Agents Go Shopping: The Rise of AI-Powered Commerce and the Unresolved Question of Accountability

The New Era of Shopping

Consider the transformation e-commerce has undergone in the last two decades. From physical stores to online shopping, from desktop computers to mobile devices, from individual product searches to personalized recommendations. Each transition has fundamentally changed consumer behavior. Now we are on the cusp of the next major transition: an era where AI agents shop for you.

This isn't just a vision of the future. By October 2025, the world's largest payment companies are actively building the infrastructure that will enable AI agents to search for products, compare prices, and complete purchases on behalf of consumers. Visa and Mastercard announced their separate frameworks. OpenAI launched direct purchase functionality within ChatGPT. Google released its own protocol. PayPal, Stripe, American Express, and dozens of other companies joined this ecosystem.

This trend, dubbed "agency commerce," represents the shift of artificial intelligence from providing information to taking action. It's no longer enough to simply ask a chatbot, "What are the best wireless headphones?" Soon you'll be able to say, "My budget is 10,000 TL, find and buy the best price-performance headphones." And the AI agent will do it for you.

Visa's Trusted Agent Protocol

In October 2025, Visa, along with more than ten partners, introduced the Trusted Agent Protocol. This open framework was built on top of existing web infrastructure and designed to help merchants distinguish legitimate AI agents from malicious bots.

The protocol's primary function is to allow authorized agents to securely transmit payment credentials and other data to merchants. This reduces friction in the payment process and prevents legitimate agents from being identified as bots and blocked.

Cloudflare played a critical role in the development of the protocol. "Securing the future of commerce is a shared responsibility, especially as AI agents begin to act on behalf of consumers," said Stephanie Cohen, the company's Chief Strategy Officer. "Our work with Visa on the Trusted Agent Protocol is a vital step in building the necessary safeguards for this new ecosystem."

The protocol is currently live on Visa's Developer Center and GitHub. Microsoft, Nuvei, Shopify, Stripe, and Worldpay are among the partners supporting the protocol. Akamai also recently joined the protocol, providing integration of edge-based behavioral intelligence, user recognition, and bot protection.

Visa announced it has successfully completed hundreds of secure, agent-initiated transactions with its partners. The company predicts that millions of consumers will be shopping with AI agents by the 2026 holiday season.

"We're seeing impressive progress in how AI will transform commerce," said Rubail Birwadker, Visa's Senior Vice President of Products and Solutions. "This holiday season marks the end of an era. By 2026, AI agents won't just be helping you shop, they'll be completing your purchases."

Mastercard's Agent Pay System

Mastercard announced its Agent Pay system in April 2025. This system allows AI agents to make purchases on behalf of users and is built upon the company's existing payment infrastructure: contactless payments, registered cards, and tokenization.

Agent Pay's key feature is that it requires agents to be registered and verified before they can make transactions. Users can set specific rules for which purchases are authorized. The platform also supports authentication tools, including on-device biometric verification.

Mastercard's Chief Digital Officer, Pablo Fourez, summarized the key questions facing merchants as follows: "With the proliferation of agent AI, merchants are increasingly facing critical dilemmas. How will they distinguish legitimate AI agents from malicious bots? How will they know if the consumer has authorized the agent to make a purchase? How will they know if the agent is correctly following the consumer's instructions?"

Agent Pay is designed to answer these questions. The system replaces payment information with digital tokens for security. Consumers can define the parameters by which the AI agent is allowed to make purchases.

Fiserv recently partnered with both Visa and Mastercard, gaining access to their agency commerce tools. This partnership makes Fiserv one of the first major processors to adopt Mastercard's Agent Pay Acceptance Framework.

OpenAI's Instant Checkout Feature

OpenAI launched the Instant Checkout feature for ChatGPT on September 29, 2025. This feature is built on the Agentic Commerce Protocol, developed in collaboration with Stripe.

ChatGPT Plus, Pro, and free users in the US can now shop directly from Etsy sellers without leaving the chat. Over one million Shopify sellers, including brands like Glossier, SKIMS, Spanx, and Vuori, will be added soon. Salesforce customers also joined this integration in October 2025.

The system works as follows: When a user asks a shopping-related question (for example, "best running shoes under $100" or "gift suggestions for someone who loves ceramics"), ChatGPT shows the most relevant products from across the web. For products with Instant Checkout enabled, users can make purchases directly from within ChatGPT.

