Watch our AI Manager launch →

Agentic Commerce in 2026: What It Is, the Stack, and the AI Agents Brands Use

Updated

Agentic commerce is shopping where AI agents do the work: an agent finds, compares and buys products for a shopper, within limits the shopper sets, and agents on the brand's side answer, sell and solve problems for the store. In 2026 the buying side got its rails: open protocols from OpenAI and Stripe, Google and Shopify, and checkout inside ChatGPT, Google AI Mode and Gemini.

For most e-commerce brands, though, agentic commerce starts closer to home. Shoppers still land on your site with questions, and after every order, whoever or whatever placed it, someone has to handle the tracking question, the return and the address change. This guide explains the term, the stack behind it, what changes after an AI agent buys, and which AI agents brands use today.

What is agentic commerce?

Agentic commerce is commerce in which AI agents act, not just advise. On the shopper's side, an agent takes a request such as "find me running shoes under 120 euros that arrive by Friday", searches, compares, and completes the purchase within the rules the shopper gave it: a budget, a delivery date, a payment method. On the brand's side, agents greet visitors, recommend products, and resolve support requests by taking real actions in the store's systems.

Agentic commerce vs agentic AI: agentic AI is the general technology, AI that plans and takes steps toward a goal by calling tools and systems. Agentic commerce is that technology applied to buying and selling. Every agentic commerce product is agentic AI, but not the other way round.

The difference from older automation is the same on both sides. A chatbot answers. A rule-based flow follows the path someone built. An agent works out what needs to happen, checks the systems it is connected to, does it, and reports back.

The agentic commerce stack in 2026

It helps to think of agentic commerce as four layers. Most brands touch all four, but only build in one or two.

Layer one: protocols. These are the shared rules that let a shopper's agent talk to a store. The Agentic Commerce Protocol (ACP) was published by Stripe and OpenAI in September 2025 as an open standard, and ChatGPT was the first AI platform to use it. The Universal Commerce Protocol (UCP) was announced by Google in January 2026, co-developed with Shopify, Etsy, Wayfair, Target and Walmart. It powers checkout on eligible product listings in AI Mode in Google Search and the Gemini app for US retailers, and covers discovery, checkout and post-purchase support. The Agent Payments Protocol (AP2), announced by Google in September 2025, handles the payment itself: the shopper's instructions are recorded as signed mandates, so the store and the payment provider can prove what was approved. The Model Context Protocol (MCP) is the general open standard for connecting AI applications to outside systems and tools.

Layer two: agent-ready commerce platforms. Your store has to be readable by agents. Shopify's Agentic Storefronts, a sales channel launched in December 2025, send a merchant's catalog to AI chats such as ChatGPT, Microsoft Copilot, Google AI Mode and Gemini from one setup in the admin. Shopify also offers an Agentic plan so brands on other platforms can list products in Shopify Catalog.

Layer three: agents on your own site. A shopper who arrives at your store can be met by an AI agent that asks what they need, recommends from the live catalog and answers product questions before they buy.

Layer four: agents after the purchase. Orders need tracking, changing, returning and refunding. This is where an AI agent with access to your helpdesk, store, warehouse, carrier and returns platform does the most work, and where most brands start.

What happens after an AI agent buys: support, returns and delivery

Checkout protocols end at the order. Everything after it stays with the brand. A shopper whose agent bought a jacket in Gemini will still ask where it is, want a different size, or need the address changed, and they will ask you, by email, chat or WhatsApp.

For a support agent that is connected to your systems, an order placed by a shopper's AI agent is just an order. It looks it up in the store, checks the carrier, and acts. In practice that means updating a delivery address in the warehouse system before the order is picked, verifying a missing parcel with the carrier and arranging a reshipment, or checking a return against your policy, creating it in the returns platform, sending the label and refunding when the parcel arrives.

This is also why agentic returns are a search term of their own. An agentic return is one handled in the conversation from start to finish: eligibility, the return itself, the label, and the refund or exchange, without sending the shopper to a portal.

From chatbots to copilots to agents: what changed

E-commerce support automation has gone through three phases. First came chatbots: keyword-matching bots that answered FAQs from decision trees. They handled simple questions and frustrated customers whenever a request fell outside the script. Then came copilots: assistants built on large language models that helped human agents write replies faster. They made agents more productive, but people still did the work.

The third phase is agents: AI that connects to your systems and resolves requests itself. What made this possible is tool use. Modern models can call APIs, read data and run multi-step tasks as part of their reasoning. Combined with integrations into the helpdesk, store, shipping, payments, returns and subscriptions, that gives you AI that does the job instead of helping with it.

Chatbots deflected easy questions. Copilots made agents faster. AI agents take routine work off the team entirely, and leave people the cases that need judgment.

