The Agentic Pay Ecosystem: Where Each Player Fits
Part IV · Adding Agentic Pay (Chapters 11–12) Builds on: Chapter 11 (six-step delegated-spending loop), Chapter 8 (issuing), Chapter 6 (acquiring), Chapter 10 (stablecoins); Payment Systems, Chapter 30 (four approaches), Chapter 28 (network neutrality) New concepts in this chapter: four player layers, six protocols and their responsibilities, three developments to track
(Companies, protocols and figures reflect public information available before September 2026, including developments from mid-2026. The names will change. Retain each player's step in the loop and position in the PSP architecture.)
1. The question from the previous chapter¶
The previous chapter defined six steps: mandate, candidate spend, policy decision, scoped credential, existing payment execution and evidence. Who performs each?
Start with the layers, then locate the companies and examine three developments from mid-2026.
2. Four layers: mandate and identity, credentials, execution and protocols¶
| Layer | Steps in the loop | Function | Players |
|---|---|---|---|
| Mandate and identity | 1–3 | Where people set boundaries, how agents identify themselves and who decides approval | Agent platforms: OpenAI, Google, Anthropic, Microsoft, Perplexity; enterprise spend software: Ramp, Brex; identity services: Visa Agentic Directory, Cloudflare Web Bot Auth, Skyfire |
| Credentials | 4 | Turn an approved spend into a restricted payment credential | Visa Intelligent Commerce, Mastercard Agent Pay; Stripe Shared Payment Token; single-use cards through Stripe Issuing, Lithic and Marqeta; wallets such as PayPal and Coinbase |
| Execution | 5 | Put the credential onto a rail and enable merchant acceptance | Existing card, bank and on-chain infrastructure; merchant platforms and acquirers including Shopify, Adyen, Checkout.com and Stripe |
| Protocols | Across the loop | Let the providers in the other layers interoperate | ACP, UCP, AP2, x402, MPP, Visa Trusted Agent Protocol |
Credentials and execution contain the PSP opportunities identified in the previous chapter. Agent platforms are prominent at the mandate entry point, while protocol design spans the competing providers.
3. What ACP, UCP, AP2, x402, MPP and TAP each cover¶
Similar names, close launch dates and overlapping supporters make protocols hard to compare. Their responsibilities are more informative:
| Protocol | Originator | Introduced | Steps covered | Settlement asset | Governance | Status described around mid-2026 |
|---|---|---|---|---|---|---|
| ACP | OpenAI and Stripe | September 2025 | 2 and 5: agent–merchant product, checkout and order interaction | Mainly cards | Open specification published by Stripe | In March 2026, OpenAI scaled back its original Instant Checkout approach toward discovery with merchant-controlled checkout |
| UCP | Google and Shopify | January 2026 | 2, 5 and 6: discovery through fulfillment | Asset-agnostic | Partly public | Connected to Google shopping and Gemini, covering Shopify merchants |
| AP2 | Google with more than 60 organizations | September 2025 | 1, 3 and 6: verifiable mandate and approval evidence | Cards and stablecoins | Google-led, with schemes, PayPal, Amex and Coinbase participating | Card schemes have aligned credential formats with AP2 |
| x402 | Coinbase | May 2025 | 5: HTTP resource request, price response and machine payment | Initially stablecoin-focused; foundation scope extends to cards | Moved to the Linux Foundation's x402 Foundation in July 2026, with 40 members | About 75 million payments and $24 million over one reported month, averaging about $0.32 |
| MPP | Stripe and Tempo | March 2026 | 5: HTTP 402 payment flow, with Stripe SPT for cards and Tempo for stablecoins | Cards and stablecoins | Published by Stripe | Stripe supports both MPP and x402 |
| Trusted Agent Protocol | Visa | October 2025 | Between 3 and 5: merchant recognition of trusted agents and verification of signed intent | Cards | Visa | Aligning with OpenAI's protocol |
Two relationships stand out.
Checkout protocols and machine-payment protocols address different ends of the use-case spectrum. ACP and UCP serve shopping flows with purchases of tens to thousands, often on cards. x402 and MPP support machine interactions, including sub-dollar payments suited to low-cost rails. The reported x402 figures imply $24 million ÷ 75 million = $0.32 per payment, illustrating the economics of small machine payments.
AP2 and Trusted Agent Protocol establish authorization and identity evidence. They answer who authorized what and which agent is acting, rather than replacing the settlement rail. They can therefore coexist with payment protocols and different settlement assets; support for AP2 and x402 is not inherently contradictory.
4. Place each company in the model¶
Map the principal players onto the PSP model and the delegated-spending loop:
| Player | Position in the PSP model | Steps in the loop | Incentive |
|---|---|---|---|
| Visa and Mastercard | Card rails | Agent-specific credentials in 4; agent verification after the policy decision | Retain agent spending on card rails while also participating in the x402 Foundation |
| Stripe | Acquiring, issuing and stablecoin endpoints | 4 through SPT and virtual cards; 5 through acceptance and MPP; protocols through ACP and MPP | Combine credentials and execution, including on-chain capability through Bridge |
| OpenAI and Google | Agent platforms outside the payment architecture | Entry point for 1–3 | Shape how intent becomes a purchase; merchant resistance limits control over checkout |
| Coinbase | On-chain execution | 5 | Broaden adoption through neutral governance of x402 |
| Shopify | Merchant platform on the acceptance side | Merchant-facing part of 5 | Make millions of merchants accessible to agents while supporting ACP and UCP |
| PayPal | Wallet with acceptance capabilities | User approval and credentials in 4; merchant network in 5 | Connect existing wallet relationships to agent platforms |
| Spend-management providers such as Ramp and Brex | Issuing plus balance core | Existing components for 1–4: enterprise policy, approval and virtual cards | Extend business spend controls into procurement agents |
| Startups such as Skyfire and Payman | Agent services outside the traditional model | Identity in 1 and wallet credentials in 4 | Build agent-specific identity and wallet products |
Agentic Payments describes the recurring opportunity to set network rules without holding the money. The incentives differ: schemes defend their role, agent platforms seek influence at the entry point, Coinbase trades exclusive standard control for broader participation, and Stripe combines protocol work with credentials and execution.
