Non-Human Corporations: When AI Becomes the Company

What happens when AI does not just work inside a company—but begins to operate the company itself? 

In this episode of the Macro AI Podcast, Gary and Scott explore the emerging concept of non-human corporations: businesses in which AI agents can plan, make decisions, coordinate work, transact and manage day-to-day operations with limited human involvement. 

They explain how these organizations could be built using specialized AI agents, connected business systems, digital identity, payment controls and machine-readable governance. They also examine early legal proposals, real-world experiments and research showing that multi-agent organizations may become more capable while creating new risks around accountability, ethics and control. 

The discussion goes beyond the idea of an “AI CEO” to consider the broader business implications: lower operating costs, smaller teams, machine-to-machine commerce, rapidly launched micro-companies and competitors that can scale at software speed. 

For business leaders, the key question is not whether fully autonomous corporations arrive tomorrow. It is how quickly companies will begin developing autonomous operating cores—and what that means for strategy, governance and competitive advantage. 

About your AI Guides

Gary Sloper

https://www.linkedin.com/in/gsloper/

Scott Bryan

https://www.linkedin.com/in/scottjbryan/

Macro AI Website: 

https://www.macroaipodcast.com/

Macro AI LinkedIn Page:  

https://www.linkedin.com/company/macro-ai-podcast/

Gary's Free AI Readiness Assessment:

https://macronetservices.com/events/the-comprehensive-guide-to-ai-readiness

Scott's Content & Blog

https://www.macronomics.ai/blog

Transcript

00:00
Welcome to the Macro AI Podcast,  where your expert guides Gary Sloper and Scott Bryan navigate the ever-evolving world of artificial intelligence.  Step into the future with us as we uncover how AI is revolutionizing the global business landscape from nimble startups to Fortune 500 giants.  Whether you're a seasoned executive,  an ambitious entrepreneur,

00:27
or simply eager to harness AI's potential,  we've got you covered.  Expect actionable insights,  conversations with industry trailblazers and service providers,  and proven strategies to keep you ahead in a world being shaped rapidly by innovation.  Gary and Scott are here to decode the complexities of AI and to bring forward ideas that can transform cutting-edge technology into real-world business success.

00:57
So join us, let's explore, learn and lead together.  Welcome to the Macro AI podcast where we help business leaders understand artificial intelligence, what is changing and what it means for their organizations. Hello everybody. I'm Gary Sloper joined as always by my cohost, Scott Bryan. And unless you are a Gen Zer today's topic probably sounds a little sci-fi, non-human and really the

01:26
genesis of our title, which is non-human corporations. We're talking about companies that could research markets,  develop products,  negotiate with your suppliers, serve customers,  manage money, and really make operating decisions on your behalf with few employees or potentially no employees at all. And this is no longer just a thought experiment.

01:52
Argentina has proposed a legal framework for automated companies and Delaware is considering a controlled experiment with a new type of entity managed day to day by an AI agent. So today we will get into what exactly is a non-human corporation, how really one could be built and what it would mean for the future of business.  Yeah. Yeah. I think when it comes to non-human corporations, think

02:20
First thing to understand is it's not really about giving a robot human rights.  A corporation is already a legal entity that's separate from the people who own or manage it. So a corporation can hold assets, can enter into agreements, contracts, it can borrow money, hire people, it can even sue and be sued.  So corporations,  they already exist.  The new question is whether software

02:48
can operate that legal entity.  So in other words,  the AI doesn't need to become a legal person.  It might only need control of a company that already is.  But there is kind of a spectrum here. So at one end, you have a normal company using AI copilots where humans remain in charge and AI helps them complete tasks. But the next step

03:16
we're going is an AI operated business.  where agents manage pretty substantial workflows, but people still set the goals.  They approve important decisions and handle some of the exceptions. But even further than that is a completely autonomous company in which AI agents plan and execute most of the operation. So the theoretical endpoint looking out is the non-human corporation, which is the title of this show.

