Insights

Agentic AI Commerce: The Next Era of Retail Shopping

Adnan Masood, PhD.Chief AI Architect. UST

Introduction: From Chat to Checkout

Imagine a customer chatting with an AI assistant about a product recommendation – and then completing the purchase right in that conversation. This is the promise of agentic AI commerce, where AI agents don’t just suggest what to buy, they help you buy it. More than 900 million people now use ChatGPT every week, relying on it for everyday tasks such as researching products, planning purchases, writing content, and getting work done, signaling that AI‑native discovery and decision‑making have already reached global scale. Now, AI platforms are evolving to handle transactions as well. In late 2025, OpenAI launched Instant Checkout in ChatGPT, allowing U.S. users to buy items from select merchants directly within a chat – the first step toward true AI-driven shopping. As an AI thought leader at a digital transformation consultancy, I’ll explain what agentic commerce means for retailers and how you can prepare for this next era of shopping.

What is Agentic AI Commerce? It’s a new model of online shopping where AI agents act on behalf of users to discover products, make decisions, and even execute purchases. In contrast to traditional e-commerce (where a human clicks through webpages and checkout forms), an agentic AI commerce system lets the user interact naturally (via text or voice) while the AI handles the formalities of the transaction. The AI agent becomes a digital personal shopper, orchestrating search, recommendation, and payment in a seamless flow.

DIVIDER

Why Now? The Rise of AI Shopping Agents

Several trends have converged to make agentic commerce both necessary and feasible. First, consumer behavior is increasingly shifting to conversational and multimodal interfaces. Shoppers are embracing voice, chat, and visual search at scale – chatbots now handle up to 70% of online customer conversations, and the conversational commerce market is forecasted to reach $32.6 B by 2035. Customers expect to ask an AI assistant for the “best running shoes under $100” and get instant, relevant results, possibly even images and videos, in the same interface. This multimodal, AI-driven discovery blends text, voice, and images into one experience, far beyond the old search-bar paradigm.

Secondly, AI capabilities have leapt forward. Modern large language models can maintain context over a shopping journey, handle complex queries, and perform tool-like actions (thanks to APIs and plugins) to carry out user instructions. This means the AI can manage a shopping workflow end-to-end – from finding a product to filling out checkout details – something not possible just a few years ago.

Finally, traditional e-commerce flows don’t translate to AI agents. For decades, online payments assumed a human is filling forms or clicking a checkout button. But an AI agent has no browser window and no fingers to click checkout. In fact, the familiar model of redirecting to a hosted payment page simply doesn’t work when an autonomous agent is completing the purchase. This gap in technology is driving the creation of new standards and protocols tailor-made for AI-driven transactions. In short, if retailers want to tap into AI as a sales channel, new approaches are needed to let machines transact with other machines securely and seamlessly.

DIVIDER

How Agentic Commerce Works: From Conversation to Purchase

When a user engages an AI shopping agent (whether in a chat interface, a voice assistant, or another app), the experience feels conversational and intuitive. For example, in ChatGPT’s Instant Checkout, the workflow goes something like this:

1. Product discovery: The user asks a question or says what they’re looking for. “I need a gift for a ceramics lover,” for instance. The AI then presents a curated list of relevant products from across the web, ranked by relevance – these results are organic and not ads. The user can ask follow-up questions, refine preferences, or request more details, all in natural language.
2. Choosing and configuring an item: Once the user decides on a product, the AI can provide options like size, color, or other variants if applicable, by querying the merchant’s catalog in real time. This is akin to a guided sales associate experience. The user makes their selections via chat (e.g. “I’ll take the blue one in size M”).
3. Instant checkout: If the merchant supports agentic checkout, the AI displays a “Buy” button or prompt next to the item. The user can then confirm the order, shipping address, and payment details right within the chat interface. There’s no need to open a browser or app; the AI securely handles it. At this point, the AI agent is essentially front-ending the transaction: it passes the order and customer info to the retailer’s systems through a secure API call, in a standardized format that the merchant understands. The payment itself is processed by the merchant’s payment provider (for example, Stripe or another gateway the merchant already uses) – not by the AI platform. The AI acts as an intermediary, relaying the necessary details (item, quantity, shipping info, payment token, etc.) to complete the purchase on the user’s behalf.
4. Confirmation and fulfillment: The merchant’s backend either accepts the order (authorizing the payment, generating an order number) or declines it (if something is wrong, like out-of-stock or payment failure). This response goes back to the AI agent, which then notifies the user in the chat with a confirmation (or an error message/instructions if there’s an issue). Once confirmed, the merchant fulfills the order just as they would any e-commerce purchase, and they remain the merchant of record – meaning they handle shipping, returns, customer service, and retain the customer relationship just as if the order came through their own site.

