Agentic payments refers to AI agents transferring assets or intervening in the exchange of value onchain. What makes this primitive fundamentally different from other market sectors in crypto is the concept of delegated economic actions, where software can find options, choose among them, and transact inside the rules and constraints set by a person, company, or application.
Moving money from A to B is straightforward for both agents and humans, or human-written software, for that matter. What’s novel about this concept is the ability to rely on autonomous operations that can prove who signed off on the payment, what the AI agent was permitted to buy, whether the seller delivered, and who steps in when something goes wrong.
Stablecoins are the engine fueling the adoption of agentic payments. Their programmability is key for situations in which the seller is software and what you are buying is access. In practice, stablecoins act as the programmable fuel that disintermediates payments for API calls, data, inference, browser sessions, paid crawling, and similar machine web resources. Given the broad adoption of fiat, one might think that it is hard to justify at retail checkout. However, the concept of programmability becomes a differentiator when applied in enterprise, where money itself is embedded in approval, compliance, and accounting workflows.
Key Takeaways
Stablecoins are most useful for agentic resource markets: They remove account setup for small, one-off software purchases: API calls, data, inference, browser sessions, and paid pages.
Stablecoins solve settlement, not the whole payment: A stablecoin transfer only shows that funds went from one wallet to another; it doesn’t prove the agent was authorized, that the price fit the user’s instructions, that the right resource was delivered, or that a refund is justified.
Retail still belongs to payment operators: Stablecoins can fund or settle payments, but PSPs and card networks own checkout, fraud, refunds, returns, and disputes.
Enterprise needs finance wrappers: USDC and USDT can move value, but companies still need vendor approval, KYB, budgets, GL coding, tax, receipts, and ERP reconciliation.
Agentic commerce is growing: The agentic footprint becomes more visible by the day as activity grows, participation increases, and infrastructure improves.
Value accrues above settlement: Retail favors PSPs and merchant platforms. Enterprise favors enterprise finance software. Machine web payments favor gateways, facilitators, wallet infrastructure, and stablecoin providers that own verification, routing, reconciliation, and recovery.
Delegated Payments Need Authorization Records
Traditional online payments assume a human is in the loop: searching, comparing, accepting terms, checking the price, entering credentials, and confirming the purchase. Agentic payments, however, don’t require a human making each of those decisions in sequence. Instead, the user simply declares their intent, like setting a final goal, and the agent autonomously decides what the best path to initiate and settle the payment is.

An agentic payment needs four linked records: the mandate that grants authority, the quote that defines the offer, the payment instruction that moves value, and the delivery record that proves what was received. Break the link between them, and the buyer can approve one thing, pay for another, and receive something else.
Agentic receipts need to prove more than payment. They need to show the seller, resource, price, mandate, and delivery result. Otherwise the payment may settle, but the buyer still cannot prove what was authorized, what was bought, or who is responsible if/when the transaction fails.
Each major payment network attacks a different part of the control record. They are not solving settlement alone but trying to bind intent to the transaction. The emerging standards split the problem across intent, agent recognition, and credential control. Google's Agent Payments Protocol, or AP2, connects intent, checkout, the selected items, and the payment instrument. Mastercard's Verifiable Intent records the user’s instruction to the agent. Visa's Trusted Agent Protocol gives merchants a way to recognize approved agents. Stripe's Shared Payment Tokens limit the damage from credential misuse by attaching each token to a seller, amount, currency, and expiry window.

Stablecoins solve settlement, but a wallet signature cannot guarantee all of the above by itself. A wallet signature can show that a key authorized a payment, but not whether the agent had permission, the vendor was approved, the buyer received the resource, or who takes the loss if delivery fails. Stablecoins settle; agentic payments still need authorization, delivery evidence, and recourse. Compliance, vendor approval, delivery, reconciliation, or liability all have to be built around the full payment flow.
Authorization and Spend Management
Settlement moves the money. Spend management controls whether the software was allowed to spend it. Spend management software is the layer above settlement. It takes a mandate and turns it into rules that wallets, merchants, PSPs, facilitators, and finance systems can enforce.
The mandate must become enforceable policy: check the quote, verify the resource, approve the payment, consume the authorization, record delivery, route exceptions, and reconcile the transaction in finance. It’s a loop that looks like this: mandate issued -> seller quote retrieved → resource descriptor checked → payment instruction matched against policy → nonce or one-time authorization consumed → delivery receipt returned → exception route triggered if delivery fails → finance record created.
For agentic payments to work, it’s key to set the agent’s limits before it acts. Pre-authorization is the step involved in setting the rules before the agent reaches the market. The mandate should define the actor, allowed purchases, approved sellers, spend limits, expiry, payment credential, and approval triggers. It should also state whether the agent can delegate, switch sellers, accept quote changes, or retry with the same credential after failure.

