Who is assembling the agentic-commerce market?
The organisations entering agentic commerce do not begin from equivalent positions. AI platforms, commerce ecosystems, wallets, payment companies, networks and national rails are extending different assets towards the places where customer intent becomes authorised commercial action.
This date-stamped Reference Map accompanies The alphabet soup of agentic commerce. The Signal explains why the technical activity matters. This Map tracks who originated or governs the principal initiatives, who is contributing or implementing them, their maturity as at 3 September 2026, and where organisations are building commercial positions.
A draft, proposal, released protocol, technical integration, controlled pilot, production-system validation and scaled commercial adoption are not equivalent. The status descriptions below distinguish them. Participation in an initiative is not evidence that an organisation has adopted it at scale or secured the corresponding market position.
SignalThe alphabet soup of agentic commerceThe Signal this Reference Map accompanies: why the technical activity matters and what it signals about the emerging market structure.The market at a glance
The groups below describe starting assets, not permanent categories. Several organisations already span more than one.
| Starting asset | What it can provide | Where participants are moving |
|---|---|---|
| Customer interface | Access to expressed intent, search, conversation or advertising demand | Product discovery, checkout, merchant participation and payment choice |
| Integrated commerce environment | Customers, merchants, catalogue, transactions and sometimes payment or fulfilment already connected | Agent-mediated choice and purchase inside an existing commercial system |
| Wallet or super app | Credentials, identity, authentication, loyalty, customer context and service access | Continuing authority, cross-service orchestration and control of the path to payment |
| Merchant and payment infrastructure | Checkout, merchant integrations, tokenisation, risk and processing | Aggregation across agentic surfaces and protocols |
| Payment network or national rail | Acceptance, credentials, scheme rules, authentication and settlement reach | Trusted-agent recognition, portable evidence and delegated payment |
| Machine-native infrastructure | Programmatic payment, software wallets and digital-asset settlement | Paid API calls, tool use and other machine-consumed services |
Two questions cut across all six.
Who controls the route to choice? This is the position from which a request is interpreted and the reachable merchants, products or services are determined.
Who controls the authority to act? This is the position from which credentials, permissions, policy and evidence allow a decision to become a transaction.
Wallets and super apps are important because they may bring those positions together. Independent agents and portable standards are important because they may allow customer context and authority to travel across them.
What the principal initiatives cover
The seven initiatives named in the Signal remain the organising set. The additional entries are included because they reveal materially different wallet-, network- or market-oriented approaches. The groupings identify each initiative’s principal contribution to this Map rather than a rigid boundary: ACT combines delegation, commerce, payment and trust, while AMP connects wallet environments to agent-mediated payment.
Intent, authority and evidence
| Initiative | Origin and governance | Current scope and maturity | What its development reveals |
|---|---|---|---|
| Agent Payments Protocol (AP2) | Created by Google; contributed to the FIDO Alliance in April 2026. | Public v0.2 defines roles, signed mandates, verification duties and human-present and human-not-present payment flows. FIDO work is continuing; AP2 is not a final FIDO standard and scaled adoption is not established. | Customer authority is being positioned as portable trust infrastructure rather than evidence held by one platform. |
| Verifiable Intent | Developed by Google and Mastercard; contributed alongside AP2 to FIDO’s member-led work. | Published, AP2-compatible and protocol-agnostic contribution providing tamper-resistant evidence of what a person authorised. Not a final standard. | Networks, merchants and issuers want independently verifiable evidence without relying solely on the agent’s account of the transaction. |
| EMV Digital Payment Credential (DPC) | Draft developed through EMVCo, which is overseen by American Express, Discover, JCB, Mastercard, UnionPay and Visa. | Draft schema framework for a verifiable credential used in card-payment authentication. It is not a wallet, complete agentic-payment system or deployed credential. | A capability concentrated in proprietary wallet environments may become more portable, but the draft does not determine which wallet or verifier will control its use. |
| EMV Agentic Payments Framework | Draft developed through EMVCo’s Agentic Payments Task Force. | Released for public review on 1 September 2026. Proposes Intent Services for registering, retrieving and managing consumer-authorised intent over time. It may inform later specifications; it is not an operating standard. | The card ecosystem is moving beyond transaction authentication towards a shared state of continuing intent. |
| Agentic Commerce Trust Protocol (ACT) | Alipay-hosted, openly documented framework with a named partner ecosystem. Its development remains sponsor-led rather than independently governed. | Covers delegation, commerce interaction, payment and trust. Published and open to participation, but broad interoperable adoption is not established. | Trust frameworks are also being designed around super apps, wallets, devices and integrated commerce—not only card-based web checkout. |
| Agentic Payment Open Protocol (APOP) | Released and led by China UnionPay. | Public framework with restricted implementation access. UnionPay reported five production-system validation transactions at launch and, by July, engagement with more than 40 industry participants alongside pilots in banking, merchant and device contexts. Scale is not established. | A payment network can combine agent recognition, structured intent, identity and authorisation over its existing acceptance infrastructure. |
Primary sources: AP2 specification; FIDO — AP2 and Verifiable Intent; Mastercard — Verifiable Intent; EMVCo — DPC; EMVCo — Agentic Payments Framework; ACT Protocol; China UnionPay — APOP launch; China UnionPay — July update.
