There’s a lot to unpack around agentic commerce. We (your commerce and search analysts) have teamed up to bring you clarity around agentic commerce, how it differs in various experiences, what’s happening now, and where we’re headed. Read on for your primer on agentic commerce to get ready for what’s here now and what’s coming!

What is agentic commerce?

To be clear, agentic commerce only sort of exists, is just launching for the first time in many arenas — and it might not be … good. Are you hearing that it’s a thing, that you need to have it right now, and that you’re already behind the eight ball? First, don’t panic. No one has this figured out yet.

But to get clear on what “agentic commerce” means, we need to establish a few basics first:

  1. What is agentic in the first place?
  2. Where is this experience happening (owned or non-owned channels)?
  3. What does “commerce” mean in this context? (No, it’s not obvious.)

1. What is agentic in the context of commerce?

Forrester’s working definition for “agentic AI” is:

Systems of foundation models, rules, architectures, and tools that enable software programs to flexibly plan and adapt to resolve goals by taking action in their environment, with increasing levels of autonomy.

So we can loosely state that agentic commerce means that those models, rules, architectures, and tools taking action are the “agents” in question. And in this case, the agents are conducting or enabling commerce at some stage of the customer lifecycle.

  • Agents might take an action on behalf of users (such as ordering a specific type of product within a specified budget). More commonly right now, agents respond to shoppers’ queries as the shopper is researching or considering a product or service for purchase.
  • Vendors have lots of names for this experience. We can’t guarantee what vendors mean when they use terms. But in general, if they say terms like “AI shopping assistant,” “conversational chat” (which is a mode of interaction, like a chat, but which may or may not be agentic), “genAI-guided selling,” “conversational commerce,” “agentic commerce,” or some combination of these terms, it might be what we’re talking about here.
  • An agentic experience often occurs through a conversational interface, but just because it’s conversational doesn’t mean it’s agentic. Case in point: Chatbots with canned responses are conversational — but not agentic.
  • Agentic experiences often involve multiple specialized agents, each trained for a distinct task. For instance, in product discovery, the agent that analyzes an image that a customer uploaded to find visually similar products is different from the agent that interprets a customer’s written query about product features or preferences. These agents work behind the scenes, and ideally, the user interacts with them through a single interface and likely never knows that there are multiple agents with different jobs generating responses.
  • Remember also that, in reality, these agents are systems — driven by code and data, including past human or machine interactions. Their “training” is not a form of understanding but a statistical process. Perceiving them as sentient or emotionally intelligent can obscure their limitations and lead to unrealistic expectations. (GenAI chatbots named after pets or with overly humanlike personas don’t help.)
  • Ultimately, machine agentic experiences mimic human ones. Responses are not “canned” or prewritten. They may combine images, text, and sound, and they feel to the user like a somewhat natural interaction.

2. Where are agentic commerce experiences happening?

When we hear about agentic experiences, it’s not instantly clear what type of experience it is. To understand what the agentic experience is, you’ll need to determine whether it’s an owned experience or not.

  • Owned experiences occur within digital properties that a brand controls, such as its website or app proprietary platforms. Today’s agentic commerce experiences in owned properties are typically controlled by the brand but often powered by third-party vendors. For instance, a retailer might implement a commerce search solution to deliver an agentic product discovery experience on its website. Customers engage with these agents using multimodal, natural language prompts — via voice, text, or images. In response, the agent might recommend products based on the input provided, drawing on contextual data such as stated preferences, browsing behavior, or purchase history.
  • Non-owned experiences take place outside the digital properties that brands control, such as in answer engines like Perplexity, ChatGPT, or Google Gemini. In these environments, agentic commerce might occur as part of a broader, more exploratory research process for the consumer. For instance, a customer might chat with one of these “agents” to find a gift for someone who’s hard to shop for, or they might chat to discover a skincare product that influencers frequently recommend. As answer engine agents such as OpenAI’s Operator, Perplexity’s Comet, and Google’s Assistant become more capable of remembering, reasoning, and personalizing, they’ll increasingly collaborate with merchants’ agents to facilitate transactions. In fact, today we’re beginning to see B2B agents interacting autonomously with each other to negotiate terms and discounts independently.

3. What does “commerce” mean in this context?

Everything so far has been clear, but it’s about to get muddy.

