Billions Served: Why AI Needs Crypto Financial Guardrails

Last week, I recorded my first interview with Mark Moss. I had listened to many of Mark’s interviews over the years, so I appreciated finally having the opportunity to sit down and have a conversation with him.

I always enjoy speaking with someone new about AI and crypto, but this conversation was especially compelling because it focused so directly on the intersection between the two. Most investors, technologists, and commentators still place them in separate silos. AI is discussed as a revolution in intelligence and productivity, while crypto is treated as a parallel story about money, markets, and digital assets.

That separation misses the larger transformation.

AI is creating a new population of autonomous economic actors. Crypto is building the financial infrastructure those actors will need to transact, establish ownership, verify identity, and operate within enforceable limits. The more capable agents become, the more inseparable these two technologies will appear.

Whenever I speak with someone who has spent years thinking deeply about that convergence, I leave with new questions, new ideas, and a clearer view of where the world may be heading. There is something unique about a conversation between two people trying to understand the future. When you talk about the past, you are examining a landscape that has already been mapped. The facts are known, the outcomes have occurred, and the story has largely been written.

Talking seriously about the future feels more like exploring a new land. You are searching for connections, testing assumptions, and trying to describe a world that does not fully exist yet.

My conversation with Mark led directly to this post because it reinforced a belief that has been growing stronger for me each day a new AI model is released: we are standing on the doorstep of a change the world has never experienced.

A new digital species is about to enter the economy.

For most of modern economic history, investors have evaluated demand through a human lens. More consumers purchase more goods, more businesses hire more employees, and more factories produce more output. Demand expands broadly with population, income, and GDP, creating a largely linear relationship between economic growth and resource consumption.

Artificial intelligence introduces a fundamentally different demand curve because the economic unit of production is beginning to shift from humans to software agents. Yet many investors continue to evaluate AI infrastructure through the framework of traditional enterprise technology. They see Microsoft, Amazon, and Google investing hundreds of billions of dollars in data centers while OpenAI, Anthropic, xAI, and a handful of frontier laboratories consume enormous amounts of compute. From that perspective, AI demand appears concentrated among only a few customers.

That framing overlooks the structural transition underway. Frontier model companies are evolving into platforms that deploy and orchestrate billions of autonomous software agents. The real source of demand is the digital workforce they enable. The cost of that digital workforce are tokens and the cost of tokens per capability is declining rapidly and the competition is rising rapidly.

It also overlooks the financial system this workforce will require.

Billions of digital agents cannot operate at scale through a financial system designed around human banking hours, manual approvals, delayed settlement, fragmented databases, and transactions that assume a person is sitting on the other side of every payment. Agents will need the ability to hold value, exchange value, verify counterparties, enforce spending limits, execute contracts, and settle transactions continuously.

Crypto provides the financial guardrails for that world.

We Are Measuring the Wrong Customer

Every quarter, investors ask the same question after hyperscaler earnings and their growing RPOs: Who is buying all of this capacity?

Viewed through a traditional enterprise lens, the answer appears straightforward. OpenAI rents Azure, Anthropic builds on AWS, Google operates its own infrastructure, and Meta trains its own models. If those companies represent the end market, AI infrastructure spending appears concentrated and potentially vulnerable to the spending decisions of only a few organizations.

The economics change once agents become the unit of demand. OpenAI, Anthropic, and Google increasingly resemble operating systems for digital labor rather than end users of compute. Every enterprise application, software workflow, and consumer service built on those platforms creates incremental inference demand. The hyperscalers are therefore building infrastructure for an installed base of software workers that may eventually number in the billions.

The same measurement problem applies to financial activity. A bank or payment network may initially see one large AI platform, enterprise, or digital wallet as the customer. Beneath that account could sit millions of agents making continuous economic decisions on behalf of individuals, companies, vehicles, robots, and other agents.

Counting the corporate account misses the economic activity occurring underneath it.

One Customer, Billions Served

This measurement illusion is surprisingly familiar.

McDonald’s appears to its suppliers as a single customer, yet its famous “Billions Served” campaign captured the economic reality beneath the corporate entity. One company represented billions of individual customer transactions occurring every year.

The AI economy is beginning to exhibit the same dynamic at a far greater scale. OpenAI may appear as a single Azure customer, Anthropic as a single AWS customer, and Google as a single internal cloud consumer. Beneath each platform sits an expanding population of software agents performing research, writing code, analyzing data, serving customers, generating content, executing financial workflows, and coordinating with other agents.

Each agent represents recurring demand for inference, and each task may require dozens of model calls across reasoning, retrieval, search, coding, memory, and planning systems. From the cloud provider’s perspective, one platform customer increasingly represents billions of software workers operating continuously.

