AI agents for ecommerce that draft the replies, refunds, quotes and reorders your team approves.

Three kinds, going by their own pages as read on 11 Oct 2026. Your platform ships some: Shopify has Sidekick and an Inbox agent, and BigCommerce has Smart Edits in beta. Helpdesk tools such as Gorgias and Fin answer customer messages. A custom agent, built for you, covers jobs with rules of your own, such as wholesale quotes or reorder drafts. Each one acts through the store's API, so the key you give it sets what it can change.

The work falls into four parts. A customer-facing agent handles order status, returns and product questions. Back-office agents handle inventory, reorders and marketplace listings. B2B agents draft quotes and repeat orders. The fourth part is the links those agents need to your ERP, helpdesk and marketplaces. FactoryJet designs, builds and supports all four on WooCommerce, BigCommerce, Adobe Commerce, Shopify and headless stores, and shows a first version on your own data before a contract.

Several do, going by their own pages as read on 11 Oct 2026. Kibo Commerce describes an agentic framework across commerce and order management, with order routing reachable by outside AI tools. Fluent Commerce announced an AI agent for order sourcing and documents a connection that gives customer service agents order status and tracking data. Manhattan lists a CSR Agent for order status and returns questions. If your store runs WooCommerce, BigCommerce or Adobe Commerce without one of those suites, a custom agent connects to the store's own API.

Start with the job, then pick the lightest tool that does it. For questions about your own store data, your platform's built-in assistant may be enough. For customer messages, a helpdesk AI tool is quick to switch on. For work that follows your own rules, such as checking returns against your policy or drafting wholesale quotes at each customer's price, a custom agent is the right call. On the scoping call we say which one we would choose for you.

In two places. On the storefront, AI answers shoppers: where an order is, whether a product fits, what the return policy says. Behind the storefront, it does office work. It flags low stock and drafts the reorder, turns an emailed purchase order into a draft order, or rewrites product data for a marketplace listing. The second kind is what people now call an AI agent, because it takes an action in your store as well as answering.

Ecommerce automation means software does a repeatable store task when a trigger fires. Classic automation follows fixed rules: when an order is paid, tag it and email the warehouse. An AI agent handles the tasks that need reading and judgment, such as a return request written in the customer's own words. Both act through the same doorway, which is the store's API and its webhooks, the messages a store sends when something changes.

Agentic commerce means an AI agent shops on a buyer's behalf. It searches a catalog, builds a cart and hands the buyer to checkout. Platforms are building doorways for it. BigCommerce documents an MCP server that lets agents search products and manage a cart, and Shopify routes these tools through its Universal Commerce Protocol. That is a different job from the back-office agents on this page. Our agentic commerce guide covers the shopper side.

It depends on what the chatbot must do. If it only answers from your policies and catalog, an off-the-shelf tool does that well. Shopify's Inbox agent answers questions about products, shipping, returns and store policies, and Gorgias and Fin plug into a helpdesk. If it must check an order, start a return or quote a wholesale price under your own rules, it needs to be wired to your store's API. That part is what we build.

AI can draft parts of one. Shopify's Sidekick can generate content such as product descriptions and images, and BigCommerce's Smart Edits turns a plain-language request into a plan of bulk product updates that you review before it applies. A full store still needs decisions that stay with you, such as the platform, payments, tax and shipping rules. FactoryJet builds stores as well as the agents that work inside them.

Merchandising is deciding which products show where, at what price and with what words. AI helps with the repetitive parts: writing product descriptions, fixing missing attributes, re-ranking search results and suggesting related products. Adobe documents AI-powered re-ranking in Live Search, and BigCommerce's Smart Edits drafts bulk product changes for review. An agent can also prepare a weekly list of products with thin data or low stock for your merchandiser to act on.

Yes. WooCommerce has a REST API that lets software create, read, update and delete store data, using keys you generate with Read, Write or Read/Write access. Its documentation also describes built-in support for MCP, a standard that lets AI tools call a system directly, and labels it a developer preview that may change. We connect through the REST API, with a key limited to the job, and webhooks so the agent hears about new orders as they happen.

Yes. BigCommerce has REST and GraphQL APIs, webhooks and, for B2B Edition stores, a Quotes API. Its documentation sets a request quota by plan: 150 requests per 30 seconds on Standard and Plus, and 450 on Pro. Every app on a store shares that quota, so an agent has to be a polite neighbor to your other apps. BigCommerce also documents an MCP server for storefronts, which a store owner switches on in settings.

Yes. Adobe says Adobe Commerce on Cloud, Adobe Commerce on-premises and Magento Open Source all use the same REST API framework, so one approach covers all three. Adobe Commerce as a Cloud Service is different. Adobe says it has a smaller set of REST endpoints and signs requests through Adobe's identity service. With B2B for Adobe Commerce installed, REST endpoints cover the seller's side of a negotiable quote.