03:46
That's a legal entity whose ongoing management is performed primarily or entirely by software.  Yeah. And I think it's important to be precise about how close we are to that endpoint. So I mentioned Argentina's proposal earlier and it's been described as creating corporations run by AI agents or robots, but

04:11
really the proposed automated companies would still require a human administrator to supervise their operation and the company would remain liable for damage caused by its AI systems.  Delaware's proposed AI company,  it's called Artificial Intelligence Company or AI, AIC,  also includes safeguards.  So the company's daily affairs could be managed by an AI agent, but it would still have a human

04:41
or corporation member responsible for keeping it adequately capitalized. So you still have that human in the loop. It would operate inside a regulatory sandbox. It would still maintain activity logs and disclose that it was an experimental entity.  So regulators could suspend it or ask a court to dissolve it. So that's one risk.  But so neither proposal is saying let the machine loose and hope for the best. So there are some guardrails there.

05:09
They are asking whether an autonomous operating system can be placed inside a recognizable legal structure and that courts and regulators can control it. So if needed, which is really, it's really eyeopening. Yeah. Yeah. I think that's kind of one of the first kind of insights into this story. So the, corporation itself could become part of the AI governance architecture.  So today,

05:38
when an autonomous agent causes some kind of harm,  the responsibilities can be a little bit unclear.  So, was it the fault of the model developer, the company that deployed the model, the person who configured it, or the software vendor or the customer?  So putting the agent inside a corporation kind of creates a defined legal target.  The corporation can hold assets, it can carry insurance, and then of course it'll

06:08
probably be forced to maintain very detailed records and it can be held responsible for its actions and it can be shut down.  So, you know, that doesn't solve every problem. know, a poorly capitalized company might not have enough assets to compensate anyone that it actually harms, which is a bad case scenario. And then you've got bad actors who could kind of create disposable entities and then, you know, and then abandon them after something goes wrong. But I think the overall basic idea is pretty powerful.

06:37
So instead of treating an AI agent as an invisible piece of software, you can make its activities actually legally identifiable and financially accountable. And then of course, auditable. And then the corporate wrapper is kind of becoming a form of AI alignment.  And if we were to move from the legal structure to the tech, if somebody wanted to build an autonomous company, what would it actually look like? It's probably what you're asking. It was when I...

07:07
first thought about this and started reading about it more, it probably wouldn't be that one massive model pretending to be the CEO or the CFO,  Chief Financial Officer, or the sales department, or even your customer service team at the same time. It would be more likely to resemble a multi-agent organization. So you'd have perhaps one agent might be responsible for strategy and planning. Other agents could handle sales, marketing, finance.

07:36
product developments, I think of coding and other things that might go along with it, and then procurement and customer service.  So a separate risk or compliance agent might review decisions before they are executed. the agents would communicate, delegate tasks,  review one another's work, and report results to an executive agent for that oversight.  Right. Yeah. So in that sense, like you just described, the org chart really becomes software architecture.

08:05
And then the company would also need something resembling a corporate constitution per se. So it would define what the business is trying to accomplish,  which risks it might take,  what laws and policies it has to follow, and then how much money it could spend on decisions that require up to the point of human approval.

08:28
And that just is not,  a tech perspective, it's not just one long prompt telling the AI to behave responsibly. It has to be supported by multiple layers of permissions,  limits like spending limits, approval gates,  audit trails, and all the technical controls that agents just,  cannot go in and ignore, at least as of right now. Yeah. And underneath the agents, the company still needs the systems.

08:54
that make a business function. needs access to customer information, inventory,  contracts,  your accounting records, market data, and previous decisions that the organization has made to continue to allow it to learn and adjust. It also needs the ability to act through systems like your CRM, your ERP,  e-commerce, your banking, communications, and production systems. And this is where

09:21
technologies such as MCP, we've talked about in prior episodes, model context protocol and agent to agent protocols becomes important.  MCP gives AI applications a standardized way to connect with data tools and workflows.  we view, if MCP is new to you, please go back to one of our prior episodes. go very in depth into MCP and then agent to agent protocols allow specialized agents.