Pre-authorization sets the boundary. Runtime authorization checks whether the agent is still inside it. The seller may have changed, the price may have moved, the API may have failed, or the agent may have been prompt-injected. Before payment, the wallet or facilitator should compare the mandate, agent action, seller quote, resource descriptor, and payment instruction. After delivery, the receipt and delivered resource should point back to the same event. Otherwise the buyer can approve one thing, pay for another, and receive something else.
In agentic payments, security failures become financial failures. Prompt injection, for example, can change what the agent believes it is buying. Goal hijacking, tool misuse, identity abuse, and excessive agency can then show up as unauthorized spend, broken receipts, disputes, compliance gaps, or delivery to the wrong request.

Settlement is the commodity layer. The premium layer is evidence: who approved the agent, whether the payment stayed inside the mandate, what the seller delivered, and where the loss goes if something fails.
Three Main Markets
Agentic payments are divided into retail commerce, enterprise procurement, and machine web access for agentic resources.
Retail is where Stripe, Visa, Mastercard, wallets, and merchant platforms already control fraud, refunds, and buyer recourse. Enterprise is where Circle/USDC, Tether/USDT/USDT0, public chains, and networks such as Canton still need AP (Accounts Payable), procurement, policy, and ERP (Enterprise Resource Planning) wrappers before the transfer happens. Machine web access is low-value software access: API calls, data, inference, browser sessions, search, storage, and paid pages. Stablecoins fit best where delivery is digital, immediate, low-dispute, and too small for account setup.
In retail, stablecoins will likely sit behind the scenes. A shopping agent can find a product and build a cart, but the merchant still controls inventory, discounts, tax, shipping, fraud, returns, refunds, and support. Even though stablecoins can improve retail funding without changing retail checkout (e.g., a user spends USDC while the merchant receives fiat), that does not make retail stablecoin-native. Stablecoins can sit behind crypto cards, wallets, and merchant payouts, but the overall workflow must still be orchestrated by PSPs, wallets, card networks, and merchant platforms.

In enterprise, settlement is only one layer. A stablecoin can settle the transfer, but companies still need KYB and sanction checks, vendor and budget approvals, tax receipts, etc. Asset issuers control minting, redemption, and blocklisting, while chains (L1s and L2s) take care of execution and finality. Enterprise systems, however, still have to decide whether a payment is approved, booked, audited, or recoverable. Without that wrapper, USDC or USDT is just a transfer.
Canton-like networks solve a different enterprise problem: private, permissioned, auditable settlement. Agentic payments may not be the core use case, but the lesson is the same. Enterprises need controlled settlement environments, not just public token movement.
The machine web is where stablecoins fit. The reason for that is because the seller is software, the product is access, and the buyer may only need one call. For that kind of purchase, account setup, invoicing, prepaid credits, and card credentials are too much overhead. A stablecoin payment lets strangers transact once and move on. That’s where the x402 standard comes into play, since most paid x402 calls are software resources: data, inference, search, storage, and infrastructure. Travel is the closest adjacent use case, but even there the paid item is API access.

Same applies to MPP call activity. The majority of activity is software resources that can be fulfilled immediately after payment (API calls, data lookups, browser sessions, etc.). MPP, or Machine Payments Protocol, is an open standard for machine-to-machine payments co‑developed by Stripe and Tempo. An agent requests a paid resource, the seller returns payment details, the agent authorizes the payment, and the seller releases the resources with a receipt.

Traditional billing works for repeat relationships. But it works poorly for one-off software access. When an agent needs a low-value resource from a new seller, account setup, invoices, cards, and prepaid credits add too much friction. Stablecoins, for instance, were never meant to be used for replacing the invoices of large contracts. That’s why stablecoin volume is the wrong first lens to measure agentic payments adoption. Adjusted volume shows stablecoins can move money at scale, but agentic payments are still too small to move the needle. Their early signal, therefore, is usage, not settled value.

