Agentic Commerce: More Than Just AI-Assisted Retail.
- 10Pearls Editorial Team
- 15 min read
Summary
In this blog, we are exploring agentic commerce. What it is, how it works, its examples in the market, including ChatGPT Instant Checkout that was rolled back, its benefits, challenges, protocols, and how businesses can prepare for agentic commerce.
Agentic AI gained significant traction in 2026, including the release of frontier models with increasingly native agentic capabilities that are evolving how businesses adopt and use AI in their operations. AI agents are already empowering businesses and individuals in unique and exciting ways and transforming how many of us do certain things, including online purchasing. Agentic commerce, which refers to involving an AI agent in the buying process, is still nascent. But smart business owners and market stakeholders are already preparing for it.
In this blog, we will discuss what agentic commerce is, where it’s heading, and what e-commerce businesses should know to capitalize on this emerging domain. We will also look into some agentic commerce initiatives, the agentic route they took, and the challenges they have yet to overcome.
What is agentic commerce?
Agentic commerce is when users (individuals and businesses) use AI agents for commercial activities like sales, purchases, and negotiations. This includes user-directed actions, automated purchases set up by the user, or giving agents the autonomy to conduct these activities while adhering to user-defined goals and constraints.
This broad definition of agentic commerce expands beyond the autonomy of AI to include human-guided, agent-facilitated actions. The precedent for agents making financial decisions already exists in trading, lending, and other related sectors, typically with regulatorily required human-in-the-loop. The “autonomy” evolution of consumer-oriented agentic commerce is something enterprises need to consider and plan for.
A good way to understand agentic commerce and its multiple dimensions is to separate the three levels of “agency” AI agents can have in agentic commerce (for now).
Agentic discovery
The AI agent searches for items to purchase on your behalf and, based on your requirements, compares them, compiles a list of needed items, and makes recommendations. You (the human) decide what to buy and handle the transaction. This is mostly what agentic commerce is at the time of writing this.
Agentic checkout
Agentic checkout adds to the previous layer. Instead of simply recommending the item, it can (either on its own or with your permission – based on the level of autonomy it has); an agentic AI system can complete the checkout/transaction part for the selected item on your behalf. This includes filling out forms either on the seller’s website/portal or within the interface that embeds this capability (more on this later), and it’s also where most engineering challenges associated with agentic commerce emerge.
Agentic autonomy
The agent makes the decision to buy something on your behalf, based on its interpretation of the buying goals you have set. They don’t take your approval for the purchase and complete the entire process – from discovery to transaction, on their own. It’s quite rare, and the closest example is Amazon’s AI-driven auto-buy feature when the price of a product drops.
Conversational commerce vs. Agentic commerce
Conversational commerce predates agentic by quite a bit. The earliest examples include chatbots that were rule-based, not powered by an LLM. They served pre-written answers based on certain rules and what they could infer from user messages.
In the age of LLMs, conversational commerce evolved. Since many frontier models have agentic capabilities, including accessing information directly from the internet and extracting data from manuals, the line between conversational and agentic commerce is blurring, especially until the discovery stage. You can ask them to find you something based on certain parameters (like price range and features), and they will serve you some options. However, agentic systems that can complete transactions or make purchases autonomously are a breed apart from conversational commerce.
How does agentic commerce work?
One way to understand how agentic AI commerce works is to separate stakeholders from the actual workflow.
Who is involved
Whether we evaluate the entire agentic AI commerce landscape or focus on specific agentic e-commerce interactions, there are four core stakeholders.
AI agents & LLMs: An AI agent, multiple AI agents working in concert, or an LLM interface that helps individuals or businesses complete commercial actions (buying/selling) is the defining characteristic of agentic commerce.
User: The individual or business leveraging AI’s agentic capabilities to complete a commercial transaction.
Seller/Business: An individual seller or business on approved platforms where AI agents/LLMs evaluate listings. They can create, modify, and curate their listings for AI agents/LLMs, either exclusively or alongside human buyers.
Payment Processor: The entity that establishes a secure payment flow from user to seller through an AI agent/LLM. From credential management to refunds, payment processors are fully or partially (along with the entity behind AI agents/LLMs) responsible for all the elements of the transaction pipeline.