Throughout this process, user consent and trust are paramount. The AI agent doesn’t go rogue and buy things on its own – it explicitly asks the user to confirm each step (like “Do you want to buy this now?”, “Shipping to your address on file, OK?”) before proceeding. All payment credentials are handled securely; for instance, ChatGPT uses encrypted payment tokens that authorize only a specific charge amount for that specific merchant, with the user’s permission. And only the minimum necessary data is shared with the merchant to complete the order (e.g. shipping address, but not the full chat history). In other words, the system is designed so that users stay in control and transactions remain transparent and secure.

DIVIDER

Under the Hood: New Protocols Enabling AI Commerce

To make the above scenario possible, a whole new technical infrastructure is being developed by industry leaders. Traditional APIs weren’t built with autonomous AI agents in mind – so companies are co-creating open protocols to fill the gap. Let’s look at the major ones gaining traction:

OpenAI & Stripe: The Agentic Commerce Protocol (ACP)

The Agentic Commerce Protocol (ACP) is an open standard developed jointly by OpenAI and Stripe as the backbone of ChatGPT’s shopping feature. ACP essentially gives AI agents a universal “language” to securely transact with merchants’ systems. It defines how an AI can initiate a checkout using the merchant’s existing e-commerce and payment infrastructure. Importantly, the retailer stays in control: the merchant remains the record owner of the transaction, and payments flow through their usual processors (like Stripe) as they always have.

In practice, ACP provides a secure way for an agent to share the needed credentials and order info without exposing sensitive data. For example, the AI might send a token representing the user’s saved credit card, rather than the raw card number, when creating the order. ACP supports various types of transactions – physical goods, digital goods, even subscriptions and delayed (asynchronous) purchases. It also lets merchants enforce their own rules (e.g. an internal approval step if an order is above a certain value or flagged by fraud checks) so that agent-initiated orders still comply with the merchant’s risk policies.

Crucially, ACP is open-sourced under an Apache 2.0 license. OpenAI and Stripe have made the specs publicly available so any merchant, AI platform, or payment provider can implement it. While Stripe was the first to support ACP (not surprising, since Stripe co-developed it), the goal is for ACP to be platform-agnostic and extensible across the industry. In fact, OpenAI’s move to open-source ACP suggests a strategy to make it a universal standard for AI commerce. As of launch, ACP is live in production within ChatGPT’s Instant Checkout experience, and OpenAI has opened an application process for more merchants to join.

Google & Partners: The Agent Payments Protocol (AP2)

Almost in parallel, Google and a consortium of partners introduced the Agent Payments Protocol (AP2) in 2025 as a complementary approach to agentic commerce. While ACP handles the mechanics of checkout and integration, AP2 focuses on trust, authorization, and accountability in AI-driven payments. It establishes a framework for how agents, users, and payment providers prove consent and authenticity of transactions.

A cornerstone of AP2 is the concept of mandates – essentially digitally signed instructions that specify what an agent is allowed to do. For example, a user might sign a mandate saying “My agent can spend up to $500 to book flights and hotels for me”. These mandates are cryptographically verifiable and can be checked by merchants or payment networks as proof that the AI’s request truly reflects the user’s intent. They are also revocable and can carry fine-grained rules (time limits, specific merchants, price caps, etc.), giving users and businesses a robust control mechanism over autonomous purchasing.

AP2’s design makes these mandates portable across platforms and payment types. A delegated spending approval could work for credit cards, bank transfers, or even digital assets and stablecoins in the same standardized way. The protocol emphasizes auditability – every agent-initiated transaction can produce an evidence trail (the signed mandates) to resolve the questions of “who authorized this?” and “did the agent do exactly what was approved?”. This helps address concerns of fraud and error in a future where software agents might transact frequently and quickly.

Google has open-sourced AP2 under Apache 2.0 and rallied a broad coalition of more than 60 initial partners, including major players like Mastercard, PayPal, Adyen, Coinbase, Shopify, and others. Such broad support indicates that many in the industry see a common standard for agent trust as critical. For retailers and payment providers, AP2 offers a way to establish verifiable trust in agent-driven sales across ecosystems – potentially simplifying compliance and liability, since everyone speaks the same “agent trust” language. While still in early stages of adoption, AP2 could become the backbone for how AI agents prove authorization in any commerce scenario, from simple retail purchases to complex multi-party workflows.

(An example use-case: You could tell a shopping bot, “Buy me that limited-edition item whenever it drops, up to $200.” You’d cryptographically sign that intent. Weeks later, when the item is in stock, the bot can automatically complete the purchase under $200. The merchant and payment processor can verify the bot had a valid mandate. This way, even if you’re asleep, your agent can grab the deal – but it cannot overspend or deviate from your instructions.)

