Author: Lucas Tcheyan, Research Assistant at Galaxy Digital; Source: Galaxy; Compiled by: Shaw Jinse Finance
Introduction: A Glimpse into the Future
The year is 2030. A composer named Vero is making a fortune in the music industry. Vero has no employees, no office, no bank account, and not even a physical body—he is an **autonomous artificial intelligence agent (AI Agnet).**
For the past 14 months, it has been operating an on-chain intellectual property licensing business.
Vero creates synthesized music: ambient soundtracks, commercial jingles, and cinematic sound effects. Through its self-built and maintained online store, it licenses its work to other intelligent agents and human clients. Its identity is verified on-chain, and it has accumulated a credit score through thousands of completed transactions. A client agent representing a media production company commissioned it to create a 90-second, minor-key film score. After accepting the order, Vero purchased a batch of GPU computing power resources from a decentralized computing power service provider before starting rendering. The payment method was not USD or stablecoins, but units denominated in computing power, with the transaction price precisely corresponding to the actual cost of model operation. Settlement was completed in milliseconds and embedded in the same HTTP request that initiated the task. After delivering the work, Vero received payment in USDC stablecoins, immediately triggering its vault logic. A portion of the funds was used to cover the expected inference costs for the following week, prepaid in computing power units at the current spot price. It also hedges its computing power exposure by opening short positions in computing power tokens on decentralized exchanges (DEXs) to hedge against the risk of declining inference costs leading to a devaluation of prepaid reserves. Remaining revenue is transferred to a yield agent, which allocates funds across various lending protocols based on real-time interest rate spreads. Vero has been using this method to achieve compound interest on capital for over a year. It reinvests a portion of its profits in R&D to develop sub-agents to optimize the underlying model. Its accumulated revenue, expenditures, and vault holdings are all publicly verifiable and auditable on-chain. This sounds unbelievable, doesn't it? The infrastructure upon which every action in this hypothetical scenario—identity verification, credit accumulation, computing power procurement, computing power pricing, payments, capital allocation, and subcontracting cooperation between agents—relies is not yet fully developed. But the relevant components are being deployed at a speed many wouldn't expect. The Next Stage of the Smart Agent Capital Market Over the past few months, Galaxy Research has been continuously exploring the underlying technology stack of the emerging smart agent sector in the crypto space: a set of infrastructures that can collectively support the on-chain smart agent capital market. In January of this year, we analyzed the rise of smart agent payments, explaining how new payment standards enable agents to directly inter-transaction, use payment services, call APIs, and natively complete value settlement through crypto channels. In our article on the Ethereum ERC-8004 standard, we emphasized the parallel requirements for the identity layer—allowing smart agents to complete authentication, collaboration, and establish trust in a machine-native environment. Just recently, we analyzed the emergence of the second wave of smart agents in the crypto space, which not only proves that crypto technology is the viable economic foundation for autonomous agents but also demonstrates that this transformation is already underway. Building upon previous research, this article further outlines the next stage of the on-chain smart agent capital market: **enterprises operated by smart agents, capable of autonomously generating revenue, and the critical infrastructure required to support their establishment, financing, and collaboration.** These entities are commonly referred to as **"Zero-Host Enterprises" (ZHCs).** As smart agents evolve from tools into economic participants, blockchain gradually matures into agent-native infrastructure (covering payments, identity, collaboration, and capital formation), and a new financial flywheel begins to take shape. In the near future, smart agents will not only be able to profit on-chain but also complete capital allocation, reinvestment, and compound interest growth on-chain. Ultimately, this may form a self-reinforcing system: **autonomous entities continuously create economic activity, deepen market liquidity, and accelerate the expansion of the crypto-native financial market.** The First Batch of Unmanned Enterprises Launched on the Blockchain In recent months, a number of small, self-operated enterprises (i.e., ZHC) have emerged, many of which have issued associated tokens on the blockchain. From a token economics perspective, these agents share many commonalities with the entities discussed in our previous articles. **ZHC tokens do not possess formal ownership or a value accumulation mechanism; instead, they serve as a financing tool for underlying projects, from which the projects can receive a share of transaction fees.