The flow
A typical agentic transaction moves through six stages:
Intent: The buyer communicates a need in natural language, usually with constraints like budget, timing, features, certifications, and delivery window attached to the request.
Discovery: The agent interprets buyer intent and searches for the product where it can. For some agentic systems, it can be any online store on the internet, while others might be more limited. For sellers, it’s critical to ensure that the agents they want discovering their products are able to read their listings and communicate all the relevant information.
Evaluation: The agent compares products against user preferences and constraints, whether explicitly conveyed or agent-interpreted from context. This is usually a far deeper comparison than a filtered search results page, since it carries a significant amount of context.
Decision or shortlist: Depending on the level of autonomy it has, the agent either presents multiple options to the user to choose from or selects one and serves it to the user to make a buy or pass decision.
Checkout: Once the purchase decision is made, the agent can complete the checkout formalities on your behalf (if it has the authority), though it mostly takes place on the merchant’s own site or app, although there is precedent for this happening inside the agentic/conversational interface.
Post-purchase: Order status, returns, and refunds – these things remain relatively unsolved for now, primarily because the order was placed and processed by an agent while the service relationship belongs to a human.
Lessons from ChatGPT Instant Checkout
OpenAI announced its “Buy it in ChatGPT” capability and instant checkout feature on September 15th, 2025. For users, it was as simple as asking ChatGPT to find a product within a specific price range. They could ask for products with specific features, capabilities, certifications, etc. Sellers/merchants from two approved platforms–Etsy and Shopify were already discoverable on ChatGPT once they connected to the relevant APIs on their platforms. Users with a Stripe account could make the payment for the item they liked with one click. The feature was powered by the agentic commerce protocol that OpenAI co-developed with Stripe.
Through late 2025 and early 2026, the ecosystem around it grew quickly. Walmart, Target, and Instacart joined, PayPal came in on the payments side, and connected enterprise brand catalogs. In February 2026, the feature went generally available to US users.
Then it stopped. In March 2026, OpenAI moved away from instant checkout and began working with retailers on dedicated apps instead, letting merchants use their own checkout experiences while OpenAI concentrated on product discovery.
Understanding the challenges OpenAI faced is useful for businesses preparing for or offering agentic commerce.
- Conversion: Buyers who reached checkout inside the chat completed at a materially lower rate than buyers who clicked through to the retailer’s own site, even though the assistant was driving strong referral traffic.
- Merchant onboarding: Getting merchants technically live was slower and harder than the announcements implied.
- Missing commerce basics: No multi-item carts, no loyalty program connections, and inconsistent product data.
- Tax and compliance: As of early 2026, there was no mechanism in place for collecting and remitting US state sales tax on in-chat transactions, which is not an edge case for a US retail channel.
While OpenAI’s retreat on Instant Checkout doesn’t indicate a failure of agentic commerce, it did endorse the pattern that currently holds: discovery in AI, buying on merchant’s site/platform.
Agentic commerce examples
The clearest agentic commerce examples in the market today can be divided into four categories.
Platform-level catalog distribution: Shopify has taken the infrastructure route by activating their platform for agentic commerce. They co-developed an open standard with Google called the Universal Commerce Protocol (UCP) and adopted it to make the Shopify Catalog open for agentic discovery. This includes ChatGPT, Microsoft Copilot, Google AI mode, and Gemini.
Retailer-owned shopping agents: Amazon has built inside its own walls rather than joining the open protocols. Its shopping assistant (Rufus) is now officially merged into Alexa for Shopping, which handles discovery, comparison, and price-triggered auto-buying across Amazon’s catalog. Its “Buy for Me” feature takes a user’s buying journey outside the Amazon walls, with an agent completing the purchase on a third-party brand’s own website on the buyer’s behalf while the experience and financial information stays inside the Amazon app. It is worth noting that this triggered complaints from independent brands who found their catalogs listed and discoverable by Amazon buyers, without their permission. This is a live reminder that consent and control are unresolved questions in agentic AI commerce.
Payments infrastructure: Mastercard Agent Pay and Visa’s Intelligent Commerce can be considered agentic commerce enablers, providing the ecosystems, tools, and standards for secure agent-handled payments. The card networks and processors have moved unusually fast here, building tokenization, spending controls, and agent identity building blocks well ahead of consumer demand.
