Agentic Commerce

What Changes When AI Can Act on Buyers' Behalf?

By Nune Igityan, AI Market Strategist and Founder, Kensai · Published 5 October 2026 · Last reviewed 5 October 2026

In Brief

AI is moving from helping people navigate buying decisions towards performing parts of the transaction itself. Agentic commerce refers to commerce in which AI can research, compare and complete transactions on behalf of consumers or businesses, with varying levels of human involvement.

• AI already influences discovery and selection. McKinsey found that 38% of consumers surveyed across France, Germany and the UK use AI to research products or services or decide what to purchase.
• New systems can increasingly browse, interact with websites and complete approved transactions.
• Retailers and commerce platforms are developing different approaches to AI-mediated shopping, including Target, Walmart and Shopify, while Amazon has restricted access by Meta's Muse.
• Payments, permission and cybersecurity are becoming part of the infrastructure required for AI to transact.
• If AI increasingly controls discovery, selection and transaction, control over the decision layer could affect distribution, customer relationships and economic value.
• Capability is developing faster than consumer delegation. How much purchasing becomes genuinely autonomous, and what it is worth, remains to be established.

1. What Is Agentic Commerce?

AI has influenced commerce for years through search, recommendations and increasingly generative AI. Agentic commerce introduces a further capability: an AI system can carry a commercial decision into the transaction itself.

An agent might find suitable options, compare them against defined requirements, select an option within those conditions, interact with a merchant and complete a purchase using an authorised payment method.

The defining shift is therefore from helping someone make a decision to executing the transaction on their behalf.

Autonomous purchasing is a more advanced form of this capability, where an agent can search, select and transact with relatively limited intervention. That remains an emerging form of commerce rather than an established consumer norm.

2. From Recommending to Transacting

The first stage of AI's impact on buying decisions has largely been cognitive. AI can interpret information, compare alternatives and recommend what a person might choose.

The next stage is operational. Agents can increasingly perform tasks that previously required the buyer to navigate websites, forms, payment interfaces and customer-service systems themselves.

This extends the market shift examined in How AI Is Reshaping Market Dynamics, which focused on discovery, interpretation, evaluation and selection. Agentic commerce follows the decision into action and transaction.

A person might ask an agent to find and purchase a suitable product within a defined budget, arrange a booking according to specific preferences or complete a recurring purchase under predetermined conditions. The person establishes the objective and permissions; the system carries out the work.

Current evidence does not show widespread autonomous purchasing. McKinsey's 38% figure measures AI use in research and purchase decisions (McKinsey, 2026); BCG's research measures AI's influence on which brands enter consideration (BCG, 2026). Neither measures completed autonomous transactions.

What AI can technically do, whether people use it, how much authority they give it, and how much commerce ultimately flows through it are separate measurements.

3. How an Agent Participates in a Transaction

An agentic transaction can involve several stages:

• Intent: A person establishes what they want.
• Discovery: The agent searches across available products, services or merchants.
• Evaluation: It assesses alternatives against the person's requirements and preferences.
• Selection: It identifies an option that fits those conditions.
• Permission: The person authorises the agent to proceed, either for a specific transaction or within predefined limits.
• Transaction: The agent interacts with the merchant and payment infrastructure to complete the purchase.
• Fulfilment: The agent may track delivery, manage changes or initiate returns.

The boundary between assistance and agency becomes clearest at the transaction stage. An AI system can recommend a flight without being able to book it. It can identify a product without being authorised to purchase it.

4. Infrastructure for Agentic Commerce

An agent that can transact requires more than a capable model. It needs access to accurate commercial information, systems through which it can interact with merchants, mechanisms for establishing identity and permission, payment credentials and ways to intervene when something goes wrong.

This also extends the interpretability problem identified in that earlier analysis of market dynamics. As AI moves from selecting to transacting, agents need reliable commercial information they can interpret: products, prices, availability, policies, permissions and other conditions of sale. The transaction layer therefore depends on commercial information being structured clearly enough for systems to act on it.

Meta's Muse illustrates this direction. The personal AI agent can browse websites and fill forms, with purchases requiring user approval (Meta, 2026). Stripe has developed infrastructure allowing Muse to use Link across more than one million businesses that accept it, while a single-use virtual card can support purchases elsewhere (Stripe, 2026).

Payment networks are developing related infrastructure. Visa's Intelligent Commerce work addresses credentials, controls, authentication and agent identity (Visa, 2026a). Mastercard's Agent Pay initiative focuses on tokenised credentials, agent identity and mechanisms for establishing what an agent has been authorised to do (Mastercard, 2025).

The infrastructure is therefore developing around the agent as an intermediary between human intent and execution.

