The Complete Guide to Machine-to-Machine Payments and Why Tempo Changes Everything
AI agents cannot open bank accounts. They cannot pass KYC checks. They have no legal identity. Yet within the next 18 months, autonomous software systems will handle somewhere between $300 billion and $500 billion in transaction volume, according to industry projections. The infrastructure to support this does not exist in traditional finance. As of March 18, 2026, Tempo and the Machine Payments Protocol represent the most significant attempt to build it.
This guide breaks down exactly what Tempo does, how the Machine Payments Protocol works, what alternatives exist across the ecosystem, and what businesses need to understand about enabling their AI agents to transact autonomously.
This guide is written by Yuma Heymans (@yumahey), founder of o-mega.ai, the AI workforce platform where autonomous agents learn to use business tool stacks and execute workflows. His work focuses on the infrastructure required for agents to operate as genuine economic actors.
Contents
- What Happened on March 18, 2026
- The Problem Tempo Solves
- How the Machine Payments Protocol Works
- The Agentic Payments Ecosystem
- Know Your Agent: Identity and Verification
- Spending Controls and Budgets
- The Economics of Machine Commerce
- Legal Liability and Accountability
- Integration Requirements for Businesses
- What This Means for the Future
1. What Happened on March 18, 2026
Tempo, a blockchain startup incubated by Stripe and the venture firm Paradigm, launched its mainnet and simultaneously unveiled the Machine Payments Protocol (MPP). This represents the first production-grade infrastructure specifically designed for AI agents to send and receive money autonomously - (Fortune).
The timing matters. Tempo had operated in testnet for three and a half months, allowing developers to experiment with payments using USDT and USDC stablecoins. The mainnet launch means the infrastructure is now live for real transactions. The simultaneous release of MPP provides the standardized protocol that agents and services need to coordinate payments programmatically.
Matt Huang, cofounder and managing partner of Paradigm, serves as Tempo's CEO while maintaining his role at the venture firm. The company raised $500 million at a $5 billion valuation in 2025 from investors including Thrive Capital. The team currently comprises 15 full-time employees, though the design input came from a broader coalition including Anthropic, OpenAI, Shopify, Mastercard, Visa, Deutsche Bank, Revolut, Nubank, and Standard Chartered - (Paradigm).
The partner list signals the ambition. These companies collectively serve billions of users worldwide. Their participation in Tempo's development suggests that major financial institutions view agentic payments not as an experimental curiosity but as critical infrastructure for the next decade of commerce.
2. The Problem Tempo Solves
Traditional payment systems were designed for humans. Every bank account requires identity verification. Every credit card requires a human applicant. Every wire transfer assumes a person initiated it. AI agents break all of these assumptions.
Coinbase CEO Brian Armstrong articulated the problem directly: AI agents cannot open bank accounts because they cannot satisfy Know Your Customer requirements. Banks verify identity through government-issued documents, biometric data, and proof of address. Software has none of these. Yet the trajectory of AI development points toward agents that need to purchase API access, pay for compute resources, acquire datasets, book services, and transact with other agents - (Coinbase).
The scale of this mismatch is growing rapidly. Gartner predicts that 40% of enterprise applications will include integrated task-specific agents by 2026, up from less than 5% in 2025. Each of these agents potentially needs to spend money. An AI agent monitoring your competitors might need to pay for premium data feeds. An agent managing your advertising might need to purchase media. An agent handling customer support might need to issue refunds or credits.
Without native payment infrastructure, developers have resorted to workarounds. They embed personal credit cards into agent systems. They pre-fund accounts and track balances manually. They build custom integrations for each payment scenario. These approaches do not scale. They create security vulnerabilities. They make audit and compliance nearly impossible.
Tempo and MPP attempt to solve this by building payments infrastructure from first principles, assuming the transacting entity is software rather than a human. The blockchain handles settlement. The protocol handles coordination. The extensions handle compatibility with existing payment rails. Together, they create a system where an agent can request a resource, receive a payment request, authorize payment from its wallet, settle instantly, and receive the resource, all without human intervention at any step.
3. How the Machine Payments Protocol Works
MPP provides a standardized method for agents and services to coordinate payments programmatically. At its core, the protocol defines how to request, authorize, and settle payments between machines. The technical architecture introduces several key innovations that distinguish it from traditional payment APIs - (mpp.dev).
The sessions primitive is perhaps the most important. Rather than requiring an on-chain transaction for every interaction, MPP allows agents to authorize a spending limit upfront and then stream micropayments continuously within that session. An agent might open a session with a $10 limit and then make hundreds of sub-cent payments to various services without each payment requiring separate blockchain confirmation. This dramatically reduces transaction costs and latency.