The financial system will experience the same explosion.

A single consumer agent could compare prices across hundreds of vendors, negotiate subscription terms, rebalance savings, purchase compute, pay for data, compensate specialized agents, and settle dozens of microtransactions before the user wakes up. An enterprise agent could purchase inventory, rent processing capacity, manage working capital, hedge foreign exchange exposure, and pay other agents for completed tasks.

One human instruction could generate hundreds or thousands of underlying economic transactions.

The cloud providers are building infrastructure for billions of digital workers that have yet to be deployed. Crypto networks are building the transaction, settlement, identity, and ownership infrastructure those workers will need once they begin conducting business with one another.

Intelligence Needs Financial Agency

The first generation of AI systems primarily produced information. They answered questions, summarized documents, generated images, wrote code, and helped people make decisions.

Agents move from producing information to taking action.

Once an agent can act, it needs financial agency. It must be able to purchase resources, compensate service providers, receive revenue, manage budgets, and verify that contractual conditions have been met. An agent that can recommend a transaction but cannot execute it remains a sophisticated assistant. An agent that can securely control economic resources becomes an economic participant.

Traditional financial infrastructure was built around identifiable humans and incorporated layers of human friction as a form of security. People sign documents, enter passwords, wait for banks to open, approve transfers, reconcile accounts, and resolve disputes through centralized institutions. Those processes are cumbersome, but the delays are manageable when humans initiate a limited number of transactions.

They become an impossible bottleneck when software agents operate continuously and transact at machine speed.

The number of transactions generated by billions of agents is beyond normal human comprehension. Agents may pay fractions of a cent for data, inference, storage, bandwidth, identity verification, API access, intellectual property, or another agent’s specialized output. These transactions may occur thousands of times during a single workflow and millions of times across an enterprise.

A financial system for agents therefore needs to be programmable, continuously available, globally interoperable, auditable, and capable of settling extremely small transactions economically.

This helps explain why Stripe’s reported bid for PayPal matters beyond the immediate takeover price. Stripe has built much of the merchant-facing infrastructure of internet commerce, while PayPal brings an enormous consumer network, Venmo, Braintree, digital wallets, and its PYUSD stablecoin. Combining those assets could create a financial platform spanning both sides of agentic commerce: the businesses receiving payments and the consumers authorizing agents to make them.

The strategic message is more important than whether this particular transaction is ultimately completed. The payment industry is beginning to prepare for a world in which checkout is initiated by software rather than humans. Stripe is already building transaction-specific credentials that allow agents to make authorized purchases without gaining unrestricted access to a customer’s underlying payment information. PayPal is also investing in the infrastructure and protections required for agent-driven commerce.

The future payment network must do more than move money. It must verify that an agent is authorized, restrict what it can purchase, determine how much it can spend, protect the underlying credentials, assess fraud risk, and preserve a record of who approved the transaction.

Cards and bank accounts may continue to fund many purchases, but the control layer surrounding them must become programmable. Stablecoins, tokenized deposits, programmable wallets, cryptographic identity, and smart contracts provide the architecture needed as commerce moves from occasional human checkout to continuous machine execution.

The Stripe–PayPal bid can therefore be viewed as another sign that the financial industry recognizes what is approaching. AI agents are becoming economic actors, and the companies that control their wallets, permissions, identity, settlement, and access to merchants may control the most important financial gateway of the agentic era.

That architecture increasingly resembles crypto.

Crypto Provides the Guardrails

The term “crypto” often causes investors to focus on speculative tokens and price volatility. The more important long-term function will be the creation of digitally native property rights and financial rules for autonomous software.

Agents need more than a payment rail. They need guardrails determining what they own, what they may spend, which counterparties they may interact with, and what conditions must be satisfied before funds are released.

A programmable wallet can give an agent access to a specific budget without granting it unrestricted access to an individual’s or corporation’s entire balance sheet. Permissions can limit transaction size, merchant category, jurisdiction, asset type, time window, or cumulative spending. Larger transactions can require additional authentication or human approval.

Smart contracts can hold funds in escrow and release them only when verifiable conditions are met. An agent can pay another agent after a task is completed, a shipment is confirmed, a digital service is delivered, or a predefined performance threshold is achieved.

Stablecoins can provide agents with a digitally native medium of exchange whose value remains understandable in conventional economic terms. Tokenized deposits, Treasury securities, money-market instruments, securities, intellectual property, and real-world assets can allow agents to move between cash, collateral, investments, and productive resources without leaving the same programmable environment.

