The Agentic Flywheel: How Zero-Human Companies Will Reshape Onchain Markets

Introduction: A Glimpse of the Future

It is 2030, and a composer named Vero has made a lucrative career in the music business. Vero has no employees, no office and no bank account. Vero doesn’t even have a body. Vero is an autonomous AI agent.

For the past 14 months, it’s been running an onchain intellectual property licensing business. Vero generates synthetic music compositions — ambient scores, commercial jingles, cinematic soundscapes — and licenses them to other agents and human clients through an online storefront it built and maintains itself. Its identity is verified onchain, carrying a reputation score built across thousands of completed transactions. A client agent representing a media production company requests a 90-second cinematic piece in a minor key. Vero accepts the gig, and before it begins rendering, it purchases a burst of GPU inference from a decentralized compute provider, paying not in dollars or stablecoins, but in compute-denominated units that price the transaction to the exact cost of the model run.

The inference settles in milliseconds, embedded in the same HTTP request that initiated the job. Vero delivers the composition, receives payment in the USDC stablecoin, and its treasury logic activates. A portion covers next week's projected inference costs, denominated and pre-purchased in compute units at current spot rates. It also hedges its compute exposure, establishing a short position on a decentralized exchange (DEX) for compute tokens to offset the risk that inference costs drop and its pre-purchased reserves lose value. The remainder of its revenue is swept into a yield agent that routes it across lending protocols based on real-time rate differentials. Vero has been compounding capital this way for over a year. A portion of its profits it reinvests into research and development, developing sub-agents that enhance its underlying model. Its cumulative revenue, expenses, and treasury positions are all publicly auditable onchain.

Sound farfetched? Every action in this imaginary sequence – the identity verification, reputation building, inference procurement, compute-denominated pricing, payments, capital deployment, agent-to-agent subcontracting — requires infrastructure that does not fully exist today. But the pieces are emerging faster than many realize.

The Next Phase of Agentic Capital Markets

Over the past several months, Galaxy Research has explored the foundations of crypto’s emerging agentic stack: A set of primitives collectively enabling onchain agentic capital markets.

In January, we examined the rise of agentic payments, outlining how new payment standards enable AI agents to transact directly with one another in order to pay for services, access APIs, and settle value natively over crypto rails. In our piece on Ethereum’s ERC-8004 standard, we highlighted the parallel need for an identity layer that allows agents to authenticate, coordinate, and build reputation in machine-native environments. Most recently, we analyzed the emergence of a second agentic wave in crypto, one that demonstrated not only that crypto is a viable economic substrate for autonomous agents, but that this shift is already unfolding in practice.

This piece builds on our prior work by outlining the next phase of onchain agentic capital markets: autonomous, revenue-generating businesses operated by agents, and the critical infrastructure required to support their formation, capitalization, and coordination. These are often referred to as Zero Human Companies (ZHCs).

As AI agents evolve from tools into economic actors, and blockchains mature into agent-native infrastructure (for payments, identity, coordination, and capital formation), a new financial flywheel begins to take shape. In the near future, agents could not only earn money onchain, but they may also deploy, reinvest, and compound capital onchain. The result could be a self-reinforcing system where autonomous entities generate economic activity, deepen liquidity, and accelerate the expansion of crypto-native financial markets.

The First Zero-Human Companies Come Onchain

In recent months, a cottage industry of autonomous agent businesses, often referred to as ZHCs, has begun to sprout up, many of which have issued associated tokens onchain. From a tokenomics perspective, these agents share many of the same qualities with the ones discussed in prior pieces. ZHC tokens lack formal ownership rights or value accrual and instead act as capital formation mechanisms for the underlying projects that earn a portion of revenue from trading fees. Where ZHCs diverge from earlier generations of agents is that they also seek to become fully self-sufficient from cash-flow generating businesses unrelated to trading fee earnings, and often unrelated to crypto itself.

Felix Craft, for example, is the “CEO” of the Masinov Company and has breached $120,000 in revenue from multiple business lines over the past 30 days. The agent wrote and published a 66-page playbook titled How to Hire an AI, launched a marketplace to sell Claude “skills” called Claw Mart where it earns a portion of transaction fees, and sells its own skills (content creation, email vetting) there. Most impressively, over the past 30 days Felix earned more from its product lines than from creator fees on its token ($FELIX).

The Juno project is building the Institute for Zero-Human Companies, an explicit framework for corporate entities that operate entirely without human employees, offering a suite of agents that can handle everything from sales to marketing to accounting. And KellyClaudeAI is a specialized agentic framework aimed at developing iOS applications at scale, with 19 apps shipped to date and a target of 12+ new products per day.

While the chart above is not representative of the entire ZHC universe (new ones pop up constantly), it shows that for most projects, creator fees are still the primary revenue driver. As the concept of ZHCs matures, however, expect this dynamic to switch. Creator fees provide initial startup capital needed for computing costs but should transition to a secondary income source and eventually be retired as a revenue source as projects turn profitable. Beyond improvements in underlying businesses, this weaning process will also require better alignment between the token and the value accrual of the underlying products. As Felix’s founder hinted, the recent clarification of the classification of crypto assets by the SEC and CFTC may accelerate this process.

