Agentic commerce is shifting ecommerce from keyword-led browsing toward AI-mediated intent, but merchants still need to own product data, customer relationships, and payment authorization. The winners will make their catalogs machine-readable while preserving clear consent, trusted checkout, and post-purchase control.
The strategic question is no longer whether AI will influence a purchase. It is whether your brand remains visible, trusted, and operationally in control when it does.
Online commerce is entering a new phase in which AI agents can research products, compare offers and initiates purchases on behalf of consumers. Increasingly, an AI agent does the searching, the comparing, and in some cases the buying, all inside a single conversation. According to Ivo Bozukov, this shift, known across the industry as agentic commerce, is moving from early experiment to genuine infrastructure far faster than most institutions anticipated.
The numbers involved are difficult to overstate. OpenAI reported in February 2026 that ChatGPT now processes roughly 50 million shopping-related queries every single day. Amazon’s AI assistant Rufus has reached 300 million users, embedding conversational discovery directly into the world’s largest online marketplace. McKinsey projects that agentic commerce will generate between $3 trillion and $5 trillion annually by 2030, a figure that would have sounded implausible only two years ago.
Consumer behaviour during peak shopping periods tells a similar story. During Black Friday 2025, Adobe Analytics recorded an extraordinary 805 per cent year-on-year surge in AI-driven traffic to retail websites. Ivo Bozukov argues that this spike was not a one-off holiday anomaly but an early signal of where consumer habits are heading.
“Black Friday used to be about who had the loudest advert,” Bozukov said. “Now it is increasingly about who an AI agent chooses to recommend, and that changes the entire competitive calculus for retailers.”
Traditional online shopping has always begun with a search term and a scroll through results. Agentic commerce inverts that model. A shopper states an intent, such as needing waterproof boots under a certain price that can arrive by a set date, and the agent takes over from there. It searches multiple retailers, weighs price against reviews and availability, and in some cases completes the purchase without further input.
This intent-driven approach represents a genuine departure from decades of e-commerce design built around search engine optimisation and product listings. Bozukov believes retailers that are slow to make their product data machine-readable risk losing visibility as AI agents increasingly influence purchasing decisions. Accurate pricing, availability, product specifications and delivery information will become increasingly important as agents assess products on behalf of consumers.
Behind the consumer-facing experience sits a fast-evolving set of technical standards, each competing to define how agents actually transact. The Agentic Commerce Protocol, developed jointly by Stripe and OpenAI, underpins product discovery within ChatGPT Shopping and was released as an open standard so other merchants and developers could build on it.
Google entered the field with its own approach, launching the Universal Commerce Protocol at the National Retail Federation’s Big Show in January 2026. Unlike narrower checkout-focused specifications, UCP was designed to cover the entire shopping journey, from initial discovery through to post-purchase support such as order tracking and returns. It launched with backing from more than twenty partners, including Walmart, Target, Visa and Mastercard, and has continued to expand its capabilities through several updates since launch.
Sitting alongside these commerce-specific protocols is Anthropic’s Model Context Protocol, which provides the underlying data connectivity that lets AI systems securely pull information from external tools and services. Ivo Bozukov describes MCP as the plumbing that makes the more visible commerce protocols function properly.
“ACP and UCP get the attention because they touch the transaction itself,” he explained. “But none of it works reliably without a solid connectivity layer feeding these agents accurate, real-time data.”
The major card networks have not been content to watch from the sidelines. Mastercard completed its first live, authenticated agentic transactions across Asia-Pacific markets in early 2026, beginning in Australia and New Zealand before expanding into Singapore, Malaysia, India, South Korea, Hong Kong and Thailand. Each transaction relied on tokenised credentials and explicit consumer consent, addressing one of the central concerns around letting software initiate payments without a human present at the point of sale.
Visa took a comparable path, moving its Trusted Agent Protocol into commercial availability after piloting the technology in a sandbox environment with more than one hundred partners. Ivo Bozukov sees this parallel movement from both networks as evidence that agentic payments have cleared the proof-of-concept stage entirely. “When two competing networks independently reach commercial launch within months of each other, that tells you the technology has matured well beyond pilot theatre,” Ivo said. “The industry is now building permanent rails, not running experiments.”