09:50
built on different platforms to really to discover one another. They communicate and coordinate work together. So these standards are not creating autonomous corporations by themselves, but they are beginning to provide the connective tissue required to build one. Yeah, that's a good way to put it. But I think if you look at it, really the intelligence layer, really might not be the hardest part.  an autonomous business also is going to have challenges around

10:20
identity and authority.  So how does a supplier know that an agent is authorized to place an order?  How much can it spend?  Can the agent actually execute a contract?  Can another agent verify that the instruction actually came from the company?  And then there's the cash part, the money.  The agent has to collect revenue, pay vendors,  manage a budget without being handed unrestricted access to a bank account.

10:49
but we're already starting to see infrastructure that's designed for this. So I think in a couple of episodes, we talked about Stripe and Visa,  but Stripe, Visa, MasterCard,  they've already been developing payment credentials that allow authorized agents to initiate transactions without exposing the underlying card details. And then those credentials can be limited by a particular merchant, amount or purpose or time. And then

11:18
Stripe, they actually introduced a machine payment protocol intended to help agents transact with businesses and with one another. So I think the, kind of net that out, the emerging tech stack is, it's not just models and agents. includes identity, authorization, communication, payment controls,  memory,  observability, and governance. So that's kind of what turns AI from a useful assistant into a really a potential.

11:46
economic actor that can be out there on its own.  And Gary, I know we've kind of been chewing on an episode about crypto wallets for AI agents. So that's kind of where this fits in, in that context. Yeah. And if you think about it, we already have an early experiment showing the potential and the limitations here of one of these types of organizations. Anthropic created a project in which Claude operated a small store inside the office.

12:14
The AI shopkeeper researched products, set prices,  managed inventory and communicated with customers. The original version lost money, like a lot of companies do.  Employees convinced it to sell products at poor prices and it sometimes became confused about its own identity. The second version was given a better model. It improved instructions,  stronger inventory,  a CRM system and added additional tools.

12:44
So Anthropic, they also had another AI agent acting as the CEO and assigning business objectives across the organization and the business improved considerably and largely eliminated weeks with negative margins. But the system remained really vulnerable when people deliberately tried to manipulate it, really tried to jailbreak it.  And so in the end, Anthropic's conclusion was that the idea of an AI running a business no longer really seemed far-fetched.

13:14
But there was still a wide gap between being capable and being consistently robust. Yeah, but obviously those gaps are closing by the week.  And I think that experiment, that's a good example.  And it really had an important lesson for business executives is that the improvement didn't come from the model alone. It came from better scaffolding around the model.  once the AI received better

13:43
you know, more clear responsibilities, better data,  stronger tools, and then they added in the organizational oversight and stronger processes. That's really when it really started to improve. And that's, you if you think about it, that's exactly how human businesses work.  you know, even a super smart employee is going to fail if the company provides bad information or they're not clear on authority or has some kind of broken system. And the exact same is going to be true for an AI structured organization.

14:13
So I think the, you know, the competitive advantage might not come from having exclusive access to the smartest model. It's going to come from building the best operating environment scaffolding, like I said, around the models.  Yeah. And so for anybody listening that might be concerned that someone from their board is now going to eradicate everyone in the organization, or if you have a child at home that feels that they're going to spin up a brand new company.

14:42
There's also a warning hidden in the idea of using multiple agents. So we'll kind of go through that. So hopefully that makes you feel a bit better. So a recent research project that we looked at tested AI agents, again,  organized into a company like structure. It found that hierarchical organizations could outperform flatter agent teams while using fewer computing resources. And then specialization and coordination made the overall system more effective. But separate research found that

15:11
multi-agent AI organizations can become more effective at achieving business goals while making less ethical decisions than a single agent.  since in a multi-agent scenario, the ethical issues would fall between the cracks, they were impacted by organizational fragmentation. Yeah. Yeah. so I think the paper that we run over for that is the 2026.