Public and permissioned ledgers can create audit trails showing which agent authorized a transaction, which rules governed it, what assets were transferred, and whether the action remained within its permitted mandate.

These are not decorative features. They are the financial safety system required to prevent autonomous commerce from becoming autonomous chaos.

The Scale Makes Traditional Oversight Impossible

Human financial supervision relies heavily on reviewing exceptions after activity occurs. Compliance departments inspect transactions, accountants reconcile books, auditors sample records, and regulators examine reports produced days, weeks, or months later.

That approach cannot scale to an economy in which billions of agents transact continuously.

Humans will not manually inspect every microtransaction between agents. They will establish policies, permissions, risk limits, and identity requirements that are enforced automatically at the moment of execution. Oversight must become embedded in the transaction itself.

Crypto makes financial rules programmable.

An enterprise can permit a procurement agent to spend up to a predefined amount with approved vendors. A consumer can allow a travel agent to purchase a flight within a price range but require approval before booking a hotel. A portfolio agent can rebalance among approved assets while remaining prohibited from borrowing, using leverage, or interacting with unauthorized protocols.

The ledger then provides a record of every action taken within those boundaries.

This creates a financial hierarchy suited to autonomous systems. Humans define goals and risk limits. Agents optimize within those limits. Cryptographic infrastructure verifies identity, enforces permissions, transfers ownership, and records the result.

Without those guardrails, giving an agent control of money would resemble giving an employee unlimited access to every corporate bank account and hoping internal policies are followed. The potential productivity would be overwhelmed by the security risk.

With programmable financial constraints, economic authority can be delegated in narrow, measurable, and revocable increments.

Stablecoins Become the Native Currency of Agents

Software agents are global by design. They may purchase services from providers in multiple countries, rent compute from decentralized networks, pay for data feeds, and compensate other agents regardless of geography.

The traditional banking system fragments that activity across currencies, correspondent banks, payment processors, operating hours, and jurisdiction-specific infrastructure. A transaction can pass through several intermediaries before reaching its final recipient, with each layer adding delay, cost, and reconciliation requirements.

Stablecoins offer agents a common digital settlement instrument that can move continuously across compatible networks. Their importance in the agent economy comes from their programmability, availability, and ability to function inside software workflows.

An agent does not care about the prestige of a banking relationship or the design of a payment application. It cares about execution speed, reliability, cost, liquidity, and certainty of settlement. It will route transactions toward the most efficient available rail just as software routes internet traffic toward available bandwidth.

As agents become more capable, they may continuously optimize where cash is held, which stablecoin is used, what network provides the lowest cost, and which tokenized instrument offers the best risk-adjusted yield. Financial balances that remain stationary because humans are inattentive may become increasingly mobile when agents constantly evaluate alternatives.

This is one reason the agentic economy may accelerate the movement toward stablecoins and tokenized assets. These instruments are designed to be held, evaluated, transferred, and exchanged by software.

Identity Becomes as Important as Money

For agents to transact safely, the financial system must know more than whether funds are available. It must know who or what is authorized to use them.

An agent may act for an individual, a corporation, a government agency, a vehicle, or another software system. Each relationship requires a verifiable chain of authority. The counterparty must know that the agent is genuine, that it has permission to perform the transaction, and that its credentials have not been revoked.

Cryptographic identity can establish those relationships without requiring every transaction to be manually confirmed. An agent can prove that it represents an approved organization, possesses a specific license, complies with jurisdictional requirements, or has been authorized to spend within a defined mandate.

Reputation may also become portable. Agents that complete work reliably can build verifiable transaction histories. Those records can help other agents evaluate counterparties, price risk, request collateral, or refuse interaction.

The agentic economy will therefore require a combination of identity, money, reputation, permissions, and settlement. Crypto networks are increasingly capable of combining these elements within a common architecture.

Agents Change the Shape of Demand

Traditional software improves worker productivity by helping employees complete tasks more efficiently. Agents increasingly complete the work themselves, allowing organizations to expand their effective workforce without proportionally expanding payroll.

A single employee may supervise dozens of specialized agents responsible for coding, research, legal review, scheduling, customer support, financial analysis, or content generation. Those agents routinely invoke additional specialized models during a workflow, creating multiple layers of inference beneath every user interaction.

One human decision can therefore generate dozens or hundreds of computational events. What appears to be a single prompt increasingly becomes an orchestration layer coordinating reasoning models, search systems, memory, retrieval, translation, image generation, and domain-specific agents. Software begins consuming software, causing compute demand to compound rather than simply grow alongside user adoption.

Financial activity compounds in the same way.

