Agentic commerce for sellers: what it means, the protocols, and what Amazon sellers should do now (2026)

Short answer: agentic commerce is commerce in which AI agents complete multi-step tasks on their own while a person sets the goal and approves the result. On the buyer side, agents search, compare and pay on behalf of shoppers. On the seller side, agentic workflows analyze market and product data, produce content and check it before a person signs off.

For sellers the consequence is plain: your product data is increasingly read by software first and by a person second. Complete attributes, claims you can back up and clean translations make you easier to understand for both. This guide explains the two meanings of the term, lists the key protocols with dates and sources, shows what changes for Amazon sellers and ends with a checklist. Current as of October 2026.

Two meanings: the buyer side and the seller side

Search for “agentic commerce” and almost everything you find is about the buyer side: payment networks, protocols, checkout inside a chat. That is what the press and analysts usually mean. For sellers the second meaning matters at least as much, because it is about your own work: back-office workflows that complete several steps on their own and ask for approval at the end.

The buyer side: the key protocols, with dates

Since 2025, payment networks, platforms and AI companies have launched a series of open standards and programs. The list below is chronological. Every item comes from the provider's own announcement or from trade press, linked under “Sources” at the end.

• Mastercard Agent Pay (29 April 2025): an agentic payments program built on “Agentic Tokens”, which let an agent pay without ever seeing the underlying card number.
• Visa Intelligent Commerce (30 April 2025): opens Visa's network to developers building AI agents. Partners named by Visa include Anthropic, IBM, Microsoft, Mistral AI, OpenAI, Perplexity, Samsung and Stripe.
• Agent Payments Protocol, or AP2 (Google, 16 September 2025): an open standard that lets an agent prove a person authorized a specific purchase. More than 60 companies backed it at launch, including Mastercard, American Express and PayPal.
• Agentic Commerce Protocol, or ACP (OpenAI and Stripe, 29 September 2025): the open standard behind Instant Checkout in ChatGPT. In March 2026 OpenAI scaled Instant Checkout back. According to CNBC, ChatGPT now sends shoppers to merchants' own stores and apps to pay.
• Stripe Agentic Commerce Suite (11 December 2025): designed to make merchants ready to sell through AI agents, among other things by sharing products, prices and availability with agents in near real time.
• Universal Commerce Protocol, or UCP (Google, 11 January 2026): co-developed with Shopify, Etsy, Wayfair, Target and Walmart. It covers discovery, cart, checkout and post-purchase. Google currently lists checkout in AI Mode and Gemini for products in the United States, Canada and Australia, and it requires well-maintained Google Merchant Center data.

How big this gets is an open question. McKinsey estimates that by 2030 agentic commerce could orchestrate up to $1 trillion in US B2C retail revenue and $3 to $5 trillion globally (October 2025). That is a projection, not a measurement, and the retreat from Instant Checkout shows that buying inside a chat is arriving more slowly than announced.

What about Amazon?

Amazon is taking its own route and relies on its own assistants. On 13 May 2026 Amazon folded Rufus into Alexa for Shopping, an assistant built into the search bar and switched on by default for signed-in US customers. Amazon is wary of outside agents: in November 2025 it sued Perplexity, alleging that Perplexity's Comet browser agent shopped on Amazon without identifying itself as AI.

For an Amazon seller this means that on Amazon, it is mostly Amazon's own AI that reads your listing. Off Amazon, in Google AI Mode or ChatGPT, what counts is your own store, your Merchant Center feeds and your presence in comparisons. Both reward the same work: complete, provable, clearly written product data.

The seller side: agentic workflows instead of one-off prompts

An agentic workflow is more than a chat window. It breaks a task into steps, fetches the data it needs, produces a result, checks it against rules and hands it to you for approval. The difference from a single prompt is that the workflow plans which steps are needed and works through them in order.

Amazon has taken this step in its own back office. On 17 September 2025 Amazon announced that Seller Assistant can set goals, develop strategies and, with the seller's permission, take action, for example on inventory, shipments and account health. According to Digital Commerce 360, it launched for all US sellers at no additional cost, with other countries to follow.

The second development is technical. The Model Context Protocol (MCP), which Anthropic released as an open standard in November 2024, lets AI assistants such as Claude use external tools and data. For sellers, that means an agent you already use can operate research, listing or catalog tools directly instead of you copying data between windows.

The boundary matters. Agentic does not mean unsupervised. Good workflows show what every claim rests on and let nothing out without approval.

What changes for Amazon sellers and brands

Whether a shopper's agent recommends your product or a workflow writes your listing, both work from what is in your data. That makes five things more important.

• Structured attributes: material, dimensions, compatibility, audience and use case belong in their dedicated fields, not just in running text. An agent filters on attributes before it reads copy.
• Claims you can prove: an AI that summarizes your listing repeats your exaggerations. Every claim about material, certification or effect needs a basis, or returns and suppressions follow.
• Listings in every language you sell in: an agent asked in Spanish reads your Spanish listing. Patchy or literal translations cost visibility exactly where you want to expand.
• Machine-readable catalogs: keep product data in one place and distribute it from there to channels and feeds. Google requires maintained Merchant Center data for UCP, and Stripe shares product, price and availability data with agents. Machines spot inconsistencies between channels faster than people do.
• Plain language instead of keyword strings: copy that answers questions (what for, for whom, under which conditions) is understood equally well by assistants and by people.

Checklist: agentic commerce for sellers

• Audit your 20 most important ASINs: are all required and recommended attributes filled in?
• Keep evidence for every product claim, such as a spec sheet, test report or certificate.
• Drop claims you cannot prove instead of merely softening them.
• Write titles and bullets as readable sentences, not keyword chains. Since 27 July 2026 titles in all non-media categories are capped at 75 characters anyway.
• Localize and review translations for each target market instead of relying on raw machine translation.
• Maintain product data in one place and export from there instead of typing it into every channel.
• Keep prices and availability consistent across your store, Merchant Center and marketplaces.
• If you run your own store, set up Google Merchant Center properly. That is where Google's agentic features plug in.
• Pilot an agentic workflow on one task with a clear approval step first, such as one listing or one image set.
• Measure the result: click-through, conversion and returns before and after the change.

Where Torch Technology fits

Torch Technology is the agentic e-commerce suite for Amazon sellers and brands. It works on the seller side: Torch Technology does not build shopping agents or a checkout.

Agentic, at Torch Technology, means: a multi-step AI workflow reads the market, reviews and competitors, produces the content and checks every claim. You approve. Through MCP and the REST API, Claude, Codex and other AI agents can use Torch's tools too.

The six capabilities of the suite and where each stands today:

The easiest place to start is a listing you already have. Check it for free with the Amazon listing check, or try Torch Listings for seven days without being charged.