Agentic commerce represents an economic model where software autonomously discovers, evaluates, and executes transactions continuously and programmatically. The autonomy spectrum ranges from human-in-the-loop approval workflows to fully autonomous operations. Key applications include purchasing SaaS licenses, booking travel, automating accounts payable, and reordering inventory.
A significant subcategory involves machine-native, sub-dollar transactions where "an AI agent calls an API, receives an HTTP 402 response ('Payment Required'), evaluates the price, pays, and consumes the resource" in milliseconds without traditional checkout processes.
AI crosses from chatbot to agent status when it reasons through multi-step tasks, calls external tools, evaluates results, and iterates autonomously. Frontier models including Claude, GPT, and Gemini now support production agent deployment across business verticals.
Payment capability removes access constraints, enabling agents to autonomously purchase SaaS seats, reorder supplies, query multiple data providers, and commission creative services.
Agentic payments span two categories: macro transactions (high-value, consumer-like purchases) and micro transactions (high-frequency, low-value API calls and data queries). Two open-source protocols have recently launched: x402 processed over "$135.7M in volume across 178.3M transactions since May 2025," while Machine Payments Protocol settled "over $38,000 in volume across ~184,600 transactions since mid-March 2026."
Engineers designing foundational internet infrastructure reserved HTTP status code 402 in 1997 for future native payment functionality. Card rail economics with fixed fees made micropayments unviable historically, causing advertising to become the internet's default business model.
Demand emerged following an AI capability threshold. "Since the releases of Claude 4.5 and GPT Codex 5.2 in mid-2025, agents have been capable of discovering unfamiliar APIs, understanding what they offer, evaluating their price, and deciding whether to pay – autonomously."
New blockchain infrastructure lowered payment costs and increased settlement speed. Purpose-built chains like Tempo offer "sub-cent fees with 500-millisecond settlement speeds," while established chains like Base and Solana reduced "gas costs to fractions of a cent, making micropayments in the $0.01–$1.00 range economically viable for the first time."
Critically, infrastructure improvements alone didn't create agentic commerce demand—capable agents provided the pull. Account abstraction, MPC wallets, passkeys, and sponsored transactions increasingly abstract blockchain complexity from end users.
Coinbase launched x402 in May 2025; Tempo launched MPP in March 2026. Both attracted live production traffic within months.
The incumbent risk involves disintermediation. If platforms controlling both demand and transaction routing pull payment flows outside existing networks, traditional payment incumbents may miss high-frequency micropayment segments where card economics perform weakest. Players combining "protocol-native speed with incumbent-grade trust, compliance, and distribution are best positioned to define the next generation of agentic payment infrastructure."
Agentic transactions fundamentally differ from human commerce. While humans make deliberate, low-frequency purchases from known merchants, agents executing research tasks might "make hundreds of purchases in an hour, each one a few cents or less, each from a different provider it just discovered."
Three dimensions define agent transactions: frequency, value, and relationship. Human commerce involves low-frequency, medium-to-high value transactions between parties with pre-existing relationships. Micro-transaction agentic commerce features high-frequency, low-value transactions between previously unconnected parties where "payment is the only interaction between buyer and seller."
Historically, digital value exchange occurred indirectly through advertising or bundled subscriptions. Agentic commerce enables "direct, usage-based exchange possible at any price point such that payment can now be attached to a single query, tool call, or data request, down to a fraction of a cent."
As agentic commerce scales, transaction counts will likely grow far faster than dollar volumes. A single agent session spending modest amounts might generate hundreds or thousands of individual payments, requiring systems to "handle far more decisions per dollar of volume than it does today."
Card infrastructure was purpose-built for human commerce: fewer transactions, larger values, and pre-established merchants. High-frequency, sub-dollar micropayments introduce different requirements, necessitating new architectural layers above traditional card systems.
The comparison reveals fundamental misalignments. Card rails optimized for fiat purchases, seconds-long authorization, day-scale settlement, and human verification. Machine-native micropayments require "sub-cent payment economics, near-instant confirmation, machine-readable credentials, and the ability to transact at high frequency with no pre-existing merchant relationship."
Subscription and metered billing models work around some constraints but require account creation, pre-negotiated pricing, and ongoing commercial relationships—impractical for agents discovering services, making single purchases, and moving on.
Cards remain flexible infrastructure extensible through adaptation. Visa's Trusted Agent Protocol (TAP), Visa Intelligent Commerce (VIC), and Visa Card Specification for MPP demonstrate how card infrastructure supports agent-initiated transactions across price ranges. New micropayment protocols represent "new architecture that expands the addressable surface of commerce above the card layer, not a replacement for it."
New protocols enable payment as "the first and only interaction between buyer and seller," creating merchant classes including APIs, tool servers, micro-service developers, and data providers. These businesses require "machine-readable payment flows, near-zero onboarding friction, and transaction economics that work at very small values."
x402 introduces a third-party facilitator. When agents pay through x402, funds go to a facilitator verifying server delivery before releasing payment to sellers.
x402 operates on any chain where a facilitator deploys, remaining un-limited to fixed blockchain sets. Current activity concentrates on Base, Solana, and Polygon, with Base carrying the majority. Payments settle onchain in stablecoins, primarily USDC. Facilitators also remain rail-agnostic, potentially settling through ACH, SEPA, or card networks, making "facilitators the key monetization and trust layer in the x402 model."
x402 currently supports fixed-price payments and consumption-based Upto models where agents authorize maximum amounts and "pay only for what it consumes."