Commerce connection
| Initiative | Origin and governance | Current scope and maturity | What its development reveals |
|---|---|---|---|
| Agentic Commerce Protocol (ACP) | Co-developed and maintained by OpenAI and Stripe. | Maintained beta protocol connecting merchant commerce capabilities to AI applications. ChatGPT supports product discovery and, for approved partners, checkout rendered in ChatGPT while the merchant maintains checkout state and remains merchant of record. Commercial scale is not established. | OpenAI can extend from interpreting demand towards conversion without assuming the merchant’s order and fulfilment obligations; Stripe supplies reusable merchant and payment infrastructure. |
| Universal Commerce Protocol (UCP) | Announced by Google with Shopify, Etsy, Wayfair, Target and Walmart; developed through a multi-company Tech Council. | Released and actively developed. The 25 August 2026 release expanded payment security, identity, consent, loyalty and additional commerce contexts. Implementations remain early and broad adoption is not established. | A common capability-negotiation layer is also a governance contest among platforms, merchants, commerce providers and payment participants. |
Primary sources: Stripe — ACP integration; OpenAI — ACP and checkout; OpenAI — product discovery; Google Developers — UCP; UCP releases; UCP Tech Council.
Payment permission and wallet connection
| Initiative | Origin and governance | Current scope and maturity | What its development reveals |
|---|---|---|---|
| Shared Payment Tokens (SPT) | Proprietary Stripe capability. | Released, time-limited and transaction-scoped token allowing an agent to pass a selected payment method and risk signals without exposing the underlying credential. Stripe names Etsy and URBN brands as adopters and supports network and buy-now-pay-later connections. | A proprietary abstraction can connect several agentic surfaces and established payment methods while its provider controls enforcement and integration. |
| Agentic Mobile Protocol (AMP) | Open-source, Ant International-led protocol. | Released in April 2026 for wallets, banking applications, super apps, wearables and in-car systems. Ant says more than 40 Alipay+ wallet partners are working on implementation and names Visa, Mastercard and Google collaborations. Scaled protocol adoption is not established. | Wallet reach and mobile trust infrastructure can provide an alternative centre for agentic commerce rather than serving only as the final payment method. |
| Agent Autonomous Payment Protocol (A2P2) | Published, sponsor-led framework from JD. | Early agent-payment framework and developer tooling connected to JD’s commerce, logistics and payment assets. Broad external implementation has not been established. | An integrated commerce company can develop a reusable interface while retaining the option to deploy it first across assets it already controls. |
| Pine Labs Payment Protocol (P3P) | Created and operated by Pine Labs; publicly documented but sponsor-led. | Released in June 2026 on UPI mandate infrastructure. Pine Labs reports Gullak live and Vijay Sales in an active proof of concept. Independent scale is not established. | A payment provider can add delegated agent authority above a national real-time rail without replacing its settlement system. |
Primary sources: Stripe — SPT; Ant International — AMP; JD A2P2; JD ClawTip; Pine Labs — P3P launch; P3P documentation.
