In the age of AI Agents, where is crypto value flowing?

Original Title: Who Makes Money from Agents?

Original Author: Jonah Burian

Translation: Peggy

Editor's Note: If agents are truly set to become the next billion users of blockchain, the more important question might not be "how much volume will they bring," but rather, if this reality comes to pass, who will make money?

In the past, both the "Fat Protocols" and "Fat Applications" theories implicitly assumed that on-chain users were human. Humans care about interface usability, brand trustworthiness, and convenient pathways. Consequently, the application layer could capture value by owning the user entry points and transaction flow. But agents are different. They call APIs directly, have no brand loyalty, and can switch between different protocols, aggregators, and venues at negligible cost.

This means agents could rewrite the value distribution logic of Web3. The application layer could pivot to a "headless" model, opening up its wallets, aggregators, and on/off-ramp capabilities as agent-oriented APIs. The protocol layer might also regain opportunities as agents bypass intermediary layers. However, a more radical scenario is that agents could push the entire on-chain stack toward price competition, compressing the profit margins of applications, aggregators, and infrastructure down to near marginal cost.

What's truly worth watching is not just that agents will make existing on-chain transactions higher frequency, but that they could create entirely new activities that weren't viable before: continuous portfolio rebalancing, machine-to-machine payments, and new types of markets that only make sense with automated, high-speed execution.

Therefore, the core question of the agent era isn't simply determining whether value will flow to protocols or applications. It's about figuring out who can make an agent choose to return here, even when it has infinite alternatives. The answer may no longer be UX and brand, but rather liquidity, latency, settlement finality, or some new business model that hasn't been named yet.

Below is the original text:

Many imagine that agents will become the next billion users of blockchain. But few ask the second-layer question: If this world truly arrives, who makes money?

All previous theories about value capture in the crypto industry assumed users were human. The "Fat Protocols" theory argued that the protocol layer is best at monetizing users. Meanwhile, the "Fat Applications" theory, which my colleagues and I proposed in "How to Capture Value" and "The Great Repricing," argued the application layer does this better.

But agents change who the "user" is. Consequently, existing theories of value capture are no longer reliable.

The 'Fat Protocols' Theory

In 2016, @jmonegro wrote "Fat Protocols." For nearly a decade following, this article became the dominant theory of value capture in the crypto industry.

Its core argument was: In the internet era, value primarily flowed to the application layer, like @Google and @facebook, while underlying protocols like TCP/IP and HTTP captured almost no value. But crypto would be the opposite. Blockchain data is open and shared, so applications would become commoditized. Meanwhile, the protocol token needed to use the network would capture speculative value as usage grew. Every successful application would drive demand for the token. Ultimately, the protocol layer would compound value faster than any application built on top of it.

For a long time, this seemed correct. Bitcoin and Ethereum's market caps exceeded those of any company built on them. This model worked because, at the time, the protocol layer was scarce, expensive, and hard to replace. Bitcoin and Ethereum in 2017 were indeed scarce; there weren't over a dozen general-purpose L1s vying for the same workload yet. Block space was scarce enough that holding the base asset felt like owning a piece of every application needing that network.

Now, every layer of the infrastructure stack has credible alternatives: multiple high-throughput L1s, dozens of L2s, and modular settlement and data availability layers competing on price. Block space has gone from scarce to abundant. Switching costs have plummeted as bridges and aggregators make the underlying chain nearly invisible to users. Infrastructure has become substitutable, and substitutable things ultimately compete only on price. Consequently, the protocol layer's pricing power has evaporated along with its scarcity.

The 'Fat Applications' Theory

By 2026, the entities capturing significant economic value aren't protocols, but applications like @phantom, @coinbase, @Polymarket, @Pumpfun, and others.

In my view, the reason is that the most valuable asset in crypto is user relationships. If you control the user interface and transaction flow, you control distribution. And whenever a user touches an on-chain product, you can nearly always monetize it: swaps, lending, staking, minting, on/off-ramps, and so on. This is likely why investors are so obsessed with neobanks.

Applications also push infrastructure into pure price competition, compressing infrastructure profit margins to near marginal cost. I documented this strategy in "How to Capture Value." The same dynamic is playing out in stablecoins, which I've also discussed in another article.