The protocol follows a straightforward flow. An agent requests a resource from a service. The service responds with a payment request specifying the amount and payment details. The agent authorizes payment from its wallet. The transaction settles instantly on the Tempo blockchain. The service delivers the requested resource. The entire cycle can complete in under a second.
MPP is designed to be rail-agnostic. While it runs on Tempo natively, the protocol itself does not require Tempo. Stripe has extended it to support cards, wallets, and other payment methods through their platform. Visa contributed specifications for letting agents pay with credit or debit cards. Lightspark has extended it for Bitcoin payments over the Lightning network. This means agents can use MPP regardless of whether they are paying in stablecoins, traditional currency, or cryptocurrency.
At launch, the payments directory already includes integrations with more than 100 services spanning model providers, developer infrastructure, compute platforms, and data services. This initial ecosystem means agents can immediately use MPP to pay for practical resources rather than waiting for adoption to build.
The blockchain underneath handles the settlement. Tempo targets 100,000+ transactions per second with sub-second finality. It supports multiple stablecoins, primarily USDC and USDT, with built-in exchange capabilities. Users pay gas fees directly in any supported USD-denominated stablecoin, eliminating the need to hold native tokens for transaction fees.
4. The Agentic Payments Ecosystem
Tempo and MPP do not exist in isolation. A broader ecosystem of protocols, platforms, and standards has emerged to address different aspects of agent payments. Understanding this landscape helps businesses choose the right infrastructure for their specific needs.
x402 Protocol (Coinbase) revives the HTTP 402 "Payment Required" status code that was reserved in HTTP/1.1 but never implemented. x402 embeds payments directly into HTTP requests, allowing websites, APIs, and applications to request and collect payments natively within the request-response cycle. In the last nine months, AI agents have executed over 140 million on-chain transactions generating more than $43 million in volume with an average transaction size of $0.31 through x402. The protocol recently expanded to support any ERC-20 token and introduced Sign-in-with-X (SIWX) to reduce friction for repeated payments - (Coinbase).
Agent Payments Protocol (AP2) from Google focuses on the trust and verification layer. AP2 addresses authorization (proving a user gave an agent specific authority), authenticity (ensuring requests reflect true user intent), and accountability (determining responsibility when transactions go wrong). The protocol uses verifiable digital credentials (VDCs), which are tamper-evident, cryptographically signed objects that serve as the building blocks of a transaction. AP2 supports different payment types from credit cards to stablecoins to real-time bank transfers - (Google Cloud).
Universal Commerce Protocol (UCP) from Google addresses the discovery and checkout problem. Without standardization, retailers wanting to sell through AI agents need to build separate integrations for each AI platform. UCP provides a common language for commerce journeys between consumer surfaces, businesses, and payment providers. It launched January 2026 with support from Shopify, Etsy, Wayfair, Target, Walmart, and is endorsed by over 20 partners including Adyen, American Express, Mastercard, Stripe, and Visa - (Google Developers).
Mastercard Agent Pay represents the traditional payment network's response. Mastercard completed Australia's first fully authenticated agentic transactions in January 2026 and announced Europe's first live end-to-end payment executed by an AI agent in March 2026 with Banco Santander. Agent Pay integrates into Microsoft's Copilot Checkout and OpenAI's Instant Checkout program in ChatGPT. The Mastercard Agent Suite, available Q2 2026, combines customizable AI agents with consulting support - (Mastercard).
Visa Intelligent Commerce takes a similar approach, with pilots across Asia Pacific, Europe, and Latin America anticipated in early 2026. Visa reports working with more than 100 partners, with over 30 actively building in their sandbox and over 20 agents integrating directly. Their Trusted Agent Protocol, supported by Akamai, delivers identity, authentication, and fraud controls for merchants to confidently accept AI agent purchases - (Visa).
Stripe's Agentic Commerce Suite focuses on the merchant side. It enables businesses to make their products discoverable to AI agents, simplifies checkout for agent transactions, and handles agentic payments through a single integration. Brands including URBN, Etsy, Ashley Furniture, Coach, Kate Spade, and Revolve have already onboarded. JD Sports became the first retailer to use the suite for AI-driven discovery through LLMs - (Stripe).
Ramp Agent Cards, launched March 2026, provide a corporate card infrastructure specifically for AI agents. Agents request tokenized card credentials on a per-transaction basis, eliminating the need to share actual card numbers. Ramp automatically applies user-defined controls including spend limits, approval workflows, and category restrictions. The cards use Visa's Intelligent Commerce protocol to scope each purchase to the intended transaction - (Ramp).
