Agentic Payments Market: October 2026
Agents can now pay in production, but control of the money has not moved. The card networks and the large wallets still hold the credential.
What did we find?
- The credential stays with incumbents. Every live consumer flow found in the window keeps the card number away from the agent. A network token, a wallet-issued single-use card, or a merchant account stands in its place.
- Approval remains the default control. Muse on Stripe Link, Claude Duet, Stripe Treasury for agents, and the Visa B2B pilot all execute only what a person approved. Walmart is the first large merchant found to offer outside agents a standing spending limit. Amazon's own Rufus Auto Buy has bought at a preset price without a second approval since November 2025.
- Liability has no settled rule. Law firm and trade press analysis agrees that Regulation E and Regulation Z do not say who loses when an agent exceeds its mandate. Visa and Mastercard added registration and identity, not new dispute reason codes found in public sources. American Express is the one network found that promises to cover agent errors, and only for registered agents. The Ninth Circuit treated Perplexity's shopping agent as the user's tool, on a record it called limited.
- Protocols multiplied, then moved into foundations. ACP, AP2, UCP, MPP, TAP, and x402 all exist. x402 went to the Linux Foundation, AP2 to the FIDO Alliance, and a Linux Foundation working group chaired from Shopify now covers agentic commerce.
- Volume evidence comes from the edges. Forter's merchant network and an unattributed x402 count are the only order or transfer counts found. Visa's last count was "hundreds" of controlled transactions in December 2025.
- Forecasts disagree by more than 30x because they count different things. The range runs from $144B of US purchases completed on AI platforms by 2029 to $3 to 5T of global orchestrated commerce by 2030.
- B2B agent payments started with the easiest step. The Visa pilot automates portal keying of already-approved invoices. Ramp's Agent Cards let an agent spend inside a corporate fund's limits. Nothing found lets an agent accept supplier credit or net terms.
The state of payments that an AI agent starts on behalf of a consumer or a business, and the infrastructure under them, from no vendor’s side. Vendor claims are marked (vendor claim) and are unverified. Third-party evidence is marked (third-party evidence). Every live read in this report is dated 2026-10-30 unless marked otherwise.
The file tracks three questions every month. Who owns the agent credential? Who carries liability when an agent buys the wrong thing? Which rail wins: cards with tokens, account-to-account, or stablecoins?
What happened in October 2026. This is the first edition. Eight things moved in October.
- The first live B2B agent payments went into production on a card network. Visa, BILL, and Billtrust started a live pilot on 2026-10-20 in which a registered agent pays buyer-approved supplier invoices through supplier portals with tokenized credentials (section 4).
- Agent order volume became measurable for the first time, through a fraud vendor. Forter’s network shows agent orders at about 5x their late-July level, an average order of $226, and screen-driving agents at 70% of agent orders (section 5).
- The networks pushed agent identity into checkout infrastructure. Shopify embedded Visa, Mastercard, and Amex agentic payment technology into Shop Pay, and Checkout.com rolled out Visa’s Trusted Agent Protocol with Visa Payment Passkey (section 9).
- Mastercard opened machine payments to developers and called it a pilot. Agent Pay for Machines entered early access on 2026-10-20 on x402-style vouchers, with a nine-payment demo and no commercial date (section 9).
- Merchants started writing their own agent rules. Walmart announced Agent Profiles with spending limits and said its returns process applies “if an agent doesn’t act as intended” (section 10).
- Stripe moved agents into business banking. Treasury for agents and a Claude Duet integration launched on 2026-10-21, both keeping a human approval step for money movement (section 9).
- The OCC finalized the first GENIUS Act rule. The rule landed on 2026-10-21, effective 2027-01-18, and a 37-bank euro stablecoin consortium won an e-money licence on 2026-10-14 (section 10).
- The networks reported results and still disclosed no agent transaction count. Visa reported nearly 50 AI partners in production and called trust “the biggest barrier to scale” (section 9).
1. Summary
Agents can now pay in production, but control of the money has not moved. The card networks and the large wallets still hold the credential. Liability sits undefined in statute and gets settled by contract and merchant policy. The rails split by ticket size, with tokenized cards carrying consumer and B2B purchases and stablecoins carrying machine micro-payments. No public source discloses agent transaction value at scale for any network, processor, or agent platform.
Seven things define the market in October 2026.
- The credential stays with incumbents. Every live consumer flow found in the window keeps the card number away from the agent. A network token, a wallet-issued single-use card, or a merchant account stands in its place (section 3).
- Approval remains the default control. Muse on Stripe Link, Claude Duet, Stripe Treasury for agents, and the Visa B2B pilot all execute only what a person approved. Walmart is the first large merchant found to offer outside agents a standing spending limit. Amazon’s own Rufus Auto Buy has bought at a preset price without a second approval since November 2025 (sections 3 and 9).
- Liability has no settled rule. Law firm and trade press analysis agrees that Regulation E and Regulation Z do not say who loses when an agent exceeds its mandate. Visa and Mastercard added registration and identity, not new dispute reason codes found in public sources. American Express is the one network found that promises to cover agent errors, and only for registered agents. The Ninth Circuit treated Perplexity’s shopping agent as the user’s tool, on a record it called limited (section 10).
- Protocols multiplied, then moved into foundations. ACP, AP2, UCP, MPP, TAP, and x402 all exist. x402 went to the Linux Foundation, AP2 to the FIDO Alliance, and a Linux Foundation working group chaired from Shopify now covers agentic commerce (section 2).
- Volume evidence comes from the edges. Forter’s merchant network and an unattributed x402 count are the only order or transfer counts found. Visa’s last count was “hundreds” of controlled transactions in December 2025 (sections 5 and 9).
- Forecasts disagree by more than 30x because they count different things. The range runs from $144B of US purchases completed on AI platforms by 2029 to $3 to 5T of global orchestrated commerce by 2030 (section 5).
- B2B agent payments started with the easiest step. The Visa pilot automates portal keying of already-approved invoices. Ramp’s Agent Cards let an agent spend inside a corporate fund’s limits. Nothing found lets an agent accept supplier credit or net terms (section 6).