Machine-payment execution
| Initiative | Origin and governance | Current scope and maturity | What its development reveals |
|---|---|---|---|
| Machine Payments Protocol (MPP) | Co-authored by Stripe and Tempo. | Released in March 2026 with early live services. Payment-method and currency agnostic; Stripe’s implementation supports stablecoins and fiat methods through SPT. Broad adoption and durable independent governance are not established. | Payment can become part of an API request or tool call, allowing services to be sold and recombined at a smaller economic unit. |
| x402 | Created by Coinbase and contributed to the Linux Foundation. | Released protocol based on HTTP 402 Payment Required. The x402 Foundation launched operationally in July 2026 with 40 members. Current execution remains substantially crypto-native; broad commercial adoption is not established. | A sponsor-led protocol is seeking vendor-neutral governance and a wider role as internet-native payment infrastructure. |
MPP and x402 are moving towards some of the same territory. MPP is expanding across conventional and digital-asset methods; the x402 Foundation’s stated remit extends from stablecoins towards broader payment types. The remaining contest concerns which request model, wallets, facilitators, policy controls and settlement arrangements become easiest for machines to use.
Primary sources: Stripe — MPP; MPP multi-method discovery; Coinbase — x402; Linux Foundation — x402 Foundation.
Connective protocols
Model Context Protocol (MCP). Anthropic created MCP to connect AI applications to tools, data and workflows. It is general agent infrastructure, not a commerce or payment-authorisation standard. Commerce and payment providers are nevertheless using it to expose capabilities to agents. Source: MCP documentation.
Agent2Agent (A2A). Google created A2A for communication between agents and contributed it to the Linux Foundation. The current project is under foundation governance. A2A can carry or coordinate commerce-related interactions, but does not itself provide checkout, payment authority or settlement. Source: A2A project.
These protocols lower the cost of reaching tools and other agents. They do not remove the need for merchant participation, trusted authority, payment credentials or correction when a transaction fails.
Where organisations are building positions
Actors are grouped by their most relevant starting asset and alphabetically within each group. The order does not imply importance, maturity or likely success.
Customer interface and orchestration
| Actor | Existing position and current movement | Present maturity |
|---|---|---|
| Combines search and AI distribution with AP2, UCP and established wallet and payment capabilities. It is participating across discovery, commerce connection, authority and payment rather than making a single-protocol bet. | Multiple released protocols and products; implementation and adoption vary by component. | |
| Meta | Begins with advertising, social, messaging and merchant reach. A Facebook flow built on ACP moves from an advertisement to product detail, AI-generated review summaries and in-app purchase. Its position is embedded commerce rather than a distinct foundational protocol. | Introduced implementation; availability and scale are not established. |
| Microsoft | Uses Copilot, Bing, Edge, MSN and advertising distribution to extend into product discovery and conversational checkout through Shopify, PayPal and Stripe routes. Merchants remain merchant of record. | US implementation live or rolling out; broader coverage and scale are not established. |
| OpenAI | Controls a conversational surface in which intent is interpreted, products presented and checkout can occur. ACP allows it to approach conversion while merchants retain order and transaction obligations. | Live product discovery and approved-partner checkout; scale not established. |
Sources: Microsoft — Copilot Checkout; Stripe — Meta’s ACP-based Facebook checkout flow.
Integrated commerce environments
| Actor | Existing position and current movement | Present maturity |
|---|---|---|
| Alibaba, Qwen and Alipay | Combine an AI assistant with marketplaces, local services, travel, maps and payment. Alipay lists Qwen among the services using its Agent Pay infrastructure; ACT provides an accompanying trust architecture. | The integrated Qwen commerce path is early. Alipay reports scaled use of its wider AI-payment infrastructure, but that is not evidence of scaled ACT adoption or interoperability beyond the ecosystem. |
| Amazon | Already connects large-scale discovery, merchants, payment and fulfilment. Alexa for Shopping extends Rufus and Alexa+ across Amazon surfaces; Buy for Me can purchase selected products from external brand sites within Amazon’s application. Amazon also participates in UCP governance and AWS implements machine-payment capability. | Live proprietary shopping capabilities; external purchasing remains selective. |
| JD | Combines retail, logistics and payment capabilities with A2P2 and early developer tooling. | Early framework and integrated experimentation; broad external adoption not established. |
| Naver | Extends AI across search and shopping through its Shopping AI Agent and AI Tab. | Discovery and shopping implementation; independent payment execution not established. |
| Rakuten | Brings Rakuten AI into a marketplace supported by loyalty, payment, travel and service relationships. | Live AI discovery and selection; autonomous payment not established. |
| Sea and Shopee | Developing an agentic-shopping prototype with Google across an established Southeast Asian marketplace and payments ecosystem. | Announced prototype, not a generally available product. |
Sources: Amazon — Alexa for Shopping; Amazon — Buy for Me; Alipay Agent Pay; ACT Protocol; Rakuten AI; Naver — Shopping AI Agent; Naver — AI Tab; Reuters — Google and Sea.
