Agentic commerce: How to build agent-aware digital ecosystems for AI-powered retail

Insights

• Digital commerce is moving from visually-rich storefronts and manual checkouts toward more autonomous, intent-driven experiences.
• Agentic commerce protocols allow consumers to delegate discovery and purchase to AI agents, reducing friction for the user.
• Retailers must evolve their e-commerce to benefit from these shifts.

Agentic commerce in retail is gaining traction in the US and is expanding into Europe and Asia-Pacific. Artificial intelligence (AI) agents with a high degree of autonomy work with software and platforms to carry out defined tasks and achieve specific goals. They are able to work with generative AI platforms such as ChatGPT; Google’s universal commerce protocol - a common language for platforms, agents, and businesses compatible with the agent payments protocol ecosystem; and Perplexity to automate many of the digital commerce tasks that consumers traditionally perform.

These AI agents scan multiple e-commerce sites like a consumer would normally do, apply consumer preferences, and evaluate products, pricing, and availability to recommend purchase-ready options, without requiring customers to leave the chat session. It is estimated that nearly one in four Americans made a purchase through an AI-powered experience in November 2025, reflecting increasing consumer exposure to AI-driven commerce. However, not all retailers’ product feeds, real-time inventory, fulfillment options, content structures, loyalty programs, promotions, and fraud systems are optimized for AI agents. Retail businesses will need to design their e-commerce systems to be agent-ready across discovery, decisioning, and execution.

How agentic commerce works

In OpenAI's agentic commerce protocol (ACP), two distinct entities orchestrate every transaction using Instant Checkout: the buyer agent and the seller agent. Each serves a specific function, yet they work toward a collective goal: to support the consumer in finding and buying the desired product.

The buyer agent functions as the consumer's intelligent proxy, acting on the consumer’s behalf, rather than the consumer interacting directly with multiple retailer websites or applications. When a user makes a purchase request, the buyer agent does a search, queries multiple seller agents simultaneously, and compares prices, shipping speeds, return policies, and inventory availability.

Each query parses tokens — or complex data or text broken down into smaller units for easier pattern detection — processing product descriptions, product-related reviews, inventory data, and specifications through large language models (LLMs), creating computational costs that can scale with catalog complexity and the number of merchants queried. The buyer agent maintains conversation context, remembering that the user mentioned needing a particular product by a specific time, and prioritizes results accordingly.

Meanwhile, the retailer or the merchant can connect the seller agent that is part of the merchant's infrastructure to product catalogs, inventory availability and location allocations, positive reviews, loyalty membership, and checkout capabilities through application programming interfaces (APIs).

When the buyer agent makes an inquiry, the seller agent responds with structured data containing product specifications, real-time availability, personalized pricing based on the consumer's loyalty status, and payment options. The seller agent can also handle the actual transaction processing, working with payment service providers to tokenize payment credentials and complete purchases, without exposing sensitive card data to the AI system.

The handshake between these agents follows a typical sequence: discovery — where the buyer agent identifies compatible seller agents; query — involving structured requests for product and offer information; negotiation — during which options are compared and consumer preferences are applied; authentication, where the consumer’s identity is verified through OAuth-based flows; authorization — which confirms payment methods and spending limits, and finally, execution, where the transaction is completed and confirmation is sent to the buyer agent.

PayPal is extending its existing Checkout SDK to support transactions initiated by agents, positioning it as a default safe payments layer for agentic commerce. This capability is in the process of rolling out and is not yet available to all users. Agentic commerce has potential to generate nearly $1 trillion in US business-to-customer retail revenue by 2030, with global impact projected at between $3 trillion and $5 trillion.

Several retail brands are already charting their course toward ACP. In the US, OpenAI has rolled out Instant Checkout in ChatGPT, beginning with Etsy sellers, and plans to extend the capability to more than a million Shopify merchants. Retail giant Target, in partnership with OpenAI, recently launched a dedicated Target app inside ChatGPT. Walmart is a first mover, integrating voice-based shopping, real-time agentic search, store-level inventory availability, and connected services like list management and replenishment. Its collaboration with OpenAI illustrates what all large retailers could achieve with agent-driven commerce.

Beyond Target and Walmart, several other retailers are also experimenting in this space, including Best Buy with AI-powered guidance, Carrefour with agentic promotions and search, and Instacart with agent flows that move from ideas for recipes to purchasing ingredients for specific recipes or meal plans. This points to a broader, cross-industry shift rather than a trend limited to a single retailer. However, retailers can participate in ACP-enabled commerce only if their e-commerce systems meet certain foundational criteria.

Technology challenges

Retailers operating on legacy systems could struggle to keep pace with this shift, as most retailers’ e-commerce stacks are designed for human-led journeys and are not optimized for agent-led interactions. As a result, several critical components are poorly suited for AI agents that need structured, real-time, and machine-interpretable access.