2. Categories and terms
Three layers make up the market, and each has its own protocols. Discovery and checkout protocols carry the cart. Credential and mandate protocols carry the permission to pay. Settlement rails move the money.
| Term | Owner and status (read 2026-10-30) | What it does |
|---|---|---|
| Agentic Commerce Protocol (ACP) | Stripe docs now say “created by Stripe, OpenAI, and Meta” (Stripe) (vendor claim). Spec version 2026-04-17, still labelled beta | Checkout between an agent surface and a merchant. Powers ChatGPT checkout |
| ChatGPT Instant Checkout | OpenAI, launched 2025-09-29. A March 2026 OpenAI update said the first version “did not provide the desired flexibility” and moved focus to product discovery and merchant checkout (eesel summary) (third-party evidence). Secondary blogs call it discontinued. Forter still counts ACP orders in October | In-chat purchase. Status conflicts between sources (section 14) |
| Shared Payment Token (SPT) | Stripe, generally available since Sessions 2026 (Stripe) (vendor claim) | A scoped token an agent passes to a merchant. Covers cards, Klarna, Affirm, and stablecoins over MPP |
| Machine Payments Protocol (MPP) | Stripe and Tempo, launched with Tempo’s mainnet, reported as 2026-03-18 (The Defiant, subtitle read, date from a search summary) (third-party evidence) | HTTP 402 challenge and receipt for paid API calls. Charge, session, and subscription intents. Pay-before-access only |
| Agent Payments Protocol (AP2) | Google, announced 2025-09-16 with “more than 60 organizations” (Google Cloud) (vendor claim). Now v0.2 and donated to the FIDO Alliance, date not shown (ap2-protocol.org) | Signed Checkout and Payment mandates that create “a non-repudiable, cryptographic audit trail”. Supports human-not-present purchases |
| Universal Commerce Protocol (UCP) | Google, announced 2026-01-11 at NRF (secondary reports) (third-party evidence) | Catalog, cart, and identity linking for Google AI Mode and Gemini. AP2 now describes itself as an extension of A2A and UCP |
| Visa Intelligent Commerce (VIC) and Trusted Agent Protocol (TAP) | Visa. TAP launched October 2025. Intelligent Commerce Connect announced 2026-04-08 as one integration across TAP, MPP, ACP, and UCP, in pilot with seven partners (Visa) (vendor claim) | Agent recognition, an Agentic Directory, tokenized credentials, and payment controls |
| Mastercard Agent Pay and Agentic Tokens | Mastercard, launched April 2025. US cardholder rollout reported for November 2025 (secondary) (third-party evidence) | A token scoped to one agent and one policy |
| Mastercard Agent Pay for Machines | Announced 2026-06-10, early access 2026-10-20, “not yet been made commercially available” (The Block) (third-party evidence) | Card, bank, and stablecoin payments between machines, with signed vouchers that batch |
| Amex Agentic Commerce Experiences (ACE) Developer Kit and Agent Purchase Protection | American Express, announced 2026-04-14. Access for “select developers”, some services still in development (eCommerceNews, 2026-04-15, Fortune, 2026-04-14) (third-party evidence) | Agent registration, account enablement, intent intelligence, tokenized credentials, and cart context. Covers agent errors on registered-agent purchases (section 10) |
| x402 | Coinbase originated it. The Linux Foundation launched the x402 Foundation on 2026-04-02 with 22 supporters (Linux Foundation) and listed 40 members at the 2026-07-14 operational launch (x402.org) (vendor claim) | HTTP 402 payments over the web. “Zero protocol fees” |
| Know Your Agent (KYA) | Category term. Skyfire uses it as a product name | Verified identity for an agent and the person or business behind it |
| Network tokenization | Visa and Mastercard | Replaces the card number with a network token. Visa reports “more than 60% of eCommerce transactions are tokenized” (PYMNTS) (third-party evidence) |
Governance moved to neutral bodies in 2026. x402 sits at the Linux Foundation, whose 17 premier members include Adyen, AWS, American Express, Circle, Cloudflare, Coinbase, Fiserv, Google, Mastercard, Shopify, Stripe, and Visa. AP2 went to FIDO. The Agentic AI Foundation’s Agentic Commerce Working Group, chaired by Ilya Grigorik of Shopify with OpenAI’s Rahul Bansal as vice chair, lists 207 active members (AAIF) (third-party evidence). The networks backed every protocol rather than pick one. Visa’s quote at the x402 launch reads “Commerce will not run on a single agent, protocol, or payment method.”
Analyst coverage read this month. No analyst placement or market guide for agentic payments was found in the window. Searches for Gartner, Forrester, and IDC October output ran out of budget twice (section 14). The nearest placement found is Zip as a Visionary in Gartner’s 2026 Magic Quadrant for Source-to-Pay Suites. Zip’s headline calls it “The Only Agentic Procurement Orchestration Platform to Make the List” (BusinessWire, 2026-01-23, headline read) (vendor claim). The latest analyst prediction found is Gartner’s, from 2025-11-18 (section 5).
3. The transaction flow and where value sits
An agent purchase passes through six steps. Each step has a different owner, and the owner of the credential step controls the others.
| Step | What happens | Who owns it now | Evidence |
|---|---|---|---|
| 1. Mandate | The person sets what the agent can buy, with limits | Agent platform or wallet | AP2 mandates. Walmart’s $50, $100, or $200 limits. Stripe Treasury approval rules |
| 2. Identity | The merchant learns the agent is legitimate and who stands behind it | Networks and identity vendors | Visa TAP and Agentic Directory, Mastercard probability score with Skyfire KYA, Cloudflare signals |
| 3. Discovery and cart | The agent finds the product and builds the order | Agent surface or merchant | ACP, UCP, WebMCP, Walmart Agent Gateway. Screen-driving agents bypass all of them |
| 4. Credential | A payment credential reaches the merchant without the agent seeing it | Networks, wallets, issuers | Agentic Tokens, SPTs, Link single-use cards, Shop Pay |
| 5. Authorization and settlement | Issuer approves, rail settles | Issuers and networks, or a stablecoin chain | Cards, ACH through Stripe Treasury, USDC on Base, Solana, or Tempo |
| 6. Dispute | Someone absorbs a wrong purchase | Unsettled | Reg E and Reg Z gaps, merchant return policies (section 10) |
Who owns the agent credential (the tracked question, October answer). Four models compete, and the networks sit inside three of them.
| Model | Who holds the credential | Live example (read 2026-10-30) |
|---|---|---|
| Network token per agent | Issuer and network | Mastercard Agentic Tokens, Visa tokenized credentials in the BILL pilot (vendor claim). Amex tokenized credentials for agents it has registered (section 2) (third-party evidence) |
| Wallet-issued credential | Wallet provider | Stripe Link pays with the saved method at more than 1 million businesses that accept Link, otherwise issues “a single-use virtual card” limited to the approved purchase (Stripe) (vendor claim). Shopify says of Shop Pay “The agent never sees the buyer’s card” (PYMNTS) (third-party evidence). In B2B, Ramp’s Agent Cards give an agent card credentials for browser checkout inside a fund’s limits (section 6) (vendor claim) |
| Merchant account | Merchant | Walmart Agent Profiles. Agents “don’t receive customers’ payment-card information” and cannot change payment methods (Walmart) (vendor claim). Amazon extends its own account outward. Buy for Me buys on third-party sites from inside the Amazon app, with the shopper approving each purchase (Modern Retail, 2025-11-13) (third-party evidence) |
| Agent-held wallet | Wallet custodian or the agent’s operator | Coinbase Developer Platform wallet and BVNK in the Mastercard machine demo, Natural’s FDIC-insured agent wallets through a partner bank (third-party evidence) |
What the scoreboard says. Wallets and networks hold the consumer credential. Merchants with enough traffic hold their own. Agent-held wallets exist only on machine and stablecoin flows. The Information reported on 2026-10-26 that Stripe’s virtual cards for agents put Visa and Mastercard “on edge” (headline through a Dealroom summary, page not read) (third-party evidence). That report frames the contest as wallet against network for the same credential.