"Our vision for ChatGPT is not just to provide information, but to help get things done in the real world," said Michelle Fradin, OpenAI's Head of Commerce Products. "A large portion of users' questions are somehow related to shopping and commerce."

An important point: Sellers remain registered sellers. Orders, payments, and deliveries are managed by the seller's existing systems. ChatGPT only acts as an AI agent for the user, securely transferring information between the user and the seller.

Sellers pay a small fee for completed purchases, but the service is free for users and does not affect prices. Instant Checkout items are not preferred in product results. When ranking multiple sellers offering the same product, ChatGPT considers factors such as stock availability, price, quality, and whether Instant Checkout is enabled.

Google's Agent Payments Protocol

Around the same time, Google launched the Agent Payments Protocol (AP2). This open standard allows AI agents to conduct secure transactions even without direct shopper intervention.

AP2's supporters include Mastercard, PayPal, American Express, Coinbase, Salesforce, Shopify, Cloudflare, and Etsy. Klarna later joined the protocol, stating that its role "reflects the evolution from a payment provider to a key infrastructure contributor for the next generation of commerce."

AP2 standardizes authorization parameters (such as price limit, vendor preference, and returnability) and creates a secure and traceable record of cryptographically signed "Mandates". This provides a technical framework to prove what the user authorized in the event of a dispute.

Market Dynamics and Growth Data

The rise of agency commerce is fueled by profound shifts in consumer behavior. According to Visa research, 47% of US shoppers are currently using AI tools for at least one shopping task, including finding gifts, comparing prices, and receiving personalized recommendations.

Adobe Insights data reveals an even more striking picture. During the 2024 holiday season, productive AI traffic to US retail sites increased by 1,300% (between November 1 and December 31, year-on-year). On Cyber Monday alone, traffic rose by 1,950%. By July 2025, AI-driven visits had increased by 4,700% year-on-year.

According to KPMG's Q1 2025 AI Pulse Survey, 99% of executives surveyed said their organizations plan to deploy AI agents. Sixty-five percent indicated they were piloting AI agents in Q1 2025, an increase from 37% in the previous quarter.

According to Salesforce's Connected Shoppers Report, 66% of shoppers are interested in using AI agents to secure high-demand items before they sell out. Sixty-five percent want agents who make purchases when the products reach the target price.

Use Cases

Payment managers say one of the early use cases will be flight and vacation bookings. For example, a user might say to the AI shopping agent: "Find the cheapest overnight flight from Singapore to Tokyo for under $500, nonstop." The agent will search, present options, book the ticket, and process the payment using the user's registered payment credentials. All of this will happen within a chat interface.

Mastercard executives said the technology could allow shoppers to authorize agents to make purchases even when they are offline. For example, automatically purchasing a product when its price falls below a predetermined threshold.

Amazon is also testing its "Buy For Me" feature. This feature uses an AI agent to purchase products not available on Amazon from external sellers, while keeping customers within the Amazon ecosystem. However, Amazon is also trying to prevent external AI agents from crawling its website. This suggests that large sellers are concerned about price pressure and losing direct access to customers.

In November, Perplexity partnered with PayPal to begin offering a free agency shopping product to US users.

Security and Verification Mechanisms

One of the main focuses of payment companies has been creating "agent tokens." These tokens use cryptographic authentication to distinguish authorized AI agents from human users and differentiate them from malicious bots.

Visa plans to add "payment signals" that provide banks with more transaction details and strengthen agent authentication with behavioral intelligence.

At the heart of Visa and Mastercard's agentic commerce systems lies the idea of tokenization. This involves creating a new 16-digit number sequence or cryptographically protected token linked to the consumer's original card. These tokens essentially act as a credit card for AI agents, linked to the consumer's original card.

Prove launched its "Verified Agent" solution in October 2025. This solution creates an "end-to-end custody chain" linking verified identity, human intent, payment credentials, and approval, all backed by cryptographic proof.

Unresolved Responsibility Question

Despite the early momentum, agentic commerce still harbors many potential structural, security, and accountability concerns. And perhaps the most critical question is: who is responsible when AI agents make mistakes?

Traditional dispute resolution involves four parties: the consumer, the card-issuing bank, the merchant bank, and the retailer. This chain becomes more complex when agents get involved.