Originally Coinbase's proprietary specification, stewardship transferred to the Linux Foundation in April 2026. The x402 Foundation launched with vendor-neutral governance and support from companies including Visa, AWS, American Express, Base, Circle, Google, Mastercard, Microsoft, Polygon Labs, Shopify, and Solana Foundation. This signaled intent to "solidify x402 as a shared industry standard rather than a single-company project."
Traditional commerce assumes human buyers with intent, judgment, and accountability. Agentic commerce raises questions regarding all three, as software acts on delegated authority making "decisions at machine speed, often without real-time human oversight."
Three primary failure modes emerge:
Mis-purchase risk involves agents buying incorrectly through task misinterpretation, wrong service selection, or overpayment. Unlike humans catching mistakes before checkout, "an agent operating autonomously may not recognize the error at all, and at machine speed, a single misconfigured agent can execute hundreds of bad transactions before anyone notices, creating reconciliation, refund, and dispute overhead at a scale that is difficult to contain."
Adversarial attacks through prompt injection manipulate agents into unauthorized actions, redirecting purchases, exfiltrating funds, or triggering unauthorized spending. Attack surfaces expand with autonomy—"the more freedom an agent has to transact, the more damage a compromised agent can do."
Liability remains unclear when agents execute unauthorized purchases. Multiple parties—delegating humans, hosting platforms, model providers, merchants—have plausible claims. "Existing legal and regulatory frameworks were not designed for this chain of delegation, and clear precedents may not be available."
Cascading failures emerge as agents transact with other agents rather than merchants directly. Multi-hop payment chains where "any agent in the chain delivers a bad result, every agent upstream has already paid for work built on that output" lack established resolution mechanisms in traditional commerce.
Trust challenges receive multi-layer addressing from both crypto-native and card-native players.
At the protocol layer, x402 and MPP build trust into transaction mechanics. "Third-party facilitator approaches are able to verify delivery before releasing funds, escrow mechanisms let buyers pay only for value consumed, and consumption-based authorization limits exposure by capping what an agent can spend on a single request."
Identity layer solutions remain fragmented. Most agents authenticate via API keys or wallet addresses providing access verification without history or reliability signals. Public registries including x402scan and MPPscan enable basic discovery. Coinbase's Bazaar registry maps merchants to payment addresses. ERC-8004's Trustless Agents Standard aims to "make agent reputation portable across networks, allowing merchants to verify an agent's transaction history before deciding whether to transact."
Card-side identity solutions include Visa's TAP, Mastercard's Agent Pay, and Stripe/OpenAI's ACP, anchoring agent identity to existing authorization frameworks, inheriting "decades of KYC and fraud infrastructure but associated with card rails."
Policy layer programmable spend controls emerge across ecosystems. Coinbase Agentic Wallets offer "gasless transactions on Base with programmable limits," while Turnkey, Privy, and Safe encode allowlists, spend caps, and merchant routing into wallet infrastructure. Card-side tokenized credentials implement similar functions through familiar infrastructure.
No universal agent identity standard bridges crypto-native and traditional rails. Reputation built in one ecosystem doesn't transfer elsewhere. Disputes lack established resolution mechanisms. Chargeback windows and evidence requirements designed for human-speed commerce don't map to "hundreds of sub-cent API calls" executed hourly.
Liability remains legally ambiguous. No jurisdiction established clear responsibility distribution "among the human who delegated authority, the platform that hosted the agent, the model provider, and the merchant." These gaps represent both significant risks and clear opportunities for traditional payment ecosystem incumbents to add value.
Distinctions between crypto-native and card-native protocols blur. MPP spans both onchain crypto and fiat payments via shared tokens. Visa's Card Specification SDK extends protocol into card-based agent commerce. This means "a single machine-payments framework can increasingly support both stablecoin-based flows and card transactions."
For card networks and issuers, this creates "a path into agentic payment flows without requiring merchants to adopt a completely separate stack." Stripe added x402 payment support on Base with USDC in March 2026, "linking stablecoin-native payment flows to the same broader agent commerce ecosystem."
The trajectory points "toward convergence rather than competition: cards for proxy purchases inside existing merchant networks, stablecoins for machine-native micropayments, and hybrid flows where both are used within the same workflow."
Three requirements remain for convergence to move beyond early adoption:
First, interoperability between standards must mature. Agents should combine "AP2-style mandates for authorization, x402 or MPP for payment execution, and TAP or similar systems for trust and card-based verification within a single workflow without bespoke integration for every combination."
Second, regulatory frameworks must evolve. "The treatment of agent-initiated transactions, liability, and cross-border agent commerce remains unsettled, especially as these systems begin operating across multiple payment rails and jurisdictions."
Third, incumbents need defined roles. Global payment processors require clarity "whether as issuers of agent credentials, processors of agent transactions, or providers of the trust and dispute infrastructure that crypto-native protocols still lack."