The Floor of the Next Economy

The agentic economy is happening. Nobody serious is debating that anymore. What almost nobody is debating yet is the more important question, the one being answered right now by default, in code, by the teams shipping the protocols the next decade will run on: who owns the data the agents run on.

This is not a question for policy to handle later. It is a question being answered now, whether or not we are paying attention, in the engineering choices being made at a handful of companies and labs over the next twenty-four months. By the time the answer is visible it will already be locked in, because the architectural decisions that determine it compound fast and are hard to reverse.

Most of the writing about the agent transition is about what agents will do. Jobs they will replace. Industries they will reshape. Capabilities they will reach. These are real questions and they get most of the attention. They are not the question that matters most. The question that matters most is who controls the data the agents need to do their work, and the answer to that question is being decided at a layer almost nobody is writing about.

What platforms actually own

Every platform you have ever paid a subscription to has one real asset. It is not the software. Software is cheap and getting cheaper. It is not the brand. Brands come and go. It is the data. More specifically, it is the data and the connections between the data, the accumulated history of what happened when, who did it, what changed, how it related to everything else that was happening at the same time. The software is the surface. The connected data is the asset.

This is the thing that makes net revenue retention go up and to the right. It is the thing that makes migrations fail. It is the thing that makes every renewal negotiation tilt the way it does. You are not paying for the features; you are paying to keep access to the shape of your own operational history, which the platform holds and you do not. The platform's pricing power is exactly equal to how hard it would be to rebuild that shape somewhere else. For most real businesses running on modern software, that number is effectively infinity.

This arrangement was tolerable, even productive, for most of the last twenty years. The platforms built good software. They gave us interoperability and distribution at scales we could not have reached otherwise. The cost of the arrangement was the growing concentration of operational data in a small number of private databases owned by a small number of companies. That cost was mostly invisible, because the alternative was worse, and because the data was not doing much besides sitting there being queried by humans.

Agents change what the data is worth.

Why agents make the concentration worse

When a human queries the data in their own silos, the data is valuable because it helps the human make a decision. When an agent queries the data, the data is valuable because it enables the agent to take an action that produces surplus. The first is a passive good. The second is a productive asset. The same database that was inert under human use becomes a factor of production under agent use. Every silo the agents need to operate against is a toll booth on a road that was not profitable before and suddenly is.

This is the structural move almost nobody is naming clearly. The platforms are not racing to build agents. They are racing to make sure that when agents arrive, their silos are the places agents have to go through. Every new API surface, every MCP server, every "AI integration" you see a major SaaS company ship is an extension of their existing data moat into the new medium. They are not competing with the agents. They are positioning themselves as the necessary intermediary between the agents and the data the agents need.

If this strategy works for them at the scale they are betting on, the endgame is not hard to describe. A handful of platforms own the connected operational data of the global economy. Every agent that wants to do commercially consequential work has to go through them. Every transaction that agents mediate pays a toll to the platforms holding the relevant data. The platforms do not need to build good agents themselves, or even good software anymore. They need to own the data the agents cannot function without. That is enough to extract an enormous share of the surplus the agent transition produces, indefinitely.

This is a multi-corporation data monopoly. It is not the result of any single company being villainous. It is the compound interest of an arrangement that already existed, newly valuable because agents made the data productive in a way it was not before. It is the default outcome if nothing changes.

The alternative

The alternative is not "agents get regulated" or "platforms get broken up" or "smart contracts distribute profits fairly." These are downstream responses that arrive after the concentration has already happened, and they work on what the concentration allows. The alternative is upstream of all of them. It is a different answer to the question of who owns the data in the first place.

In that alternative, data is owned by the entities whose activity produced it. Companies own their own operational data. People own their own personal data. The platforms are clients of that data under granted capabilities, not custodians of it by default. The connections between data, which is where most of the value lives, run on open protocols rather than private systems. Agents operate on this distributed landscape by receiving bounded, revocable grants from the actual owners of what they need to read. No single actor controls the connective tissue, because the connective tissue is a commons rather than a product.

This is not a business-model preference. It is an architectural property. Either the rails the agents run on treat data as sovereign to its owner and capabilities as bounded grants from that owner, or they do not. If they do, the data monopoly does not form, because there is nothing to monopolize. If they do not, the monopoly forms by default, regardless of what anyone later tries to legislate.

The specific unlocks of this alternative (faster transactions, lower friction, fairer competition, new categories of business that are not currently possible) are real, and later pieces in this series will look at them up close. But the unlocks are downstream of the architectural choice. The choice itself is the thing. It is being made right now. It is being made by a small number of people who are aware that it is the choice, and a much larger number of people who are making it implicitly by shipping into the existing shape.