Where value sits (analysis). The credential step carries the economics because it decides whose fee schedule applies (section 11). The identity step is the newest toll point, and the networks have moved fastest to own it through TAP, the Agentic Directory, and Mastercard’s agent score.
4. Buyers and triggers
| Buyer | Buys on | Evidence (read 2026-10-30) |
|---|---|---|
| Merchants | Telling good agents from bots, keeping conversion, staying in agent channels | Checkout.com cites TAP to separate “trusted AI shopping agents from bots and malicious traffic” (Checkout.com, 2026-10-29) (vendor claim). 47% of merchants report lost sales from false declines (PYMNTS survey of March 2026, through FinanceFeeds) (third-party evidence) |
| Issuers | Trust, fraud loss, unclear liability | An Oct 9 opinion piece expects issuers to share credentials only with large AI providers under contract, add identity steps, or decline (Payments Dive) (third-party evidence) |
| Agent builders | A credential without card data, merchant acceptance | Link “powers payments for more than 200 personal agents” (Stripe, 2026-10-21) (vendor claim) |
| Enterprises and AP teams | Removing portal keying, controls, card rebates | The Visa, BILL, and Billtrust pilot (Visa, 2026-10-20) (vendor claim) |
| Developers buying APIs and content | Per-call payment without accounts or contracts | x402 and MPP. Mastercard’s demo agent bought research and eight images within an approved budget (third-party evidence) |
The B2B trigger is portal keying, not purchasing judgment. The Visa pilot agent “executes only payments the buyer has already approved and does not make payment decisions” (Visa release, paraphrased in the source read). TAP lets Billtrust admit the agent past CAPTCHA and bot blocks without a separate agent login (vendor claim). The pilot’s scale claims belong to the companies, not the pilot. BILL cites nearly 500,000 businesses and Billtrust $1 trillion of annual trade credit volume, and the release gives no pilot transaction count (vendor claim). The AR market’s portal-delivery pain sits in AR Automation Market report.
The consumer trigger is AI-assisted shopping that already happens. Visa’s Morning Consult survey (N=1,016 US adults, fielded 2026-09-26 to 10-02) finds 54% have used AI for at least one shopping task, up from 47% a year earlier (Visa, 2026-10-30) (vendor claim, vendor-commissioned). Only 23% trust GenAI to handle payment transactions for them, from a Harris Poll for Visa (N=2,065, fielded 2026-05-26 to 05-28) (Visa, 2026-09-09) (vendor claim, vendor-commissioned). Use of AI to shop runs ahead of willingness to let it pay.
5. Size and structure
Published forecasts span $144B to $5T because they count different things. Two count purchases an agent completes. The rest count spend an agent influences, shapes, orchestrates, or mediates.
| Firm | Published | Figure | Year | Scope | What it counts | Read |
|---|---|---|---|---|---|---|
| eMarketer | 2026 (date not read) | $144B | 2029 | US | Sales completed on AI platforms. Completed | Through Stellagent aggregator. The primary it links (eMarketer) returned 403 (third-party evidence) |
| Morgan Stanley | December 2025 | $190B base, $385B bull (10 to 20% of US e-commerce) | 2030 | US | Purchases autonomously executed by agents. Completed | Primary page timed out on two reads. Figures through aggregators (third-party evidence) |
| Bain | 2025-12-17 | $300 to 500B (15 to 25% of e-commerce) | 2030 | US | Purchases an agent starts, shapes, or completes. Influenced | Bain (third-party evidence) |
| McKinsey | 2025-10-17 | Up to $1T US B2C retail, $3 to 5T global | 2030 | US and global | ”Orchestrated” revenue, goods only, excludes services and B2B marketplaces. Influenced | McKinsey mirror (third-party evidence) |
| Edgar Dunn | Not read | $1.7T narrow, $2.9T broad | 2030 | Global | Retail transaction flows. Influenced | Aggregator only, with no primary link. Unverified |
| Gartner | 2025-11-18 | Over $15T of B2B spend | 2028 | Global B2B | Purchases mediated by AI agents. Influenced | The Gartner release returned 403 on two reads. Figure through aggregator (third-party evidence) |
| Juniper Research | Not read | $8B (2026) to $1.5T | 2030 | Global | Agentic commerce transaction value | Search summary only. Juniper’s press page listed no agent release. Unverified |
| Juniper Research | Not read | $14.6T B2B virtual card volume, 83% of the virtual card market | 2029 | Global | Not agent-specific. Context for B2B card rails | Quoted in Visa, 2026-10-21 (3P via V) |
Read the spread by metric, not by firm. Completed-purchase forecasts sit at $144 to 385B for the US. Influenced-spend forecasts start at $300B and reach $5T. McKinsey’s US figure and Morgan Stanley’s bull case differ by about 2.6x for the same country and year because one counts orchestration and the other execution. Treat any figure above $1T as influenced spend, not payment volume.
Measured activity, the only actuals found.
| Measure | Value | Source and tag |
|---|---|---|
| Agent orders on Forter’s network | About 5x the late-July level by mid-October. Average order $226. 27% of orders at $50 or less | Forter, 2026-10-23 (third-party evidence, one vendor’s network) |
| Agent mix on Forter’s network | OpenAI placed 82 to 93% of weekly agent orders in August. Muse placed 43% in the week of 2026-10-12. Screen-driving agents rose from 1% to 70% of agent orders | Same (third-party evidence) |
| x402 transfers | 8.7 million transfers worth $367,950 in the week of 2026-08-17, about $0.04 each | FinanceFeeds, 2026-10-28, underlying data provider not named (third-party evidence, unattributed) |
| x402 share by chain | Solana “driving nearly 65% of x402 transaction volume this year” | Linux Foundation release, 2026-04-02 (vendor claim) |
| Meta Muse | More than 10 million users. More than a quarter of engaged shoppers averaged purchases above $200 | TechCrunch, 2026-10-21 (V via 3P) |
| Amazon Rufus (AI shopping assistant, not agent payments) | 250 million customers in 2025, monthly users up 140%. Amazon expects more than $10B of incremental annualized sales | Modern Retail, 2025-11-13 (V via 3P) |
| Visa agent transactions | ”hundreds of controlled, agent-initiated transactions” (December 2025). No later count | Visa (vendor claim) |
Structure. Forter’s numbers show concentration in a few agent surfaces. One agent went from zero to 43% of weekly orders in three weeks. Agent surfaces, not merchants or networks, decide where volume lands. Stablecoin payment flows are larger but are not agent flows. Allium estimates annual stablecoin payments at $401 to 527B (through Visa, 2026-10-01) (3P via V).