Travel and ticketing: Travel is structurally the most primed domain for agentic commerce, partly because inventory is already exposed through standardized APIs. Also, because multi-step itineraries, fare changes, and rebooking are exactly the kind of work buyers are happy to delegate. One business question that is still open in this domain is who gets the commission for a booking that an agent executed (where relevant)?
Benefits of agentic commerce
The benefits of agentic commerce split cleanly between the two sides of the transaction – buyers and sellers.
For buyers
A compressed journey: Discovery, comparison, and decision can all happen within one conversation and in a single window, instead of sprawling across a dozen tabs.
Single-window access across multiple marketplaces: An agent can search through and pull products from every catalog it has access to, instead of remaining limited to the platform or marketplace it’s native to. But it depends upon agent permissions and consent of third-party marketplaces.
Deeper intent matching: A buyer can express constraints, tradeoffs, and context that search filters cannot accommodate. Roughly 58% of consumers already turn to LLMs for product and service recommendations, indicating a growing level of trust in LLMs when it comes to purchase decisions.
Reduced decision fatigue: Agents minimize decision fatigue for both repeat and high-involvement purchases.
For sellers
An unsaturated organic channel: Agent-driven exposure to new customers is not yet a fully pay-to-play avenue, which makes this an unusually attractive window for organic visibility.
Higher-intent traffic: When an agent has already pre-qualified a product against the buyer’s stated constraints, the visitor who arrives and makes the final call is highly unlikely to deviate.
Differentiation through trust signals rather than ad spend: Depth and authenticity of reviews, certifications, and third-party validation can carry more weight with an agent than placement can.
A new discovery channel: OpenAI’s own research indicated that 2.1% of ChatGPT queries concern purchasable products. It may seem small as a percentage, but if you see it in the context of the massive message volume that ChatGPT entertains, it’s a large absolute number.
Agentic commerce challenges
There are multiple open questions when it comes to agentic commerce, which businesses must answer to become ready for agentic commerce. These questions are:
How will your products be discovered when AI makes the recommendations?
It’s important to understand that agentic discovery of a product online is a machine-to-machine process. Agents do not browse your product listing the way humans do. They fetch and go through machine-readable structured data covering product attributes, features, delivery time, policy terms, etc. If this data isn’t present or is too thin for an agent to make a judgment, your product might not get noticed. Focus on broad schema coverage and feed accuracy while adopting good data engineering and product data hygiene practices to ensure agents have easy access to relevant data.
How will trust, pricing, and differentiation influence an AI agent's decision?
One great thing about agentic commerce is that they don’t have any bias. Even when your preferences are guiding their search, they will remain objective in their product analysis and focus on verifiable facts. Businesses with genuine review depth, clear policies, consistent pricing across channels, and verifiable third-party credentials are likely to be noticed more and weighed better than their more “visible” counterparts. Unless it’s conveyed explicitly, it’s a repeat user preference, or carries clear differentiation in its market, brand premium may not influence an AI agent’s decision as it does a human’s.
Is your digital experience ready for machine-to-machine commerce?
Machine-to-machine commerce introduces certain technical challenges – like the ability of an e-commerce platform or an online store to distinguish between a bot that’s there to scrape and a bot that visited to make a purchase. Other challenges include rate limits designed for human buyers, session logic behind a page rendered for humans, and a visually cued checkout flow. Making digital experiences ready for machine-to-machine commerce is not just about front-end changes; it requires new buyer journeys and custom software development at the architecture level.
How do you manage trust, fraud, and verification risk?
Agent-based transactions aren’t aligned with many of the assumptions built into the current fraud prevention systems. New authentication and verification models focus on bounded autonomy and cryptographic delegation. KYC has also evolved to KYA – Know Your Agent. Fraud prevention is also evolving for agentic speed. This is being addressed, at least to an extent, by payment processors through agent identity, delegated payment credentials, and other agentic controls, but the mechanism and standards aren’t commonplace yet. One way many e-commerce companies that could benefit from agentic commerce have closed themselves off for safety is to block all agent traffic. A much better alternative would be for you to start building and deploying AI security and governance controls now so when agentic commerce gains even more traction, you are ready.
