Amazon's Seller Assistant Workflows: What Launched and What Amazon Hasn't Said

Amazon's own assistant can now watch your account around the clock and act on instructions you write in plain English. Amazon announced Seller Assistant workflows at Accelerate on September 23, with examples ranging from an alert when a top product drops below four stars to a rule that watches a category for competitive openings, adjusts your pricing, and refreshes your listings.

Before a workflow touches a real decision, it's worth knowing what Amazon has said about approvals, audit trails, and what these workflows optimize for, and what it hasn't.

Quick Answer

• What launched: On September 23, 2026, Amazon added workflows to Seller Assistant: plain-language instructions it monitors around the clock, inside guardrails you set, with an audit trail of every action. Persistent memory and a plugin for Amazon Quick and Claude shipped alongside.
• How control works: each workflow either surfaces recommendations or takes actions, and Amazon says you review and approve actions before they're carried out.
• What it covers: Amazon's examples and templates span ratings, account health, orders, pricing, restocking, listings, and product expansion.
• Where it stops: Amazon hasn't said workflows can act on other channels, apply across multiple accounts, or account for costs that live outside Seller Central.

What Amazon Announced

In its announcement, Amazon describes workflows as "custom automations that run continuously," watching for conditions and responding even when you aren't logged in to Seller Central. Until now, Seller Assistant acted when you asked it to. Workflows are meant for the checks that follow the same pattern every time, such as inventory thresholds, competitive pricing, and account health alerts, so you no longer have to remember to ask.

How a workflow is built

You write the instruction in plain language and set guardrails. Amazon's simple example alerts you when any of your top 10 products drops below four stars and drafts a response plan. Its complex example watches your top category for competitive openings and, when it finds one, adjusts your pricing and refreshes your listings. Prebuilt templates cover account health, order summaries, pricing trends, and product expansion opportunities, and you can build custom workflows from scratch.

Recommendations or actions, with approval

For every workflow, you choose whether it only surfaces recommendations or takes actions on your behalf. Amazon says that when a workflow takes actions, you review and approve them before they're carried out, and every action is logged with an audit trail. The announcement doesn't describe a mode that runs actions without that approval, and Amazon says workflows come at no additional cost.

Memory and the plugin

Seller Assistant now keeps persistent memory of your pricing patterns, inventory cycles, and growth goals, and Amazon says that context follows you into Amazon Quick and Claude. The new Amazon Selling Partner plugin connects Seller Assistant to Quick and, in beta for sellers in Amazon's US stores, to Claude, so those tools can read and act on your Amazon data with human approval on actions. That gives the "should I let AI touch my account" question we've covered before a new route, alongside the MCP connections sellers already use. Seller Assistant's ability to act on your behalf dates to Accelerate 2025.

Amazon also reports that Seller Assistant has reached over 90% of selling partners, has hundreds of thousands of active users, and sees its recommendations accepted over 90% of the time. Those figures cover the assistant as a whole rather than workflows, they come from Amazon, and they measure acceptance rather than results.

A Familiar Design: Convergent Evolution

Biologists use the term convergent evolution for unrelated species that develop the same trait independently because they face the same pressures, the way birds and bats both arrived at wings. Amazon's workflows are a case of it in eCommerce software. Plain-language instructions, guardrails, a choice between recommending and acting, approval before anything changes, and a log of every run are the design Qore, our product for turning an operator's reviews into workflows that run on a schedule, was built around.

When two teams reach the same design separately, the design is probably right. The differences that matter then come down to where the workflow lives and what it can see.

The Questions Amazon Hasn't Answered

A launch that lets an assistant act on your pricing, inventory, and listings raises more questions than one announcement can answer. These are the ones that decide how much of your business you should route through it.

How does approval work at volume?

Amazon says actions wait for your approval. It hasn't said whether that approval happens per action, in batches, or as a standing rule. If a category-watch workflow proposes 40 price changes over a weekend and each one needs sign-off, the round-the-clock monitoring still ends in a queue someone has to clear on Monday morning.

What can a guardrail enforce, and whose goal wins?

Amazon doesn't define what a guardrail can hold. Take its own pricing example. When a workflow spots a competitive opening and lowers your price, Amazon benefits either way, because shoppers see a better price and the store looks more competitive. You benefit only if the new price still clears your margin after fees and landed cost. Amazon's interests and yours overlap much of the time, but the announcement doesn't say what a workflow optimizes for when they don't.

What does the audit trail record, and can you take it with you?