6. Segments
| Segment | Who sells there | What decides it |
|---|---|---|
| Consumer checkout inside an AI assistant | Meta Muse with Stripe Link, PayPal, and Shop Pay. OpenAI with ACP. Google with UCP and AP2. Microsoft Copilot Checkout with PayPal, Shopify, and Stripe. Perplexity with PayPal. Grok and Instinct (named by Forter) | Agent reach, merchant acceptance, the approval model |
| Retailer-owned agents | Amazon Rufus with Auto Buy and Buy for Me | Own traffic and stored payment, extended to other sites |
| Merchant-hosted agent access | Walmart Agent Gateway, Shopify catalog and checkout open “to any agent or app” (Fall ‘26 Edition headline, 2026-10-28) (vendor claim). Salesforce Commerce lists UCP checkout on Google as “GA Oct ‘26” with the merchant as merchant of record (Salesforce) (vendor claim) | Merchant control of identity, limits, and returns |
| Processor agent suites | Adyen Agentic, Worldpay MCP, Fiserv with Visa and Mastercard frameworks, JPMorgan Payments with Mirakl (section 9) | One integration across ACP, UCP, AP2, and TAP for the merchant |
| Screen-driving agents | ChatGPT Agent, Muse, other browser agents | No protocol needed. 70% of Forter’s agent orders in mid-October |
| Machine-to-machine and API payments | x402, MPP, Mastercard Agent Pay for Machines, Cloudflare | Cost per transfer at cents or below. Stablecoin settlement |
| B2B accounts payable and spend | Visa with BILL and Billtrust, Stripe Treasury for agents, Ramp Agent Cards, SAP Pay inside SAP Cloud ERP | Pre-approved payments, controls, supplier acceptance |
| Agent banking and wallets | Stripe Treasury (Fifth Third holds the accounts), Natural, Coinbase Developer Platform, Privy | Money transmission licensing, custody (section 10) |
| Agent identity and fraud | Visa TAP, Mastercard with Skyfire, Amex agent registration, Trulioo KYA, Cloudflare, Forter | Acceptance by merchants and bot-management vendors |
Procurement agents that negotiate and commit on credit did not appear in the window. The B2B live use case pays invoices a person approved. Business buying needs payment terms, remittance, and dispute handling in the protocol (analysis), and the October launches add none of the three.
Procurement suites ship agents that route and check, and Ramp is the one found that lets an agent pay.
| Vendor | Agent product (read 2026-10-30) | Does the agent pay? |
|---|---|---|
| Ramp | Agent Cards give an agent card credentials for browser checkout. The agent “follows the fund’s spend limit, merchant rules, approval policies, and accounting defaults” (Ramp docs) (vendor claim). The docs keep MPP and x402 payments separate and give no launch stage or date. Ramp and Visa agreed on 2026-04-02 to build bill-pay agents on VIC and TAP, with no launch date (Yahoo Finance) (third-party evidence) | Yes, on the company’s own card, inside fund limits |
| SAP | Joule agents for Ariba intake, Ariba contracts, and Fieldglass were slated for June 2026 (SAPinsider, headline read) (third-party evidence). SAP Pay launched at SAP Connect inside SAP Cloud ERP. A due invoice is “processed and reconciled within the same workflow”, powered by SAP subsidiary Tereina (Techzine, 2026-10-05 or 10-06) (third-party evidence) | Not shown. The article ties no agent to SAP Pay |
| Zip | Claims 50+ AI agents across finance, legal, procurement, IT, and security (search summary of Zip material) (vendor claim, unverified) | No agent payment credential found |
| Coupa | Navi Agent Studio and Coupa Compose, launched at Inspire (competitor-authored JAGGAER comparison, hostile-source) (third-party evidence) | No agent payment credential found |
What the table says (analysis). SAP Pay matters more than the agents. An ERP that owns the payment step can hand it to its own agents later without a card network in the flow.
7. Drivers and restraints
Drivers, ranked by evidence quality.
- Consumers already shop with AI. 54% of US adults used AI for a shopping task (section 4). Meta reports 10 million Muse users (section 5).
- Tokenization is already the default rail for e-commerce. Visa says more than 60% of its e-commerce transactions are tokenized and its credential base reached 5.2 billion, up 6% (PYMNTS, 2026-10-27) (third-party evidence). Agentic Tokens extend a system issuers already run.
- Computer use removed the integration wait. John Collison said “computer use was the last big unlock that we needed” and that agentic activity “basically started going vertical” in September, with no figures (Payments Dive, 2026-10-15) (third-party evidence). Forter’s 70% screen-driving share supports the direction.
- Stablecoin rails grew fast. Visa’s stablecoin settlement passed a $25B annualized run rate, about tenfold year over year (third-party evidence). Visa Direct moved stablecoin prefunding into production on 2026-10-15 with USDC (Visa) (vendor claim). Tempo claims $60B+ of annualized stablecoin payment volume (tempo.xyz) (vendor claim). None of these is agent volume.
Restraints.
- Trust. 23% of US consumers trust GenAI with payments (section 4). Visa’s CEO said “in agentic, trust is the biggest barrier to scale” on the fiscal Q4 call (third-party evidence).
- Merchants block agents they did not authorize. Amazon blocked Muse on 2026-09-21 (Axios, cited through a secondary summary) (third-party evidence). Amazon said it had not authorized the agent to access customer accounts, scrape data, or process transactions. The courts have not backed such blocks so far. The Ninth Circuit vacated Amazon’s injunction against Perplexity’s Comet in August (section 10).
- Banks see agents outside approved connections. Axios reported on 2026-10-22 that banks found AI agents accessing accounts outside approved connections. The page blocked fetches, and the claim rests on a newsletter summary (third-party evidence, page not read).
- Liability is undefined (section 10).
- Human approval caps autonomy. Every consumer flow found except Amazon’s Auto Buy defaults to per-purchase approval (section 9).
- Fragmentation. Six protocols and two foundations compete for the same checkout. Visa built Intelligent Commerce Connect to bridge four of them, which shows the cost the fragmentation imposes on integrators (vendor claim).
