What is agentic commerce? How AI agents buy complex products

Agentic commerce is commerce where an AI agent, not a person clicking through a web page, does the work of buying: it finds the product, checks that a configuration is valid, gets a price and moves to a quote or an order. For a seller of configurable products, that means exposing options, engineering rules and pricing as something an agent can call, not a PDF it can only read. Spokbee does exactly that: it publishes your configurator's products, rules and prices as tools an AI buying assistant uses to configure, price and quote.

This explainer defines agentic commerce, separates it from AI search and chatbots, names the protocols agents use, lists what a seller has to expose, and shows how Spokbee makes a complex, made-to-order product ready for the AI buyer today.

What is agentic commerce?

Agentic commerce is a transaction in which an AI agent acts on a buyer's behalf: it understands the need, searches the products, resolves the details, confirms the price and completes or hands off the purchase. The buyer states intent. The agent calls the seller's systems. The seller's rules and prices still decide what is valid and what it costs.

The name joins "agentic AI", software that takes actions toward a goal with limited supervision, to the oldest activity on the web. For a consumer product with a handful of variants, the agent compares a few listings and pays. For a configurable, engineered product, the agent has to do the work your application engineer does today: ask the right questions, apply the fitment and compatibility rules, price every option and produce a quote that production can build. That second case is the one that separates "we appear in an AI answer" from "an AI can buy from us".

Spokbee splits the buying journey into two stages on its own pages. AI product discovery is the stage where an assistant understands what you sell: published products, descriptions, specifications, selectable options and their dependencies. Agentic selling is the stage where the assistant gets actions: read configuration options, validate selections, calculate a price and prepare a quote or a checkout link. Agentic commerce is the second stage, built on the first.

How is agentic commerce different from AI search and chatbots?

AI search answers a question with text. A chatbot answers questions inside one website. Agentic commerce takes actions inside the seller's systems. The difference is the verb. A search result tells a buyer that you make 48-inch cabinets. An agent checks that 48 inches in graphite with four shelves is a valid build, prices it and asks for the quote.

The three rows build on each other. Discovery gets you into the conversation. Guided selling helps a buyer who has already landed on your configurator describe a need and get a configuration back, with the price impact of each option explained. Agentic commerce moves the whole exchange into the buyer's assistant, where the buyer increasingly starts, and brings a real configuration, price and quote back out.

Which protocols do AI agents use to buy?

Two open standards matter today. The Model Context Protocol (MCP) lets an AI application call a seller's tools, and the Agentic Commerce Protocol (ACP) standardises product feeds and checkout between merchants and agents. Spokbee implements MCP: a connected assistant sees your catalog as tools it can call to discover, configure, price and quote.

The protocol's maintainers describe MCP as "an open-source standard for connecting AI applications to external systems" and compare it to a USB-C port: one connector instead of a custom integration per assistant. Claude, ChatGPT, Visual Studio Code and Cursor all support it, according to the MCP introduction (checked October 2026). A server publishes tools. The client inside the assistant discovers and calls them.

Stripe announced ACP on 29 September 2025 as "a new, merchant-friendly open standard codeveloped by Stripe and OpenAI". Its announcement says businesses "must expose their products, pricing, and checkout in a way agents can use", and that the merchant still accepts or declines each order and keeps control of tax, fulfilment and returns (Stripe newsroom, checked October 2026).

For a product with a few SKUs, a feed is enough. For a product with thousands of valid combinations and rules between them, no feed can list the answer, so the agent has to ask. That is why Spokbee's agent interface is a set of tools rather than a catalog file. Spokbee's agentic selling page states that a configure_and_price response includes validity, completeness, constraint violations, available options and a price breakdown. Developers who build their own integration get the same actions over REST, plus webhooks that send events such as a newly created quote to an endpoint they manage, as the integrations page describes.

What does a seller have to expose for an AI agent to buy?

A seller has to expose five things in a form an agent can call: the products it sells, the options and rules that make a configuration valid, the price of each configuration, a way to turn a configuration into a quote and, where wanted, a checkout. A sixth, access control, decides which agents reach which of those.

The rules row is where configurable products differ from everything else an agent buys. A wheel that does not fit, a cabinet with more shelves than its height allows, a vehicle body with an option that excludes another: a human application engineer catches these by experience. An agent catches them only if the rules are encoded and evaluated on every call. In Spokbee the rule you write once applies to every website buyer, every dealer and every agent, because the website configurator and the agent interface run the same configuration and pricing engines. One product model, more ways to buy.