  • When we hear the word “commerce” added to the agentic experience, it might mean that a customer is shopping for something. It might mean that a customer has the ability to add something to a cart. It might also mean that the customer can enter payment information, check out, and complete the transaction within the agentic experience.
  • All of these are still fairly unusual use cases. In owned experiences, some brands have agentic chat boxes on their websites (or less commonly, but ideally, conversational search boxes that bring the experience into the search bar). But these are largely product discovery processes. In other words, they enable customers to ask about products they might like, the agentic experience suggests products, and the customer can refine the request to narrow the product options. The cart and checkout processes still transition out of the chat and into the regular website flow.
  • In non-owned experiences, commerce might include a user prompting an answer engine to help find products across many websites. This prompt can enable comparison shopping. It also enables shoppers to ask subjective questions and request answers in a certain tone. This back-and-forth has led to some concerning shopping habits, such as women asking answer engines to “honestly” criticize their appearance and recommend products based on influencers’ content.
  • Some of these answer engines support a checkout experience directly in their applications today (e.g., you can “Buy with Pro” if you’re a paid subscriber to Perplexity in the US). But the experience is still limited and less than ideal. For example, Perplexity touts free shipping on all items you can buy in its platform, but that might make it harder to find an available product if availability is limited to only those that retailers will ship for free. It enables purchasing through specific retailers, like Walmart, but your Walmart purchase via Perplexity is not connected to your Walmart account and doesn’t show in your purchase history. This scenario isn’t specifically a Perplexity problem. The market at large hasn’t yet figured out how to close the chasm between answer-engine screen scraping and direct agent access to (or integration with) merchant data such as availability, promotions, purchase history, and loyalty programs.

What these “agentic commerce” experiences do — and don’t do

  • In owned experiences, shopping assistants don’t enable actual checkouts — yet. Shopping assistants, whether agentic, conversational, or otherwise, might recommend products and enable customers to click through to the product detail page or potentially even add to their cart. But there is no checkout within the “chat” or conversational experience. In some cases, they’re still early versions of what might, someday, be well-used shopping tools. Today, what vendors call agentic commerce or conversational commerce experiences may be glorified chatbots or guided selling processes that still serve static, canned workflows rather than generative, responsive experiences.
  • In non-owned experiences, direct commerce experiences in answer engines are rare. And when they’re present, product assortment is poor, product details are inconsistent, and the experience for the customer is disjointed from the merchants’ owned experiences. Plus, they’re using imperfect work-arounds to make the payments happen, because the card networks (e.g., Mastercard, Visa) do not and cannot yet formally recognize agentic payments. The answer engines (e.g., ChatGPT, Gemini, Perplexity) are working with payment partners, commerce platforms, and merchants to build the foundations for their agents to make purchases on behalf of customers and to improve the customer and merchant experiences when they do.

Where are we now, and where are we going?

  • In the non-owned environment, we are beginning to see more planning, adapting, and taking action that earns the agentic label more accurately. Agentic commerce has arrived in that sense. Perplexity’s Buy with Pro enables paid users to check out natively, and OpenAI is out ahead with instant checkout in ChatGPT and protocols for commerce and product feeds. This will continue to move quickly as answer engines rush to market with commerce functionality and merchants work to adapt to these new protocols (including updating their product feeds for agentic commerce).
  • What we’re calling agentic commerce in the owned environment is still mainly “assistive” — meaning that it rarely acts autonomously on behalf of the shopper and instead aids shoppers in their decision-making. We are allowing some grace with the definition because there are some autonomous decisions (e.g., deciding which products to display) and we expect that more agentic functionality will roll out quickly. Despite this advancing technology, these experiences are not necessarily good, positive for customers, or successful for businesses.

What should you do now?

  • For now, our best advice for vendors building these experiences is: Keep going. We know you’re working hard to figure it out, but please ensure that development involves massive amounts of testing to be certain that you’re creating experiences shoppers want. Even if live testing is the only option, roll it out to a limited percentage of traffic (ideally, power users) first, solicit their feedback, and have internal testing teams working on it, as well.
  • Our advice for businesses selling things and worrying whether they’re behind: You’re not. Give the vendors a minute to figure it out and consider being a test subject if you’d like, but again — please don’t rush to create experiences just because you can.

To discuss this in more depth, please book a guidance session or inquiry with one of your Forrester analysts on the commerce and search teams!