The proliferation of protocols reflects the genuine complexity of the problem. Different use cases require different solutions. A consumer AI assistant buying products needs different infrastructure than an enterprise agent paying for API access. The protocols are increasingly designed for interoperability, with AP2 compatible with UCP and MPP designed to work across multiple payment rails.
5. Know Your Agent: Identity and Verification
Traditional finance operates on Know Your Customer (KYC) principles. Banks verify that account holders are real people with verifiable identities. This framework does not translate to AI agents. A new framework, Know Your Agent (KYA), is emerging to fill the gap.
KYA establishes that a digital agent is authorized to act on behalf of a verified human or organization. Where KYC establishes that a person exists and has been verified, KYA establishes that an agent has been authorized to operate within specific parameters. The chain between human identity and agent authority must remain intact across every transaction - (Sumsub).
The verification challenge has multiple dimensions. First, agents need provable identity: something that distinguishes one agent from another and persists across interactions. Second, agents need provable authorization: cryptographic evidence that a specific human or organization delegated specific authorities to this specific agent. Third, agents need provable constraints: verifiable limits on what the agent can and cannot do.
Mastercard's Verifiable Intent protocol, developed in collaboration with Google and aligned with AP2 and UCP, creates a tamper-resistant record of what a user authorized when an AI agent acts on their behalf. This addresses a core problem: when an agent makes a purchase, how does the merchant know the agent had permission? How does the payment network verify the transaction is legitimate? Verifiable Intent provides the cryptographic proof - (Mastercard).
For businesses deploying AI agents, KYA requirements mean implementing several capabilities. Agents must have stable, verifiable identities, typically cryptographic key pairs that can sign transactions. Delegation must be explicit and auditable, with clear records of what authorities were granted. Constraints must be enforceable, not merely advisory. And the entire chain from human identity through delegation to agent action must be traceable for compliance purposes.
The regulatory landscape adds complexity. Under the EU's Sixth Anti-Money Laundering Directive (6AMLD) and the incoming Payment Services Directive 3 (PSD3), verification requirements are tightening. Regulators are shifting from checking whether controls exist to evaluating whether controls actually work. Agents that transact without proper verification chains create compliance exposure for the businesses that deploy them.
6. Spending Controls and Budgets
Autonomous agents with unlimited spending authority present obvious risks. The infrastructure emerging around agentic payments includes multiple layers of controls to bound agent behavior.
Spending limits operate at multiple levels. Transaction limits cap individual purchases. Session limits cap total spending within a defined period. Period limits (daily, weekly, monthly) provide longer-term bounds. These limits are enforced at the protocol level, not merely logged for later review. An agent that attempts to exceed its limits simply cannot complete the transaction.
Merchant restrictions control where agents can transact. Category restrictions might allow an agent to purchase software subscriptions but not physical goods. Vendor whitelists might limit transactions to pre-approved counterparties. Geographic restrictions might prevent transactions with merchants in certain jurisdictions. These controls help businesses ensure agents stay within intended operational boundaries.
Approval workflows introduce human oversight at defined thresholds. Transactions below a threshold proceed automatically. Transactions above require explicit human confirmation before the agent can continue. This preserves autonomy for routine operations while ensuring humans review significant decisions.
Ramp's implementation illustrates how these controls work in practice. When a company issues Agent Cards, administrators define spending policies in the Ramp dashboard. The policies specify transaction limits, category restrictions, and approval requirements. When an agent attempts a purchase, Ramp validates the transaction against these policies before authorizing the card charge. Non-compliant transactions are blocked immediately, with no need for post-hoc review or reversal.
Prepaid approaches offer structural control beyond policy enforcement. Rather than giving agents access to credit lines or corporate accounts, businesses can fund agent-specific prepaid balances. The agent can only spend what has been pre-loaded. When the balance reaches zero, spending stops. This caps maximum exposure regardless of any policy configuration errors.
Coinbase's Agentic Wallets implement similar controls. Session caps limit how much agents can spend per session. Transaction limits control individual transaction sizes. The wallet enforces these programmatically, preventing agents from exceeding their authority even if their code contains bugs or is compromised.
For businesses implementing agentic payments, the control framework should match the risk profile. Agents performing low-value, high-frequency transactions (like paying for API calls) might operate with minimal oversight. Agents making larger or more consequential purchases should have tighter limits and more approval requirements. The infrastructure now exists to implement these distinctions at the protocol level rather than relying on agent code to self-enforce.
