Agentic commerce: how AI agents are transforming online shopping and payments

Online shopping is entering a new phase. Consumers are beginning to move from using AI to search for products towards allowing it to compare options, build baskets and, increasingly, act on their behalf.

This is agentic commerce: a model in which AI agents can perform parts of the shopping journey autonomously, based on a customer’s goals, preferences and predefined limits.

The shift is already visible. McKinsey research published in 2026, based on a December 2025 survey across France, Germany and the United Kingdom, found that 38% of respondents use AI tools to help discover or decide which brands, products or services to purchase. Among shopping-related uses, comparing brands, models, prices and reviews is the most common, cited by 63% of respondents.¹

What makes this transition significant is that it changes the nature of the buying journey itself. Rather than moving step-by-step through search, selection and checkout, customers can increasingly express an intent and delegate parts of the process to software that acts within defined parameters. In this sense, commerce is beginning to move from traditional search-to-checkout journeys towards intent-to-execution experiences.

As these capabilities extend from discovery to execution, AI shopping agents could reshape not only how products are found, but also how transactions are authorised, authenticated and paid for.

What is agentic commerce?

Traditional e-commerce still requires the customer to perform most of the journey: searching, opening websites, comparing products, selecting an item, choosing delivery and completing checkout.

Generative AI has already shortened some of those steps by answering questions and recommending products. Agentic AI commerce goes further because the technology can take actions rather than simply provide information.

A customer could, for example, ask an agent to:

• find a product matching specific requirements and budget;
• compare price, availability, reviews and delivery conditions;
• shortlist the best options;
• assemble a basket;
• select an appropriate payment method;
• complete the transaction once the required level of authorisation has been provided.

The level of autonomy will vary. Some shoppers may want an agent only to recommend products, while others may authorise purchases within predefined parameters.

This distinction matters. McKinsey’s European research shows that AI usage in the shopping journey is currently strongest during discovery and comparison and declines as consumers move closer to transaction execution, such as basket building and checkout.¹ The transition towards autonomous commerce is therefore likely to happen progressively rather than all at once.

From customer journeys to delegated journeys

Traditional digital commerce follows a familiar sequence: search, browse, compare, select, checkout and pay.

Agentic commerce introduces a different model centred on delegation. Instead of manually navigating every step, a consumer can express an objective, set parameters such as preferences or spending limits, and allow an AI agent to perform part of the process.

Today, AI is primarily helping consumers discover and evaluate products. As capabilities and consumer trust develop, agents can take on more of the journey: assembling baskets, preparing transactions for approval and, in appropriate use cases, executing purchases within predefined boundaries.

McKinsey describes agentic commerce as an automation curve, with different levels of delegation rather than a single leap to full autonomy. It also highlights that delegation is likely to advance faster in utility-driven, repetitive and relatively low-regret purchases than in categories where human judgement remains an important part of the experience.²

For merchants, this means preparing not only for AI-assisted shopping, but for a future where software increasingly becomes an active participant in commercial transactions.

AI shopping agents are creating a new path to purchase

For merchants, one of the biggest changes is where purchasing decisions are made.

In traditional e-commerce, businesses optimise websites, search rankings, advertising and checkout experiences to influence human visitors. With AI agents in ecommerce, an increasing part of product discovery and evaluation can happen before a customer reaches the merchant’s website – or without them visiting it at all.

The potential scale is significant. McKinsey estimates that, even under moderate adoption scenarios, AI agents could mediate $3 trillion to $5 trillion of global consumer commerce in goods by 2030.²

Behaviour is already changing. Adobe reported that traffic from generative AI tools to US retail websites during the 2025 holiday season increased by 693.4% year on year. AI-referred shoppers also converted 31% more than visitors from other traffic sources during that period.³

While these figures relate to the US, there is a reason they offer an interesting early signal for other markets: McKinsey describes the United States as currently setting the pace in the deployment of full-funnel agentic commerce execution, while autonomous commerce remains at an earlier stage in Europe.⁴ They therefore provide an early indication of how AI-mediated discovery can translate into commercial intent.

For merchants, visibility therefore increasingly means being understandable not only to people and search engines, but also to AI shopping agents.

Product information, pricing, availability, delivery terms and policies need to be accurate, structured and accessible enough for an agent to retrieve and evaluate them.

Agentic payments: when AI reaches the checkout

The most significant change comes when the agent moves from recommending a purchase to executing it.

Agentic payments enable an AI agent to initiate or complete a payment on behalf of a consumer or business within agreed parameters.

This raises questions that a conventional checkout does not have to solve in the same way. Payment systems need to establish:

• who authorised the agent and what it is allowed to do;
• what the customer intended to purchase and under which conditions;
• which payment credentials the agent can use;
• what spending or transaction limits apply;
• how authentication and fraud controls should work when software is acting for a human;
• how the transaction can be traced and audited if something goes wrong.

Perhaps the most important difference is that executing the payment is only part of the challenge. Agentic payments also need to prove delegation: who authorised the agent, what it was authorised to do and whether the resulting transaction remained within those boundaries.