Yes, and Shopify has its own page on our site. This page covers every platform. Our Shopify AI agents page goes deeper on the GraphQL Admin API, draft orders, Shopify Flow and Sidekick. One fact for planning: Shopify's documentation meters its Admin API by plan, from 100 points a second on standard plans to 1,000 on Shopify Plus, and a single query may not cost more than 1,000 points.

Yes. Headless means the storefront shoppers see is custom-built and talks to the commerce platform through an API. A back-office agent does not touch the storefront at all. It uses the platform's admin API, the same as on any other store. A shopper-facing agent is the part that changes. It is built into your storefront code, because a chat widget made for a standard theme may not fit a custom front end.

Yes, and it is a sensible first job. The agent looks up the order and its tracking in your store, checks the carrier status and answers in the customer's own channel. It only reads, so the risk is low. The care goes into identity: the agent confirms who is asking before it shares an address or an order detail. WooCommerce's documentation warns that order and customer operations may expose personal information.

It can check a return against your policy and prepare the refund. We set it so a person approves before money moves, until the logs show it follows your rules. This matters on WooCommerce, where the REST API's refund call uses the payment gateway to refund by default. A key with write access to orders can therefore send money back, so the agent's key and its approval step are designed together.

Yes. The agent reads the buyer's request, matches each line to a product, applies that customer's price list and saves a draft quote for your sales rep. BigCommerce B2B Edition has a Quotes API for creating and updating sales quotes. Adobe Commerce B2B has negotiable quotes, and Adobe says its REST endpoints cover the seller's side. Reorders work the same way. The agent drafts from the last order and a person confirms.

It can draft and submit listing updates through each marketplace's own API. Amazon's Listings Items API takes listing updates one at a time and can preview errors before a listing is created. Amazon also warns that a full update replaces the listing, so attributes left out may be dropped. Walmart's Marketplace APIs cover items, inventory, orders and pricing. On eBay, a published offer is what becomes a live listing. We keep a person on price changes.

It can watch stock levels and draft the purchase order. It reads on-hand quantities and recent sales from the store, compares them with the reorder points you set, and prepares the order for a buyer to approve. Where your stock lives in an ERP, the agent reads it there. BigCommerce's documentation advises absolute adjustments as the default way to update inventory, so an agent that writes stock back should use that method.

It is as safe as the key you give it. Every platform limits what a key can do. WooCommerce keys are Read, Write or Read/Write, BigCommerce API accounts carry OAuth scopes, which list what the key may touch, and Shopify apps request specific access scopes. We give each agent its own key with the least access the job needs, start it read-only, log every action, and keep refunds, discounts and price changes behind a person.

It will, so plan for it. On 11 Oct 2026, Shopify's documentation said the catalog and cart tools on its Storefront MCP endpoint were removed in favor of its Universal Commerce Protocol, and WooCommerce's MCP page called its own integration a developer preview that may change. After launch we watch the platform's change notes, re-test the agent when something moves and fix what breaks.

An off-the-shelf tool has a list price. On 11 Oct 2026, Fin's pricing page listed $0.99 per outcome, and Gorgias listed its Starter plan with AI Agent at $40 a month. A custom build is a project. CMARIX's 2026 guide puts a basic FAQ bot at $5,000 to $15,000 and a custom AI chatbot at $20,000 to $80,000. FactoryJet quotes a fixed price in writing after a short scoping call.

It depends on how many systems the agent touches and whether it changes data in them. Building a custom AI agent costs roughly $5,000 to more than $180,000, according to development firm ProductCrafters' 2026 breakdown. That range covers every kind of agent and is not specific to ecommerce. What moves a project up the range is the number of systems and the number of actions that change data. FactoryJet quotes a fixed price in writing after a short scoping call.

We quote it with the build, so you see both numbers before you decide. FactoryJet quotes the build and the ongoing support as fixed prices in writing after a short scoping call. After launch we keep managing the servers, the AI models, the API connections and the maintenance, so your team does not have to.

Buy the app when it does the job. A helpdesk AI tool answers policy and order questions from day one, and your platform's own assistant is already in your admin. Build when the job follows rules only your business has, crosses two systems such as your store and your ERP, or should start without a person typing a prompt. The two can sit side by side, with an app for chat and a custom agent for quotes or returns.

Yes. FactoryJet builds EDI integrations (the structured order files large buyers send), punchout catalogs, and Xero, HubSpot and WhatsApp integrations. We have worked on Odoo, NetSuite, SAP Business One, ERPNext and custom ERPs. For a store agent that usually means reading stock and customer prices from the ERP and writing draft orders back. Our ERP AI agents page covers that side.

Yes. Before a contract, we show working software on your own data. For a store agent that usually means a handful of your real orders, return requests or wholesale emails run through a first version, so you can compare its drafts with what your team did.

Yes. Platforms change their APIs and AI model makers retire versions, so an agent needs looking after. We watch its logs, re-test it when your platform or the model changes, and fix what breaks. Our AI agent monitoring and support page covers that service.

You do. The agent is built for you and the code is yours. We still look after it day to day: the servers, the AI models, the API connections and the updates. If you later move the work in-house or to another firm, the code and its notes go with you.