Amazon sellers can now run their store from Claude

On Wednesday morning, during Amazon's Accelerate seller conference in Seattle, market-news account @StockMKTNewz posted a summary of Amazon's update to Seller Assistant. Amazon's claim, as the post relays it: the Claude-powered assistant "can run a seller's business 24/7."

That line is marketing. The details under it are more interesting, because they show how a very large platform decided to let an AI agent act on a business owner's behalf.

What Amazon actually shipped

There are three pieces, and they are easy to blur together.

First, workflows. Seller Assistant already lived inside Seller Central and drafted recommendations. It can now watch conditions continuously and act when the seller is not logged in. The examples in the post range from simple ("alert me if a top product drops below 4 stars") to a pricing agent that watches a category and adjusts prices automatically.

Second, memory. The assistant keeps track of each seller's pricing patterns, inventory cycles and growth goals, and that memory follows the seller between Seller Central, Amazon Quick and Claude.

Third, the plugin. Amazon's own announcement says sellers can connect their listings, inventory, performance metrics and sales data to Amazon Quick, and in beta to Anthropic's Claude. The AI can take action there, not only read the data. Setup takes about 60 seconds with no coding, it is U.S.-only for now, and every primary account holder gets a free year of Quick Plus through December 31.

The adoption numbers in the post are large. Amazon says Seller Assistant has reached over 90% of its selling partners worldwide, and that sellers accept its recommendations more than 90% of the time.

The part worth copying: the control switch

The most useful detail is a small one. Per the post, sellers "choose upfront whether a workflow just surfaces recommendations or actually takes action." Every action lands in an audit trail. Amazon's announcement adds that the plugin has set limits on what it can access and needs human approval before it acts.

That is a sensible way to hand work to an agent, and it isn't specific to Amazon. You start with the agent suggesting. Once its suggestions are right often enough, you let it act on the low-risk jobs and keep a log you can check. Amazon's 90% acceptance rate is what tells it where that point is. Most businesses never measure it, so they either trust the tool blindly or never trust it at all.

Open for sellers, closed for shoppers

This also fills in a story we covered on Monday. Amazon blocked Meta's Muse shopping agent from Amazon.com over the weekend. A few days later it opened the seller side to Claude.

The two moves fit together. API Evangelist pointed out that Amazon now has solid agent connections for sellers and advertisers, and nothing comparable for shopping agents. Amazon lets agents in where it owns the merchant relationship and collects the fees. It keeps them out where an agent would stand between Amazon and the buyer. GeekWire notes that seller fees brought in $46.8 billion last quarter, and that Amazon still decides which AI platforms can plug in.

So sellers get an agent that can reprice at 2 a.m. The rules it runs under are still Amazon's.

What this means for a small business

You probably don't sell on Amazon, but the pattern still applies. The software you already pay for is starting to take instructions from an AI assistant, and that assistant keeps working after you close the tab. Whether it helps comes down to two things:

• Where the switch sits. Before you turn anything on, write down which jobs the agent only suggests and which it does on its own.
• What gets logged. If the tool can't show you what it did and why, keep it in suggest-only mode.

Ninety percent of Amazon sellers already use outside AI tools, Amazon says. The ones who get the most from this update will be the ones who already know which of their tasks are safe to hand off.

If you aren't sure which of your own tasks are ready, New Face Design's free process audit goes through them with you and sorts them into what an agent could suggest, what it could handle, and what should stay with you.