A design question, not a debate

The fork between the two worlds is not a philosophical disagreement to resolve through argument. It is a design question being answered through implementation. The primitives required to build the second world have matured in the last two or three years. Cryptographic identity is no longer a research paper. Capability-based access control has working implementations. Content-addressed systems have real tooling. Open agent protocols have a standard that is stabilizing. Each of these was theoretical a decade ago. All of them are production-ready now.

The composition of them into something coherent is what decides which world we end up in. That composition is not yet obvious. It is obvious to a small number of people building at this intersection. It is not obvious to the platforms, who are busy extending their existing moats to cover agents. It is not obvious to the labs, who are focused on capability. It is not obvious to the policy conversation, which is working on the previous problem.

In twelve to twenty-four months, the composition will be obvious to everyone. By then the teams that moved early will have either built something that shapes the category or failed to. The choice between the two worlds will have been made, in rough outline, by whoever shipped first with enough conviction to commit to a specific shape. Everyone else will be fitting into what they built.

Where the visible conversation is

The visible work on the agent transition sits on three pillars. The labs are building better agents. They are racing on capability, alignment, and deployment. The crypto world is building settlement rails. They are racing on throughput, stablecoin adoption, and on-chain primitives for programmable money. The platforms are extending their data moats to cover agent use, positioning themselves as the places agents have to pay to go through. None of these pillars, on its own, addresses the data-ownership question, which is the question underneath all three.

The labs produce agents that are more capable, but capability that operates against silos owned by other companies is capability that pays rent to those companies. The crypto teams produce settlement, but settlement without context just moves value faster along whatever paths the incumbents define. The platforms produce integration, but integration on their terms is the monopoly by construction.

The layer nobody is building publicly is the one where data ownership is distributed to its rightful owners, capabilities replace platform mediation, and connections between data run on open standards rather than private ones. That layer is the connective tissue the other three pillars need in order to compose into something other than the default. Without it, the three pillars compose into the multi-corporation data monopoly. With it, they compose into something different.

This is not an accusation. The teams building agents are doing agent work, the teams building settlement are doing settlement work, the teams building platforms are doing what platforms do. Each is rational given its position. The observation is just that none of them is, by job description, building the data-ownership layer. That layer is being left to emerge on its own. Layers left to emerge on their own tend to emerge in whatever shape is most profitable for whoever is closest to them, which is usually the platforms.

The default

If nothing changes, here is what happens.

The agents get better on schedule. They start doing commercially consequential work. To do that work they need data, and the data lives in the silos. The platforms that own the silos expose API surfaces that let agents operate within their walls, charging for access and taking a percentage of the activity that flows through them. Agents that want to operate across silos have to negotiate with each platform individually, accept the terms each one offers, and route their activity through whichever paths the platforms have approved.

The productivity gains of the transition compound. Almost none of them flow to the workers whose jobs are being displaced, the small businesses whose operations are being automated, or the customers on whose behalf the agents are acting. They flow to the handful of companies that ended up holding the data the agents could not operate without. Policy tries to redistribute after the fact and struggles, because the rails were not built for redistribution and can only approximate it from outside. The concentration of wealth from the agent transition tracks the concentration of data ownership that preceded it, and exceeds it by a wide margin, because agents made the data newly valuable in a way that compounds.

This is the default. It is not the result of anyone being villainous. It is the result of letting the default happen.

What is being decided

The floor of the next economy is being poured right now. It is being poured in the engineering decisions being made at a small number of companies, and in the protocols they commit to, and in the shape of the rails they build. Everything that runs on top of that floor will run on the rails the floor provides. If the rails distribute data ownership to its rightful owners, the monopoly does not form, the surplus from the transition flows back through the people whose activity it displaces and reshapes, and the future remains theirs to shape. If the rails route agent activity through the existing silos, the monopoly forms by default, the surplus accumulates with the few, and the future gets shaped around the people it most affects rather than by them.

There is a window to influence which one we get. The window is open because the primitives required for the alternative exist but have not yet composed into something commercial. It is closing because the default is compounding every week. The window is not open forever.

I have been writing about the visible conversation because it is visible. The more important conversation is the one almost nobody is having yet, which is the conversation about who owns the data the agents will run on. That is the conversation I care about, and it is the one I want to pull more people into, because the answer we arrive at determines whether the surplus from the transition flows back to the people whose lives it reshapes or away from them, and whether the future they face is one they can still write.

The next piece in this series looks at the pain the data-ownership concentration is already causing, even before agents arrive. The piece after that looks at what distributed data ownership has to look like technically. The piece after that describes what is being built.

But the stakes of the whole series are here, in this piece, and they are simple enough to state in one sentence. We are pouring the floor of the next economy right now, the choice is between distributed data ownership and a multi-corporation data monopoly, and the question is being answered by default unless someone chooses otherwise.

I want someone to choose otherwise. That is why I am writing this.