5. How Retailers and Platforms Are Responding

Retailers and commerce platforms are beginning to take different approaches to AI-mediated shopping.

Target has opened elements of its shopping experience to AI interfaces including Google, Microsoft and OpenAI, allowing customers to discover products, build baskets and purchase through conversational environments while retaining control over the transaction (Target, 2026).

Walmart has been developing agentic shopping capabilities alongside partnerships with major AI platforms (Walmart, 2026). Shopify has introduced Agentic Storefronts and, with Google, co-developed the Universal Commerce Protocol (UCP), an open standard for agentic commerce (Shopify, 2026). These approaches connect merchants to AI-driven commerce channels across discovery, checkout and post-purchase activity.

Amazon has taken a more restrictive approach toward external shopping agents. In September 2026, Amazon blocked Meta's Muse from accessing Amazon.com after saying Meta had not obtained permission for the agent to access the site and that third-party applications should respect a provider's decision about whether to participate. Meta said Muse does not see users' passwords or payment details (Meta, 2026; Bishop, 2026). The episode illustrates that access to merchant systems is not automatically guaranteed by the technical ability to browse or transact.

These different approaches reveal an emerging strategic tension. AI agents could become another route through which demand reaches merchants. They could also become an intermediary controlling discovery, selection and parts of the customer relationship.

6. How Agents Are Authorised to Pay

Payment is the clearest boundary between an AI assistant and an agent that can transact. Crossing it takes more than access to a card number. The person has to give the agent authority that is specific enough to complete one purchase and narrow enough that it cannot be used for anything else.

The emerging answer is to give the agent a separate credential rather than the person's own card details. The payment layer issues a substitute that carries the permission with it, recording what was approved and within what limits. Stripe's approach shows the pattern. Where a business already accepts Link, an approved purchase is completed using the person's stored payment method. Elsewhere, Link issues a single-use virtual card for that purchase alone, so the agent never holds credentials it can reuse (Stripe, 2026).

Visa and Mastercard are building the same idea into the card networks. Both are extending tokenisation so that a payment presented by an agent can be identified as agent-initiated, linked to an instruction the cardholder actually authorised, and traced back to a specific agent (Visa, 2026a; Mastercard, 2025). Authentication is also moving earlier. The person confirms with a passkey or a biometric check at the point of approval, and the agent completes the payment later on its own.

This has a consequence beyond security. If the credential is what proves the purchase was authorised, the networks that issue it are well placed to become the layer where an agent's right to act is established, and where disputes about it are settled.

The underlying challenge is to give an agent enough authority to complete an approved transaction while keeping that authority limited, revocable and visible to the person.

7. Security, Permission and Consumer Trust

Once an agent can act, security extends beyond payment credentials.

An agent may interact with accounts, addresses, purchase histories, loyalty programmes, personal preferences, websites and APIs containing sensitive information. Compromised credentials, malicious instructions, manipulated content, prompt injection or unintended actions could therefore have consequences beyond a single transaction.

The emerging security architecture needs to address identity, authentication, scoped permissions, credential isolation, secure execution, monitoring, auditability and human intervention.

There is also a fundamental trade-off around context. The more useful an agent becomes, the more information and authority it may require. That can improve its ability to act on a person's behalf while increasing the consequences of misuse or compromise.

Trust remains a constraint. Visa's 2026 U.S. survey found that 72% of consumers had used an AI assistant, while only 23% trusted GenAI to handle payment transactions on their behalf (Visa, 2026b). In a broader Visa study across the U.S., Australia and New Zealand, around 85% of respondents considered control over their data important or very important, while nearly nine in ten wanted transparency into decisions made by agents (Visa, 2026a).

8. What Changes for Discovery, Selection and Distribution?

Agentic commerce extends the market shift examined in that earlier analysis of market dynamics.

AI is already influencing which products and companies enter consideration. Adobe reported that AI-referred traffic to U.S. retail sites grew 62% year-on-year in July 2026 and was up 1,219% since Adobe began tracking it in October 2024 (Adobe, 2026b). The quality of that traffic has also changed: in March 2026, AI-referred visitors converted 42% better than non-AI traffic, while revenue per visit was 37% higher (Adobe, 2026c). By July, the conversion advantage had increased to 60% (Adobe, 2026b), while May data showed AI-referred visits generated 53% more revenue per visit (Adobe, 2026a).

As AI moves closer to transaction, being discoverable becomes increasingly important to being selectable and ultimately purchasable.

This changes distribution. If an AI system generates the shortlist, evaluates alternatives and can then purchase on the customer's behalf, the merchant may compete for inclusion in an AI-mediated decision environment rather than relying solely on its own website or traditional acquisition channels.

The customer relationship could also become less straightforward. A retailer may receive the order while having less visibility into why the customer selected it, which alternatives were considered or which information influenced the decision.

