Author: Jordi Visser, Senior Analyst at Wall Street; Compiled by Shaw, Jinse Finance
Last week, I completed my first exclusive interview with Mark Moss. Having listened to many of his podcast episodes over the years, I greatly valued the opportunity to sit down and have an in-depth conversation with him.
I have always enjoyed discussing artificial intelligence (AI) and the crypto industry with new acquaintances, but this conversation was exceptionally profound because it focused squarely on the intersection of these two domains. Most investors, technologists, and industry commentators still treat them as separate spheres: when people talk about AI, they emphasize intelligence and a productivity revolution; when discussing crypto, they view it merely as a parallel track involving currency, financial markets, and digital assets.
This fragmented perspective causes observers to overlook a far more significant transformation underway.
Artificial intelligence is giving rise to a large number of autonomous economic agents, and cryptographic technology will provide the complete financial infrastructure for these entities—enabling transactions, establishing asset ownership, verifying identities, and facilitating operations within a framework of enforceable rules. The more capable AI agents become, the more inseparable these two technologies will be.
Whenever I speak with industry professionals who have spent years working at the intersection of these fields, I always leave with fresh questions and insights that sharpen my understanding of where the world is headed. A dialogue between two individuals jointly exploring the future is inherently valuable. Discussing the past merely involves mapping a landscape that has already solidified—the facts are settled, outcomes are fixed, and the narrative arc is already written.
In contrast, seriously discussing the future is more like stepping onto an uncharted continent: we search for connections between phenomena, test various hypotheses, and attempt to sketch a vision of a world that has not yet fully taken shape.
My conversation with Mark directly inspired this article. With every release of a new large AI model, I grow increasingly convinced of one point: we stand at the threshold of a transformation unlike anything humanity has ever witnessed.
A new class of digital agents is about to formally enter economic activity.
I. The Traditional Economic Measurement Framework Has Become Obsolete
Throughout the history of modern economic development, investors have consistently evaluated demand from a human-centric perspective: more consumers purchasing goods, more enterprises hiring employees, and more factories expanding capacity. Population, income, and gross domestic product (GDP) rose in tandem, driving demand expansion. Economic growth and resource consumption were largely linearly correlated.
Artificial intelligence has fundamentally rewritten the underlying logic of the demand curve—the primary agents of economic production are shifting from humans to software-based intelligent agents. Yet many investors continue to assess AI infrastructure using analytical frameworks designed for traditional enterprise software. They observe Microsoft, Amazon, and Google investing hundreds of billions of dollars to build data centers, and OpenAI, Anthropic, xAI, and numerous cutting-edge labs consuming massive amounts of computing power. From this perspective, AI compute demand appears concentrated among only a handful of clients.
This analytical framework overlooks an ongoing structural transformation: leading large-model companies are evolving into platforms that orchestrate and manage billions of autonomous software agents. The true source of demand lies in the digital workforce incubated by these platforms. The core cost for this digital labor force is denominated in tokens, with the per-unit token cost of compute power declining rapidly and continuously, while industry competition intensifies dramatically.
At the same time, this framework also ignores the entirely new financial infrastructure required to support this digital workforce.
Billions of intelligent agents cannot operate at scale within traditional financial systems designed for humans: bank operating hours are limited, manual approval processes are cumbersome, settlement involves delays, and databases remain siloed—all transaction logic assumes the counterparty is a human being. Intelligent agents require round-the-clock capabilities to hold and exchange value, verify counterparties, enforce spending limits, and automatically execute contracts and settlements.
Cryptographic technologies happen to provide the financial risk-control safeguards needed for this digital economic ecosystem.
II. Misleading Statistical Methodology: Counting Only Enterprise Clients While Ignoring the Vast Base of Underlying Intelligent Agents
Each cloud provider earnings season, investors pose the same question: who exactly is consuming vast quantities of computing resources?
From the perspective of traditional enterprise services, the answer seems clear: OpenAI rents compute capacity from Microsoft Azure, Anthropic relies on Amazon AWS, Google builds its own infrastructure, and Meta develops its own large models. If these companies are viewed as end-demand entities, AI compute spending appears highly concentrated, making industry sentiment extremely vulnerable to budget fluctuations at just a few firms.