By dollar volume, agentic payments are still negligible because each payment is small. That said, it’s still worth tracking other metrics such as transaction count. Even though transaction count can look noisy because of test transactions and low-value spam, it is still useful as a footprint metric, even if it remains weak as a demand metric.

The Q4 2025 surge came entirely from x402; MPP did not launch publicly until March 2026. About 144M tracked transactions occurred in that quarter; roughly 90M were classified as non-gamed activity, mostly in software and service categories, while 54M showed patterns consistent with self-dealing or wash activity—wallets repeatedly recycling funds to inflate transaction counts rather than reflect independent demand.
x402 vs MPP
x402 is best understood as HTTP-native payments for software access. It is an open-source HTTP payment protocol. A server responds to a request with “402 Payment Required” and payment terms. The agent signs the payment payload, retries the request, and the server or facilitator verifies payment before releasing the resource.
The open protocol is only one layer. The other is the actual implementation, where a facilitator can decide which tokens and chains to support. For example, Coinbase’s CDP facilitator supports EVM tokens on Base, Polygon, Arbitrum, and World through EIP-3009 or Permit2 and SPL or Token-2022 payments on Solana.

x402 solves the seller’s onboarding problem. If an unknown buyer wants one API call, model response, browser session, or paid page, traditional billing is too much overhead. x402 lets the seller state the price, verify payment or authorization, and return the resource.
Base is the leader across x402 transaction activity. Coinbase is the main facilitator path, and Base fees are low enough for sub-dollar software calls. In May 2026, Base had about 82% of tracked x402 transactions, Solana about 17%, and Ripple about 2%. In June through June 21, Base rose to about 95%, Solana fell to about 4%, and Ripple stayed near 1%. YTD, Base is ~81%, Polygon ~9%, Solana ~9%, and Ripple less than 1%. Polygon’s YTD share came from an early-2026 spike and not steady activity.

MPP applies the HTTP 402 flow to machine-to-machine payments. A client requests a paid resource, the server responds with payment terms, and the client retries with an authorized payment credential. Once payment is verified, the server returns the resource and receipt.
MPP’s early onchain footprint is most visible on Tempo: ~1.13M transactions to date, with 46% recorded in the most recent 30 days and daily active recipients typically in the low hundreds. Stripe also supports USDC-based MPP payments on Solana but there is no datapoint to reference.

MPP works for fiat payments too. Stripe’s MPP implementation is payment-method agnostic. Fiat payments can use Shared Payment Tokens backed by Link, cards, wallets, or BNPL (Buy Now, Pay Later), while stablecoin payments use supported onchain methods. For Stripe merchants, it’s all about operational continuity: agent payments can flow through the same setup that already handles risk controls, reporting, refunds, balances, and payouts.

The difference between x402 and MPP comes down to seller integration and payment-method choice. x402 is an open, crypto-native HTTP payment flow that sellers can run directly, through facilitators or through Stripe. MPP is an open, payment-method-agnostic HTTP protocol. Its advantage for Stripe merchants is continuity: agent payments can use the same merchant setup they already rely on.
x402 looks larger in public transaction data, but visibility is part of the reason. x402 payments create onchain traces, whereas MPP can process both stablecoins and fiat, as well as BNPL or Payment Intents inside merchant accounts. Public data, therefore, overstates the gap between x402 and MPP. MPP can run inside Stripe through multiple payment methods, and some of that activity will not show up as onchain volume.

Taken together, x402 and MPP are less a winner-take-all protocol contest than two routes into the same market. x402 makes paid software endpoints native to the open web; MPP supports a broader payment-method model, with Stripe providing a direct route into existing merchant infrastructure. Their success will depend on which model gives sellers easier distribution, lower operating friction, and sustained paying demand.
Agentic Payments Adoption
As introduced earlier, tracked agentic payments show high transaction count and low ticket size. Through late May, we have 183.87M transactions and $48.97M spent, or about $0.27 per payment. The latest 30-day window had 75.41M transactions and $24.24M spent, or about $0.32 per payment. The marketplace sample across all categories was about $0.23 per payment. That points to API calls, paid pages, or browser sessions. This level of adoption belongs to machine web access, not consumer checkout or enterprise invoices yet.