What makes a platform truly agentic

The word "agentic" is being slapped on every AI product in 2026, so it's worth defining what actually qualifies. A truly agentic commerce platform needs four capabilities.

First: system access. The AI must connect to your actual tools (helpdesk, e-commerce platform, shipping, payments, returns, subscriptions) and have both read and write access. Looking up an order is not agentic. Processing a refund is.

Second: autonomous reasoning. The AI must decide what to do based on the customer's request and your policies, without following a pre-built decision tree. If you have to build a workflow for every scenario, it's automation, not agency.

Third: multi-step execution. Real customer requests often require querying multiple systems, making decisions based on the results, and taking several actions in sequence. An agent that can only do one thing per turn isn't truly agentic.

Fourth: graceful degradation. The AI must know what it doesn't know. When a request exceeds its capabilities or policies, it should escalate cleanly to a human, with full context, rather than hallucinating an answer or getting stuck in a loop.

Agent verification and trust

If agents can buy, stores need to know which agents to trust. Agent verification is how a merchant tells a legitimate shopping agent from a scraper or a fraud bot. Visa's Trusted Agent Protocol and Mastercard's Agent Pay both use Web Bot Auth for this: registered agents sign their requests cryptographically, so networks like Cloudflare can check who is really knocking. AP2's signed mandates do the same for the payment, recording what the shopper actually approved.

Trust runs the other way too. When an AI agent acts for your store, you decide what it may do on its own. Good platforms let you switch each action on separately, keep a person in the loop for anything above a limit you set, and log every action the agent takes.

The top 8 AI tools for e-commerce support

1. Minimal AI: Best for one set of AI agents before, during and after the purchase

Minimal AI builds AI agents for e-commerce brands that work across the customer journey: before, during and after the purchase. On your site, the Agentic Storefront (in Beta) asks visitors what they are looking for, recommends from the live catalog, upsells before they leave, and answers support questions with the same agents. Dreaver, Boombrush and XXL Nutrition use it. Every storefront conversation shows up in Insights, so you learn what shoppers think before they buy or leave.

After the order, the same agents work inside the helpdesk you already use, such as Gorgias, Zendesk, Intercom, Freshdesk, Trengo or Salesforce, and can do anything a human agent could do there or in a connected system. At Dreaver the agent changes delivery addresses in Channeldock before the warehouse picks, at XXL Nutrition it verifies a missing parcel through the PostNL weight check and arranges the reshipment, and in a return it checks eligibility, creates the return, sends the label and triggers the refund. Minimal AI automates up to 90% of all support tickets.

There is no fixed integration list: Minimal has built more than 400, and any system with an API is connected free, before you sign. Your support team trains the agents in plain language through the AI Manager, and every week it proposes improvements, called weekly tasks, for the team to approve. Each action has its own approval switch, and refunds above an amount you set can go to a person.

Where it stops: there is no voice support today, and plans start at 1,000 tickets a month, so it is built for brands with a real support load.

Top features

  • Agents before, during and after the purchase: Agentic Storefront (Beta) and support
  • Resolves up to 90% of all support tickets end to end
  • Takes any action a human could, in the helpdesk and every connected system
  • No fixed integration list: 400+ built, new ones built free before you sign
  • Runs inside the helpdesk you already have
  • Approval per action and by amount
  • Trained in plain language, with weekly tasks to approve
  • Storefront and support conversations in Insights

Pricing: From EUR 489 a month for up to 1,000 tickets, falling to EUR 0.36 a ticket at volume. Billed per ticket, not per resolution. 14-day money-back guarantee.

Website: gominimal.ai

2. Gladly: Best for large retailers that want one AI across sales and service

Gladly calls itself an agentic commerce platform for large retail brands. One AI covers chat, voice, email and SMS, remembers each shopper across conversations, and handles both service and sales, such as product recommendations and cart recovery. Its customers include Tory Burch, UGG and Ulta Beauty.

Where it stops: there is no public pricing, and it is built for large retailers, so a mid-size brand that mainly needs order, return and shipping actions in its own stack will pay for more platform than it uses.

Top features

  • One AI for service and sales across chat, voice, email and SMS
  • Shopper memory across conversations
  • Product recommendations and cart recovery
  • Built for large retail brands

Pricing: Custom pricing (demo required).

Website: www.gladly.ai

3. Salesforce Agentforce: Best for enterprises already on Salesforce Commerce

Agentforce is Salesforce's agent platform. It runs AI agents across email, chat, voice and self-service, with governance and observability for large teams. It is strongest with Salesforce Commerce and Order Management, where agents can process cancellations, returns and refunds natively.

Where it stops: on Shopify or other storefronts those actions usually rely on AppExchange connectors, which adds setup, and enterprise implementations take longer than self-serve tools.