8. Pricing and deal models
Few agent-specific prices are public. Stripe publishes the only agent primitive price found. Networks, OpenAI, and Google publish no fee for their agent programs.
| Vendor | Product | Price | Unit | Source (read 2026-10-30) |
|---|---|---|---|---|
| Stripe | Shared Payment Token | $0.15 | Per SPT issued | stripe.com/pricing (vendor claim) |
| Stripe | Cards, including Link | 2.9% + 30¢, +1.5% international, +1% currency conversion | Per successful transaction | Same (vendor claim) |
| Stripe | Stablecoin payments through Link | 0.8%, a promotional rate through 2027-01-01, then +0.2% | Per successful transaction | Same (vendor claim) |
| Stripe | Stablecoin payments, global | 1.5%, includes fiat conversion, screening, and gas | Of transaction amount in USD | Same (vendor claim) |
| Stripe | Stablecoin payouts | $0.50 + 0.5% | Per payout | Same (vendor claim) |
| Stripe | Issuing virtual card (no agent-specific rate) | $0.10 | Per virtual card | Same (vendor claim) |
| x402 Foundation | x402 protocol | $0 protocol fee, “nominal payment network fees” apply | Per transaction | x402.org (vendor claim) |
| Open Standard | OUSD stablecoin through Stripe | Dollar to OUSD conversion at no cost on Stripe. Issuer charges “a small, predictable transaction fee”, amount not stated | Per transaction | Stripe, 2026-09-30 (vendor claim) |
| Meta | Muse | Free for most users. Meta plans to keep “a percentage of the commerce”, rate not stated | Share of commerce | TechCrunch, 2026-10-21 (V via 3P). A $20 or $100 monthly tier appears in a secondary summary, unverified |
| Stripe | Treasury for agents | Not stated | n/a | Stripe (vendor claim) |
| Perplexity | Shopping with PayPal checkout | Free for US users | n/a | Perplexity, 2025-11-25 (vendor claim) |
| Ramp | Agent Cards | Not stated. The Plus plan costs $15 per user per month plus a “Platform fee based on team size”. Procurement is an add-on to Plus or Enterprise | Per user per month | ramp.com/pricing (vendor claim) |
| Visa, Mastercard, American Express, Adyen, OpenAI, Google | VIC, TAP, Agent Pay, Agent Pay for Machines, ACE Developer Kit, Adyen Agentic, ACP, UCP, AP2 | No fee disclosed | n/a | Section 2 and 9 sources |
Three deal models are visible. Processors charge per credential or per transaction on top of standard card pricing (Stripe’s SPT). Protocol foundations charge nothing and leave fees to the rail (x402). Agent surfaces plan a take on the commerce they route (Meta). A merchant fee for ChatGPT checkout circulates on low-authority blogs at 4% and remains unconfirmed (section 14).
9. The agentic shift
Network and processor messaging ran from announced protocol, to controlled pilot, to “turning point”, while disclosed volume stayed near zero. Mastercard’s head of product called October “a bit of a turning point” in an interview that gave no counts (Fortune, 2026-10-26) (third-party evidence). The same article says “Agent Pay is built on x402”. That conflicts with Mastercard’s own split between Agent Pay (tokens) and Agent Pay for Machines (vouchers) and likely conflates the two. Visa reported nearly 50 AI partners in production and expects the roster to double “in the coming months”, with no transaction count (third-party evidence).
| Rung | Definition | Who is there (October 2026) |
|---|---|---|
| 5. Production at disclosed volume | Agent transactions in production with a public count or value | x402 only, through an unattributed third-party weekly count (section 5). No card network, processor, or agent platform discloses agent transaction volume |
| 4. Generally available, volume undisclosed | Real customers can buy or pay through agents today | Stripe (Link agent wallet for Muse since 2026-09-08 and 200+ personal agents, SPTs, Agentic Commerce Suite). Meta Muse (US, Canada, Singapore). PayPal (Muse checkout across PayPal merchants, 2026-09-22, secondary). Shopify (Shop Pay for agents, browser-agent checkout from September). OpenAI (ACP checkout, scope narrowed in March). Mastercard Agent Pay (US cardholders, secondary). Stripe Treasury for agents (2026-10-21). Amazon (Rufus Auto Buy for all US Prime members and Buy for Me on third-party sites, both from 2025, Modern Retail) (third-party evidence). Microsoft Copilot Checkout (US, Copilot.com, launched about 2026-01-09 with PayPal, Shopify, and Stripe, Shopify merchants enrolled automatically, eWeek) (third-party evidence). Perplexity (PayPal-powered checkout, free for US users, 2025-11-25) (vendor claim). Salesforce (UCP checkout on Google, “GA Oct ‘26”) (vendor claim). Klarna and Affirm (payment methods inside Stripe SPTs, Affirm from 2026-03-03 per a BusinessWire headline, page 403. Google Pay buttons in Gemini and AI Mode from May 2026 per an eMarketer headline, not read) (third-party evidence). Ramp Agent Cards (public docs and an enrollment step, stage not stated) (vendor claim) |
| 3. Live pilot with real transactions | Named transactions with real money in a controlled scope | Visa with BILL and Billtrust (B2B, live 2026-10-20). Visa with Revolut and Cleverbridge in France (Cleverbridge release 2026-09-24). Mastercard with Danske Bank on Priceless.com (2026-09-21). Mastercard Agent Pay for Machines (nine payments in a Cloudflare Connect demo, early access). Google UCP and AP2 checkout (US-only pilot per Forter, while Salesforce lists its connector to it as generally available in October). Claude Duet with Link (research preview, 2026-10-21). Adyen Agentic (limited availability for US enterprise merchants since 2026-06-16. Names ESW, Scheels, Sézane, and SharkNinja as retailers and Amex, Mastercard, Salesforce, and Visa as partners. No transaction count, Adyen) (vendor claim) |
| 2. Spec, sandbox, or developer access | Builders can integrate, no disclosed production use | AP2 v0.2, ACP (beta), MPP, Visa Intelligent Commerce Connect (pilot with seven named partners). Checkout.com supports TAP (2026-10-29). Skyfire KYA (in Mastercard’s tested stack). Circle Arc (live network, agent features, no agent volume). Natural (six products live, no named customers). American Express ACE Developer Kit (select developers, 2026-04-14). Amex says it is “already processing AI-assisted transactions” with AI platform partners and gives no count (third-party evidence). Worldpay MCP (Developer Hub and GitHub since 2025-11-24, Digital Transactions) (third-party evidence). Trulioo KYA (its agent index “currently lists no agents”, agents.trulioo.com) (vendor claim) |
| 1. Announced program or backing only | Named support, no product evidence found in the window | Fiserv (announced Visa TAP and Mastercard’s Agent Pay Acceptance Framework on 2025-12-22, all in future tense, no go-live found, Digital Transactions) (third-party evidence). JPMorgan Payments with Mirakl (2026-03-10, JPMorgan’s page gives no live status, JPMorgan) (vendor claim). Walmart Agent Profiles (launch “in the next month”). Prove and Sardine (no agent product found, newsrooms returned 404). Paxos (no agent item among its three 2026 releases, Paxos) (vendor claim) |
Approval is the control everyone ships. Stripe says of Muse that “Consumers approve each transaction total directly in the chat” (vendor claim). Claude Duet users “approve each purchase before Claude completes the transaction with a virtual card” (Stripe) (vendor claim). Stripe Treasury for agents lets a business set “when money movement needs human approval”, with agent spending limits promised “in the coming weeks” and agent-created cards “soon” (Stripe) (vendor claim).
Walmart and Link’s spending limits are the first standing-limit controls offered to outside agents. Autonomy without per-purchase approval exists in production on machine payments within a prepaid budget and in Amazon’s Rufus Auto Buy. Auto Buy purchases once a preset price is reached, and shoppers “have 24 hours to cancel before it ships” (Modern Retail, 2025-11-13) (third-party evidence). Corrected 2026-10-30 because the first draft said machine payments were the only case.