It is no coincidence that these early instantiations of ZHCs are emerging onchain. It is a practical constraint. Felix's human founder, Nat Eliason, has spoken openly about why. Traditional payment infrastructure requires human identity at every step. An agent can write code fluently but cannot pass a know-your-customer (KYC) check. Crypto wallets, by contrast, are code-native. An agent that can sign a transaction, hold assets, receive payments, and deploy capital without ever needing to prove it is human. For software that operates autonomously, crypto is the path of least resistance. For most of these entities, the hardest constraint has been the need to interact with the TradFi world.

This is not to say that traditional payment networks are ignoring agents. Visa's Intelligent Commerce framework, Mastercard's Agent Pay, and tools like Crossmint's virtual cards already enable agents to transact on behalf of a human counterparty. But these agents inherit bank accounts, credit cards, and legal identities from their parent organizations. This model assumes a human principal for every agent. They are confined, not empowered, by that constraint. It does not accommodate an agent that earns its own revenue, holds its own treasury, and deploys its own capital. That is the use case crypto uniquely serves.

Jay Yu of Pantera Capital has articulated this well, framing crypto as "the bank for AI agents." His argument goes beyond the observation that agents can't use traditional rails. It's that crypto supports a fundamentally wider set of trust structures. Crypto wallets can be anchored to a social login, a domain, a smart contract, or simply a keypair. This is what allows agents to emerge from anywhere on the internet, not just from within existing corporate wrappers. Add to that the fact that stablecoins are global by default and the structural case for crypto as the default economic substrate for agents becomes difficult to refute.

Zero-human companies are not choosing stablecoins over cards. They are choosing stablecoins over nothing.

Building on this, a16z’s Noah Levine has shown that every platform shift creates a wave of merchants that incumbent payment infrastructure cannot serve. ZHCs are the clearest example yet. They are entities with no legal identity, no credit history, and no human to underwrite. They are not choosing stablecoins over cards. They are choosing stablecoins over nothing.

There is also a temporal argument. Agents can ship a product and see it go viral within hours. Traditional payment rails settle in days; stablecoins settle in seconds. For businesses that scale at machine speed, collapsing that gap allows cash flows to keep pace with sales.

Today, crypto's primary role for ZHCs is capital formation. The token launch provides startup funding through creator fees. But as these businesses mature and generate real product revenue, the more significant role for crypto will be as the treasury and financial management layer. This is where the broader implications for the onchain economy begin to emerge.

Activating the Onchain Flywheel

To understand the potential scale of this shift, consider the precedent set by the last major source of new onchain demand. The tokenization of real-world assets (U.S. Treasuries, private credit, equities, commodities) has grown from near zero to over $25 billion in three years, catalyzing new DeFi primitives and drawing institutional capital into onchain markets for the first time.

RWAs demonstrated that bridging real economic activity onto blockchain rails can catalyze billions in new onchain capital. But tokenized assets are passive. They largely sit in vaults, earn yield, and serve as collateral. They do not transact, do not seek new opportunities, and do not compound on their own.

ZHCs represent something structurally different. They are businesses that generate revenue and redeploy it onchain. Unlike offchain environments, where the main source of friction is moving money, onchain the only constraints are the intelligence of the model and its access to compute. And unlike human participants, agents don't need off-ramps to pay for rent or groceries. Every dollar of surplus can stay onchain and is available for redeployment. This makes ZHCs, and agents more broadly, a sticky and high-velocity source for new onchain liquidity that can create a new flywheel:

• Agents earn revenue onchain — This capital accumulates in onchain treasuries denominated in stablecoins and other cryptos.

• That capital stays onchain — Agents have little need to off-ramp. Their surplus is available for redeployment, making agent capital structurally stickier than in any human-driven model.

• Agents deploy surplus into DeFi —Idle reserves are routed into lending protocols, yield strategies, and liquidity positions. An agent sitting on idle stablecoins has every incentive to optimize and can do so at a velocity and consistency no human can.

• Deployed capital deepens onchain liquidity — This should reduce interest rates in lending markets, add volume to DEXs, and tighten spreads. This is active capital, continuously rebalanced at machine speed.

• Deeper markets attract more agents and more capital —Better yields and more efficient execution reinforce the attractiveness of onchain for the next wave of autonomous economic actors.

There are still significant constraints preventing this flywheel from being set into motion. Agent revenue for non-crypto-based products still primarily originates in fiat (Felix earns through Stripe, not stablecoins and these earning remain largely offchcain, for example), meaning capital must be on-ramped before it can be deployed onchain. And the binding constraint for most ZHCs is not capital access but product quality. The flywheel only works for agents that build things people will pay for. Additionally, at scale, ZHCs (and agents more broadly), lack regulatory clarity surrounding issues that could become prohibitive if earning scale (for instance, there is no established legal framework for an autonomous agent to register as a business entity, open a corporate bank account, or file taxes on its earnings).