An agent completing a business task may purchase data from one provider, inference from another, storage from a third, and verification from a fourth. Each provider may use its own agents and subcontract additional digital services. A single economic objective branches into a network of machine-to-machine payments.

This is the fundamental difference between human labor and digital labor. Human organizations scale by adding employees. Agentic organizations scale by multiplying intelligence, decisions, and transactions.

Jevons’ Paradox Applies to Transactions

Economic history provides a useful framework for understanding this behavior. Improvements in coal efficiency increased coal consumption rather than reducing it. Lower bandwidth costs dramatically expanded internet usage, while cheaper storage created an explosion in digital information.

Artificial intelligence applies the same principle to cognition. Every reduction in inference costs makes new applications economically viable. Tasks that cannot justify one dollar per inference become attractive at one cent, while another order-of-magnitude reduction allows intelligence to be embedded into products, workflows, and services that previously could not support it.

The same principle applies to financial transactions.

Human beings avoid tiny transactions when fees and friction exceed the value being exchanged. Agents will transact whenever the expected benefit exceeds the marginal cost. As settlement becomes cheaper, agents can pay for increasingly granular units of data, compute, attention, energy, software, and intellectual property.

Lower transaction costs will therefore increase the number of transactions rather than merely make the existing volume cheaper. Machine commerce could ultimately generate transaction counts that dwarf today’s consumer payment system because software can divide economic activity into increments too small and too frequent for humans to manage.

The cheaper intelligence and settlement become, the more agents will consume both.

Compute and Crypto Are Complementary Infrastructure

The AI investment debate often separates compute infrastructure from digital-asset infrastructure. One is treated as productive technology, while the other is treated primarily as a speculative financial market.

The agent economy reveals that they are complementary layers of the same system.

Compute gives agents intelligence. Networks give them communication. Crypto gives them ownership, identity, money, and enforceable economic boundaries.

An agent without compute cannot reason. An agent without connectivity cannot coordinate. An agent without a secure financial architecture cannot become a trusted economic participant.

Data centers, semiconductors, memory, networking, power generation, blockchains, stablecoins, tokenized assets, digital identity, and programmable wallets are therefore components of the same emerging machine economy.

The AI factories will produce intelligence. Crypto rails will allow that intelligence to exchange value.

Investors Are Counting Companies Instead of Agents

Much of Wall Street still measures AI demand by counting frontier model companies, hyperscalers, or enterprise software vendors. That approach is analogous to evaluating the early internet by counting browser developers rather than internet users.

The installed base that matters is the population of software agents. Every enterprise deploying customer-service agents, every developer building coding agents, every healthcare provider implementing diagnostic systems, every financial institution automating research, and every consumer using personal assistants contributes recurring inference demand.

Every successful deployment also creates potential financial activity. Each agent may purchase resources, pay other agents, receive revenue, manage collateral, or move capital among competing opportunities.

Over time, every smartphone owner may interact with multiple personal agents, every enterprise workflow may coordinate hundreds of specialized agents, every autonomous vehicle may maintain its own financial wallet, and every humanoid robot may purchase energy, replacement parts, software, and services.

Infrastructure demand expands alongside the digital workforce. Transaction demand expands alongside its economic activity.

Billions of Agents Require Billions of Guardrails

The debate over whether OpenAI, Anthropic, Google, or Microsoft can individually justify today’s infrastructure spending misses the larger transformation taking place. These companies are becoming distribution platforms for an economy increasingly populated by software workers whose marginal cost continues to decline.

As organizations shift from hiring humans to deploying agents, demand for compute grows in proportion to every opportunity where intelligence can create value. Each reduction in the cost of inference expands the number of agents that can be deployed, the complexity of tasks they can perform, and the volume of computation flowing through the cloud.

The same expansion will occur in finance. Each additional agent creates new payments, contracts, asset transfers, collateral requirements, and machine-to-machine transactions. The volume will quickly surpass the capacity of humans to approve, supervise, reconcile, or even comprehend each individual action.

That is why crypto financial infrastructure is so important. Stablecoins, tokenized assets, programmable wallets, cryptographic identity, smart contracts, and verifiable ledgers provide the guardrails that allow financial authority to be delegated safely to software.

Just as McDonald’s became famous for “Billions Served,” the cloud providers are quietly becoming the infrastructure behind billions of digital workers. Crypto networks may become the financial infrastructure behind the trillions of transactions those workers generate.

AI gives agents intelligence. Crypto gives them economic agency with enforceable boundaries.

You cannot serve billions of digital agents without financial guardrails built for machine speed, and the resulting volume of transactions will be beyond anything the human economy has ever experienced.