First-party data and direct customer relationships could therefore become more valuable even as external AI systems become important sources of demand.

9. Market Structure, Intermediaries and Economic Value

Digital commerce has traditionally brought several functions together: the interface, customer behaviour, product presentation, transaction and fulfilment.

Agentic commerce could separate them. McKinsey estimates that agentic commerce could orchestrate $3 trillion to $5 trillion globally by 2030, though that is a projection rather than observed activity (McKinsey, 2026).

An external AI agent may control discovery and selection. A retailer may control inventory and fulfilment. A payment network may control transaction infrastructure. An AI platform may hold information about customer preferences and purchasing history.

This raises the possibility that a retailer could receive demand without controlling the interface through which that demand was generated, and could become more dependent on external systems that allocate customer attention and purchasing decisions.

Retailers may respond by building their own agents, strengthening direct relationships or controlling more of the customer experience. The different approaches being taken by Target, Walmart, Shopify and Amazon suggest that the boundary between retailers and external agents is already becoming a strategic issue.

The economic implications follow from who controls access to demand. Customer acquisition, conversion, repeat purchase and lifetime value are partly shaped by who controls the decision environment. If an external agent becomes the primary route through which customers discover and select products, the distribution of economic value across merchants, platforms, infrastructure providers and buyers is likely to change.

For a consumer-facing agent, defensibility could emerge in the relationship with the user. A preferred agent may accumulate knowledge of preferences, purchasing history, decision criteria and a record of reliable execution, and trust would be part of what accumulates. This is a hypothesis rather than an established moat. Interoperability may reduce switching friction at the infrastructure layer, while accumulated context may increase it at the relationship layer. Monetising that position through merchants, recommendations or transaction economics could also create conflicts with the user's interests and weaken the trust the advantage rests on.

No monetisation model is yet established, whether through transaction fees, merchant-funded placement, subscriptions or adjacent services.

Open standards could produce the opposite outcome. If agents can move across protocols and merchants can expose their systems through common standards, parts of the transaction layer may become low-margin infrastructure, and value could depend less on owning the pipes than on controlling demand or the user relationship. Whether value accrues to interoperable infrastructure, to agents controlling demand, or to agents users find hard to replace remains open.

Agent-to-agent interaction could extend this further. A buyer's agent might eventually negotiate with a retailer or supplier agent over price, availability, delivery, loyalty or terms. This remains an emerging direction rather than an established feature of commerce at scale.

10. What We Know — and What We Don't Yet Know

We know that:

• AI already influences discovery, evaluation and selection.
• AI-referred traffic to commerce sites is growing.
• Agents can increasingly browse, interact with websites and perform commercial tasks.
• Payment, identity and permission infrastructure is being developed specifically for agents.
• Major retailers, platforms and payment companies are responding to AI-mediated commerce.

We do not yet know:

• How many consumers will routinely authorise AI to purchase on their behalf.
• How much purchasing will become genuinely autonomous.
• What share of commerce will flow through agents.
• Whether consumers will prefer retailer-owned, platform-owned or independent agents.
• How much customer data retailers will retain when agents mediate transactions.
• How bargaining power and margins will evolve between agents, platforms, merchants and payment networks.
• Whether agent-to-agent commerce becomes commercially significant.

The central uncertainty is therefore not technical capability alone. Adoption, delegation, transaction volume and economic impact remain separate questions.

11. Where This Leaves Markets

Agentic commerce extends AI's role in markets from influencing decisions towards executing them.

AI has already entered discovery, interpretation, evaluation and selection. The emerging development is its movement into action: interacting with merchants, negotiating deals, initiating transactions and potentially managing parts of fulfilment on a person's behalf.

That creates new questions around distribution, customer relationships, payment infrastructure, security and control. It also raises a broader question about where economic value sits when the system controlling the decision is separate from the company fulfilling the transaction.

The infrastructure is developing. Consumer behaviour, market power and economic value are still being determined.

AI first influenced the decision. It is now beginning to participate in the transaction.

Key Definitions

Agentic commerce commerce in which AI can research, compare and complete transactions on behalf of consumers or businesses, with varying levels of human involvement.

Autonomous purchasing a more advanced form of this capability, where an agent can search, select and transact with relatively limited intervention.

Permission the authority a person grants an agent to proceed, either for a specific transaction or within predefined limits.

Agentic transaction a transaction in which an agent interacts with the merchant and payment infrastructure to complete the purchase.

Questions This Article Answers

What is agentic commerce?

An AI system researches, compares and completes a transaction on a person's or company's behalf, rather than only recommending what to buy.

Why are payments important?

An AI system that can transact needs secure, controlled access to payment infrastructure and credentials.