Merchant and payment infrastructure
| Actor | Existing position and current movement | Present maturity |
|---|---|---|
| Adyen | Positioning itself as a protocol-agnostic merchant layer across UCP, AP2, ACP and AI surfaces, preserving merchant payment data and operations while aggregating emerging routes. | Limited initial availability; broad scale not established. |
| Ant International | Uses Alipay+ wallet and merchant reach to develop AMP, while participating in Visa, Mastercard and Google initiatives. | Released protocol and implementation work across named partners; scale not established. |
| PayPal | Combines a consumer wallet, branded checkout and merchant processing. It has announced ACP-based merchant access and ChatGPT checkout, and support for UCP checkout with Google. | Announced integrations and infrastructure; availability varies and scaled agentic-commerce adoption is not established. |
| Pine Labs | Extends its merchant and payment infrastructure into delegated agent payments through P3P over UPI mandates. | Pine Labs reports one live named implementation and an active proof of concept; independent scale is not established. |
| Salesforce | Is implementing native UCP support for Agentforce Commerce merchants and participates in UCP governance. | Announced integration and governance participation; rollout timing had not been finalised. |
| Shopify | Supplies merchant commerce infrastructure and is a founding force behind UCP, helping merchants become reachable across multiple customer surfaces while retaining Shopify’s position in checkout and operations. | Released protocol and expanding implementation ecosystem. |
| Stripe | Participates across ACP, SPT and MPP and supports network-led agentic capabilities. Its position spans merchant connection, payment permission and machine-native execution. | Multiple released capabilities with different adoption levels. |
| Worldline | Exposes payment functions through MCP and has demonstrated agentic transactions with banks and networks in production-system environments. | Production proof points, not broad commercial availability. |
Sources: Adyen Agentic; PayPal — ACP and ChatGPT; PayPal — UCP and Google; Pine Labs — P3P; Salesforce — UCP support; UCP Tech Council; Worldline — agentic commerce; Stripe agentic commerce.
Payment networks and national rails
| Actor | Existing position and current movement | Present maturity |
|---|---|---|
| China UnionPay | Combines domestic and international acceptance with APOP, a later agent value-added services protocol (AVOP), and agentic card, acquiring and device pilots. | Released network-led framework, industry-participant engagement and restricted production-system validation and pilots; scale not established. |
| Mastercard | Extends tokenised credentials, policy controls and network rules through Agent Pay, Agent Pay for Machines and Verifiable Intent, with connections to SPT and MPP. | Released programmes and controlled authenticated transactions; scale varies and is not established. |
| NPCI and UPI | Reuters reported on 1 September 2026 that NPCI was preparing a Unified Agent Protocol for bounded agent payments over UPI-related capabilities. | Reported development only. No authoritative specification or launch had been published by 3 September. |
| Visa | Extends token, authentication, agent-recognition and acceptance capabilities through Visa Intelligent Commerce and Trusted Agent Protocol, alongside platform integrations including OpenAI. | Released capabilities and early deployments; individual programmes differ in availability and scale. |
Sources: China UnionPay — APOP development and AVOP; Mastercard — Agent Pay for Machines; Visa — Trusted Agent Protocol; Reuters — reported NPCI work.
Wallets and super apps are cross-layer positions
A wallet is no longer relevant only at the moment of payment. It can hold credentials, identity, authentication, loyalty, policy and evidence of continuing authority. A super app can add customer context, merchant reach and service orchestration.