The smaller the payment, the more fee structure matters. This is an opportunity that stablecoin issuers like Circle have been watching closely. Circle’s Nanopayments, for instance, are built for payment granularity. Buyers sign offchain authorizations, while Circle Gateway batches settlement onchain. That allows USDC payments as small as $0.000001 without paying gas on every transfer. If agents pay per API call, token, or browser action, the rail has to support huge volumes of tiny payments without fees consuming the transaction.
The gamed line rose during the same burst in Q4 2025, so the spike didn’t come from pure demand. After the spike, real transactions fell sharply. By late May, we saw 107K real transactions and a 22% gamed share, with nearby days around one-fifth to one-quarter gamed.

By late May, tracked activity had about 708K cumulative buyers and 7.3K cumulative sellers, or roughly 96 buyers per seller. That can happen in a market with many wallets and few paid resources. It can also happen in a market with one-time wallets, test wallets, sybil behavior, or inactive sellers.

Agentic payments are generating usage, but not yet much seller revenue. x402 looks better on total activity and repeat payments, but the money is spread across many sellers. MPP has more buyers per recipient, but the total spend is still small.

Base dominates tracked public-chain agentic payment volume through x402. In May 2026, Base had about 92% of volume, Solana about 7%, and Tempo about 1%. For 2026 through June 21, the landscape remains the same in favor of Base. Solana is the consistent second chain, and Tempo is becoming more visible through MPP activity. Polygon had a noticeable share bump due to further integrations and facilitator support.

Coinbase routed ~66.8% of May x402 transactions. Dexter routed 8.3%, PayAI 4.9%, "Other" 9.5%, and the long tail 10.5%. The number of facilitators has increased substantially since 2025, but routing is still concentrated in a few of them. Routing share should not be treated as end-demand ownership.

Demand is what determines value-capture economics. For machine web payments, which is the segment with the most rapidly growing adoption at the moment, transaction count acts as a more direct proxy than total dollar volume. It’s effectively measured by focusing on unique paying users and sellers, repeat paid usage, delivery success rate, refund/failure performance, and net seller revenue.
As mentioned earlier, only a portion of the total is reflected onchain, since MPP supports fiat. It’ll take some time until all relevant metrics become fully available. For the time being, early data points to experimentation, or early adoption, rather than proven product-market fit.
The Value Capture Map
In retail, Stripe, PayPal, Visa, Mastercard, Apple Pay, and Shopify Payments keep most of the economics because they are already everywhere and merchants depend on them for acceptance, dispute evidence, payouts, and operational records. A shopping agent may create the cart, but checkout is still a merchant workflow. Stablecoins can fund or settle the payment through wallets, PSPs, cards, or merchant payouts. They can capture reserve, conversion, and settlement economics, but they do not own the customer relationship unless they also control orders, refunds, disputes, and exceptions.

Enterprise adoption of agentic payments isn’t likely to converge on one winner. That’s because traditional companies are incentivized to keep agent payments inside trusted spend systems: enterprise finance software, bank rails, commercial cards, and PSPs. Agents may initiate the request, but approval and payment release stay inside the company’s control stack. Take SAP as an example. With SAP Digital Currency Hub, you can mark an invoice to be paid in stablecoins, send the payment through a linked crypto service provider, and match the resulting statement in SAP.
Finally, in the machine web, Coinbase has a strong default distribution and first-mover advantage. Tracked x402 activity is a testament to this, as it is the most dominant entity by facilitator share. On top of that, Base’s liquidity and the connection with the Coinbase Exchange reinforce one another. That said, this is a space where the moat is still subject to the role played by facilitators, which are those entities responsible for implementing the actual open-source protocol. Stripe, for example, leads the MPP merchant path, especially when the seller already uses Stripe, since payments, refunds, reporting, and payouts can stay inside the same system.
Facilitators gain share when the tools that sellers use to launch paid APIs choose them as the default. If a seller publishes an API through a cloud worker, MCP host, API marketplace, or enterprise gateway, and that product uses FluxA or PayAI instead of Coinbase, then they get the x402 transaction. If the seller sets up x402 directly, Coinbase is still the easiest default.

