Agentic Payments: The Complete Guide (2026)

What are agentic payments?

Agentic payments are transactions initiated and completed by AI agents on a user's behalf, under rules the user defines in advance — instead of a human clicking "buy."

Every payment system built over the past three decades assumes a live human is present at checkout: typing card numbers, approving 3-D Secure prompts, clicking confirmation buttons. Agentic payments replace that moment-of-purchase approval with policy-driven delegation: you tell an agent what it may buy, how much it may spend, and when — and cryptographic protocols prove to the merchant and the bank that the agent really is acting within your authorization.

Three shifts make this different from "just another checkout button":

1. From authentication to delegation. The network no longer verifies you at every purchase; it verifies your agent's right to act under a standing mandate you signed once.
2. From sessions to mandates. Instead of a checkout session, the unit of trust is a cryptographically signed contract (an "Intent Mandate" or "Cart Mandate") that records exactly what the user authorized.
3. From human-speed to machine-speed. Agents don't abandon carts, don't get distracted by banners, and expect checkout to complete in under a second. Merchants optimizing for agents optimize for structured data and zero friction.

Market size: Juniper Research projects agentic commerce spend will reach $1.5 trillion globally by 2030, growing from pilot-only deployments in 2025–2026. Bain estimates the US market alone at $300–500 billion by 2030 (15–25% of domestic e-commerce); McKinsey's global projection runs $3–5 trillion. Even the conservative numbers imply the fastest infrastructure build-out in payments since contactless.

How an agentic payment works, step by step

A typical flow under Google's AP2 (the most widely backed open protocol) looks like this:

USER                    AGENT                    MERCHANT                NETWORK/ISSUER
 │                        │                         │                        │
 │ 1. Intent Mandate      │                         │                        │
 │ "Buy running shoes     │                         │                        │
 │  under $120, size 10"  │                         │                        │
 │ (signed, with limits)  │                         │                        │
 │───────────────────────>│                         │                        │
 │                        │ 2. Discovery &          │                        │
 │                        │    negotiation          │                        │
 │                        │────────────────────────>│                        │
 │                        │ 3. Merchant-signed cart │                        │
 │                        │<────────────────────────│                        │
 │ 4. Cart Mandate        │                         │                        │
 │ (approval — or auto if │                         │                        │
 │  pre-authorized)       │                         │                        │
 │───────────────────────>│                         │                        │
 │                        │ 5. Payment Mandate ─────┼───────────────────────>│
 │                        │    (signals "an AI      │                        │
 │                        │     agent did this")    │   6. Risk check &      │
 │                        │                         │      authorization     │
 │                        │                         │<───────────────────────│
 │ 7. Receipt + full audit trail (who authorized what, when, within which limits)
      

The three mandate types are the heart of the system:

If anything goes wrong — wrong item, wrong amount, a hallucinating agent — the mandates form a non-repudiable audit trail that determines who's liable: user, agent developer, or merchant. That accountability question, not the technology, was the main blocker to agent-led checkout. The protocols below are competing answers to it.

The six competing protocols (comparison table)

This is the part every vendor page conveniently leaves out. Here's the neutral map, current as of July 2026:

How to read this table

• This is not (yet) a winner-take-all war. AP2's donation to FIDO and Mastercard/Visa chairing FIDO's payments working group signal convergence at the trust layer, while UCP and ACP compete at the shopping-experience layer. x402 owns the machine-to-machine/crypto lane.
• The pragmatic split: if your customers arrive via ChatGPT, ACP touches you first. If they arrive via Google Search/Gemini, UCP does. The card networks' frameworks (TAP, Agent Pay) run underneath both.
• What could change: FIDO standardization could merge AP2 and Verifiable Intent into the single identity/consent layer everything else plugs into. Watch that working group more than any press release.

What this means for merchants and SMBs

Almost everything written about agentic payments targets banks and payment networks. Here's the part for people who actually sell things.

Agents are already shopping. The question is whether they can find you.

Stripe's agentic-commerce team describes agents as "brutally honest shoppers": they don't see your banner, your pop-up, or your brand story. They parse structured product data and complete checkout in under ~800ms — or they leave. Three practical priorities, in order:

1. Discoverability (do this now)

• Structured product data: complete schema.org markup, accurate machine-readable descriptions that directly answer questions ("waterproof to 2m" beats "adventure-ready").
• If you're on Shopify, Etsy, or a major platform: UCP support is arriving through your platform — check what's enabled before building anything.
• Test how ChatGPT and Gemini describe your products today. That's your new storefront.

2. Checkout friction (this quarter)

• Every extra step loses agent traffic entirely — an agent doesn't "persevere" through a clunky flow, it routes to a competitor.
• If you use Stripe: turn on their agentic commerce features and Link. If you're on card rails: ask your processor about tokenization readiness for Agent Pay / Intelligent Commerce.

3. Policy decisions (this year)

• Will you treat agent traffic differently from human traffic? (Rate limits, pricing, offers?)
• How do loyalty and offers survive when an agent sits between you and your customer? Google's Direct Offers pilot (discounts surfaced in AI Mode at the moment of purchase intent) is an early answer.
• Fraud posture: agentic commerce hasn't raised fraud rates so far, but "Know Your Agent" checks will become table stakes. Ask vendors how they verify agent identity.

The B2B side is bigger than checkout

Consumer shopping gets the headlines, but the near-term money is in B2B flows: agents that monitor inventory and trigger procurement, chase unpaid invoices and collect payment, manage subscription renewals, select optimal payment rails per transaction, and reconcile automatically. If your business sends invoices and waits to get paid, agentic payments are relevant to you this year — not in 2030.

Risks and open questions (the honest section)

• Liability is still settling. Mandates create audit trails, but no new network dispute rules exist yet for "my agent bought the wrong thing." Same old chargeback rules, new actor.
• Misaligned optimization. An agent optimizing for "cheapest" may exploit coupon loopholes or pick options you'd never choose. Constraint design matters more than model choice.
• Fraud arms race. If good bots can pay, bad bots will try. Enumeration attacks and synthetic agents are already appearing; agent identity verification (FIDO's focus) is the countermeasure to watch.
• Trust adoption gap. Surveys consistently show wide gaps between AI curiosity and comfort with autonomous spending (~2% of consumers comfortable with fully autonomous financial decisions). Growth will be delegation-with-guardrails, not full autonomy.

FAQ

Sources & further reading

• Google Agentic Commerce — AP2 specification and reference implementation (github.com/google-agentic-commerce/AP2)
• FIDO Alliance — Agentic Authentication & Payments Working Groups
• Juniper Research, Agentic Commerce Market 2026–2031
• Stripe Sessions 2026 — Agentic Commerce Suite, Link Agent Wallet
• Mastercard — Agent Pay framework & Verifiable Intent
• Coinbase — x402 protocol documentation