The strategic contest is not settled. Three broad outcomes remain plausible:
- an independent agent carries the customer’s context and authority across multiple commercial environments;
- a portable wallet becomes the trusted control point used by several agents;
- or a platform or super app absorbs the agent and keeps more of the route from intent to payment inside its environment.
Current activity supports all three possibilities. AP2, Verifiable Intent and DPC pursue forms of portability. AMP and ACT are being built around wallets and super-app environments. OpenAI and Google are extending customer interfaces towards commerce. Amazon, Alibaba and JD can build across systems they already control.
The battleground is therefore not simply “top of wallet.” It is whether intent, authority and commercial access remain portable—and which participant sets the defaults when they meet.
Competitive movement
In July 2026, Reuters reported that Stripe and Advent International had offered more than $53 billion to acquire PayPal. On 28 August, Reuters reported that the consortium had abandoned the pursuit. Stripe, PayPal and Advent did not publicly confirm the negotiations.
As at 3 September, this is not evidence of consolidation; Stripe and PayPal remain competitors. It is nevertheless strategically suggestive. A combination would have joined Stripe’s merchant and infrastructure reach with PayPal’s consumer wallet, Venmo, checkout and merchant-processing assets. The inference that these cross-layer positions contributed to the attraction is QURKI analysis, not a stated rationale from the companies.
Sources: Reuters — reported offer; Reuters — pursuit reportedly abandoned.
Developments checked but not promoted
| Development | Treatment |
|---|---|
| Apple | Its device and wallet position could become strategically important if authority and customer context move into agent-mediated commerce. No sufficiently evidenced, distinct agentic-commerce initiative was found by the currency date, so it is a position to monitor rather than a principal assembler in this edition. |
| Kimi and Moonshot AI | General-agent and browser capabilities are established, but no adequate first-party evidence was found of a material merchant, checkout, payment or commerce-protocol position. |
| Tencent and WeChat Pay | MCP services and restricted testing indicate an emerging position, but the evidence is not yet sufficient for a principal profile. |
| Doubao | Reported commerce testing was not matched by adequate first-party confirmation. |
| European sovereign and regional payment infrastructure | The digital euro, European instant-payment infrastructure and Wero/EPI may reduce dependence on non-European payment providers. No material agentic-commerce layer was established by the currency date, so they are not principal entries in this Map. Their strategic effect could nevertheless be altered by whichever agentic layer influences access and payment choice above the rail. |
| Mercado Libre, Grab and M-Pesa | Each has potentially relevant regional commerce or payment assets; no sufficiently developed and authoritative agentic-commerce initiative was found by the currency date. |
| Brazilian Pix tooling | Iniciador has released an MCP-based toolkit through which an agent can prepare a Pix payment for biometric user authorisation and bank execution. Useful evidence of an early account-to-account route, but not yet a principal market formation. |
Source for Pix tooling: Iniciador — agentic Pix toolkit.
These exclusions are deliberate. An actor enters the principal map only when its activity changes the reader’s understanding of the emerging structure, not merely because it has announced an AI feature.
What the map reveals
The market is not converging on one globally uniform stack.
In fragmented environments, protocols and infrastructure providers are connecting separate customer interfaces, merchants, wallets and payment systems. Integrated marketplaces and super apps can build across assets they already control and use standards selectively to extend beyond them. Networks and national rails provide further routes.
The direction of movement is nevertheless consistent:
- customer interfaces are extending towards checkout and payment;
- commerce platforms are extending towards agent-mediated choice;
- wallets are extending towards intent, policy and cross-service authority;
- payment providers are aggregating multiple protocols;
- networks and national rails are extending recognition, credentials and rules;
- machine-payment systems are extending towards broader methods and commercial uses.
Protocol openness does not settle the resulting structure. Distribution, defaults, acceptance, commercial terms, evidence, liability and implementation remain decisive.
The most valuable position may belong neither to the organisation with the most visible agent nor to the author of the winning protocol. It may belong to the participant that becomes difficult to bypass when customer intent has to become authorised, trusted and correctable commercial action.