Agentic commerce has a trust problem, and American Express just put a fix on the table. On October 13, 2026, Amex introduced its "Intent Intelligence Framework," a way to capture and verify what a customer actually asked for before an AI agent pays. TechCrunch reported the launch after an interview with Amex's Hannah Bozian.

I am Nitin Kumar, Lead Project Coordinator at Debugged Pro. We build AI agents for clients in India and the USA, and payment questions come up in almost every scoping call. This guide covers what Amex announced, how the flow works, and what your team should build next.

Person typing on a laptop with an Agentic AI interface and icons floating above the keyboard — Agentic AI concept image used by American Express Technology for its ACE Developer Kit post. Image credit: American Express (americanexpress.io).

Key Takeaways

Intent comes first. Amex now captures the customer's purchase instructions and limits before the agent pays.
The customer signs off. They authenticate and approve the intent the agent recorded.
No real card number at checkout. The agent uses a secure digital credential tied to that approved intent.
Mismatches can be blocked. Ask for Nike sneakers, and an agent that tries to buy Adidas can be stopped.
It builds on ACE. The framework is part of the Agentic Commerce Experiences developer kit Amex launched in April.
Agencies should design for it now. Structured intent, approvals, and logs will soon be table stakes.

What Amex Announced on October 13

AI agents built on large language models do not always behave the same way twice. That makes a wrong purchase a real risk. Amex wants the customer's instructions to travel with the transaction from start to finish.

Amex calls this the Intent Intelligence Framework. The company told TechCrunch it hopes the approach becomes an industry standard for AI-powered transactions.

Hannah Bozian is Amex's VP of Agentic Partnerships and Strategy. "These new transactions will have a consumer, a merchant, and an agent," she said. Amex wants those transactions to stay "seamless, trusted, and secure, the way our card members expect them to be."

Why now, if disputes are not rising yet

Amex said it has not seen more disputes because of agentic commerce so far. It is building ahead of demand. Its own Trendex survey suggests that demand is coming.

66% of surveyed U.S. consumers said they are likely to use AI agents to research the best price for a product or service.
61% said they are likely to use agents for product or service recommendations.
Both figures cover the next 12 months, according to what Amex told TechCrunch.

How the Agentic Commerce Intent Flow Works

The flow is easier to follow as steps. Here is the order Amex described, in plain terms.

Create Intent API. The agent sends the customer's request and limits, such as merchant name, maximum budget, and product preferences.
Customer approval. The customer verifies their identity and signs off on the intent.
Intent ID and token. Amex creates a unique Intent ID plus a Proof of Intent Token that carries the high-level limits.
Scoped credential. The token is used to get a secure digital payment credential. The agent pays with that, not the real card number.
Checks after checkout. The same Intent ID and token can support fraud checks and dispute handling later.

Five-step agentic commerce flow from Create Intent API to customer approval, Intent ID, scoped credential, and dispute checks — How the Amex intent flow fits together. Credit: Debugged Pro original diagram based on details Amex shared with TechCrunch and on its ACE pages.

The sneaker example

Bozian gave a simple case. A customer asks for white Nike sneakers in size 9. The agent tries to buy shoes from Adidas instead. Because the approved intent says Nike, Amex could stop that transaction.

"Our approach differs from the implementations of others in that it pulls intent earlier into the transaction flow," she said. "And this is important because it helps mitigate some risk before the risk occurs."

Where it goes next: less hand-holding

Many agents ask for consent before every purchase today. Bozian expects some to act with much more freedom later. Think of an agent that reorders paper towels when you run low.

In that case, only price and size really matter. A machine check in the loop makes sense there. A log of what each agent was allowed to do, and when, also helps catch fraud carried out through an agent.

Inside the ACE Developer Kit

The new framework sits inside the Amex Agentic Commerce Experiences (ACE) developer kit. Amex launched ACE on April 14, 2026. It has five services.

ACE service	What it does
Agent Registration	Lets only verified AI agents transact on the Amex network
Account Enablement	Lets Card Members link their cards to verified agents
Intent Intelligence	Captures purchase intent for authentication, authorization, and disputes
Payment Credentials	Lets verified agents pay with tokenized credentials
Cart Context	Uses cart-level details to improve checks and dispute reviews

Amex's ACE page says the Account Enablement, Intent Intelligence, and Payment Credentials specs were available from April 14. Agent Registration and Cart Context were still listed as under development.

The security design, in brief

An April post on the American Express Technology blog explains the engineering. A few points stand out for developers.

Intent covers what to buy, where to buy it, how much to spend, and when extra approval is needed.
All API calls use OAuth and mutual TLS.
The system treats client input as untrusted. Each step uses short-lived signed artifacts with replay protection.
Partners use regular APIs plus event notifications for status changes.
Merchants keep processing payments on existing rails.

Amex lists Adyen, Cloudflare, Fiserv, Google, Microsoft, OpenAI, PayPal, and Stripe among the companies supporting its agentic commerce work. TechCrunch said Amex is in talks with partners across payments about putting ACE to use.