7. The Economics of Machine Commerce
The economic projections for agentic commerce range from significant to transformational. McKinsey projects the market will reach $3 trillion to $5 trillion globally by 2030. Gartner estimates that AI "machine customers" could influence or control up to $30 trillion in annual purchases by the same date. Even conservative estimates suggest AI agents will handle 15% to 25% of all US e-commerce by 2030, translating to $300-500 billion in transaction volume - (Antler).
Current volumes are smaller but growing rapidly. The x402 protocol has processed over 140 million transactions generating $43 million in volume. Virtuals Protocol reports an aggregate "agentic GDP" of $477 million from over 15,800 live AI agent projects, with a target of $3 billion for 2026. If current growth continues, agent-initiated micropayments alone could exceed $500 million in monthly volume by Q4 2026.
The economics differ fundamentally from human commerce. AI agents generate hundreds of sub-cent micro-actions per interaction. Traditional percentage-based payment fees or minimum charges make these transactions unprofitable. A payment processor charging 2.9% plus $0.30 per transaction cannot economically process a $0.01 API call. This is why crypto and blockchain rails have gained traction: they can settle transactions at fractions of a cent.
The x402 protocol specifically addresses micropayment economics. Transactions as small as $0.01 can now be profitable because blockchain settlement costs are measured in fractions of a cent rather than tens of cents. This unlocks genuine pay-per-use monetization for APIs, content, and services that could never work with traditional payment rails.
For service providers, agent commerce opens new revenue models. An API that previously offered only monthly subscriptions can now charge per-call. A content provider that required annual licenses can now charge per-article. A compute provider that needed minimum commitments can now bill per-second. The ability to meter and charge at arbitrarily fine granularity changes what business models are viable.
For businesses deploying agents, the economics shift operational spending from human labor to compute and API costs. A company running 50 AI agents making hundreds of purchases weekly might generate more payment volume than their entire human workforce. Unlike humans, agents do not forget to submit receipts, do not book out-of-policy travel, and do not require reimbursement workflows.
Visa's Jack Forestell has called the agentic web "the biggest payments opportunity in two decades." The statement reflects both the volume potential and the infrastructure gap. Traditional payment networks were not designed for machine-to-machine commerce. Those who build the infrastructure to serve it stand to capture significant value as volumes grow.
8. Legal Liability and Accountability
When an AI agent makes a purchase that turns out to be fraudulent, unauthorized, or simply wrong, who bears responsibility? The legal framework for agent liability remains unsettled, but certain principles are emerging.
AI agents cannot be held liable themselves. They have no legal identity. They cannot enter contracts. They cannot be sued. Responsibility necessarily flows to humans or organizations. The question is which humans or organizations.
Courts and regulators generally look across the entire chain: who designed the system, who deployed it, who controlled its outputs, who profited from the conduct, and who had the ability to prevent the harm. In practice, the deploying company typically faces the most direct exposure because it puts the AI in front of consumers or uses it to conduct business - (Lathrop GPM).
The Air Canada case established important precedent. Air Canada was required to honor a discount promised by its customer service chatbot, even though the chatbot's promise contradicted official policy. The tribunal ruled that a company is liable for all outputs of AI systems it deploys, regardless of whether those outputs are accurate. Companies cannot avoid responsibility simply because software acted autonomously.
This creates significant implications for agentic payments. If your AI agent purchases something it was not authorized to purchase, your company likely cannot disclaim responsibility by pointing to autonomous agent behavior. If your agent enters a contract on terms you did not intend, you may be bound by those terms. If your agent commits fraud, your company may face fraud liability.
The liability landscape differs by region. In Europe, the revised Product Liability Directive extends traditional product liability rules to cover software and AI. A developer or producer of a defective AI system can be held strictly liable for harm regardless of fault. In the US, liability questions are generally left to existing tort law principles applied case-by-case.
For businesses deploying transacting agents, risk mitigation requires several approaches. First, implement robust controls that actually prevent unauthorized actions rather than merely logging them. Second, maintain clear audit trails showing what authorities were delegated and what constraints were imposed. Third, ensure agents operate within scopes that the business is prepared to stand behind. Fourth, consider insurance products that may emerge to cover agentic liability exposure.
The regulatory picture continues to evolve. Payment Services Directive 2 (PSD2) and Strong Customer Authentication (SCA) regimes require clear human authorization for payment orders. How these requirements apply to agent-initiated payments remains unclear. Businesses should expect regulatory guidance and potentially new rules as agentic commerce scales.