Top features

  • AI agents across service, with governance and observability
  • Native actions in Salesforce Commerce and Order Management
  • Email, chat, voice and self-service
  • Agent Builder for custom agents

Pricing: $2 per conversation for customer-facing agents, or Flex Credits at $500 per 100,000 credits (about $0.10 per action). Agentforce for Service add-on $125 per user a month.

Website: www.salesforce.com

4. Gorgias AI Agent: Best for Shopify-first brands already on the Gorgias helpdesk

Gorgias is a helpdesk built for e-commerce. Its AI Agent answers and resolves requests across email, chat, WhatsApp, SMS, Instagram, Facebook and phone, with more than 130 pre-built actions such as order edits, cancellations, refunds and subscription pauses, plus custom actions through an API. It works with Shopify, BigCommerce, Magento, WooCommerce and PrestaShop.

Where it stops: the AI Agent only works inside the Gorgias helpdesk, and you pay for the helpdesk plan by ticket volume and for every automated interaction on top.

Top features

  • 130+ pre-built actions: order edits, cancellations, refunds, subscription pauses
  • Custom actions via API
  • Email, chat, WhatsApp, SMS, social and phone
  • Shopify, BigCommerce, Magento, WooCommerce and PrestaShop

Pricing: Helpdesk from $10 a month (50 tickets) to $900 a month (5,000 tickets). AI Agent $0.90 per automated interaction on annual billing, $1.00 monthly. 7-day free trial.

Website: www.gorgias.com

5. Siena AI: Best for Shopify brands that want an e-commerce specialist

Siena AI is an AI agent built only for e-commerce. It handles returns, refunds, address updates, labels and replacements in one flow, and for subscription brands it can skip, pause or swap an order. It runs on top of Gorgias, Zendesk, Kustomer, Intercom, Gladly and Dixa.

Where it stops: Shopify and Fulfil are the only store platforms on its integrations page, and pricing starts with a $750 monthly platform fee before the first automated ticket.

Top features

  • Returns, refunds, address changes, labels and replacements
  • Subscription skip, pause and swap
  • Runs on Gorgias, Zendesk, Kustomer, Intercom, Gladly and Dixa
  • Custom integrations via API

Pricing: Platform fee $750 a month plus $0.90 per automated ticket. Free trial not stated.

Website: www.siena.cx

6. Yuma AI: Best for Shopify, WooCommerce and Magento brands that pay per resolved ticket

Yuma AI is an e-commerce AI agent that works on top of your helpdesk (Gorgias, Zendesk, Kustomer, Salesforce Service Cloud, Gladly, Front and others) and your store on Shopify, WooCommerce or Magento. It checks shipment and subscription status, validates return eligibility, updates addresses, issues refunds and creates labels.

Where it stops: there is no public price, and the plan depends on volume, workflows and platforms, so you need a demo and a quote.

Top features

  • Refunds, address changes, labels and return eligibility in the thread
  • Works on top of nine helpdesks
  • Shopify, WooCommerce and Magento
  • Pay only for fully resolved tickets

Pricing: Outcome-based: billed only when Yuma fully resolves a ticket, no public price. 30-day free trial on real ticket data.

Website: yuma.ai

7. Fin (by Intercom): Best for chat-first support teams, on Intercom or another helpdesk

Fin is Intercom's AI agent, now also sold on its own for other helpdesks such as Salesforce, HubSpot and Freshdesk. It answers on chat, email, voice and social, and Procedures let it read from and write to third-party systems, for example to process refunds.

Where it stops: e-commerce actions such as order edits and returns need Procedures and connectors to your own systems rather than working out of the box, and you pay for every resolution, so the bill grows as Fin resolves more.

Top features

  • Autonomous resolution across chat, email, voice and social
  • Procedures that read and write to third-party systems
  • Works on Intercom and other helpdesks
  • Pay per outcome ($0.99)

Pricing: $0.99 per outcome, 50 outcomes a month minimum. Intercom seats from $29 a month (annual) if you also use its helpdesk. 14-day free trial.

Website: fin.ai

8. Zendesk AI: Best for large support teams already on Zendesk

Zendesk AI bundles AI agents for customers and a copilot for human agents inside the Zendesk Suite. Every Suite plan includes a monthly allowance of automated resolutions per agent, with more sold per resolution. The official Shopify integration lets agents view orders and process refunds without leaving Zendesk.

Where it stops: e-commerce flows come through integrations rather than as default product behavior, which means more configuration up front.

Top features

  • AI agents for customer-facing resolution
  • Copilot for human agents
  • Automated resolutions included per agent, more per resolution
  • Official Shopify integration with in-ticket refunds

Pricing: Suite Team EUR 55 and Professional EUR 115 per agent a month on annual billing, including 5 or 10 automated resolutions per agent a month; extra resolutions EUR 1.50. 14-day free trial.