The position generally follows what a company already controls. Rail operators extend credentials; merchant platforms extend checkout; user platforms capture mandates; new entrants seek protocol, identity or wallet roles.
5. Three mid-2026 developments and what they imply¶
Between the first announcements in 2025 and mid-2026, three developments gave more concrete signals than product plans alone.
First, merchants retained checkout control. OpenAI launched Instant Checkout in September 2025, letting users purchase within ChatGPT. By March 2026, with only dozens of merchants live in the original approach, it shifted emphasis toward discovery and comparison while checkout remained in merchants' systems. Walmart's approach used its own account and payment experience. In the language of acquiring, checkout contains the merchant's customer relationship, pricing and data. The acceptance-side opportunity is to make merchant-controlled checkout usable by trusted agents. Visa's June 2026 work with OpenAI follows that pattern: tokenized Visa credentials enter the agent flow with user-set spending and merchant-category limits, while merchants retain existing authorization and fraud processes.
Second, machine payments showed substantial counts at small average values. The x402 Foundation's reported monthly figures — 75 million payments and $24 million — produce a $0.32 average. They provide an observed example of the small-payment economics discussed in Agentic Payments. The relevant PSP capability is an on-chain endpoint that lets a business agent pay per use from stablecoin funds.
Third, authorization approaches remained differentiated while transport governance became more neutral and compatibility increased. Schemes aligned with AP2, Visa worked toward OpenAI compatibility, Stripe supported both MPP and x402, and schemes and Stripe joined the x402 Foundation. This resembles the separation in a remittance message: common transport can coexist with separate authorization implementations. For PSPs, it favors credentials and execution that can connect to several protocols.
6. Three developments to track¶
Agentic Payments proposed tracking liability allocation, payment size and regulatory divergence. From a PSP's perspective, add three questions.
Who issues the accepted credential? Schemes, PSPs, issuers and wallets all want step 4. Track which credential merchants actually accept by default: an agent token, Stripe SPT, a single-use virtual card or wallet approval. Default acceptance can create pricing power at that layer.
Who retains checkout? The developments after March 2026 point toward merchant-controlled checkout in consumer shopping. That creates work for acquirers and merchant platforms: machine-readable checkout, agent verification and scoped-credential acceptance. Watch how Shopify, Adyen and Checkout.com implement those capabilities.
Does enterprise use work before consumer use? Business mandates are complex, but invoices, contracts and budgets already provide structured evidence. Spend-management companies possess many components of steps 1–4. If this use case works first, the PSP's existing business customers may be the earliest adopters.
7. Review of the course¶
We began with one relationship: a PSP provides its customers with one balance core, connected to pay-in and payout endpoints across countries. Each part added something to it:
| Part | What it added | Reasoning tool |
|---|---|---|
| I · Architecture | Licensing, engineering and build-or-partner choices for three components | Permission and correct operation; endpoints act on funds, the core records customer balances |
| II · Extensions | Acquiring, local methods and issuing | Push and pull allocate responsibilities; scheme sponsorship creates intermediaries on both sides |
| III · Stablecoins | Customer balances, on-chain endpoints and a treasury bridge | Who holds the token, for how long, and how costly is the existing corridor? |
| IV · Agentic Pay | Six-step delegated-spending loop and four player layers | Agents change pre-payment decisions; existing capabilities shape each player's role |
The final exercise is to choose a payment company and explain it using these four rows. Does it own or source its balance core? Where does it build or source endpoints? Does it offer acquiring, issuing and local methods? Which stablecoin positions does it use? Which steps of the agentic loop does it occupy? If you can reconstruct those relationships, you can apply the course beyond its examples.
8. Self-check questions¶
- A PSP announces full support for ACP, UCP, AP2 and x402. Which steps can those protocols connect? Does protocol support establish complete Agentic Pay capability?
- How does a move back toward merchant-controlled checkout affect the opportunity for the providers in acquiring?
- A PSP offers only bank payout endpoints, with neither issuing nor acquiring. Where can it enter the delegated-spending loop, and what credential can it provide?
9. Answers¶
Try answering before reading on.
ACP and UCP connect agent platforms and merchants around candidate purchases and execution; UCP also reaches fulfillment evidence. AP2 connects mandate, approval and evidence. x402 supports machine-payment execution. But interoperability alone does not supply scoped credentials or funds movement. The PSP still needs credential-issuance authority and working balance and endpoint capabilities, or partners providing them.
It increases the need for merchant-side work. If checkout stays with the merchant, someone must let it recognize trusted agents, accept scoped credentials and join orders to payments. Those functions fit acquirers and merchant platforms. The merchant-facing portion of execution remains an essential part of the agent purchase.
Step 4, using a scoped bank-payment instruction limited to a beneficiary, amount and expiry, followed by its existing payout execution in step 5. Its strongest fit is enterprise invoice payment: an agent proposes a supplier payment, policy approves it, the PSP issues the limited authorization and executes it, and evidence links back to the invoice. Without card capabilities, it cannot directly provide the same credential for card-only shopping.
Previous: Chapter 11 · Agentic Pay: What Problem Does It Solve?