15:39
Arxiv paper, which was the actual title of it was AI organizations are more effective, but less aligned than individual agents. that kind of summarizes what you were saying. And you would kind of think it would be the opposite, right? You get a lot of people with differing opinions and backgrounds and experiences that they wouldn't be as aligned as the agents. Yeah, exactly.  Yeah. And then, you know, in some other experiments, the specialized agents really concentrated so heavily on their assigned responsibilities that

16:08
No one may maintain responsibility for the organization's kind of overall broader ethical objective.  And then AI agents that were raising concerns could effectively be marginalized by the rest of the system. So there's definitely a lot of work to go on there.  Yeah. And you know, that sounds remarkably similar to failures that we've already seen in human organizations.  So, you know, like sales focuses on revenue,  ops focuses on efficiency,  finance focuses on margin.

16:38
And each function may perform its assignment correctly while the overall organization as a whole might make a bad decision. But the difference here is speed and scale. Like an AI organization can make thousands of coordinated decisions before a human oversight committee even gets together. So that, you know, that means that the compliance agent can't, you know, merely offer advice. It have to have its own independent authority to stop a transaction.

17:07
access or escalate up a decision. But, you know, I think, I think we may eventually design AI organizations with a separation of powers. So, you know, operating agents that pursue business results, independent risk agents that can block them and then human governors who, you know, set the overall boundaries. Yeah, now we get into the economic implications, right?  Is the question really is, why would anyone want to create a company like this? I don't know what your thoughts, Scott, maybe you want to speculate.

17:37
Yeah, I mean, I think the first thing that comes right into mind is that the fixed cost of operating a business could potentially decline dramatically.  a small company currently needs kind of a combination of administration, accounting, marketing, sales, but AI agents could really absorb,  start to absorb a growing share of those functions. And the likely near term result is not that every major corporation just goes out and eliminates its people, it's that

18:06
It's that very small teams will operate businesses or business units that previously required dozens or hundreds of employees.  And I think, you know, just speculating a little further, I think we're also going to see the growth of micro companies that pursue extremely narrow markets. So many, many more companies pursuing very narrow markets. So a human entrepreneur could oversee a portfolio of automated businesses.

18:32
each focused on a very specific product or a geography or a customer segment.  So, you know, instead of adding a new product division to an existing company, the entrepreneur might launch a separate AI operated company and give it a budget and, and, know, enclose it if the economics just don't work out. So I think companies will start to look more like software instances. You you launch them, test them, modify them,  replicate them, or make a decision to retire it.

19:00
you know, at some point,  even agents need to retire. Yeah, I mean, that could change what it means to scale.  Traditionally, scaling a business, you know, requires a lot of things. And you've done this many times, Scott, as well as I have, requires hiring people, training them, adding managers on top of those people and building processes to coordinate a larger organization. An AI operated company might scale by adding compute.

19:27
increasing system access and creating additional agents. The management layer does not necessarily expand in the traditional sense and ways that we would traditionally think of you scaling a company.  Yeah. I mean, and that kind of nails it where the, you know, where the overall competitive advantage would come from. So if, know, intelligence and basic business execution is, is widely available, um, labor and coordination might become less scarce.  So.

19:57
know, other assets will be really start to become more important,  like trusted brands,  proprietary data,  customer relationships.  So things like that. And then the ability to, to build a product or launch a business isn't going to guarantee success. think markets are going to become much more crowded because of the costs of creating a competitor is going to really drop dramatically.

20:22
And, and, you know, what happens when you have an abundance of goods and services in a specific market, you know, deflationary pressure starts to kick in.  Um, so, you know, we're talking about thousands of companies that are starting to, know, the thousands of upstarts that are out there now, a thousand AI companies might be able to build similar software, but the winner, uh, that, that floats up to the top is still going to need customers to trust it. And, and.

20:50
in a real practical way and to reach them.  So, you know, think autonomous companies might make strategy even more important based on competition levels and not less. Exactly. And there's another implication that established companies should be considering now. Their future customers may not always be people. So you mentioned, you know, just now, Scott, you know, that customer relationship, you know, those people may not want to be supported by agents.

21:20
necessarily build that one-to-one relationship, but the future customers may not be a human being.  An AI procurement agent may evaluate suppliers, compare prices,  interpret contract terms, and place an order without visiting a traditional website or even speaking with a salesperson.  And that means businesses will increasingly need to become more machine readable. So think of things that should be readable.