How does Spokbee make configurable products ready for AI agents?

Spokbee turns a published configurator into four agent tools: find_products to discover published products, configure_and_price to check options and calculate a price, quote_product to create a configured quote and checkout to prepare a buyer checkout link. Every call runs through your configuration and pricing engines, so an agent-built configuration is as valid as one built on your website.

Spokbee's agentic selling page walks through a sample workshop cabinet. The data is illustrative and no orders are placed, but the shape of the exchange is exactly what a connected assistant does.

The tool call behind the second step looks like this. Product slugs and selection IDs come from your own catalog.

{
  "tool": "configure_and_price",
  "arguments": {
    "product": "workshop-cabinet",
    "selections": {
      "finish": "graphite",
      "width_in": "48",
      "shelves": "4"
    }
  }
}

The third step is the one that separates a configurator from a chatbot. The six-shelf request is not refused with a generic error. Spokbee returns the rule and the remaining options, and the assistant corrects course without guessing.

You stay in charge of the channel. The agentic selling page lists three controls. Choose your reach: enable public agent access for published products, or connect integrations with scoped agent keys. Scope the actions: give each key the permissions it needs, with rate limits and the ability to revoke access. Keep purchasing intentional: enable checkout per product, with quotes validating the configuration and checkout running through your connected payment setup. Agent analytics then tracks requests, configuration activity, quotes and sessions, so you can see which assistants are active and what they configure.

Why does agentic commerce matter for manufacturers now?

Buyers are already sending agents. Gartner expects 40 percent of enterprise applications to include task-specific AI agents by the end of 2026, up from under 5 percent in 2025, and Adobe measures AI-referred retail traffic growing and converting better than other traffic. Configurable products are where an agent earns its keep, because the answer depends on the details.

All three figures were checked in October 2026. The retail numbers come from consumer sites, and the point for a manufacturer is not the percentage but the direction: a growing share of buying starts inside an assistant, and the assistant will favour sellers whose answers it can verify.

Compare that with the way complex products have been quoted for decades. A buyer fills in a request form. Sales forwards it to an application engineer. The engineer checks the drawing, works the spreadsheet and sends back a PDF a week later, and every change of mind restarts the loop. Spokbee customers have already replaced that loop on the web: Stor-Loc's routine quoting now takes about half the time it did, as told in how Stor-Loc streamlined sales with Spokbee, and PL Custom's time-to-quote for emergency vehicles went from about two years to a few days. Agentic selling puts that same engine where the buyer's assistant is, so the first answer an agent gets from you is valid, priced and ready to quote.

How do you get started with agentic selling on Spokbee?

Start with one published product. Spokbee's agentic selling page lists what you need: a published Spokbee product with its configuration rules, public agent access or a scoped agent key, and a compatible client. Checkout also needs product-level enablement and a connected payment account. The free Starter plan is where the first configurator gets built.

Step 1 decides how good the agent's answers are. The guided selling documentation notes that the AI helper works best when parameters and options carry rich descriptions, and the same holds for an external agent reading your catalog. Encode the rules your engineers apply by hand, price every option, and the agent inherits all of it.

The same published product serves every other channel too. It runs as a 3D configurator on your own website, in a dealer experience and in the buyer's AI assistant, from one product model. Agentic commerce is not a separate catalog to maintain. It is the configurator you already have, made callable.

Sources

• Spokbee: Agentic selling and MCP tools, the four agent tools, the sample cabinet exchange, the seller controls and the "what do we need to get started" answer. Checked October 2026.
• Spokbee: Get found by AI, the two stages of the buying journey and what product discovery exposes. Checked October 2026.
• Spokbee: CAD and API integrations, REST APIs, webhooks and MCP tools. Checked October 2026.
• Spokbee docs: AI features, the guided selling chat. Constraints, Quotes and Commerce and checkout. Checked October 2026.
• Model Context Protocol: introduction. Checked October 2026.
• Stripe newsroom: Stripe powers Instant Checkout in ChatGPT and releases Agentic Commerce Protocol, 29 September 2025. Checked October 2026.
• Gartner: 40% of enterprise apps will feature task-specific AI agents by 2026, 26 August 2025. Checked October 2026.
• Adobe: AI traffic trends report, 19 August 2026. Checked October 2026.
• McKinsey: Five fundamental truths: how B2B winners keep growing, 2024 B2B Pulse. Checked October 2026.