Mike Lozanoff, JPMorgan Chase’s global head of merchant services, summarized the situation at the Money 20/20 conference: “Could an agent hallucinate and buy something we didn’t say? If that happens, it’s unclear who will fix the problem.”

The Early Warning Services manager gave a concrete example: A consumer tells an AI agent to order a navy blue t-shirt, but receives a red one. They call the bank, but the bank has no record of that particular consumer ordering a t-shirt because the agent made the purchase. Then they call the store, but the store has no record either. "All of these are operational headaches."

Industry experts predict that the responsibility will most likely fall on the sellers. Jamie George, Ravelin’s Vice President of Account Management and Partnerships, put the situation plainly: “Card schemes won’t take responsibility. Neither will the customers, the card-issuing banks, or the AI model that isn’t yet generating revenue. That leaves very few options. It will be the merchant.”

George also made a prediction: "I'd bet fraud in agentic commerce will increase. I'll give everyone good odds." Nobody on the panel accepted the bet.

Consumer Rights and the Current Legal Framework

Current laws provide certain protections. The Truth in Lending Act and Regulation Z (for credit cards), and the Electronic Funds Transfer Act and Regulation E (for debit cards and other electronic transfers) limit consumer liability for unauthorized transactions. Financial institutions are required to reimburse consumers for unauthorized transactions exceeding modest legal caps. These caps are typically up to $50 for credit cards and $500 for debit cards.

Card network rules also provide cardholders with "zero liability" protection against unauthorized transactions, subject to certain conditions. As a result, card-issuing banks generally bear the upfront cost of the refund to the cardholder. However, under network rules, unless the merchant can provide sufficient evidence that the transaction was properly authorized and verified, they can shift this loss to the merchant's merchant bank (and ultimately to the merchant) through the refund claim process.

One of the questions that arises now is: When an AI agent initiates a payment transaction, how can the merchant, not the cardholder personally, prove proper authorization and verification?

Possible Solutions and Future Scenarios

Some AI platforms could offer taking responsibility for errors as a competitive advantage. This could become a premium feature: "If our agent makes a mistake, we'll cover your return." This transforms responsibility from a legal requirement into a product feature, creating differentiation opportunities between platforms.

Three main protocols (Visa's Trusted Agent Protocol, Mastercard's Agent Pay, and Google's AP2) shift the dispute resolution model from post-transaction resolution to pre-transaction proof of authorization. However, it remains unclear whether this is sufficient to properly link the intent of a transaction to a properly informed consumer.

Basis Theory recently secured $33 million in Series B funding to expand its processor-independent chassis and leads the Agentic Commerce Consortium, where more than 20 companies design interoperable standards for "trust."

Strategic Assessments for Sellers

Agency-based commerce presents both opportunities and threats for sellers. On the one hand, AI agents are opening up a new sales channel. ChatGPT's more than 700 million weekly active users demonstrate the size of its potential customer pool.

On the other hand, AI agents could potentially create price pressure and reduce sellers' direct access to customers. When agents scan multiple sources simultaneously to find optimal prices and options, this revolutionizes price discovery. However, it can squeeze margins for sellers.

Areas where sellers need to prepare include: adopting agent verification systems, developing their own bots for negotiation, creating new loyalty programs, and developing cross-selling strategies to compete with efficient AI agents.

Conclusion: Rules Being Written

E-commerce took 10-15 years to develop mature dispute resolution frameworks. Agency commerce players are trying to build equivalent infrastructure in 1-2 years as transaction volumes increase.

The 2025 holiday season will be the first meaningful test of agent-led payments at scale. Visa says, "2025 will be the last year consumers shop and pay alone."

The infrastructure is being built. Transaction volume is increasing. However, the rules that will apply when problems arise are still being written.

This is a critical period for the payments industry, AI companies, merchants, and regulators. The decisions made and standards established will shape the future of commerce. And the key to that future lies in building trust: consumers trusting AI agents, merchants trusting these agents, and everyone trusting that a fair solution will be found if something goes wrong.

The era of agentic commerce is beginning. Whether we are ready is another question.

First published: LinkedIn, 2025-12-29 23:49. Original article: https://www.linkedin.com/pulse/ajanlar-al%C4%B1%C5%9Fveri%C5%9Fe-%C3%A7%C4%B1k%C4%B1yor-yapay-zeka-destekli-ticaretin-melih-%C3%B6zhan-gxmef