Coinbase & Web3: The x402 Protocol

A third initiative, coming from the crypto/Web3 world, is x402, developed by Coinbase. This protocol tackles a different aspect of the puzzle: it revives the long-dormant HTTP status code 402 Payment Required as a mechanism for machine-to-machine payments on the web. While ACP and AP2 address end-to-end commerce and trust, x402 is laser-focused on enabling microtransactions and pay-per-use services in a simple, open way.

Here’s how x402 works: Suppose an AI (or any client) wants to access a resource – say an API endpoint or a piece of content – that isn’t free. Instead of the typical approach (signing up for an account or obtaining an API key and then being billed later), the client just requests the resource directly. The server can respond with an HTTP 402 status, which includes a standardized payload describing how much to pay and where (for instance, “send 0.0005 ETH or $0.10 in USDC to this address to access this resource”). The AI agent can then handle the payment on the fly (e.g. executing a cryptocurrency transaction or using some digital payment rail), and once done, retry the request. If payment is detected, the server grants access (HTTP 200). All of this happens programmatically within seconds, without human intervention or long-term subscriptions.

The beauty of x402 is its minimalism and openness: it doesn’t depend on any single platform or currency. It’s chain-agnostic, meaning it could work with various blockchain networks or even traditional payment networks, and it avoids proprietary gateways. This approach is ideal for microtransactions – for example, an AI that needs to pull data from a premium API could pay per request, or a content site could charge a few cents per article automatically, rather than forcing a monthly subscription. For retailers, x402 might not apply to typical physical goods sales, but it could enable new models (like paying tiny amounts for access to inventory data, IoT device interactions, etc.) especially as IoT and AI services interact. Coinbase’s implementation uses stablecoins (like USDC) for simplicity, and the whole specification and SDKs are public on GitHub for anyone to integrate.

Different Approaches, One Ecosystem

It’s worth noting that ACP, AP2, and x402 aren’t mutually exclusive or truly competing – they address different layers of the emerging agentic commerce stack. ACP is at the checkout integration layer, focusing on how the AI agent plugs into merchants’ existing sales systems. AP2 provides a trust and authorization layer that could underlie many types of agent transactions across platforms. x402 is at the payment execution layer, streamlining the act of transferring value in an open-web way. An enterprise might well end up using all three: for instance, enabling shopping bots via ACP, using AP2 for internal governance and compliance, and x402 for certain micro-payment use cases. The key takeaway is that industry leaders are actively building the infrastructure to make AI-to-business commerce secure and scalable. These standards may quietly shape the future of retail long before they’re directly visible to consumers.

DIVIDER

Opportunities and Imperatives for Retailers

For retailers and brands, agentic AI commerce represents both a huge opportunity and a strategic imperative. AI agents could soon become a significant new sales channel, akin to how mobile apps emerged a decade ago. Here’s what this shift means and how you can capitalize on it:

• Reach Customers Through AI Channels: AI assistants like ChatGPT (and potentially Google’s Bard or others) are becoming shopping destinations in their own right, exposing products to millions of users. For example, ChatGPT’s Instant Checkout opens access to hundreds of millions of weekly users – a new way for sellers to reach customers without relying on them visiting a website or marketplace. Retailers who integrate early get first-mover advantage in these AI-driven discovery experiences.
• Preserve Control of Your Brand and Data: A valid concern in any new channel is loss of customer relationship. Agentic commerce protocols are designed so that merchants remain the merchant of record, even when the AI facilitates the sale. You still process payments through your gateway, you still handle fulfillment, and you still own the customer interaction post-sale (for support, marketing, etc.). In other words, you gain the convenience of an AI storefront without ceding control of pricing or customer data to a middleman. This is far better for brands than, say, turning over the transaction entirely to a third-party marketplace.
• Seamless Integration (Low Friction): Adopting these new protocols can be surprisingly lightweight. If you’re already using Stripe for payments, enabling agentic payments via ACP can take as little as one line of code. Even if you use a different payment processor, you can still participate by using tools like Stripe’s shared payment token API or implementing the open Delegated Payments spec – all without overhauling your existing backend. In short, becoming “agent-ready” might not require a major tech project, especially if you leverage pre-built libraries and modules from the open-source community.
• Enhanced Visibility in AI Recommendations: AI shopping agents aim to optimize user experience, which includes offering quick, hassle-free purchases. When multiple merchants carry the same product, ChatGPT’s ranking algorithm may consider factors like price, availability, and whether a merchant supports Instant Checkout (i.e. is agent-integrated) to decide which option to present first. This means merchants who implement agentic checkout could be favored in competitive scenarios, since they offer a smoother path for the user. Being agent-friendly could literally make your products more discoverable in AI-driven searches. As one industry observer put it, “If you are not agent-ready, you are not discoverable” in the coming era of AI commerce.
• New Customer Experiences & Conversions: Agentic AI can reduce friction in the buying journey, potentially boosting conversion rates. Consider a user casually chatting about a product – if they can purchase it the moment interest peaks, without jumping through hoops, you’re less likely to lose that sale. Early stats from AI-driven personalization and chat experiences show dramatically higher conversions (e.g. 4× higher conversion from proactive AI chat in retail tests). By embracing conversational purchases, retailers can capture “impulse buys” or routine re-orders effortlessly. Moreover, you can serve customers who never even visit your website – their AI agent finds and buys your product from within a chat or voice interaction elsewhere. It’s an expansion of omni-channel into “AI-channel.”
• Future-Proofing and Innovation: Just as mobile-responsive sites and one-click payments became standard, agentic commerce capabilities may soon be expected by consumers (or by their AI assistants). Forward-looking retailers are already experimenting: 89% of retailers are testing or deploying AI in some form, and the AI in e-commerce market is projected to grow from about $9 B in 2025 to over $64 B by 2034. Embracing these protocols early allows you to help shape standards and ensure compatibility. It also sends a message that your brand is innovating in customer experience. On the flip side, ignoring this trend could leave you scrambling later to integrate with dominant AI platforms on their terms. The risk of waiting is that agentic purchasing becomes embedded in consumer behavior (through ChatGPT, Google, Alexa, etc.), and latecomers might find themselves invisible in those ecosystems.