However, when agents become the primary demand entities, the entire economic logic is fundamentally rewritten. OpenAI, Anthropic, and Google resemble operating systems for digital labor rather than end consumers of computing power. Enterprise applications, software workflows, and consumer-facing services built on these platforms will continuously generate new demand for inference computing capacity. The infrastructure provided by cloud service providers ultimately serves a software workforce numbering in the billions.
Similar statistical blind spots exist in the financial sector. On the surface, banks and payment networks count AI platforms, enterprises, or digital wallets as their major clients, but beneath those accounts may lie millions of agents—making ongoing economic decisions on behalf of individuals, businesses, vehicles, robots, and even other agents.
Counting only corporate accounts would entirely miss the vast volume of underlying economic activity.
A single enterprise client serves billions of underlying entities.
This kind of statistical misconception is actually not unfamiliar.
To suppliers, McDonald's appears as just one procurement client, yet its iconic slogan—'Serving billions of people'—accurately captures the true scale of economic activity hidden beneath the corporate facade: behind a single company lie tens of billions of individual consumer transactions annually.
The AI economy is replicating this logic on an even larger scale: OpenAI appears to be merely a single customer of Azure, Anthropic just one customer of AWS, and Google’s internal cloud usage is counted solely as internal demand. Yet behind each platform lies an ever-expanding cluster of software agents—conducting research, writing code, performing data analysis, serving users, generating content, executing financial workflows, and collaborating across agent networks.
Each agent generates continuous inference demand; a single task may invoke dozens of models to complete the full pipeline of inference, retrieval, search, coding, memory access, and planning. For cloud providers, one platform client represents billions of round-the-clock digital workers.
The financial system will likewise experience a transaction surge of equivalent magnitude.
An individual agent can, before its user wakes up, compare quotes from hundreds of merchants, negotiate subscription terms, rebalance investment portfolios, purchase computing power, acquire data services, and pay to invoke specialized agents—completing dozens of micro-transactions. Enterprise agents, meanwhile, can autonomously procure inventory, lease computing resources, manage working capital, hedge foreign exchange risk, and pay other agents for task completion.
A single, simple instruction from a human may underlie hundreds or even thousands of economic transactions.
Cloud providers are building the computational infrastructure to support billions of digital workers that have yet to be widely deployed; meanwhile, encrypted networks provide the foundational infrastructure—enabling transactions, clearing, identity verification, and asset ownership—for these agents to engage in commercial activities with one another.
III. Intelligence Requires Financial Autonomy
The core function of first-generation AI systems is information output: answering questions, summarizing documents, generating images, writing code, and assisting humans in decision-making.
Agents, however, represent a leap from 'outputting information' to 'acting autonomously.'
Once agents possess the ability to act, they must also be granted financial execution rights: autonomously procuring resources, paying service providers, collecting revenues, managing budgets, and verifying contractual performance conditions. An AI that can only offer transactional recommendations but cannot execute them remains merely an advanced assistant; only agents capable of securely managing economic assets qualify as genuine market participants.
Traditional financial infrastructure is built around natural persons, with layered manual review processes serving as inherent security mechanisms: signing paper documents, entering passwords, waiting for banks to open, manually approving transfers, reconciling accounts by hand, and resolving disputes through centralized institutions. Given that humans conduct a limited number of transactions individually, such cumbersome procedures remain tolerable.
However, agents operate continuously 24/7 and execute transactions at machine speed, rendering traditional processes an insurmountable bottleneck.
The volume of transactions generated by billions of agents far exceeds human comprehension. An agent might spend mere fractions of a cent to purchase data, inference compute, storage, bandwidth, identity verification, API access, intellectual property, or specialized services from other agents; a single business process could trigger thousands of microtransactions, and a single enterprise might generate millions of transactions per day.
A financial system designed for agents must be programmable, available 24/7, globally interoperable, auditable, and capable of processing ultra-microtransactions at low cost.
This also explains why market rumors suggest Stripe is interested in acquiring PayPal—an acquisition whose significance extends far beyond the purchase price itself. Stripe has built a comprehensive payment infrastructure for merchants in internet commerce, while PayPal holds a vast base of consumer users, along with Venmo, Braintree payment channels, and the PYUSD stablecoin. Integrating the two would create a financial platform covering the entire smart commerce value chain: on one end, merchant acquirers; on the other, everyday users authorizing AI agents to make payments on their behalf.