9. Integration Requirements for Businesses
Enabling AI agents to transact requires infrastructure changes at multiple levels. The specific requirements depend on whether a business is deploying agents that spend money, accepting payments from agents, or both.
For businesses deploying spending agents, the fundamental requirement is wallet infrastructure. Agents need cryptographic identities (key pairs) that can authorize transactions. They need access to funds, whether through direct wallet balances, credit lines, or prepaid accounts. They need integration with payment protocols (MPP, x402, AP2) appropriate to their transaction types.
Coinbase's Agentic Wallets provide one approach. The infrastructure includes pre-built functions for authentication, funding, sending, trading, and earning. Programmable spending limits and session controls are built in. Over 50 million transactions have already processed through the platform, demonstrating production readiness.
Ramp's Agent Cards provide another approach, specifically for enterprises using traditional card rails. Agents receive tokenized credentials rather than actual card numbers. Spending policies are defined in the Ramp dashboard and enforced at transaction time. The integration requires modifying agent code to request credentials from Ramp's API rather than using embedded card numbers.
For businesses accepting agent payments, the requirements center on discoverability and checkout. Agents need to find your products and understand your pricing. They need to complete checkout programmatically without human-oriented interfaces like CAPTCHAs or multi-step forms.
Stripe's Agentic Commerce Suite addresses these needs. Connecting your product catalog to Stripe and selecting which AI agents can access your store enables discovery. Stripe handles checkout, payments, and fraud detection, sending order events to your existing commerce stack. The Checkout Sessions API manages shipping, taxes, and payment processing.
Google's Universal Commerce Protocol provides standardized integration for businesses wanting to sell across multiple AI surfaces. Rather than building separate integrations for each AI platform, UCP provides a common language. The protocol integrates via APIs, Agent2Agent (A2A), and Model Context Protocol (MCP).
For both scenarios, security requirements are significant. Agents should never have access to raw card numbers or account credentials. Tokenization and session-scoped credentials minimize exposure from compromised agents. Audit logging must capture the full chain from delegation through authorization through transaction. Fraud detection systems need updating to recognize agent transaction patterns, which differ from human patterns.
The integration effort is not trivial, but the protocols and platforms have matured to the point where implementation is engineering work rather than research. Most businesses should start with a single use case, validate the integration, and expand from there.
10. What This Means for the Future
The launch of Tempo and MPP represents infrastructure maturation rather than technology breakthrough. The underlying capabilities, blockchain settlement, cryptographic authorization, programmable constraints, have existed for years. What changed is that major financial institutions now view agent payments as production-ready and commercially significant.
The next 18 months will likely determine whether agentic commerce remains a niche capability or becomes standard infrastructure. Several factors will drive adoption.
Protocol convergence or competition will shape the market. Currently, businesses face choices between MPP, x402, AP2, UCP, and proprietary solutions from Visa and Mastercard. If these protocols converge on interoperable standards, adoption accelerates. If they fragment into competing ecosystems, adoption slows as businesses wait for clarity.
Agent capability expansion will drive transaction volume. Today's agents handle relatively narrow tasks. As agents become more capable and take on broader responsibilities, their payment needs expand correspondingly. An agent that can only search for products generates no transactions. An agent that can compare, negotiate, and purchase generates significant volume.
Regulatory clarity will either enable or constrain growth. Current regulations were not designed for agent commerce. Regulators may adapt existing frameworks, create new agent-specific rules, or create uncertainty that chills adoption. The EU's AI Act obligations taking force in 2026 will provide early signals.
For businesses, the strategic question is timing. Early movers can build competitive advantage by enabling agent commerce before competitors. They can capture market share among AI-first consumers and businesses. They can learn operational lessons while volumes are manageable.
Late movers avoid implementation costs for immature infrastructure. They benefit from protocol convergence and regulatory clarity. They implement against proven patterns rather than experimental approaches.
Platforms like o-mega.ai represent the next generation of this infrastructure, providing AI workforce platforms where agents already have the identity, authorization, and tool integration required for agentic commerce. The underlying infrastructure shift toward agents as economic actors opens opportunities for businesses to automate not just tasks but entire transaction chains.
The fundamental trajectory is clear. AI agents are becoming economic actors. They need to spend money. The infrastructure to enable this is now live and backed by major financial institutions. Businesses that depend on commerce, either as buyers or sellers, will eventually need to engage with agent payments. The question is not whether, but when.
This guide reflects the agentic payments landscape as of March 18, 2026, the day Tempo launched its mainnet. Protocols, platforms, and regulatory requirements change rapidly. Verify current details before making implementation decisions.