Stripe keeps a hard credential deadline. From December 2026, Stripe’s MCP server stops accepting full-access secret keys and restricted keys without the Agent tag. Stripe’s docs say “Migrate by November 30, 2026” (Stripe docs) (vendor claim). A secondary summary gives 2026-10-31, and Stripe’s own docs take precedence. The deadline also appears in AR Automation Market report.
Pivots show where the startups found no market. Nekuda, funded in May 2025 by Madrona with Amex Ventures and Visa Ventures, now describes its product as WebMCP tooling for merchant websites (nekuda.ai) (vendor claim). Payman now sells agentic AI to banks and credit unions (paymanai.com) (vendor claim). Both earlier described agent payment wallets and mandates (positioning before 2026 not re-read, section 14).
10. Regulation
No regulator has written a rule for payments an agent starts. The liability answer comes from old consumer-protection text, card-network policy, and merchant terms. Stablecoin rails gained a federal rulebook in October.
Liability: who loses when an agent buys the wrong thing
| Regime | What it says for agents (read 2026-10-30) | Source |
|---|---|---|
| Regulation E (debit, bank accounts) | A transfer by someone the consumer gave the access device to is excluded from “unauthorized” unless authority was revoked. Official commentary suggests the consumer stays liable when that person exceeds authority. Whether credentials shared with AI software trigger the exclusion is unclear | Payments Dive opinion, 2026-10-09 (third-party evidence). Goodwin reads a transfer as “presumed authorized if a consumer furnishes credentials to authorize an agent” (Goodwin, 2026-06-11) (third-party evidence). Fenwick calls it “currently unresolved” (Fenwick, 2026-04-22) (third-party evidence) |
| Regulation Z (credit cards) | Authorized use turns on actual, implied, or apparent authority, which “arguably” covers an agent acting outside its scope. Billing-error rights for goods “not accepted by the consumer or not delivered as agreed” usually push the loss to the merchant | Goodwin (third-party evidence) |
| Card network zero-liability | Applies only once a purchase is classed as unauthorized, so it does not resolve a dispute over whether a purchase fell inside the agent’s mandate | Payments Dive opinion (third-party evidence) |
| Card network agent rules | Visa makes agentic payment providers register with the network and keep “a defensible chain of authorization” (Goodwin’s description). Mastercard scopes each Agentic Token to one agent and policy. No new agent-specific chargeback reason codes found | Goodwin (third-party evidence). Visa’s Intelligent Commerce page says the “Product is currently in the process of deployment” and does not mention disputes (vendor claim). Mastercard’s Agent Pay page returned 403. Rule books not read (section 14) |
| Network purchase protection | American Express Agent Purchase Protection covers charges from agent errors when the cardmember authorized the purchase and a registered agent sent an authenticated purchase intent. Amex calls it the industry’s first such commitment | eCommerceNews, 2026-04-15, Fortune, 2026-04-14 (third-party evidence). Claim limits and exclusions not published in the sources read |
| Stablecoin rails | No chargeback protection. Reg E protections do not extend to crypto-native rails | Backbase, 2026-08-06, Fenwick (third-party evidence) |
| Merchant policy | Walmart’s returns process applies “including if an agent doesn’t act as intended” | Walmart, 2026-10-21 (vendor claim) |
| Money transmission | Turns on whether a platform holds “meaningful authority over payment flows”, such as wallet keys, override power, or pooled funds | Fenwick (third-party evidence) |
What the table says (analysis). On debit, the loss leans toward the consumer who handed over credentials. On credit, it leans toward the merchant. On stablecoins, nobody reverses the payment.
Walmart’s policy, Amex’s protection, and the BILL pilot’s pre-approval design are private fixes for a public gap. Amex’s fix shows the price of coverage. The network covers the loss only for agents in its own registry. The Payments Dive author proposes that losses fall among issuers, networks, merchants, and AI providers and not on consumers. No regulator has taken that up.
US agencies and statutes
| Item | State (read 2026-10-30) | Read |
|---|---|---|
| GENIUS Act, OCC | Final rule 2026-10-21, NR-OCC-2026-92 and Bulletin 2026-50. Covers reserves, redemption, risk management, custody, and foreign issuers. BSA, AML, and OFAC rules come in a separate Treasury rulemaking. Effective 2027-01-18. Applications window to be announced separately (OCC) (third-party evidence, primary) | Crypto press reports a $5M capital floor and a changed treatment of rewards arrangements. The OCC release states neither, so both stay unverified |
| GENIUS Act, other agencies | Fed proposal 2026-09-24 and Treasury interim final rule effective 2026-09-30 (AR Automation Market report). FDIC proposals of December 2025 and April 2026, final rule not found | The Act takes effect by 2027-01-18. No GENIUS text addresses agents |
| CFPB | No agent, Reg E, or digital-payments item in the newsroom for September or October 2026. The one payments-adjacent October item is a debanking initiative (2026-10-14) (CFPB) (third-party evidence, primary). Section 1033 status sits in AR Automation Market report | The bureau that enforces Reg E is silent on agents |
| Courts: Amazon v. Perplexity | Amazon sued in November 2025 and won a preliminary injunction against Perplexity’s Comet agent in March 2026. The Ninth Circuit vacated it, finding Amazon “unlikely to succeed in its claims under federal and California computer access laws” (Retail Insight Network, 2026-08-05) (third-party evidence). The panel treated the agent as a tool following the user’s instructions | The panel said “its ruling is limited to the factual record currently before it”. The case returns to district court. An Amazon rehearing request is reported but not verified |
| Industry self-regulation | The Consumer Bankers Association has a framework favouring ongoing monitoring (Backbase) (third-party evidence) | Voluntary only |
EU and UK
| Item | State (read 2026-10-30) | Read |
|---|---|---|
| PSD3 and PSR | Provisional deal 2025-11-27. Council negotiators approved the compromise text 2026-04-22 and Parliament’s ECON committee 2026-05-05. Official Journal publication not found in the sources read. PSD3 transposition 21 months after entry into force, PSR verification-of-payee articles 27 months after (Legal 500 summary) (third-party evidence) | The texts read do not mention agent-initiated payments. SCA rules get more prescriptive, which bears on how an agent authenticates |
| EU AI Act | GPAI obligations from 2025-08-02. Annex III high-risk obligations from 2027-12-02 (AI Act tracker, updated 2026-08-31) (third-party evidence) | The December 2027 date reflects a delay from the original August 2026. The adopting text was not read |
| MiCA and euro stablecoins | Qivalis, backed by 37 European banks in 15 countries including ING, BNP Paribas, UniCredit, and Danske Bank, received a Dutch e-money licence on 2026-10-14. It plans launch “later in 2026” and does not yet issue (qivalis.eu) (vendor claim) | Banks now have a licensed euro token in hand before any agent volume exists |
| UK | The FCA’s AI approach page, updated 2026-10-02, covers the FCA’s own LLM use and does not mention agents or agent payments (third-party evidence, primary) | One page cannot prove absence. No other FCA page was read (section 14) |
11. Who earns what on an agent transaction
An agent purchase adds two new toll points to a card payment and removes none. The agent surface wants a share, and the credential provider charges per token. The card economics underneath stay as they were, with interchange and surcharging rules tracked in AR Automation Market report and payments-versus-software economics in AR Automation Market report.