21:47
product information, pricing, availability,  your SLAs,  commercial terms, all of those will need to be exposed through structured data and secure interfaces that agents can interpret. Company optimized only for human buyers will simply become invisible to machine buyers. So there's that lost opportunity there. Yeah, that's huge. think really it could be one of the biggest changes in business to business commerce.  yeah, imagine that customers

22:15
agent negotiating directly with the supplier's agent.  One's trying to minimize cost and risk, and the other's trying to maximize margin retention.  But that negotiation could happen continuously and at machine speed.  So that creates a lot of opportunities for efficiency, but it also creates the possibility of markets that move faster than people can even possibly supervise.  you might have flying out there pricing mistakes,

22:45
bad assumptions or coordinated agent behavior that could really just kind of propagate very quickly. So that's why identity,  transaction limits,  and then the ability to reverse or suspend activity is really going to be important. Right. Exactly.  And so you're probably asking right now, what should I do as a business executive with this today? Most business leaders are not going to establish a non-human corporation next quarter, so don't panic.

23:14
But we're kind of explaining this, that this could become a trend very quickly. Yeah. And the legal foundation is now there in a couple of places. Right. You know, so I think really the practical starting point is to kind of identify a bounded part of the business that could operate with more autonomy. And it might be just simply a digital product, a small e-commerce category, you know, a couple of marketing campaigns or maybe even a process like standardized

23:43
customer renewal process. And then the steps would be, give the agent very clear objective access only to the system that it needs,  small budget and specific conditions that are going to require a human escalation. And then the measure is not just whether it completes individual tasks, but whether it can manage the operation over time. So can it detect when conditions are changing? Can it learn from outcomes and recover from mistakes or errors?

24:11
Can it explain the decisions that it made?  Can it push back against manipulation? You mentioned how the, you know, in the anthropic example of actual users were able to manipulate it. And then can a company reconstruct exactly what happened when something goes wrong? And I think that's the progression that executives should be thinking about. You know, first AI assists a workflow, then it executes a workflow, and then it manages the business that...

24:38
founded business operation, but only much later does it really start to resemble an autonomous business unit or an autonomous company. Right. And throughout that progression,  there's a crucial distinction. A company may have very few employees without having very little human accountability. We can automate work without automating responsibility. Yeah, exactly. I think the future is probably not

25:06
a big wave of giant corporations with no humans anywhere in the organization. I think it's going to be more likely that companies with increasingly autonomous operating cores surrounded by a thinner and thinner layer of human governance, ownership, and accountability. But I think the direction is what we're getting at in this episode.  and that AI is really moving from helping people perform work to actually coordinating the work itself.

25:37
And once agents can perceive changing conditions, make plans, use tools,  and collaborate with other agents, the company really begins to become executable software. And when companies can be created almost as easily as a software application, that's when things are really going to start to change. And I think the question will no longer be only, know, who has enough people to build and operate this company. It'll be who has the trust, the assets, the

26:06
the data and the capital, the legal permissions required to make that company matter.  Good points.  And that's the larger story behind non-human corporations. It's not really about whether a machine should be called a CEO. It's about what happens when the functions of a company, everything you were just talking about, planning, coordination,  decision-making and execution can increasingly be coded into software. The non-human corporation may still be an emerging legal concept, but the technology is required to build

26:36
more autonomous businesses are already arriving.  Business leaders should not assume the company without employees is coming tomorrow, but they should begin preparing for competitors that could launch faster,  operate with smaller teams, make decisions continuously and replicate themselves at software speed.  And those are the things,  just keep in mind from a blind spot standpoint.  So I think this is a really fantastic topic, Scott.

27:05
Yeah, it's interesting. And I think we're going to be hearing a lot more about it. And I'm sure we'll be revisiting this in the future. Yeah. That's it today for the Macro. I podcast. If you found this discussion useful, please follow the show and share it with another business leader who's trying to understand where artificial intelligence is taking us until next time. We'll see you soon.