Whether or not this acquisition ultimately materializes is less important than the strategic signal it sends to the industry: the payments sector is already positioning itself for an era in which software autonomously initiates settlements. Stripe has introduced transaction-specific authorization credentials that allow agents to complete compliant purchases without gaining full access to users’ underlying payment information; meanwhile, PayPal continues to invest heavily in the supporting infrastructure and risk-control systems required for smart commerce.
Future payment networks must do more than simply transfer funds—they must perform multiple layers of verification: authenticating agent permissions, restricting purchasable categories, capping individual transaction amounts, safeguarding underlying payment credentials, assessing fraud risk, and maintaining records of human authorization.
Bank cards and bank accounts will remain as sources of funds, but the outer control layer must be programmable. Stablecoins, tokenized deposits, programmable wallets, cryptographic identities, and smart contracts together form a new foundational architecture—one that enables commerce to shift from sporadic human-initiated settlements to continuous, machine-driven automatic settlements.
Thus, the rumor of Stripe’s potential acquisition of PayPal serves as yet another signal that the financial industry is recognizing a fundamental shift: AI agents are emerging as independent economic entities. Whoever controls agents’ wallets, operational permissions, digital identities, clearing channels, and merchant relationships will hold the most critical financial gateway of the intelligent era.
This foundational architecture aligns closely with cryptographic systems.
IV. Cryptographic Technology: Financial Risk Control Assurance for AI Agents
When most investors think of cryptography, they associate it only with speculative tokens and price volatility. However, its long-term core value lies in natively establishing digital property rights and financial constraint frameworks for autonomous software.
AI agents require not just payment channels, but also a robust risk control framework: one that defines the assets they hold, their allowable expenditure scope, eligible counterparties, and the preconditions for releasing funds.
Programmable wallets can allocate dedicated budgets for AI agents rather than granting access to an individual’s or enterprise’s full account balance; permission rules can restrict transaction amounts, merchant sectors, applicable regions, asset types, usage time windows, and cumulative spending limits; high-value transactions can trigger secondary verification or manual approval steps.
Smart contracts can hold funds in escrow and automatically release payments only after verifiable conditions are met: upon task completion, logistics confirmation, delivery of digital services, or achievement of predefined performance metrics, at which point the agent automatically settles compensation with the collaborating agent.
Stablecoins provide agents with a native digital medium of exchange whose value is pegged to traditional fiat currencies, facilitating economic calculations. Tokenized deposits, U.S. Treasury bonds, money market funds, securities, intellectual property, and physical assets enable agents to seamlessly switch between cash, collateral, investment vehicles, and production resources within the same programmable environment—without requiring cross-system operations.
Public blockchains and permissioned ledgers maintain complete audit trails, clearly recording the initiating agent, execution rules, and detailed asset flows for every transaction, enabling verification that all operations remain within authorized boundaries.
These are not merely supplementary features but essential components of a financial safety infrastructure, preventing autonomous commercial activities from descending into disorder and chaos.
V. Transaction volumes surge, rendering manual oversight entirely ineffective
Human-led financial supervision is primarily ex-post: compliance departments conduct spot checks on transactions, accountants manually reconcile accounts, auditors sample and archive records, and regulatory authorities review reports days or even months after the fact.
This model cannot scale to an economic system involving billions of agents engaged in continuous transactions.
It is impossible for humans to manually verify every micro-transaction between agents. Instead, policies, operational permissions, risk limits, and identity access rules must be predefined and automatically enforced at the moment of transaction execution. Regulatory logic must be embedded directly within the transaction workflow itself.
Cryptographic technology enables financial rules to be programmatically implemented and enforced.
Enterprises can restrict procurement agents to spend only pre-approved amounts with compliant suppliers; users can authorize travel agents to book flights within a specified price range but require manual approval for hotel reservations; investment agents may rebalance portfolios within an approved asset pool but are prohibited from engaging in borrowing, leveraging, or interacting with unauthorized protocols.
The ledger retains a complete record of all operations within the scope of the rules.
This gives rise to a financial hierarchical architecture tailored for autonomous systems: humans set objectives and risk thresholds, while agents autonomously optimize within those constraints, and cryptographic infrastructure verifies identities, enforces permissions, transfers assets, and preserves operation records.
Without this risk-control framework, granting agents discretionary control over funds would be equivalent to giving employees unrestricted access to corporate accounts—relying solely on internal policies for self-regulation. Any productivity gains would be entirely eroded by security risks.