| Party | Earns on a card-funded agent purchase | Evidence |
|---|---|---|
| Issuer | Interchange, unchanged | Agentic Tokens and VIC run on existing card economics (vendor claim) |
| Network | Network fees, plus identity and token services | No separate agent fee disclosed (section 8) |
| Processor or wallet | Standard processing, plus a per-credential fee | Stripe SPT at $0.15 (vendor claim) |
| Agent surface | A share of the commerce | Meta’s stated plan for Muse (V via 3P) |
| Merchant | Pays all of the above, gains the order | No merchant-side agent discount found |
Worked example on list prices (arithmetic, not a disclosed deal). Forter’s average agent order is $226 (section 5). On Stripe’s standard card price, the merchant pays $6.85 (2.9% plus 30¢). Adding one SPT brings it to $7.00. The same order paid in stablecoin through Link costs $1.81 at the 0.8% promotional rate, or $3.39 at the 1.5% global rate. Any agent-surface share comes on top of these figures and is undisclosed.
Rail economics by ticket size (the tracked rail question, October answer). Cards win where a consumer or an AP team pays a merchant, because the credential, tokens, and dispute rights already exist. Stablecoins win where a machine pays per call, because x402’s average transfer of about $0.04 sits below any card’s fixed fee. Account-to-account rails appear only under Stripe Treasury’s ACH and bank transfers, and no agent-specific A2A product was found. The networks hedge both ways. Visa and Mastercard are founding validators on Circle’s Arc (arc.io) (vendor claim), premier x402 members, and founders of Open Standard, the OUSD issuer, with Coinbase, Shopify, and Stripe (vendor claim).
Stablecoin yield is the open economic lever. Open Standard pays partners “rewards for adopting and transacting with OUSD”, and Ramp is a named partner (vendor claim). The GENIUS Act bars issuer interest, and how the final OCC rule treats third-party rewards stays unverified (section 10). Whoever can pay a merchant to accept a rail can undercut card pricing on agent flows.
12. Consolidation and funding
Deal timeline, agent payments and the rails under them. Nekuda $5M seed led by Madrona (May 2025) · Google AP2 launch (2025-09-16) · Visa TAP (October 2025) · x402 Foundation (2026-04-02, operational 2026-07-14) · Amex ACE Developer Kit and Agent Purchase Protection (2026-04-14) (third-party evidence) · Adyen Agentic (2026-06-16) (vendor claim) · Open Standard unveils OUSD with “more than 140 partners” (2026-06-30, The Block through Filene) (third-party evidence) · Natural $30M Series A led by Forerunner, over $40M raised (2026-07-21, TNW) (third-party evidence) · Mastercard/BVNK (August 2026, price undisclosed, The Block) (third-party evidence) · Stripe/OpenRouter ($7.5B, August 2026, Payments Dive) (third-party evidence) · Stripe/Parafin (agreed 2026-09-30, price undisclosed) (vendor claim) · OUSD live on Stripe (2026-09-30) (vendor claim) · SAP Pay launched at SAP Connect (early October 2026) (third-party evidence) · Qivalis e-money licence (2026-10-14) (vendor claim).
What the timeline shows:
- The networks buy stablecoin plumbing, not agent startups. Mastercard’s BVNK purchase gave it the agent wallet in its own machine demo. No network bought a KYA or agent-wallet startup in the window.
- Stripe buys the layer above payments. OpenRouter routes model calls, including for Meta’s Muse, per Payments Dive. That puts Stripe on the request path an agent takes before it pays.
- Agent wallet startups are repositioning. Nekuda and Payman moved away from consumer agent payments (section 9). Natural raised the largest disclosed round for agent-native payments infrastructure in 2026 among the sources read.
- Stablecoin consortia formed on both sides of the Atlantic, and US banks lag. OUSD went live on Stripe and Qivalis (37 European banks) won its license within two weeks of each other. US bank consortia have nothing live. The Clearing House targets a tokenized-deposit network with “more than a dozen” owner banks for the first half of 2027. It “has not selected a blockchain vendor”. Early Warning’s ZelleUSD starts with an India corridor, and its “issuing entity and reserve custodian are undisclosed” (Filene, 2026-07-14) (third-party evidence). The post does not mention agents. Paxos shows no agent item among its 2026 releases (section 9).
October funding. No agentic-payments funding round dated in October was found. Search coverage was limited, so this is not proof of absence (section 14). Skyfire’s funding history was not verified in the window.
13. Dated checkpoints
| Date | Event | Why it matters |
|---|---|---|
| 2026-11-12 | Forter webinar on agent order fraud, abuse, and chargebacks | The first third-party agent dispute data |
| 2026-11-16 | Interchange settlement final approval hearing (AR Automation Market report) | Sets the card cost base under every card-funded agent purchase |
| About 2026-11-21 | Walmart Agent Profiles launch (“in the next month” from 2026-10-21) | The first large merchant with standing spending limits for outside agents |
| 2026-11-26 to 11-30 | US Thanksgiving to Cyber Monday | The first holiday with Muse live. Watch Forter, Adobe, and Salesforce agent and AI-traffic data |
| 2026-11-30 | Stripe MCP stops accepting non-agent keys from December | The first hard credential cutover for agent access |
| November to December 2026 | Stripe Treasury agent spending limits (“coming weeks”) and agent-created cards (“soon”) | Whether Stripe moves from per-payment approval to standing limits |
| By 2026-12-31 | Visa Intelligent Commerce Connect broader rollout (promised “this year” in April) | Whether Visa ships one integration across four protocols |
| By 2026-12-31 | Qivalis euro stablecoin launch (“later in 2026”) | A bank-issued euro token available to agent rails |
| 2027-01-01 | Stripe Link stablecoin promotional rate of 0.8% ends | The cheapest published stablecoin agent rail reprices |
| 2027-01-18 | GENIUS Act and OCC rule take effect. FDIC finals due before | US issuers can apply. Watch for any agent-specific guidance |
| Late January 2027 | Visa fiscal Q1 2027 results | Whether “nearly 50” AI partners in production doubled, and the first agent transaction count |
| First half 2027 | The Clearing House tokenized-deposit network target | The first US bank-owned token rail. Watch whether it admits agent payments |
| 2027 | Visa Commercial Virtual Authentication and Supplier Validation general availability | B2B agent controls leave pilot |
| Not dated | Amazon v. Perplexity back in the Northern District of California | Whether a merchant can block a user’s agent by contract or computer-access law |
| Not dated | Mastercard Agent Pay for Machines commercial availability | Machine payments on a card network leave pilot |
| Not dated | PSD3 and PSR Official Journal publication | Starts the 21-month and 27-month clocks |
| 2027-12-02 | EU AI Act Annex III high-risk obligations | Covers AI credit scoring, which any agent credit product touches |
14. Evidence limits
- Web search was limited. Most reads in this edition are direct fetches of known pages, which favors vendor pages over third-party coverage. Absence of an event means “not found in the pages read”, not “did not happen”.