Agent Purchase Protection: Who Pays When an Agent Gets It Wrong

This is the part clients will ask about first. Amex has announced Agent Purchase Protection for agent mistakes. The ACE page describes it as a future benefit.

Here is how the published terms read today:

It applies to eligible U.S. credit cards issued by American Express.
The agent must be registered with Amex and use the ACE kit.
Amex must receive the customer's authenticated purchase intent from the agent.
An eligible error is a purchase that does not match that intent.
The customer must first try a return with the merchant, where possible.
Vague requests may not be covered. Amex gives "best" or "really nice" as examples.

That last point matters for product design. If your agent accepts fuzzy instructions, the protection may not help. Clear, structured intent is now a business requirement, not just good UX.

What Agentic Commerce Means for India and USA Agencies

Most agencies will not build payment networks. We will build the agents, apps, and storefronts that plug into them. That changes what clients expect from us.

For USA agency teams

Retail, travel, and dining clients will ask for "AI that books and buys." Scope intent capture from day one.
Merchants will care about fewer disputes. Show how your flow sends clean intent data upstream.
Legal teams will ask who is liable for agent errors. Point them to the network terms and keep your own logs.

For India agency and delivery teams

Many India teams build for U.S. clients. Learn the ACE flow now so you can deliver it when a client qualifies.
For domestic launches, India has its own card tokenization and payment rules from the Reserve Bank of India. Check them with your payment partner first.
Strong QA matters here. Test cases for "wrong brand," "over budget," and "expired intent" should be standard.

If you are comparing approaches, our breakdown of Google's AI shopping and commerce protocol covers how another big player is shaping agent checkout.

A Practical Agentic Commerce Build Checklist

You do not need ACE access to start. These steps make any shopping or booking agent safer, and they map well to what Amex is asking for.

Checklist for building an agentic commerce flow with structured intent, approval, scoped payment, matching checks, and audit logs — A suggested build checklist for agency teams. Credit: Debugged Pro original diagram, not an official Amex specification.

1. Turn requests into structured intent

Do not pass raw chat text to checkout. Convert it into fields: item, brand, size, merchant, budget cap, and expiry. Ask a follow-up question when a field is missing.

2. Show a clear approval screen

Display the exact limits before the customer approves. Use plain words like "Nike, white, size 9, up to $120." Avoid hiding limits behind a "confirm" button.

3. Never store raw card numbers in the agent

Use tokenized or scoped credentials from your payment provider. Keep secrets out of prompts, logs, and model memory.

4. Check the cart against the intent

Before paying, compare the cart with the approved fields. Block on any mismatch and ask the customer. This is the same idea Amex is pushing at the network level.

5. Log everything an auditor would ask for

Who approved the intent, and when
Which agent and model version acted
What the cart held at checkout
Why the agent made each choice, in short notes

6. Govern it with a known framework

The NIST AI Risk Management Framework gives a structure clients already trust. Map where the agent touches money, measure error and dispute rates, and set clear rules for when a human must step in.

Is Agentic Commerce Ready for Your Clients?

For low-risk, repeat purchases, it is getting close. Reorders, simple bookings, and price checks are good first projects. High-value or emotional purchases still need a human in the loop.

The bigger lesson is about design. Payment networks are starting to reward agents that capture intent clearly. Teams that build that habit now will have an easier time plugging into Amex, and likely others, later.

If you are planning a shopping or booking agent, our agentic AI development team can help you scope intent capture, approvals, and logging. For a wider review of your product and payment stack, talk to our software development consulting team.

FAQ: Agentic Commerce and the Amex Intent Framework

What is agentic commerce?

Agentic commerce is shopping where an AI agent finds, picks, and pays for a product or service on a person's behalf. The customer sets the goal and the limits, and the agent carries out the purchase.

What is the Amex Intent Intelligence Framework?

It is an American Express framework that captures a customer's purchase instructions before an AI agent pays. The customer approves that intent, and Amex issues an Intent ID and a Proof of Intent Token. The token is used to get a payment credential tied to those limits, so a mismatched purchase can be stopped.

Does the agent see my real card number?

According to Amex, no. The Proof of Intent Token is used to get a secure digital payment credential, and the agent uses that at checkout instead of the customer's actual card number.

What is Amex Agent Purchase Protection?

It is a planned protection for eligible U.S. Card Members. If a registered agent sends Amex the customer's authenticated intent and then buys something that does not match it, Amex says it will protect eligible customers from those charges. Vague requests like 'best' or 'really nice' may not qualify.

Should agencies in India and the USA build for agentic commerce now?

Yes, but start small. Add structured intent capture, clear approval screens, and logs to any shopping or booking agent you build. Then connect to network programs such as Amex ACE once your client qualifies and the specs fit the use case.

About the author: Nitin Kumar is Lead Project Coordinator at Debugged Pro. He works with India and USA agency teams on AI agent projects, client scoping, and safe rollouts of new payment and automation tools.