Agentic commerce vs. traditional e-commerce
Even in its developing stage, agentic commerce has the potential to significantly disrupt traditional e-commerce. Below are some key differences between agentic commerce and conventional e-commerce.
| Agentic Commerce | Traditional E-Commerce | |
|---|---|---|
| Buyer’s journey | Compressed from discovery to purchase into one exchange. The financial transaction can take place during the exchange or on the seller's website or app. | Spread over multiple discovery and decision points within the platform or across multiple seller stores. |
| Single-window access to multiple marketplaces | Yes. For all catalogs that the agent can access. | No. Limited to one marketplace. Aggregators exist but divert buyers to seller’s websites. |
| Buyer’s product search | Descriptive, prompt-based, and far more granular than what e-commerce platforms allow. | Limited to platform’s search engine capabilities and filters. |
| Product search depth | Agents may access and process far more product data than what’s visible in a listing. | Platform search feature is developed for humans and reflects their searchability limitations. |
| Intent awareness | May be grounded in a deeper understanding of buyer’s purchase intent, spending behavior, and previous discussions. | Only what is explicitly conveyed by the buyer. |
| Autonomy | Partial to full. Agents can stop at recommendations with human making the purchase decision but agentic buying decisions exist too. | None. Buyers make all decisions. |
| Trust signals | Can be gathered from multiple external sources and cross-checked. | Only what is conveyed in listing and via reviews posted on the platform. |
Agentic commerce market size and growth
Agentic commerce is still relatively nascent, so projections right now are cautious. They may be reevaluated significantly once it gains more momentum.
- Agentic commerce could drive up to $17.5 trillion in commerce by 2030. (Source: Deloitte)
- Total agentic commerce transactional value predicted for 2026 is $8 billion. (Source: Juniper Research)
- Agentic commerce market size is expected to reach $65.5 billion in 2033, at a 35.7% CAGR from 2026. (Source: Grand View Research)
- There are about 4.5 million agentic search queries a month on average. (Source: Commerce)
$17.5T
Potential agentic commerce by 2030 (Deloitte)
$8B
Predicted transactional value for 2026
(Juniper Research)
$65.5B
Market size by 2033, at 35.7% CAGR (Grand View Research)
4.5M
Agentic search queries per month (Commerce)
The growing ecosystem of agentic commerce protocols
Multiple agentic commerce protocols are already in the market and while some offer end-to-end coverage, others focus on a specific phase of the buyer’s journey.
Agentic Commerce Protocol (ACP)
Agentic Commerce Protocol, or ACP, was developed by OpenAI and Stripe and shared as an open standard, so it sits outside these two ecosystems. Any AI agent/LLM, payment processor, or merchant can leverage ACP to establish secure payment channels between the user and the merchant. It focuses on two aspects of the buyer’s journey: Checkout and payment.
Universal Commerce Protocol (UCP)
The Universal Commerce Protocol (UCP) was developed through the collaboration of Google and Shopify. It’s an open standard that provides the building blocks for agentic commerce and spans across the entire buyer’s journey – from discovery to checkout. It claims to accommodate any payment processor and any wallet, and is built with interoperability in mind.
The fragmentation problem
While ACP and UCP get all the limelight, these aren’t the only partial or full protocols available. There are multiple payment-specific protocols, while Model Context Protocol (MCP) may allow agentic connectivity across a massive range of platforms and marketplaces. Interoperability across different protocols is emerging and may take some time to mature.
How to leverage the full potential of agentic commerce
If you are an enterprise with e-commerce offerings, it’s important to start looking into the role and potential impact of agentic AI in e-commerce. If you are a seller on platforms that AI agents are already searching, this is ever more urgent. Either way, there are five areas sellers and enterprises should start focusing on if they wish to succeed in the era of agentic AI commerce.
Integration
Most LLM interfaces possess search capabilities, but a proper integration between them and your platform is crucial to enable search and match capabilities via metadata. It also enables users to leverage platform-specific refund and return features. Businesses that are officially connecting to a specific agentic platform must adhere to their integration requirements and make sure they are exposing the right data because even if they have access to dedicated connectors, improper or incomplete data flow may deter agentic commerce activities. Sellers that want to make their data available to multiple platforms and a wider range of agents should rethink their entire AI integration ecosystem.