Retailers’ product data is built to support images, banners, layouts, and marketing copy that look good on a website. Humans can infer meaning from visuals and descriptions even if details are implied or inconsistent, but LLMs struggle with vague product descriptions, while AI agents need structured, semantically-rich data that they can reliably interpret and compare across brands. The problem is amplified when data is fragmented across systems.

Brand voice and content are crafted for persuasion and storytelling, not for precise, machine-readable representations of value, constraints, and differentiators that agents require for decision-making. The content that convinces human shoppers to make a purchase means nothing to a LLM parsing specifications. For example, humans can gather meaning from phrases like “premium feel,” “built for performance,” or “great for everyday use.” But these phrases don’t have measurable definitions for an LLM that needs quantifiable features, certifications, and performance benchmark-based content to make a decision.

Data on inventory availability across stores and warehouses is frequently delayed, fragmented, or exposed only through internal systems, making near real-time validation difficult for agents operating in transactional contexts. For agents making instant decisions, even small delays can lead to inaccurate recommendations. Agents need real-time availability and not end-of-day batch updates to avoid recommending products that appear unavailable at checkout. Content architecture prioritizes page-based navigation and search engine optimization (SEO) for humans, rather than the API-first, schema-driven access needed for agentic discovery.

Fraud detection systems flag agent-initiated transactions as suspicious because they don't recognize the behavioral patterns, or the content delivery network (CDN) isn’t optimized. Most fraud systems rely on signals such as mouse movements, click patterns, session duration, and page navigation. AI agents, however, do not behave like humans — they don’t scroll, hesitate, or click. Their interactions are fast, deterministic, and API-driven, which often causes fraud models to misclassify legitimate agent-initiated transactions as automated abuse.

To address these issues, retailers need to build a foundation that makes their digital presence discoverable and transactional in the new agentic economy. Retailers require digital ecosystems that are agent aware, capable of interacting with third-party AI agents and machine readability, and structured in a way that AI systems can easily interpret the data.

The path forward

While agentic commerce protocols are still in their early stages, retailers need to act now: Preparing for this shift requires a deliberate, phased approach.

Assess readiness
The first step is to conduct an agentic engine optimization (AEO) and ACP readiness audit to identify gaps across critical areas such as product catalog data structures, inventory management, fraud detection models, CDN configuration, payment processing, privacy and consent enforcement, content quality, user reviews, loyalty programs and promotions, and API accessibility. Agentic commerce introduces fundamentally different technical, data, and trust requirements than traditional e-commerce. The goal is to identify every breakpoint that prevents AI agents from engaging smoothly. An upfront audit allows retailers to pinpoint where their existing stacks fall short of AI agents’ requirements, before committing to costly redesigning, integrations, or platform changes.

Shift from SEO to AEO
Retailers should redesign product content for generative comprehension using rich metadata, contextual storytelling, and machine-readable formats so that AI agents index products, interpret, reason, summarize, and recommend them. Content that is optimized only for human browsing or traditional SEO is often opaque, incomplete, or misleading for generative systems.

By contrast, AEO can help structure, enrich, and clarify content so AI agents can accurately interpret, reason over, and recommend it. Retailers need to clearly define attributes such as size, material, compatibility, and performance specifications, use consistent naming and standardized formats that explain what the product is and how it relates to other products, so agents can understand it. They also need to redesign their loyalty and promotion engines for agent queries — by replacing coupon codes or static offers such as buy one get one free, or a certain discount with rule-based benefits that include price tiers, free delivery, and bonus points — and expose real-time entitlements that an agent can reason over.

Enable machine-readable systems
AI agents can only act on information that is consistent, structured, and machine-interpretable. Retailers should audit for and build flexible, well-organized and fast-response APIs, and consistent data models with rich metadata, taxonomies, and ontologies so AI systems can easily access and use their information.

Build the right partnerships
Retailers cannot build agentic commerce capabilities in isolation. They should form strategic partnerships with technology providers, including AI platforms, payment networks, fraud specialists, and cloud infrastructure vendors, to develop AI-first commerce frameworks purpose-built for agent interactions. By working with the right partners, retailers gain access to shared standards, scalable infrastructure, and continuously improving AI models, reducing the cost and complexity of building these capabilities internally. The result is a more resilient, flexible, and future-ready e-commerce stack that can safely support autonomous agents while preserving trust, compliance, and customer choice.

The winners in the emerging ACP economy will not be the retailers with the best-looking websites. They will be the ones with agent-aware digital ecosystems, whose systems speak fluent AI, whose seller agents respond in milliseconds with complete and accurate data, and whose architectures anticipate questions before they are asked. As AI agents increasingly mediate discovery, comparison, and purchasing on behalf of consumers, retailers that are not agent-ready risk becoming invisible or unselectable in this new commerce flow.