to form.
An organization can carry an innovation into practice. It can also provide the environment in which someone first forms the intention to pursue it.
Of course, some companies have long produced one innovation after another. Google comes to mind. Incubators and venture studios, too, exist expressly to support innovation.
Even so, in the traditional firm, organizational design has seldom begun with the question of how to create an environment in which intentions of a particular kind are likely to form and acquire economic value.
AI gives us fresh grounds for asking that question.
If AI can do much of the work of execution, the structures firms have built around execution may take on a different economic significance. At the same time, the role an organization plays in the formation of innovative intentions, and in turning them into economic value, may become increasingly important.
Such a firm still has to consider how to achieve the economic goals it has already set. But it faces a more fundamental set of questions. Which people should it bring together? What kind of environment should it create for them? What relationships and institutions would make a particular kind of valuable innovation more likely to occur, and more likely to yield economic value?
Suppose this “economy of intent” has economies of scale. That is, suppose that organizing people in a particular way allows them to create more economic value than they could by working independently. Those economies of scale could themselves give the firm a new economic reason to exist.
How far those gains extend, and at what point the costs of organizing people offset them, could in turn help determine the boundaries of the firm.
This is what I mean by an organization built around intent in the age of AI.
Such firms would still have to carry out the work of production. And there is every reason to expect them to differ. Different kinds of innovation may call for different combinations of people, different working environments, and different institutional arrangements.
The central question of organizational design may shift. Firms have traditionally asked how to carry out a given productive task as effectively as possible. Increasingly, they may ask how best to help intentions for a particular kind of innovation take shape and become a new economic reality.
We can now return to the question of one AI or many inside a firm.
If all the firm needs to do is carry out an intention that has already been settled, the work of its various AIs can, in principle, be brought together in a single system of execution.
But suppose an important part of the firm’s economic value comes from different people continually forming new intentions, from the way those intentions influence one another, and from the choices people make about which to pursue or combine. In that case, preserving an independent relationship between each person and the AI that acts for them may give the organization something it could obtain in no other way.
A personal AI would then help a person do more than get their work done. It would be a means by which they could form an intention, give it expression, and bring it into economic reality through the organization.
Cooperation among personal AIs would have a significance well beyond the technical problem of assigning tasks. It could become one of the basic mechanisms through which a firm organizes innovation.
All of this, of course, is a hypothesis put forward in the midst of the AI revolution. How far the firms of the future will come to organize themselves around intent remains an open question.
But if there is something to the argument we have been making, then we can ask a further question.
How should a firm organize itself, and in particular how should it design the relationships between people and their AIs (which, in the sense developed here, means to a large extent the relationships among multiple AIs), so that high-value intentions can take shape and be turned into economic value more quickly?