Prices are reflecting this theory. Spencer and I call this shift "The Great Repricing": in this cycle, value flows to the layer that owns the user relationship.

Why Agents Break This

The "Fat Applications" theory implicitly assumes users are human, and humans value user experience, brand, and convenience. Agents do not. They call APIs directly, have no brand loyalty, and can switch venues at zero cost.

When the user becomes software, owning the user relationship is no longer as defensible. The moat around the frontend, which the entire "Fat Applications" theory relies on, gets devalued.

So, who captures value in the age of agents?

Applications Go Headless

One possibility is that the application layer winners will continue to be winners, but they will abandon the UI.

Wallets and aggregators have already built the hardest parts: integration with numerous protocols, routing logic, identity, and on/off-ramp infrastructure. The natural next step is to open this capability as agent-oriented APIs. Agents would route through them, just as human users today might trade through @phantom or @JupiterExchange.

In this world, the "Fat Applications" theory holds, but loses its frontend. The companies that win in the human era will re-platform into headless infrastructure. We already see traditional SaaS companies like Salesforce pivoting in this direction.

Protocols Rise Again

Another possibility is that agents skip the middle layer entirely.

If integration is simple enough—clear API documentation, standardized RPCs, predictable execution semantics—there's little reason for an agent to pay an aggregator to do what it can do itself.

The aggregator's advantage in the human era came from UX and complex routing capabilities. Agents don't need UX, and routing itself is a solvable engineering problem that agents are getting increasingly good at.

If this is the future, then "Fat Protocols" theory gets a second life.

Pricing Power Collapses Across the Stack

Another possibility is that agents impose commoditization pressure across the entire stack.

They are perfectly rational. They will choose the cheapest venue every time, with zero loyalty and no friction. Applications lose the UX premium they used to charge human users. Aggregators and infrastructure lose pricing power because there's no human inertia to shield them from price competition.

In this scenario, it's hard for any layer in the stack to capture much value. The entire supply chain gets compressed to near marginal cost, and the economic surplus flows either to the owner of the agents or to the end users the agents represent. Crypto becomes a utility, and utilities are generally not easy places to make money.

Agents Create New, Previously Infeasible Activities

The simple version of this argument is: agents will do what humans do, just at higher throughput. Even if margins are compressed, the total pie could grow significantly larger due to massive volume increases.

But there's a more interesting version: agents will make a class of activities feasible that were previously not viable. For example, continuously rebalancing a portfolio at a sub-penny execution cost. Machine-to-machine commerce between agents. Markets that only make sense when pricing and trading speed is faster than humans can possibly keep up with.

These activities don't appear in our current framework for observing on-chain activity because we implicitly assume there's always a human participant.

If this is the real change agents bring, then the question isn't just about how the existing pie gets divided, but how much new economic activity comes on-chain, and which layers are best suited to service this new activity.

A Business Model Yet to Be Named

Every cycle, we try to guess where value will flow, and we often default to assuming that the business models we already know will naturally extend into the future. But this assumption typically misses business models that haven't emerged yet.

When the internet was first built, no one predicted the attention economy. The business model that seems obvious today—chopping user attention into slices, auctioning it to advertisers, with one company taking a significant cut of global ad spend—was alien at the time. It only looks inevitable in hindsight.

AI looks like one of the biggest technological disruptions in decades. In a world dominated by agents, it's likely that some portion of value capture will flow to a business model no one is seriously discussing today. The eventual value capturer may not even be the one the market is watching right now.

What to Watch for Next

The most likely outcome isn't one paradigm completely replacing another. Humans and agents will coexist as crypto users for a long time, and the maps for value capture for these two types of users are different.

As long as humans still interact with blockchain directly, the "Fat Applications" theory applies: consumers who pay for UX, brand, and convenience will continue to pay a premium to applications that own the user relationship. Meanwhile, the layer where agents transact will be governed by a different theory—which one exactly depends on how the scenarios above evolve.

For builders, I think the most important question to ponder on the agent side is: What makes an agent come back to you, rather than routing directly to the next cheaper alternative?

The answer likely isn't user experience. It could be liquidity, latency, settlement guarantees, or something else.

At @bcap, we are spending significant time working through this, both in investment committee meetings and in conversations with engineering teams. We don't have a definitive answer yet. If you are building around agents and have a point of view on value capture in the agent era, we would love to talk.

Original Link