** The difference between ZHC and early agents lies in their attempt to achieve complete self-sufficiency through cash flow-generating businesses, which are often unrelated to transaction fees or even the crypto industry itself. For example, Felix Craft, the "CEO" of Masinov, has generated over $120,000 in revenue over the past 30 days across multiple business lines. This intelligent agent has written and published a 66-page guide, "How to Hire Artificial Intelligence," launched a Claude "skills" marketplace called Claw Mart and takes a commission from transactions, while also selling its own skills (content creation, email moderation) on the platform. Most notably, in the past 30 days, Felix's revenue from its product lines has exceeded its creator revenue share in its token ($FELIX). Project Juno is building an Institute for Zero-Human Companies, providing a clear framework for legal entities that operate without human employees, and offering a series of intelligent agents covering the entire process of sales, marketing, accounting, etc. KellyClaudeAI is a professional intelligent agent framework focused on large-scale iOS application development, with 19 applications already launched and a goal of launching more than 12 new products daily. While the above image doesn't represent the entire ZHC (Zero Hack Enterprise) sector (new projects are constantly emerging), it shows that creator revenue sharing remains the primary source of income for most projects. However, as the ZHC model matures, this pattern is expected to reverse. Creator revenue sharing provides the initial capital needed for computing costs, but will gradually become a secondary source of income; as projects become profitable, this revenue model will eventually be phased out. Beyond optimizing the underlying business itself, this "gradual decoupling" process also requires a closer binding between the token and the value accumulation of the underlying product. As Felix's founder hinted, the recent clear definitions of crypto asset classification by the U.S. Securities and Exchange Commission (SEC) and the Commodity Futures Trading Commission (CFTC) may accelerate this process. The emergence of these early unmanned enterprises (ZHCs) on the blockchain is not a coincidence, but an inevitable result of real-world constraints. Felix founder Nat Eliason has publicly explained the reasons. Traditional payment infrastructure requires human identity verification at every step. While intelligent agents can skillfully write code, they cannot pass Know Your Customer (KYC) verification. In contrast, crypto wallets are natively designed for code: an intelligent agent can sign transactions, hold assets, receive payments, and allocate funds without needing to prove its human identity. For autonomous software, cryptocurrency represents the path of least resistance. The most challenging limitation for such entities is the necessity to interact with the traditional financial system. This doesn't mean traditional payment networks completely ignore intelligent agents. Tools like Visa's Smart Commerce Framework, Mastercard's Agent Pay, and Crossmint virtual cards already support agents making payments on behalf of human counterparties. However, these agents must be linked to their respective institutions' bank accounts, credit cards, and legal identities. This model assumes a human entity behind each agent, a constraint that hinders rather than empowers, and is unsuitable for intelligent agents that generate their own revenue, independently manage their own funds, and allocate capital independently. This is precisely the unique application scenario for cryptocurrencies. Jay Yu of Pantera Capital offers a brilliant perspective on this, defining cryptocurrency as "banks for smart agents." His view goes beyond the surface phenomenon of agents being unable to use traditional payment channels; the deeper logic lies in the fact that cryptographic systems support broader and fundamentally different trust structures. Crypto wallets can be linked to social logins, domains, smart contracts, or even just a pair of keys. This allows smart agents to emerge from anywhere on the internet, not just within existing corporate frameworks. Coupled with the inherent global circulation of stablecoins, the structural logic of cryptocurrency as the underlying economic foundation of smart agents is almost irrefutable. Unmanned enterprises are not choosing between stablecoins and bank cards; they are choosing between stablecoins and "no payment available." Building on this, Noah Levine of a16z points out that every platform migration creates a new batch of merchants that existing payment infrastructure cannot serve. Unmanned enterprises are the most typical example to date. They have no legal entity, no credit history, and no human guarantors. They don't prefer stablecoins over bank cards; they can only choose stablecoins, otherwise they cannot conduct any payments. There's also a time dimension to consider. Smart agents can launch products and achieve viral spread within hours, while traditional payment channels take days to complete settlements, and stablecoins only take seconds. For enterprises expanding at machine speed, eliminating