Why do trust and cybersecurity matter?

Agents may access sensitive information and commercial systems. Greater context and authority can make an agent more useful while increasing the consequences of misuse or compromise.

How are retailers responding?

Retailers and platforms including Target, Walmart and Shopify are developing ways to participate in AI-mediated commerce, while Amazon has restricted access by external shopping agents such as Meta's Muse.

How could agentic commerce change distribution?

AI agents could increasingly control discovery and selection before a customer reaches a retailer, changing where demand is allocated.

Could agentic commerce change market power?

Possibly. If AI platforms become significant decision layers between consumers and companies, control over demand allocation could become an important source of economic value. Open standards could also work in the opposite direction, making agent access to merchants widely available and reducing the returns to controlling any single interface. Which effect dominates is not yet established.

Is autonomous purchasing already widespread?

Current evidence supports growing technical capability and early commercial deployment, but widespread consumer delegation and large-scale autonomous purchasing remain less established.

References

Adobe (2026a) AI Traffic Trends Report: Q3 2026 AI-Sourced Traffic Insights. Adobe Digital Insights. Available at: https://business.adobe.com/resources/sdk/q3-ai-traffic-trends-report/q3-2026-ai-sourced-traffic-insights.pdf

Adobe (2026b) 'U.S. consumers are embracing LLMs to make travel plans, but many brands have AI visibility gaps', Adobe Digital Insights, 19 August. Available at: https://business.adobe.com/blog/us-consumers-are-embracing-llms-to-make-travel-plans

Adobe (2026c) 'Adobe report: U.S. retailers see surge in AI traffic, but many websites are not entirely readable by machines', Adobe Digital Insights, April. Available at: https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable

BCG (2026) Five Consumer Shifts Reshaping Growth. Boston Consulting Group. Available at: https://www.bcg.com/publications/2026/five-consumer-shifts-reshaping-growth

Bishop, T. (2026) 'Amazon blocks Meta's Muse AI assistant in new standoff over agentic shopping', GeekWire, 20 September. Available at: https://www.geekwire.com/2026/amazon-blocks-metas-muse-ai-assistant-in-new-standoff-over-agentic-shopping/

Mastercard (2025) 'Mastercard unveils Agent Pay, pioneering agentic payments technology to power commerce in the age of AI', Mastercard, April. Available at: https://www.mastercard.com/au/en/news-and-trends/press/2025/april/mastercard-unveils-agent-pay-pioneering-agentic-payments-technology-to-power-commerce-in-the-age-of-ai.html

McKinsey (2026) 'Europe's agentic commerce moment: Decision influence is here; execution is coming', McKinsey & Company, 2 March. Available at: https://www.mckinsey.com/industries/retail/our-insights/europes-agentic-commerce-moment-decision-influence-is-here-execution-is-coming

Meta (2026) 'Introducing Muse: The World's First Personal AI Agent Built for Everyone', Meta Newsroom, 8 September. Available at: https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/

Shopify (2026) 'The agentic commerce platform: Shopify connects any merchant to every AI conversation', Shopify, 11 January. Available at: https://www.shopify.com/news/ai-commerce-at-scale

Stripe (2026) 'Stripe helps Muse, Meta's new personal AI agent, shop across the internet with Link', Stripe Newsroom, September. Available at: https://stripe.com/newsroom/news/stripe-helps-meta-muse-shop-with-link

Target (2026) 'How Target is Shaping Shopping Experiences in Conversational AI', Target Corporation, June. Available at: https://corporate.target.com/press/fact-sheet/2026/06/conversational-ai

Visa (2026a) Earning Consumer Trust in the Age of Agentic Commerce. Visa. Available at: https://corporate.visa.com/en/products/intelligent-commerce/earning-trust-report.html

Visa (2026b) 'Visa Trust Index: trust powers agentic commerce', Visa, 9 September. Available at: https://corporate.visa.com/en/sites/visa-perspectives/innovation/visa-trust-index-trust-powers-agentic-commerce.html

Walmart (2026) 'Walmart and Google turn AI discovery into effortless shopping experiences', Walmart, 11 January. Available at: https://corporate.walmart.com/news/2026/01/11/walmart-and-google-turn-ai-discovery-into-effortless-shopping-experiences

About the Author

Nune Igityan is an AI Market Strategist and Founder of Kensai. Her research examines how artificial intelligence is restructuring markets, economic value and competitive advantage, with a focus on AI infrastructure, agentic AI, AI-mediated markets and agentic commerce. She previously held senior product marketing and growth roles at Google and Meta.

How to Cite This Article

Igityan, N. (2026) Agentic Commerce: What Changes When AI Can Act on Your Behalf? Kensai. Available at: https://www.kensai.uk/research/agentic-commerce