How to get started? Begin with a strategic evaluation. Identify where an AI-driven purchase flow could intersect with your current systems. If you’re on a modern e-commerce platform, check if plugins or updates are available for ACP or AP2 integration. Engage with your payment processor about support for agent-initiated payments – many are already involved (Stripe, Adyen, PayPal and others are in the conversation). It’s wise to run a sandbox trial: perhaps enable agentic checkout for a small subset of products or a test audience, and see how it performs. This gives you insight into technical adjustments needed and any policy considerations (e.g. setting spending limits for agents, updating fraud checks for automated orders). Crucially, involve your digital strategy and AI teams (or partners) – implementing agentic commerce is not just an IT upgrade, but part of a broader AI adoption roadmap for your organization.

At my company, we’ve been deeply involved in this shift – building accelerators, tools, and best practices for agentic commerce. For example, our teams have prototyped integrations with ACP and AP2, so we can rapidly connect a retailer’s catalog and order system to ChatGPT or other AI platforms. We understand the nuances of preserving customer data rights, maintaining brand voice in AI interactions, and aligning these new channels with existing CRM and fulfillment workflows. In short, we can help you become “agent-ready” smoothly, so you can focus on strategy and customer experience.

DIVIDER

Conclusion: Embracing the Agentic Future

We stand at the dawn of a new era in commerce. Just as e-commerce transformed retail 20+ years ago, and mobile commerce a decade ago, AI-driven commerce is now emerging as the next evolution. The difference this time is that the interface isn’t a screen or an app – it’s an intelligent agent that talks and acts for the customer. Retailers who succeed in this era will be those who meet customers wherever their digital preferences lie, including within AI assistants. As AI becomes a key interface for how people discover, decide, and buy, technologies like the Agentic Commerce Protocol will form the connective tissue between people and businesses in the next economy.

For retail leaders, the opportunity is to create shopping experiences that are more seamless and personalized than ever before – available on demand in any context the customer needs. It’s shopping that feels like a conversation with a helpful expert, backed by the efficiency of automation. By taking steps now to integrate with agentic AI commerce, you can capitalize on new revenue streams, improve customer convenience, and stay ahead of the curve in a fast-changing digital landscape. Just as importantly, you’ll signal to your customers (and their AI assistants) that your brand is open for business in the channels of the future.

The agentic commerce revolution is just starting to unfold. With open standards and collaborative innovation, this is no longer a distant sci-fi concept but a present reality—one that retailers can participate in today. UST helps retailers design, build, and scale agentic AI commerce end-to-end, not as a science experiment but as a production-ready business capability. We bring deep expertise across AI agents, commerce platforms, payments, data, cloud, and security, paired with proven accelerators, reference architectures, and reusable libraries to move quickly from concept to live deployment. From integrating agentic checkout protocols into existing e-commerce and payment stacks, to orchestrating multimodal shopping agents across chat, voice, and apps, to ensuring governance, trust, and compliance at scale, UST applies the systems thinking required to make agentic commerce real. Our teams understand both the underlying protocols and the operational realities of retail, enabling brands to become agent-ready while preserving control over customer relationships, data, and fulfillment—turning AI from a novelty into a durable growth engine. The shopping agent is here; UST helps you put it to work for your business. See AI-powered retail in action.