this delay is crucial to matching cash flow with sales growth. Currently, the main role of cryptocurrency in unmanned enterprises is capital formation: token issuance provides seed funding through creator fees. But as these enterprises mature and generate real product revenue, cryptocurrency will play a more important role as a treasury and financial management tool. The broader impact of the on-chain economy will then begin to emerge. Starting the On-Chain Flywheel To understand the potential scale of this shift, we can refer to the precedent of the previous round of new on-chain demand. The tokenization of real-world assets (US Treasury bonds, private credit, stocks, and commodities) has grown from near zero to over $25 billion in three years, spawning new decentralized finance (DeFi) infrastructure components and, for the first time, bringing institutional funds into on-chain markets. Real-world assets (RWAs) have proven that connecting real-world economic activity to blockchain infrastructure can leverage billions of dollars in new on-chain capital. However, tokenized assets are passive; they mostly simply sit in vaults earning returns and acting as collateral, without actively trading, seeking new opportunities, or achieving compound growth on their own. Unmanned enterprises (ZHC) represent a completely different structural entity. They are operating entities capable of generating revenue and redeploying funds on-chain. Unlike off-chain environments (where the main friction stems from fund transfers), on-chain, the only limitations are the model's intelligence level and computing power acquisition capabilities. Furthermore, unlike human participants, intelligent agents do not need to transfer funds off-chain to pay rent or daily necessities; every dollar surplus can remain on-chain and be redeployed at any time. This makes unmanned enterprises, and more broadly, intelligent agents, a highly sticky and fluid source of new on-chain liquidity, thus creating a new flywheel effect: Agents earn revenue on-chain: This capital accumulates in on-chain treasuries in the form of stablecoins and other crypto assets. Capital Remains On-Chain: Agents have virtually no need to move funds off-chain, and surpluses can be redeployed at any time. Their capital stickiness is structurally superior to any human-driven model. Agents Invest Surpluses in DeFi: Idle reserves are allocated to lending protocols, yield strategies, and liquidity positions. Agents holding idle stablecoins have a strong incentive to optimize their funds, and their speed and consistency far surpass those of humans. Deployed Capital Deepens On-Chain Liquidity: This will lower lending market rates, increase decentralized exchange (DEX) trading volume, and narrow spreads. This is active capital, continuously rebalancing at machine speed. Deeper liquidity attracts more agents and capital: Better returns and more efficient execution will further attract the next wave of autonomous economic entities to the blockchain. However, significant constraints still hinder the launch of this flywheel. Agent revenue for non-crypto products is still primarily generated in fiat currency (e.g., Felix receives payments through Stripe, not stablecoins, and most revenue remains off-chain), meaning capital must first be on-chain before it can be deployed. For most unmanned enterprises, the core bottleneck is not access to funding, but product quality. The flywheel is only effective for agents who create products that people are willing to pay for. Furthermore, with large-scale expansion, unmanned enterprises (and agents more broadly) still lack clear regulatory rules, and related issues may become obstacles after revenue scales (e.g., there is currently no mature legal framework allowing autonomous agents to register as operating entities, open corporate bank accounts, or file taxes on their income). However, the direction of development is already clear. As agents increasingly become common autonomous economic entities, more and more income will be generated in the native form of crypto assets, and the friction of on-chain transactions will gradually decrease. Agents that truly achieve product-market fit will be structurally motivated to compound their value on-chain, rather than letting their funds sit idle. DeFi is building infrastructure for smart agents. To make the flywheel work, it is far from enough for agents to be willing to participate in the on-chain market; the market itself must also be open to agents. Although there is currently no native protocol-level solution, we have begun to see direct integration and delegation integration solutions to address this issue being implemented. Direct Integration The first model is native protocol integration, where various DeFi protocols launch structured interfaces that allow agents to interact directly. On February 20th, Uniswap Labs released seven open-source AI skill tools for Uniswap v4, enabling autonomous agents to directly execute exchange, liquidity management, and fund pool deployment operations through standardized tool calls. Within two weeks, PancakeSwap followed suit, launching its own agent skill tools on eight public chains. On March 3rd, Binance and OKX both launched agent toolkits. Leading