Marketing
Since agentic commerce spaces are not yet saturated or pay-to-play, shifting the focus and budget to organic marketing can yield strong results. This would involve matching user (buyer) intent contextually, targeting pain points, understanding Agent/LLM interactions of your target audience, and hitting those pain points with your marketing content. It’s important to understand that LLMs run far more contextually aware searches than marketplaces and search engines, so a consistent and informative marketing approach across multiple channels will help.
Listing
Listings are typically listed from a marketplace’s search algorithm’s perspective or for a broader SEO perspective, but that’s no longer enough. You must understand the kind of questions people might ask an AI agent when looking for a product like yours, the features they may ask it to vet products for, or any specific signals they might embed in their prompts for evaluation criteria. Overhauling your listings for Answer engine optimization (AEO) is a good start, but a more comprehensive strategy to revise listings for agents would be better. Good data governance is critical here because agents are likely to treat consistent data across multiple channels as a trust signal.
Trust signals
In agentic search, high-quality trust signals can bump your products to the top of the results, and a lack of such signals (reviews and more) might hurt your chances, even if you check all the other boxes. It’s also worth looking into the depth, authenticity, and diversity of trust signals, not just the count.
Redefining metrics
The core performance tracking dimensions would remain the same, like visibility and conversions, but metrics will be changed. They are not yet broadly available, but as agentic commerce evolves, we may likely see metrics like the number of times your product was displayed for relevant searches, how many times your products were selected for further questions, total conversions (that’s a classic), and in-chat vs. Agent-assisted conversions. Preparing for these new metrics can make you proactive about performance tracking in the new, shifting market.
Powering the next wave of agentic commerce with 10Pearls
ChatGPT’s Instant Checkout episode has helped the market learn some important lessons. Enterprises and product engineering partners that understand this shift toward agentic commerce are already working towards harnessing this capability. As a future-focused AI and innovation partner, 10Pearls can help you leverage the full potential of agentic commerce.
Our agentic AI capabilities and deep understanding of the Model Context Protocol (MCP), multi-agent orchestration, and agent mesh frameworks allow us to spearhead a wide range of agentic commerce initiatives. The end-to-end services that start with AI readiness assessment and consultations help us identify and pursue the best approach to integrating agentic commerce in your business model. If you are weighing whether to connect to an existing agentic platform or build your own capability, start with an agentic commerce readiness review.
Agentic commerce FAQs
How to implement agentic commerce workflows?
Agentic commerce workflow implementation journey for sellers starts with integrating their online store/marketplace to the agent that buyers will interact with. They also need to adopt a protocol that enables secure payment through an agent and a payment processor to facilitate transactions through this protocol. The agent must have access to products’ metadata through integration and clear and reliable trust signals to increase the chances of visibility and conversion.
What are the best tools for agentic commerce?
The best tools for any enterprise or seller adopting agentic commerce are subject to their ecosystem and approach to agentic commerce. Sellers leveraging an LLM’s native agentic commerce capabilities will only require tools like integration APIs to start, assuming they are independent or their platform is already connected. In contrast, sellers implementing their own agentic commerce will benefit from agent-development toolkits and orchestration frameworks. They will also need tools like payment processor integration APIs, secure payment protocols, and compliance and verification tools. Performance tracking tools are critical for both types of sellers.
How can marketplaces prepare for agentic commerce?
To prepare for agentic commerce, marketplaces should ensure their infrastructure is ready for more agent traffic (dedicated APIs, rate limits, etc.), listings are enriched with metadata and schemas for AI agents, and they have adopted agentic commerce protocols for safe payment processing. Integrating trust signals and connecting with payment processors AI agents are leaning toward may help them stand out.
How can enterprises get started with agentic commerce initiatives?
Enterprises must start by determining whether their business model is even aligned with the current or evolving agentic commerce paradigm, and if it is, understanding their timeline. Then they must determine if they have the capabilities and data infrastructure for an agentic commerce initiative.