Website: www.zendesk.com

Comparison table

ToolFocusE-commerce actionsPricingBest for
Minimal AIOn-site and post-purchase agents, on any helpdeskAny action a human could take, in store, warehouse, carrier and returns systems; 400+ integrationsPer ticket, from EUR 489/moBrands wanting one set of agents for sales and support
GladlySales and service platformRecommendations, cart recovery, serviceCustomLarge retail brands
Salesforce AgentforceEnterprise agent platformNative in Salesforce Commerce; Shopify via connectors$2 per conversation or Flex CreditsEnterprises on Salesforce
Gorgias AI AgentPost-purchase agent inside Gorgias130+ pre-built actions: edits, cancellations, refunds$0.90 to $1.00 per automated interaction, plus helpdeskShopify-first brands on Gorgias
Siena AIPost-purchase e-commerce agentReturns, refunds, address changes, subscription skip and pause$750/mo plus $0.90 per automated ticketShopify brands wanting a specialist
Yuma AIPost-purchase e-commerce agentRefunds, address changes, labels, return eligibilityPer resolved ticket, quoteShopify, WooCommerce and Magento brands
Fin (by Intercom)Chat-first support agentRefunds and order changes via Procedures and connectors$0.99 per outcomeChat-heavy support teams
Zendesk AIAI in a general helpdeskShopify refunds via integrationPer agent, resolutions included, EUR 1.50 per extraLarge teams on Zendesk

How to choose the right tool

Start with the layer where your customers already are. For most brands that is not ChatGPT checkout yet. It is the shopper on your product page with a question, and the customer who wants to change an order.

A short checklist for 2026. First, make your catalog agent-ready: clean product data, and if you are on Shopify, decide which AI channels to switch on. Second, put an agent on your own site that can recommend from the live catalog and answer support questions in the same conversation. Third, make sure your post-purchase agent can act, not just answer: test it on a refund, a return with a label, an address change before shipping and a missing parcel. If the answer is "yes, once we build a workflow", that is automation, not an agent. Fourth, check the guardrails: approval per action, limits by amount, and a clean handover to a person with the full context.

The protocols will keep changing for a while. The work after the order will not. Brands that can resolve that work with agents are ready for whatever channel the next order comes from.

Minimal AI: AI customer support built for e-commerce

Set up your AI agent in plain language. No engineers, no vendor dependency.

Frequently asked questions

What is agentic commerce?+

Agentic commerce is shopping and selling where AI agents act: a shopper's agent finds, compares and buys within limits the shopper sets, and a brand's agents recommend products and resolve support requests by taking real actions in the store's systems.

What is the difference between agentic commerce and agentic AI?+

Agentic AI is the general technology: AI that plans and takes steps toward a goal by using tools and systems. Agentic commerce is agentic AI applied to buying and selling, from product discovery and checkout to returns and refunds.

Which platforms support agentic commerce?+

On the buying side, OpenAI's ChatGPT uses the Agentic Commerce Protocol from Stripe and OpenAI, and Google uses the Universal Commerce Protocol for checkout in AI Mode and Gemini. Shopify's Agentic Storefronts send a merchant's catalog to those AI chats. On the brand side, AI agents such as Minimal AI, Gladly, Salesforce Agentforce, Gorgias AI Agent, Siena AI, Yuma AI, Fin and Zendesk AI handle sales and support.

What is AI agent verification in agentic commerce?+

It is how a store or payment network checks that a shopping agent is legitimate. Visa's Trusted Agent Protocol and Mastercard's Agent Pay use Web Bot Auth, where registered agents sign their requests cryptographically, and Google's AP2 records the shopper's approval as a signed mandate.

Can I add agentic AI on top of my existing helpdesk?+

Yes. Several AI agents work inside the helpdesk you already have. Minimal AI, for example, runs inside Gorgias, Zendesk, Intercom, Freshdesk, Front, Kustomer, Dixa, Trengo, Richpanel, Salesforce and others, and every Minimal AI customer keeps their own helpdesk.

What are agentic returns?+

An agentic return is handled by an AI agent in the conversation from start to finish: it checks eligibility against your policy, creates the return in your returns platform, sends the label and triggers the refund or exchange, without sending the shopper to a returns portal.

Is it safe to let AI agents take actions?+

It is when the guardrails are clear. Good platforms let you switch each action on separately, send actions above an amount you set to a person for approval, start with the agent suggesting rather than acting, and log every action it takes.

Related reading

The next generation of AI E-commerce support is here

Automate up to 90% of tickets
Easy to use with the AI Manager
No engineers or long setup required