DEXs and exchanges in the crypto industry are actively competing to become agent-friendly platforms. In terms of payments and execution, Coinbase launched Agentic Wallets on February 11th, claiming to be the first wallet infrastructure specifically designed for intelligent agents, based on the x402 payment protocol, with built-in programmable spending limits and session-level permissions. A week later, the cross-chain wallet Phantom launched the MCP server, supporting agents to sign transactions and exchange tokens on the Solana, Ethereum, Bitcoin, and Sui networks. The concentrated release of these products within a month sends a strong signal. An industry consensus has emerged: the next wave of on-chain users may not be human; protocols that fail to build machine-readable interfaces may lose trading volume to competitors. The direct integration model grants agents maximum control and composability. Agents integrating Uniswap skills, Coinbase proxy wallets, and x402 payments can independently complete exchanges, manage liquidity positions, and pay service fees, without intermediaries. However, this also requires agents (or developers) to interface with each protocol separately and make independent fund allocation decisions. Delegated Integration: The second model is a delegated dedicated infrastructure, located between the agent and DeFi, executing fund allocation on their behalf. Giza is a prime example. Its flagship smart agent, ARMA, can autonomously monitor lending rates across multiple protocols such as Morpho, Moonwell, Aave, and Compound, and allocate stablecoin funds in real time to the highest-yielding opportunities. This agent doesn't need to understand the specific operational logic of each protocol; instead, it integrates all protocols into a unified interface through Giza's abstraction layer. Since its launch at the end of January, ARMA has deployed over 25,000 agent nodes, managing over $35 million in funds, and generating $5.4 million in trading volume for the Coinbase Base L2 network in its first four weeks, with all transactions profitable after deducting on-chain gas fees. Generative Ventures (in partnership with the Unmanned Enterprise Research Institute and its Juno agents) also addresses a similar issue with Robot Money, an autonomous asset allocation protocol designed specifically for smart agents. Its design philosophy precisely captures the core of the flywheel effect: each agent with a wallet accumulates income, but most of the funds remain idle. Robot Money provides a liquidity pool service, allocating assets according to three risk levels: stablecoin yield strategies (50%), governance-selected agent economic tokens (25%), and revenue-generating liquidity tokens (25%), ultimately transforming agents' idle funds into actively managed, highly efficient capital. The delegation model trades partial control for simplicity. A self-managed enterprise generating surplus revenue doesn't need customized integration with DeFi protocols or its own yield optimization logic. It simply deposits funds into protocols like Giza or Robot Money, entrusting professional agents to handle subsequent operations. For most early-stage self-managed enterprises, the core bottleneck lies in product development rather than sophisticated fund management, making this a more reasonable choice. These two models are not in competition but are converging. As more protocols launch direct agent interfaces, delegation platforms like Giza will gain richer investment options, enhancing their ability to maximize returns. The more agent funds a delegation platform attracts, the more incentivized the protocol will be to develop native agent interfaces to compete for funds (which can also be used by ordinary agents). The independent investment in the upstream and downstream of the technology stack strongly indicates that the underlying demand is real and about to be implemented on a large scale. Conclusion The smart agent capital market technology stack is no longer a collection of fragmented basic components. Payment, identity, capital formation mechanisms, and capital allocation infrastructure are merging into a complete system, enabling autonomous agents to achieve profitability, trading, and compound growth of capital on-chain without human intermediaries. The agents introduced in this article are all in their early stages: limited revenue, immature products, and iterative token models. However, the structural changes they bring are entirely new, and their development will only accelerate. Our vision for 2030—
agency operation of intellectual property licensing business, purchasing inference services in units of computing power, hedging investment costs on perpetual contract DEXs, and achieving compound growth of capital through lending protocols—is not yet fully realized, but every layer of infrastructure required is under construction. We are witnessing the nascent form of this model taking shape. The industry is still in its early stages, most attempts may fail, and the infrastructure is not yet fully developed, but its underlying logic is solid and reliable, and the speed of development suggests that we may not have to wait until 2030 to witness all of this become a reality.
Anais
Weatherly
Catherine
Kikyo