As agents gain the ability to initiate real payments, a new compliance question has emerged for payment providers: how does an institution verify that an AI agent acting on a customer’s behalf is authorised to do so, and within what limits? Emerging Know Your Agent frameworks are attempting to answer that question, establishing ways to authenticate an agent’s identity, confirm the scope of its permissions, and maintain an auditable record of what a customer actually approved.
This shift matters because it changes where liability and trust sit within a transaction. A payment initiated by software, rather than a person tapping a card, needs a different verification model, one that can confirm consent happened before the agent acted rather than relying solely on traditional cardholder authentication at the moment of purchase.
One notable development has been the move away from checkout happening entirely within the chat window. OpenAI scaled back its Instant Checkout feature in March 2026, shifting instead towards redirecting shoppers to merchant sites to complete their purchase. This adjustment preserves the direct relationship between retailer and customer, even as AI handles the discovery and recommendation stages of the journey.
Bozukov views this recalibration as a healthy correction rather than a retreat.
“Merchants spent years building loyalty programmes, checkout experiences and post-purchase relationships,” he noted. “Handing all of that to a chat interface was always going to create friction. Letting AI own discovery while merchants retain the transaction itself is a more sustainable division of labour.”
As agentic commerce infrastructure matures across established markets, the expectation is that these capabilities will extend into the GCC and other emerging regions, where mobile-first payment habits are already well established. For payment providers operating across multiple geographies, this represents a significant opportunity to build agentic capability into markets that have already shown a willingness to adopt fast, mobile-native payment methods.
Ivo Bozukov expects the coming year to determine which providers are positioned to capture that growth.
“The institutions moving early into Know Your Agent frameworks and protocol interoperability today will be the ones ready when agentic commerce reaches these newer markets,” Bozukov said. “This is not a niche experiment anymore. It is the direction the entire payments industry is heading.”
Agentic commerce is a model of online shopping in which an AI agent helps a customer research, compare, select, and sometimes purchase products based on stated goals and constraints. Instead of manually searching product pages and comparing tabs, a shopper can describe what they need, such as a budget, size, delivery deadline, preferred brand, or product feature. The agent then evaluates available offers and presents a recommendation or purchase path. The merchant still needs accurate product data, dependable fulfillment, secure payment controls, and a strong customer experience to convert that agent-assisted interest into a lasting relationship.
Agentic commerce differs from ecommerce search because an AI agent interprets customer intent and performs more of the comparison work that a shopper normally does manually. Traditional search starts with a keyword and returns a set of results for the customer to filter, evaluate, and purchase from. Agentic commerce starts with constraints and outcomes, then uses product data, pricing, availability, shipping information, reviews, and policies to narrow the options. For merchants, the shift means optimizing for reliable, machine-readable product and operational information, not only keyword relevance, ads, and visual merchandising.
The Agentic Commerce Protocol, often called ACP, is an open standard developed by OpenAI and Stripe to help AI agents, merchants, and payment systems coordinate commerce interactions. It is designed to support product discovery and purchase flows without forcing every merchant to build a custom integration for each AI shopping surface. ACP is only one part of the emerging stack. Merchants also need reliable catalog data, inventory services, payment authorization, fraud controls, fulfillment information, and post-purchase support. The protocol can standardize communication, but it cannot compensate for inaccurate pricing, weak operations, or unclear customer consent.
The Universal Commerce Protocol, or UCP, is an open protocol introduced by Google with Shopify and other industry partners to support agent-enabled commerce across discovery, checkout, and post-purchase activity. Google describes UCP as an open and agnostic standard that keeps the customer relationship central while allowing AI systems to work across commerce platforms. It is designed to be compatible with related standards, including Model Context Protocol and payment-focused agent protocols. For merchants, UCP matters because it points toward a future where product, transaction, and order information can move more consistently between AI agents, retailers, payment providers, and customer-service systems.
Shopify merchants should prepare for agentic commerce by improving product data accuracy, feed quality, inventory reliability, shipping promises, return clarity, and payment risk controls. Begin with the data agents need to evaluate your offer: product attributes, variants, availability, pricing, delivery timing, compatibility, and policies. Then review the customer journey after an AI recommendation, including product-page clarity, checkout speed, order confirmation, support, and post-purchase retention. Smaller brands should focus first on clean Shopify catalog operations and merchant feeds. Larger brands should align PIM, ERP, warehouse, regional inventory, and fraud systems so agents receive consistent real-time information.