Then they must choose one of the two available paths to agentic commerce – connect with an available agentic platform or develop their own agentic commerce capabilities and interfaces. The first has constraints related to platform-level permissions and integrations. The second requires technical capabilities to develop and maintain agentic commerce and business capabilities to make agentic commerce work within their model.
A mix of both can be practical, and a seasoned AI expert like 10Pearls that also brings diverse industry experience to the table can help enterprises pursue the most optimal approach to agentic commerce.
How does agentic commerce ensure security and compliance?
Agentic commerce leverages the compliance and data security guardrails of the underlying model, existing privacy and financial security controls of the payment processor, a secure payment protocol connecting all stakeholders, and merchant identification and verification at the integration layer. However, as agents become more autonomous, a new security and compliance layer will emerge with new checks, like buyer intent verification, autonomy limits, transaction provenance, and agent identity.
What industries will benefit most from agentic commerce?
Industries that have more “decision-heavy” products and services and rich metadata might benefit more from agentic commerce in the first wave. They are naturally positioned to benefit from this new customer journey. For now, this includes electronics, home goods, and travel.
How will agentic commerce transform customer experience (CX)?
Agentic commerce will compress the customer’s journey from discovery to purchase and allow customers to fetch results from multiple catalogs/marketplaces at once. Customers will also be able to convey their intents (what they wish to buy) and constraints (budget, features, etc.) more comprehensively through prompts and start trusting agents’ recommendations as their positive experiences accumulate.
What is the difference between agentic commerce and conversational commerce?
Conversational commerce is mainly about the interface, which allows a buyer to complete their journey through dialogue. In contrast, agentic commerce is about delegating parts of or the entire customer journey to AI agents, though it does include dialogue to convey the requirements to the agents. There is significant overlap between the two now as frontier models leverage agentic capabilities under the hood while keeping the commerce aspect conversational.
What is Google's Universal Commerce Protocol, and how does it compare to ACP?
Google’s Universal Commerce Protocol was developed in conjunction with Shopify and provides the building blocks for agentic commerce to connect agents, businesses, and payment processors. With a strong focus on interoperability, it covers the entire buyer’s journey – from discovery to checkout, which makes it different and much broader than ACP, which focuses on checkout and payment.
What are the benefits of agentic commerce?
For buyers, the benefits include a compressed journey, simultaneous access to multiple catalogs and marketplaces, easy communication of preferences and constraints, and a rapid, unbiased comparison of dozens or even hundreds of products. Sellers benefit from the new organic reach agentic commerce offers, and as a channel that’s not yet saturated. They also benefit from higher-intent traffic and a chance to increase sales numbers based on reviews rather than spending on visibility through ads.
What are the challenges of agentic commerce?
Existing data, digital infrastructure, and fraud and dispute frameworks will have to be overhauled to handle agent traffic. Different marketplaces may have different attitudes towards agentic commerce, and the adoption of different types of protocols increases the engineering work platforms must do to cast a wider net.
Related blogs
Fintech
Banking as a Service (BaaS): How It Works
Learn what Banking as a Service (BaaS) is, how it powers embedded finance, and how non-banks integrate accounts, cards, payments,...
Fintech
Synthetic Identity Fraud Detection and Prevention
Synthetic identity fraud is the fastest growing financial crime in the US. Understanding why that is and what its life...
Fintech
How AI creates value with open banking data
Every fintech company with an open banking license in Saudi Arabia must build the basic infrastructure to receive open banking...
Fintech
How AI fraud detection strengthens financial security and prevents real-time fraud
Discover how AI fraud detection outperforms traditional methods by identifying novel threats, boosting accuracy, and scaling with ease.
Fintech
Key strategies for navigating 1033 compliance in finance
In its final rule, implementing section 1033 of the Dodd-Frank Act, the Consumer Financial Protection Bureau (CFPB) defined requirements that...
Fintech
Explore common misconceptions of agile fatigue
Within the past few years, there has been relative fatigue in many organizations in adopting and implementing Agile practices, processes,...
Fintech
Fintech, or Financial Technology, is transforming financial services. The fintech industry has seen explosive growth over the past decade, and...
Fintech
Identify leading fintech software development partner to create secure, scalable, and innovative applications customized to meet your specific financial technology...