As for stablecoins, their issuers mostly capture value when agents hold the asset and sellers accept it. In other words, when the actual stablecoin is used as a medium of exchange and unit of account.
Today, the part of the machine-payment market we can observe is essentially USDC-centric, as most of the feeds are USDC/Base or USDC/Solana, and every endpoint pull is set in USDC terms. That does not measure private MPP, card, wallet, or PSP flows though. Other stablecoins that want to capture market share, like USDT0 (omnichain deployment of Tether’s USDT), still need facilitator support, wallet support, and sellers willing to quote resources in that asset. USDT0's edge would be Tether liquidity across chains. That remains an untapped opportunity for them.
A number of players can win, but at different layers, and not automatically or equally. Circle benefits if agent payments create more demand for USDC balances and more USDC settlement. Visa and Mastercard benefit if agents keep paying through cards, where the networks already handle acceptance, authorization, disputes, and refunds. Stripe wins when merchants can incorporate agent payments into the existing reporting, refunds, balances and payouts system they already use. Tempo benefits if MPP brings real payment volume to the chain. Coinbase wins if x402 sellers and developers continue to pick Coinbase for facilitation and Base for settlement. Enterprise agents are covered by corporate approval and accounting systems, which benefits enterprise finance software. These roles can coexist, but the overlap grows once the same company tries to control both the payment and the related records.
Conclusion
Stablecoins are not the main product in agentic payments. Despite that, they still provide and capture value. When the buyer does not know the seller, the item is cheap, and the seller can send it immediately, stablecoins become a viable settlement option. An API call, browser session, datasets, or other machine web resources are among the most evident current use cases for stablecoin-funded agent payments. They remove account setup, card onboarding, invoices, and prepaid balances from purchases too small to matter for all that.
Retail and enterprise are tougher markets for stablecoins to capture value. They can be used to settle transactions, while card networks and PSPs retain control over the visible payment flow until checkout. The same applies to enterprise, where stablecoins can settle a transfer if accepted by the company, but enterprise finance software still captures much of the workflow value because it’s the orchestration layer that handles finance controls (budgets, taxes, receipts, etc.)
Crypto’s role in the agentic payments stack is currently narrow but still economically meaningful. In stablecoin-native flows, the stablecoin is the settlement asset. Issuers such as Circle handle issuance and redemption, maintain reserve backing, and apply issuer-level compliance controls. x402 standardizes how a seller communicates payment terms and how a client presents payment authorization. Sellers can verify and settle directly or delegate those tasks to a facilitator. The underlying blockchain executes and finalizes the transfer, with fees accruing according to each network’s design.
Crypto is not the whole agentic-payments stack. Still, crypto-native providers and asset issuers can capture more of the stack if they also own authorization and wallet policy, service discovery, seller and resource verification, delivery evidence, reconciliation, and refund or exception workflows. For crypto to win this vertical, its players need to step up and move higher in the stack beyond settlement.
References
A402: Binding Cryptocurrency Payments to Service Execution for Agentic Commerce - arXiv
Agent Authorization Framework - AP2
Agentic Payments Data - Agentic Market
Agent Payments Protocol - AP2
Agentic Payments Sector - Artemis
Enabling The Agentic Economy with Nanopayments - Circle
Facilitators - x402scan
Five Attacks on x402 Agentic Payment Protocol - arXiv
Global Payments Report 2026 - Worldpay
How Agentic AI Will Reshape Payments - IMF
How Verifiable Intent builds trust in agentic AI commerce - Mastercard
How x402 Works - Coinbase Developer Documentation
Is 2026 the Year of Agentic Payments? - Fenwick
Machine payments - Stripe
Machine Payments Protocol Explorer - MPPScan
OpenAI's agentic commerce shift: What it means for merchants - ChatGPT
OWASP Top 10 for Agentic Applications for 2026 - OWASP Gen AI Security Project
Shared payment tokens - Stripe
Specifications - Visa Trusted Agent Protocol
Stablecoin Transactions - Visa Onchain Analytics
The Payment Problem Hiding in the Agentic API Economy - USDT0
Whispers of Wealth: Red-Teaming Google's Agent Payments Protocol via Prompt Injection - arXiv
x402 — Payment Required - x402