This is already reflected in emerging industry frameworks. Google’s Agent Payments Protocol (AP2), for example, uses cryptographically signed mandates to provide evidence that an agent has been authorised to complete a checkout and make the related payment.⁵ Mastercard and Google’s Verifiable Intent framework similarly aims to create a tamper-resistant record linking user authorisation, instructions and the resulting agent action.⁶

Why European markets may evolve differently

Agentic commerce is unlikely to develop identically in every market.

Within the European Union, Strong Customer Authentication remains an important part of the regulatory framework for many electronic payments under PSD2, subject to the exemptions provided by the rules.⁹ Agentic payment models therefore need to accommodate existing authentication and authorisation requirements rather than simply assume that human authentication disappears from the transaction.

Early implementations indicate that agent-driven journeys can be designed around these requirements. In June 2026, Worldline and ING announced a live end-to-end agentic payment in the Netherlands with Mastercard, using established authentication and authorisation mechanisms and requiring explicit consumer approval.¹⁰ In July, Worldline and ING completed another live agent-driven transaction with Visa that was explicitly designed to operate within existing regulatory requirements, including Strong Customer Authentication.¹¹

These examples do not establish a single model for European agentic commerce, but they show that delegated commerce and regulated authentication do not have to be mutually exclusive. Agentic journeys can combine AI-led discovery and transaction preparation with trusted mechanisms for validating customer intent when required.

The UK is following its own regulatory path. HM Treasury’s 2026 consultation explicitly recognises that current UK payment rules were developed before agentic AI and is examining whether areas including authentication, consent and liability need to adapt.⁷

For merchants operating internationally, this suggests that agentic commerce may evolve through different regulatory and technical models rather than a single universal journey.

What does agentic commerce mean for merchants?

Becoming ready for agentic commerce is not simply about adding an AI assistant to an online store. It requires looking at the commerce infrastructure behind the customer experience.

1. Make products machine-readable

Agents need reliable access to structured information about products, prices, availability, delivery, returns and promotions. Data quality therefore increasingly becomes part of customer acquisition.

2. Make payments accessible beyond the traditional checkout

If purchasing increasingly happens through AI interfaces, the payment experience cannot depend exclusively on a customer navigating a web checkout. APIs, tokenised credentials and embedded payment capabilities become increasingly important.

3. Preserve payment flexibility

An agent could eventually evaluate not just what to buy but how to pay. Merchants therefore benefit from payment infrastructures able to manage different methods and routes – from cards and wallets to account-to-account payments – without creating separate integrations for every new channel.

4. Build trust into every autonomous action

Delegation requires clear consent, controls and accountability.

Payments will need to combine convenience with verifiable customer intent, security and the ability to determine exactly what an agent was authorised to do. As agentic payments evolve, the ability to link identity, authority and transaction outcome is likely to become as important as executing the payment itself.

From checkout optimisation to commerce orchestration

The rise of agentic commerce does not mean the traditional checkout will disappear overnight. Human-led, AI-assisted and increasingly autonomous journeys are likely to coexist.

Then, what changes is the role of the payment layer?

Until now, payment optimisation has largely focused on making the final stage of a merchant-owned journey faster and more reliable. In an agentic environment, payment infrastructure increasingly needs to support transactions originating from multiple AI interfaces, identify the authority behind them, validate the relevant permissions and determine the most appropriate way to execute them.

This reflects a broader shift towards trust, identity and orchestration as important components of the payment experience.

It also makes agentic payments closely connected to payment orchestration: the ability to manage different payment methods, providers, authentication mechanisms and transaction routes through a flexible infrastructure.

The objective remains familiar – making it easy for a legitimate customer to complete a purchase securely. What changes is that the customer interacting with the payment infrastructure may increasingly be represented by an AI agent.

Fabrick Payment Orchestra helps businesses accept and manage end-to-end payments through automated, centralised and integrated management of the payment lifecycle.

The next phase of digital commerce

Agentic commerce is still developing, and fully autonomous purchasing remains less common than AI-assisted discovery and comparison. But the direction is becoming increasingly clear: commerce is beginning to evolve from entirely customer-executed journeys towards delegated journeys, in which software plays a growing role in decision-making and transaction execution.

Protocols are emerging, regulators are examining how delegated transactions should work, and consumers are becoming increasingly accustomed to using AI during the shopping journey.

For merchants, success will depend not only on creating excellent customer experiences, but also on becoming accessible, understandable and trustworthy to AI agents. Product data, payment capabilities, authorisation mechanisms and, increasingly, post-purchase processes will need to work across both human and machine-driven interactions.

In that environment, the winners will not simply have the best checkout. They will have adaptable commerce infrastructure capable of orchestrating payments, supporting trust and handling transactions wherever the buying decision ultimately takes place.

As AI becomes a new interface for commerce, flexibility, interoperability and control will increasingly shape the ability of merchants and payment providers to adapt.