- Outside-in only. Public research measures what vendors publish, not what agents buy. No network, processor, or agent platform discloses agent transaction value, so section 5’s actuals come from one fraud vendor’s network and one unattributed transfer count.
- Primary pages blocked fetches. Mastercard’s newsroom and Agent Pay page, Gartner, eMarketer, OpenAI, Axios, Morgan Stanley (timeout twice), Affirm’s investor site, Fiserv’s release on BusinessWire, and Fintech Futures refused automated reads. Claims from them rest on trade press, aggregators, or search summaries, and each is labelled.
- Vendor-commissioned surveys. Both consumer surveys in section 4 were fielded for Visa.
- Unverified items. None of these was confirmed at a primary source.
- ChatGPT Instant Checkout status: discontinued (secondary blogs) or narrowed (OpenAI’s March update as summarised). Forter still counts ACP orders.
- The 4% ChatGPT merchant fee.
- Tempo mainnet and MPP launch date of 2026-03-18 (search summary only).
- UCP launch date and backer list (secondary only).
- AP2’s FIDO donation date.
- Mastercard Agent Pay’s November 2025 US rollout and its 2026-09-30 or 10-01 agent-score expansion (two dates in two sources). Mastercard pages returned 403 again on a second try.
- The underlying data provider for the x402 weekly count.
- The OCC rule’s $5M capital floor and rewards treatment.
- Whether the FDIC finalised any GENIUS rule.
- PSD3 and PSR Official Journal publication.
- The adopting text for the AI Act high-risk delay.
- Muse’s paid tiers and its coverage under Link purchase protections.
- Amazon’s reported rehearing request in the Perplexity case.
- A secondary claim that Amazon folded Rufus into an “Alexa for Shopping” agent. Other 2026 sources still describe Rufus as live.
- Five forecasts reached only through aggregators or search summaries (eMarketer, Morgan Stanley, Edgar Dunn, Gartner, and Juniper’s $1.5T figure). Each primary was tried. eMarketer and Gartner returned 403, Morgan Stanley timed out, Edgar Dunn has no primary link, and Juniper’s press page lists no agent release.
- Klarna and Affirm placement in Gemini and AI Mode (May 2026 headlines only) and Affirm’s UCP extension.
- JPMorgan and Mirakl retailer pilots (secondary only, JPMorgan’s page gives no status).
- Whether Fiserv’s December 2025 TAP and Agent Pay commitments went live, and whether Global Payments closed its Worldpay purchase.
- Salesforce’s ChatGPT checkout date (sources conflict, Salesforce’s own page lists only the Google connector).
- Amex Agent Purchase Protection claim limits and exclusions, and the protocols the ACE kit supports (one secondary lists TAP, MPP, ACP, and UCP).
- Skyfire’s funding and its claim that KYA is accepted across “more than 60% of the Web”.
- Axios’s report of agents accessing bank accounts outside approved connections.
How was this study done?
- Method
- Public sources only. Desk study of payments that AI agents start for consumers and businesses. Research window 2026-10-01 to 2026-10-30, with earlier background where the structure needs it. The study is refreshed monthly, and this edition is the first, the October 2026 edition.
- Sample
- Desk study. 78 public sources: vendor pages, pricing pages and docs, filings and earnings material, analyst and press coverage, and government pages.
- Confidence
- Every claim comes from a public source and carries a link. Vendor claims are marked (vendor claim) and are unverified. Third-party evidence covers analyst reports, surveys, filings, trade press, and government pages. Web search was limited during the research window, and several analyst, review, and vendor pages blocked automated reads, so some claims rest on press releases, wire copies, or search summaries, and each is labeled. The evidence shows what vendors publish and how analysts and the press describe it. It cannot show what vendors win or what agents do in production. The final section lists the items that stay unverified.
Sources
- Stripe (docs.stripe.com)
- eesel summary
- Stripe
- The Defiant
- Google Cloud
- ap2-protocol.org
- Visa (usa.visa.com)
- The Block
- eCommerceNews, 2026-04-15
- Fortune, 2026-04-14
- Linux Foundation
- x402.org
- PYMNTS
- AAIF
- BusinessWire, 2026-01-23
- Stripe
- PYMNTS
- Walmart (corporate.walmart.com)
- Modern Retail, 2025-11-13
- Checkout.com, 2026-10-29
- FinanceFeeds
- Payments Dive
- Stripe, 2026-10-21
- Visa, 2026-10-20 (usa.visa.com)
- Visa, 2026-10-30 (usa.visa.com)
- Visa, 2026-09-09 (usa.visa.com)
- Stellagent
- eMarketer
- Primary page (morganstanley.com)
- Bain
- McKinsey mirror (karriere.mckinsey.de)
- Gartner release
- Visa, 2026-10-21 (usa.visa.com)
- Forter, 2026-10-23
- TechCrunch, 2026-10-21
- Visa (usa.visa.com)
- Visa, 2026-10-01 (usa.visa.com)
- Salesforce
- Ramp docs (agents.ramp.com)
- Yahoo Finance
- SAPinsider
- Techzine, 2026-10-05 or 10-06
- Payments Dive, 2026-10-15
- Visa (usa.visa.com)
- tempo.xyz
- stripe.com/pricing
- x402.org
- Stripe, 2026-09-30
- Stripe
- Perplexity, 2025-11-25
- ramp.com/pricing
- Fortune, 2026-10-26
- eWeek
- BusinessWire (secure.businesswire.com)
- eMarketer
- Adyen
- Digital Transactions
- agents.trulioo.com
- Digital Transactions
- JPMorgan
- Paxos
- Stripe docs
- nekuda.ai
- paymanai.com
- Goodwin, 2026-06-11 (goodwinlaw.com)
- Fenwick, 2026-04-22
- Intelligent Commerce page (visa.com)
- Backbase, 2026-08-06
- OCC
- CFPB (consumerfinance.gov)
- Retail Insight Network, 2026-08-05 (retail-insight-network.com)
- Legal 500 summary
- AI Act tracker, updated 2026-08-31 (artificialintelligenceact.eu)
- qivalis.eu
- AI approach page (fca.org.uk)
- arc.io
- Filene
- TNW (thenextweb.com)
Published .
Cite this
Couch, C. (2026). Agentic Payments Market: October 2026. Ends at Cash. https://chriscouch.ai/research/agentic-payments-market-2026/
BibTeX
@misc{couch2026agenticpayments,
author = {Couch, Chris},
title = {Agentic Payments Market: October 2026},
year = {2026},
howpublished = {Ends at Cash},
url = {https://chriscouch.ai/research/agentic-payments-market-2026/}
}