Amex tries to 'grease the skids' for agentic commerce

• Key insight: Amex has introduced several products to help vet and execute agentic payments.
• What's at stake: Banks and payment companies are investing heavily in agentic AI and other forms of artificial intelligence. 
• Forward look: Amex will add similar services to its agentic commerce menu, including incentives. 

With payment companies quickly adding agentic artificial intelligence, attention is turning to building trust and making the technology easy to use.

American Express has introduced several agentic commerce tools, including a developer kit that enables technology to perform several steps, including verifying agents — a key part of the growing number of agentic protocols that aim to guide transactions and vet AI agents to ensure they are valid and perform authorized tasks.

At stake is a market that's growing quickly but with some trepidation. More than half of banks plan to boost their AI spending, according to research from American Banker. And 70% of banks say agentic AI will have a "significant or game-changing impact" on the financial services industry, according to research from Sound Hound and American Banker.

But at the same time, research from Visa and Morning Consult say consumers want some form of protection, such as approval and other guardrails, before using AI agents to make payments. 

"We want to offer clarity for those who are asking AI agents to make purchases on their behalf," Luke Gebb, executive vice president and global head of innovation at American Express, told American Banker. "It's still very early days for agentic commerce and this will help move that forward."

Amex's kit

Amex says it has completed "thousands" of agentic payments through tests with AI platform partners. The developers' kit is part of a strategy to bring AI-supported commerce to a broader audience as other major payment companies such as Visa, Mastercard, Block, PayPal and Stripe also embrace agentic commerce.

To support its developers' kit, Amex will rely on what it calls a "closed loop model" — the company is a card issuer, payment network and merchant acquirer — to provide data to support and vet agentic transactions for merchants and consumers. 

"Trust and security are critical to grease the wheels for this, to put the wind at the back of agent commerce," Gebb said.

The developers' kit includes agent registration and verification, account enablement for consumers to register their cards for agentic transactions, intent analysis, verification of payment credentials and context of the digital shopping cart as an added risk management measure. 

For consumers, the kit can enable controls from the Amex app, with plans to add benefits and rewards into agent-performed transactions, Gebb said. 

"It's stuff like 'if the cost of this product is below $300, buy it,'" he said. 

The kit is designed to reduce disputes and charge-backs by adding more assurance to agentic transactions, Gebb said. If there are errors involving the AI agent, Amex will cover that, Gebb said. "We're standing behind the agents," he said. 

Building standards

Other agentic standards, or efforts to vet AI agents before making purchases, include Google's Agent Payments Protocol, which the company released in September. It enables a framework for merchants and payment processors in agentic AI. Adyen, Alipay, Amex, Mastercard, PayPal and Worldpay contributed to the protocol.

Stripe and OpenAI in October launched Agentic Commerce Protocol, a standard for agentic AI-powered transactions; and Open Issuance, which enables businesses to launch and manage their own stablecoins. Coinbase and security firm CloudFlare have also built a protocol to create standards for digital agents and AI systems to make payments through a shared format.

"It's very important in building trust," Alenka Grealish, who co-leads generative AI research at Celent, told American Banker.

Grealish, who has written about the difficulty in building trust in new commerce technology, argues the potential of agentic commerce depends on success factors that lie well outside the domain of large language models.

"The current crux is standardizing messy e-commerce data, which is a prerequisite for building consumer trust," she wrote.

As agentic AI advances, agentic commerce is bumping into a classic challenge: messy data and state-level tax requirements, according to Grealish. Merchants' product information, including back order/pre-order, are often not standardized, according to Grealish, noting shipping options are inconsistent, including thresholds for free shipping.

For example, "OpenAI's decision to scale back embedded checkout in ChatGPT is an early indicator that trust deficits and data friction are already tempering agentic commerce ambitions," Grealish wrote. "Instead, it is now focusing on positioning ChatGPT as a discovery and orchestration layer, while the actual purchase happens inside merchant or partner apps that connect to ChatGPT."

An important part of agentic commerce is semantics and workflows, which enables AI agents to handle the entire shopping process, Grealish said. Among the companies building protocols, Google, for example, is building a broader framework whereas OpenAI and Stripe are focusing on agent checkout, according to Grealish. 

Outside of the U.S., UnionPay last week issued its Agentic Payment Open Protocol, which is designed to make it easier to support agentic payments between merchants, finance institutions and technology companies. The protocol guides agent registration, user identity and standards for authorization. It is also designed to work with other protocols and can be used in and outside of China.

In 2025, agentic commerce rode the tailwinds of large language model breakthroughs, Grealish wrote. "Today, it faces the headwinds of messy merchant data and building consumer trust. In 2026, stakeholders are focusing on establishing standards and scaling their adoption."