Through programmable financial constraints, fund-disbursement authority can be delegated in a granular, quantifiable, and revocable manner across multiple layers.
VI. Stablecoins: Native Settlement Currency for Agents
Agents are inherently global—they can procure services across borders, rent decentralized computing power, purchase data and information, and pay overseas agents without geographical limitations.
Traditional banking systems fragment such cross-border transactions: different currencies, correspondent banks, payment service providers, business hours, and regional regulatory infrastructures create multiple layers of segmentation. A single payment must pass through numerous intermediaries before reaching its recipient, with each step introducing delays, fees, and reconciliation costs.
Stablecoins provide agents with a unified digital settlement instrument that operates seamlessly across compatible networks 24/7. Their core value in the agent-driven economy lies in their programmability, uninterrupted operation, and ability to be embedded directly into software workflows.
Agents do not care about the brand of the partner bank or the interface of the payment application; they prioritize execution speed, reliability, cost, liquidity, and settlement finality. They will automatically select the most efficient payment channel, much like software dynamically allocates network bandwidth.
As agent capabilities continue to advance, they will constantly optimize where funds are held, which stablecoins to use, which public blockchain offers the lowest costs, and which tokenized assets deliver the best risk-return profile. Funds long left idle due to human inattention will see significantly enhanced liquidity under continuous price and yield comparisons by agents.
This is also the core reason why the intelligent economy will accelerate the adoption of stablecoins and tokenized assets: these financial instruments are designed from the outset to be natively compatible with software-based autonomous holding, valuation, transfer, and exchange.
7. Digital Identity Is as Critical as Funds
For secure agent-based transactions, the financial system must verify not only account balances but also the authorized entity behind fund operations.
Agents can represent individuals, enterprises, government agencies, autonomous vehicles, or other software systems, and every layer of agency relationship requires a verifiable authorization chain. Counterparties must confirm the agent’s authentic identity, transactional authority, and that credentials have not been revoked.
Cryptographic identities enable trusted agency relationships without requiring manual verification for each transaction: agents can present proof of accreditation from authorized institutions, compliance licenses, jurisdictional access permits, and limited spending authority.
Creditworthiness will also become portable across platforms: agents with consistent performance records accumulate verifiable transaction histories, enabling other agents to assess counterparty risk based on those records, request collateral, or decline collaboration.
Therefore, the intelligent economy requires an integrated system encompassing identity, funds, credit, operational permissions, and clearing and settlement—capabilities that cryptographic networks are progressively consolidating.
8. Agents Reshape Demand Structures
Traditional software enhances human employee productivity by helping them complete tasks more quickly; agents, by contrast, directly assume full business functions, enabling enterprises to scale effective labor capacity without proportionally increasing headcount.
A single employee can manage dozens of specialized agents handling tasks such as coding, industry research, legal review, scheduling, customer service, financial analysis, and content generation. Within a single business process, agents further invoke multiple specialized models—meaning a single user instruction may trigger nested layers of reasoning and computation.
A single human decision triggers dozens to hundreds of computational actions; behind a simple prompt lies a coordination layer orchestrating inference models, retrieval systems, memory banks, search, translation, image generation, and domain-specific agents. Software begins serving software, and computing demand no longer scales linearly with user numbers—it expands exponentially.
The same principle applies to financial transaction demand.
An intelligent agent fulfilling a business task may procure data, inference computing power, storage, and verification services from multiple providers; each provider, in turn, invokes its own agents or outsources other digital services. A single commercial objective thus fragments into a vast machine-to-machine payment network.
This is precisely the fundamental distinction between human labor and digital labor: traditional enterprises scale by hiring more employees, whereas intelligent systems achieve exponential growth by layering intelligent decisions and transactional behaviors.
IX. Jevons Paradox Applies Equally to Transactions
The historical Jevons Paradox in economics aptly explains this phenomenon: improvements in coal-burning efficiency did not reduce coal consumption but instead increased overall demand; as bandwidth costs declined, internet usage surged; lower storage costs triggered an explosive growth in the volume of digital information.
Artificial intelligence replicates this dynamic in the realm of computing power: every reduction in inference cost unlocks a wave of new applications that become economically viable. When the cost per inference was $1, it held no commercial value; at $0.01, it enabled mass adoption; and with another order-of-magnitude cost reduction, intelligent capabilities can be embedded into products, processes, and services previously unable to bear the computational expense.