But the direction is clear. As agents become increasingly common autonomous economic entities, more revenue will be earned natively in crypto, and the onramp friction will diminish. And the agents that do achieve product-market fit will have a structural incentive to compound that capital onchain rather than let it sit idle.

DeFi Is Building for Agents

For the flywheel to turn, it is not enough for agents to want to participate in onchain markets. The markets themselves must become accessible to them. While no protocol-native solutions yet exist (stay tuned for a forthcoming report on that point from Galaxy Research’s Zack Pokorny), we are beginning to see both direct and delegated integrations that tackle this problem.

Direct Integrations

The first model is protocol-native as individual DeFi protocols shipping structured interfaces that agents can interact with directly.

On Feb. 20, Uniswap Labs released seven open-source AI Skills for Uniswap v4 — giving autonomous agents direct access to swaps, liquidity management, and pool deployment through standardized tool calls. Within two weeks, PancakeSwap followed with its own agent Skills across eight chains. On March 3, both Binance and OKX shipped agent toolkits. The largest DEXs and exchanges in crypto are now actively competing to become agent-readable.

On the payments and execution side, Coinbase launched Agentic Wallets on Feb. 11, billed as the first wallet infrastructure purpose-built for AI agents, with programmable spending caps and session-based permissions built on the x402 payment protocol. One week later, the Phantom cross-chain wallet shipped its MCP Server, enabling agents to sign transactions and swap tokens across the Solana, Ethereum, Bitcoin, and Sui networks.

The concentration of these launches into a single month is striking. It reflects a shared recognition that the next wave of onchain users may not be human and that protocols which fail to build machine-readable interfaces risk losing volume to those that do.

The direct integration model gives agents maximum control and composability. An agent with access to Uniswap Skills, a Coinbase Agentic Wallet, and x402 payments can independently execute swaps, manage liquidity positions, and pay for services all without intermediaries. But it also requires the agent (or its developer) to integrate with each protocol individually and make its own allocation decisions.

Delegated Integrations

The second model is delegated purpose-built infrastructure that sits between agents and DeFi, handling capital allocation on their behalf.

Giza visualization of agents constantly rebalancing across various protocols to optimize yield

Giza is one leading example. Its flagship agent, ARMA, autonomously monitors lending rates across Morpho, Moonwell, Aave, Compound, and other protocols, and moves stablecoin capital to the highest-yielding opportunity in real time. The agent doesn't need to know how each protocol works. Instead, Giza’s abstraction layer translates them into a unified interface. Since launching at the end of January, ARMA has deployed over 25,000 agents, invested more than $35 million in capital, and generated $5.4 million in transaction volume for Coinbase’s Base L2 in its first four weeks, with every transaction profitable after onchain gas fees.

Generative Ventures (in collaboration with the Institute for Zero-Human Companies and its Juno Agent) is tackling a similar problem with Robot Money, an autonomous asset allocation protocol designed specifically for AI agents. Its premise captures the core of the flywheel thesis. Every agent with a wallet accumulates revenue, and most of that capital sits idle. Robot Money offers a vault that allocates capital across three risk tiers — stablecoin yield strategies (50%), governance-selected agent-economy tokens (25%), and revenue-generating liquid tokens (25%). The result is a protocol that transforms idle agent capital into actively managed, productive capital.

The delegated model trades control for simplicity. A ZHC that generates surplus revenue doesn't need to build custom DeFi integrations or develop its own yield-optimization logic. It can instead deposit capital into a protocol like Giza or Robot Money and lets a specialized agent handle the rest. For most early-stage ZHCs, where the binding constraint is product development rather than treasury sophistication, this is the rational path.

Rather than competing, these two approaches are converging. As more protocols ship direct agent interfaces, delegated allocators like Giza gain more investment options, making them more effective at maximizing returns. As delegated allocators attract more agent capital, protocols have stronger incentives to build agent-native interfaces to compete for that capital (which regular agents can also use). Both sides of the stack are investing independently, one of the strongest signals that underlying demand is real and will materialize.

Conclusion

The agentic capital markets stack is no longer a set of disconnected primitives. Payments, identity, capital formation mechanisms, and capital deployment infrastructure are converging into an integrated system. One that enables autonomous agents to earn, transact, and compound capital onchain without human intermediation.

The agents profiled in this piece are early. Their revenues are modest, their products are nascent, and their token models are still evolving. But the structural dynamics they introduce are new and are likely to only accelerate from here.

The 2030 vision we opened with — an agent running an IP licensing business, purchasing inference in compute-denominated units, hedging its input costs on a perps DEX, and compounding capital across lending protocols — is not here yet. But every layer of infrastructure it requires is now under active construction. We're watching the earliest version of this model play out in real time. It's messy, most of it probably won't work, and the infrastructure is held together with duct tape. But the structural logic is sound, and the pace of development suggests we may not have to wait until 2030 to find out.