How much time do shoppers still waste opening multiple tabs, comparing products, checking reviews, tracking prices, and repeating the same steps before they can confidently buy something?
AI is beginning to reduce some of that friction. Instead of simply answering shopping questions, AI systems are increasingly helping consumers research products, evaluate options, narrow choices, and move closer to a purchase. That shift is already visible.
In a 2026 IBM Institute for Business Value study conducted with the National Retail Federation, 45% of surveyed consumers said they use AI for some part of the buying journey. The study found that 41% use AI to research products, 33% to interpret reviews, and 31% to search for deals or promotions. (Source).
McKinsey estimates that AI agents could mediate or orchestrate $3 trillion to $5 trillion in global consumer commerce involving goods by 2030. The estimate excludes services and significant areas of B2B commerce and refers to commerce influenced or handled by agents, not necessarily purchases completed fully autonomously (Source).
This emerging shift from AI-assisted shopping toward AI-driven, permissioned action is commonly described as agentic commerce.
Agentic commerce is a form of buying and selling in which AI agents act on behalf of consumers or businesses to help achieve a purchasing goal. Depending on the integrations and authority they receive, these agents can research products, compare options, recommend or select among them, and potentially initiate or complete transactions on a user’s behalf.
Unlike traditional AI shopping tools that mainly answer questions or recommend products, an agentic system can potentially pursue a goal across multiple steps and take delegated action.
For example, instead of asking, “What are the best laptops under $1,200?”
a shopper could instruct an agent: “Find a lightweight laptop under $1,200 with at least 16 GB of RAM that can be delivered by Friday.”
If the agent has access to reliable product, inventory, pricing, and delivery data, it can evaluate available options against those requirements rather than simply returning a list of recommendations.
Agentic commerce is built around four core elements:
The defining difference is not simply conversation versus action. It is the degree to which an AI system can pursue a goal across multiple steps and act under delegated authority.
A conventional shopping assistant may tell a user what they could buy. An agentic system can potentially move beyond recommendation and perform some of the work required to complete a purchase, subject to available integrations, data quality, and user permissions.
This does not mean every agent operates autonomously. Depending on the permissions provided, an agent might:
The clearest distinction is how much of the buying journey the AI is responsible for.
In simple terms:
The distinction is therefore based on the level of delegated, goal-directed action, not merely on whether the shopping experience uses a chatbot or conversational interface.
Understanding agentic commerce requires a few concepts that explain how AI agents interact with businesses, commerce systems, and payment infrastructure.
Together, these concepts illustrate a broader change in digital commerce: websites and checkout interfaces may increasingly be complemented by machine-readable data, programmatic access, permissioned actions, and payment infrastructure designed for interactions involving AI agents. MCP is one possible integration approach alongside conventional APIs, structured product feeds, and commerce-specific checkout or payment protocols.
Agentic commerce turns a shopper’s intent into a sequence of actions that an AI agent may handle.
Agentic commerce generally follows three steps:
The shopper provides an objective and the main constraints the agent should follow, such as:
The agent uses these instructions as the basis for the rest of the buying process.
The agent searches for products or services that match the shopper’s criteria and evaluates them using the commerce data available through merchant websites, APIs, product feeds, or other connected systems.
This can include:
The reliability of this evaluation depends on whether merchants provide current, structured, and trustworthy data. If an agent relies on stale webpages, incomplete feeds, or ambiguous policies, its recommendations may be inaccurate.
The agent can then filter out unsuitable options and narrow the results to those that best match the shopper’s requirements. Where the required integrations are available, it may also recheck prices, inventory, and delivery details before checkout.
Once a suitable option is selected, the agent can move the purchase toward checkout.
Depending on the system and its permissions, this may include:
A typical agentic-commerce flow moves from product discovery to cart and checkout, followed by payment authorization and order confirmation.
Where merchant and platform integrations support them, agents may also help with post-purchase tasks such as order tracking, reordering, booking management, or return requests.
Agentic commerce can operate at different levels of autonomy:
In publicly documented deployments, many current use cases emphasize assisted discovery, recommendation, cart preparation, and human-approved purchasing. Fully autonomous purchasing remains more constrained by authorization, fraud, liability, and consumer-trust requirements.
Getting started with agentic commerce does not mean giving AI agents full control over the buying process from day one. It means preparing commerce systems so agents can understand products, access reliable information, perform approved actions, and operate within clear business rules.
AI agents need structured, current data that they can interpret without relying only on visual webpages. Machine-readable data can improve an agent’s ability to identify products and compare options, but it does not guarantee accurate recommendations unless the underlying information is complete and regularly updated.
Businesses should make information such as the following easy to access:
When this information is structured and kept up to date, an agent can more accurately determine whether a product matches a user's request.
Finding a product is only one part of the journey. An agent may also need to check stock, retrieve a live price, add an item to a cart, calculate shipping, or initiate checkout.
That requires connections through technologies such as:
These connections allow agents to interact with commerce infrastructure programmatically instead of relying only on interfaces designed for people. Different technologies may support different parts of the workflow, including product discovery, tool use, checkout, authentication, authorization, payment, or post-purchase service.
Businesses also need to decide what an AI agent is allowed to do.
For example, an agent may be allowed to:
Controls around authorization, payment limits, identity, fraud prevention, and audit trails help prevent unintended or unauthorized transactions.
Agentic commerce is easier to implement when businesses begin with a narrow, measurable task rather than trying to automate the entire buying journey.
Good starting points include:
Once the workflow performs reliably, businesses can gradually expand the agent's responsibilities.
The goal is not maximum automation from the beginning. A stronger approach is to build around accurate data, secure system access, clear permissions, and controlled actions, then increase autonomy as the experience becomes more reliable.
Agentic commerce represents a shift from manual online shopping toward delegated digital action. As AI agents become better at interpreting intent, accessing current data, and working across connected systems, they may take on more of the buying journey while operating within defined permissions.
The opportunity goes beyond automating checkout. The broader change is the development of commerce systems in which AI agents can understand goals, evaluate options, use approved tools, and act on behalf of users as authorized intermediaries.
As this model develops, businesses will increasingly need accurate machine-readable data, secure integrations, scoped permissions, and clear controls that allow AI agents to participate in commerce without sacrificing trust, accountability, or human oversight.
McKinsey estimates that AI agents could mediate or orchestrate $3 trillion to $5 trillion in global consumer commerce involving goods by 2030. It estimates the US B2C retail opportunity at approximately $900 billion to $1 trillion in orchestrated revenue. These are forecasts, not current sales figures.
An AI shopping agent could compare products based on price, features, and delivery time, prepare a cart, and either request approval or complete an eligible purchase within predefined limits.
MCP connects AI applications with external tools and data. In commerce, it could provide access to product information, inventory, pricing, or selected actions. MCP does not itself provide payment authorization or fraud protection.
It can be secure when supported by authentication, delegated authorization, spending limits, fraud detection, monitoring, and audit logs. Tokenization can also reduce exposure of customers’ underlying payment details.
Agentic AI refers broadly to systems that interpret goals, use tools, and take actions. Agentic commerce applies those capabilities to activities such as product discovery, purchasing, payments, fulfillment, and returns.