This pattern applies equally to financial transactions.
Humans forgo small-value payments when fees and operational friction exceed transaction value; intelligent agents, however, will continuously initiate transactions as long as expected returns exceed marginal costs. As settlement costs continue to fall, agents can granularly procure specialized data, computing power, bandwidth, energy, software, and intellectual property.
Declining transaction costs will not merely reduce fees on existing transactions—they will catalyze a vast number of new ones. The ultimate scale of machine-driven commerce will far surpass today’s consumer payment systems, as software can subdivide economic activity into ultra-micro, high-frequency transactions beyond human capacity to manage.
The lower the costs of computing power and settlement, the greater the scale at which agents consume them.
10. Computing Power and Cryptography: Complementary Layers of Dual Infrastructure
In market discussions about AI investment, computing infrastructure and digital asset infrastructure are often treated in isolation: the former is categorized as physical productive technology, while the latter is viewed merely as a speculative financial market.
However, the intelligent economy demonstrates that they constitute two complementary layers of a unified system: computing power endows agents with reasoning capabilities; networks provide communication capabilities; and cryptography grants them asset ownership, digital identity, settlement currency, and enforceable economic boundaries.
Agents without computing power cannot reason; without networks, they cannot coordinate; and without a secure financial architecture, they cannot function as trustworthy market participants.
Data centers, semiconductors, storage, communication networks, power infrastructure, blockchain, stablecoins, tokenized assets, cryptographic identities, and programmable wallets all form core components of the emerging machine economy.
AI computing factories produce intelligent capabilities, while encrypted payment channels facilitate the flow of intelligent value.
11. Investors Mistake Enterprises for the Statistical Core and Overlook the Vast Population of Intelligent Agents
Most Wall Street institutions still measure AI demand by tracking leading large-model companies, cloud providers, and enterprise software vendors—a mindset equivalent to focusing only on browser developers during the early days of the internet, rather than the billions of internet users themselves.
The true core entities are the vast numbers of software-based intelligent agents. Enterprises deploying customer service agents, developers building code-writing agents, healthcare institutions implementing diagnostic systems, financial institutions automating investment research, and ordinary users employing personal assistants—all continuously generate demand for inference computing power.
Every deployed autonomous agent system will generate associated financial activities: independently procuring resources, paying other agents, collecting service revenues, managing collateral, and reallocating funds across assets.
In the long run, every mobile user will interact with multiple personal agents; each enterprise will orchestrate hundreds of specialized agents across its business processes; every autonomous vehicle will be equipped with an independent financial wallet; and every humanoid robot will autonomously procure energy, spare parts, software, and services.
Demand for computing infrastructure will expand in tandem with the scale of digital labor, and transaction volumes will grow alongside the economic activities of these agents.
Twelve. Billions of agents require a corresponding financial risk management framework.
Focusing solely on whether OpenAI, Anthropic, Google, or Microsoft can individually sustain current infrastructure investments misses a far broader wave of transformation. These companies are becoming distribution platforms for digital labor, whose marginal operating costs continue to decline.
As enterprises shift from employing human workers to deploying autonomous agents, any scenario where intelligence creates value will drive a corresponding expansion in computing demand. Each reduction in inference costs unlocks greater numbers of deployable agents, higher task complexity, and increased total cloud computation.
The financial sector will experience equivalent-scale expansion: each new agent introduces incremental demand for payments, contracts, asset transfers, collateralization, and machine-to-machine transactions. Total transaction volumes will rapidly exceed the capacity limits of manual approval, regulatory oversight, reconciliation, and even human comprehension.
This is precisely why crypto-native financial infrastructure is indispensable: stablecoins, tokenized assets, programmable wallets, cryptographic identities, smart contracts, and verifiable ledgers collectively form a comprehensive risk control framework, enabling secure delegation of financial authority to software.
Just as McDonald's became renowned for serving billions of customers, cloud providers are quietly supporting the operation of billions of digital workers; meanwhile, crypto networks will carry trillions of transactions generated by these agents, becoming the foundational financial infrastructure.
AI endows agents with reasoning capabilities, while cryptography grants them economically enforceable execution rights.
To serve billions of digital intelligent agents, it is essential to build a financial risk control system tailored for high-speed machine operations; the transaction volume generated by such a system will far exceed that of any previous stage in human economic history.