Amazon says every action is logged. It hasn't said what the log contains, how long it's kept, or whether you can export it or pull it through an API. That matters the first time a price change costs you margin: you need to see what the workflow saw, which rule fired, and who approved it. Agencies need the same record for client reporting, and finance teams need it to reconcile price and inventory changes against the P&L. A log you can only read inside Seller Central serves none of those well.

What sits between your instruction and the action, and can you read it?

Amazon says Seller Assistant is built on Amazon Bedrock with Anthropic's Claude models, combined with its own commerce expertise. It hasn't described what happens between the plain-language instruction you write and the action a workflow proposes. Tell a workflow to watch your top category for competitive openings, and something has to decide which competitors count, which data to pull, how far a price can move, and what qualifies as an opening. Amazon hasn't said whether your instruction becomes a fixed set of steps you can read before it runs, or whether the model interprets it fresh each time.

The difference shows up the day you open the audit trail. An action log can tell you a price moved from $24.99 to $22.49 on Tuesday. It won't necessarily tell you which data the workflow looked at, which condition it decided was met, or whether it would make the same call with the same inputs next week. Amazon says every suggestion comes with reasoning, which helps. An explanation written after the decision still falls short of logic you can inspect, test, and sign off on before the workflow touches your account.

What does memory keep, and can you correct it?

Seller Assistant now learns your pricing patterns and growth goals over time. Amazon hasn't said whether you can see what it has learned, correct it, or reset it. If it learned from a clearance quarter when you priced below target on purpose, you would want to know before that pattern shapes next quarter's recommendations.

What leaves Amazon through the plugin?

Amazon says Seller Assistant keeps seller data within Amazon's infrastructure and never shares it outside Amazon. It also says the plugin connects your listings, performance metrics, inventory levels, and sales analytics to Quick or Claude. The announcement doesn't explain how those two statements fit together: which data reaches the outside tool, which stays inside Amazon, and whose terms govern it once it's there.

Does it work across accounts?

Amazon describes workflows one selling account at a time and hasn't said whether a workflow can be written once and applied to several. For an agency running 30 client accounts, that is the difference between maintaining one standard and maintaining 30 copies that drift apart.

Will it still be here in two years?

Amazon retires seller-facing programs when its priorities move. It stopped new Amazon Posts in June 2025 and shut the program on July 31, 2025, citing declining impressions and an upcoming page redesign, Adweek reported. Inspire, its shoppable video feed, launched in late 2022 and was gone by February 2025, according to Modern Retail. A workflow that encodes how you price and restock is a far bigger dependency than a content feed. If Amazon changes or retires it, your logic and its history go with it.

Where Amazon's Workflows Run Out

Every workflow runs on what Amazon can see inside the account it's connected to. Quick can put your Amazon data next to your other systems, but the workflows Amazon announced run inside Seller Central. That covers a lot of an Amazon business, and it leaves out the parts that usually decide whether a pricing or restock decision was right.

• Your other channels. A workflow that cuts your Amazon price to win a competitive opening can't see that the same SKU now sells for more on your Walmart listing and your Shopify store, or that the new price breaks the parity you promised a retail partner. Amazon hasn't said workflows read or act on any other marketplace.
• Your true costs. Amazon knows its own fees. Your landed cost, freight, 3PL charges, and last month's supplier price increase sit outside Seller Central, and those set the margin floor a pricing rule has to respect.
• Your whole roster. An agency or multi-brand operator needs one reviewed standard applied across every account and updated once when the rule changes. Amazon has described workflows one login at a time.
• Your team's standard. The way your best operator reviews a pricing drop usually lives in their head. A workflow saved in one Seller Central login doesn't turn that review into something the whole team runs, versions, and hands to the next hire.
• Logic you can read before it runs. A workflow defined as explicit steps, covering the data it pulls, the conditions it checks, and the limits it respects, behaves the same way on every run and can be reviewed like any other process document. One that a model interprets fresh on each run is harder to test beforehand and harder to explain afterward.
• A referee with no stake. Amazon earns from your ad spend and fees, and it competes as a retailer in many categories. A workflow that runs outside the marketplace answers only to your rules.
• A record you own. When the logic and its history live outside Seller Central, they stay with you if you change tools and survive if Amazon retires the feature.

For a single-account seller who mostly wants rating alerts and order summaries, Amazon's workflows may be all you need, and they cost nothing extra. The gaps above appear as soon as the business extends past one account or one marketplace.

Recommendations or Actions: Deciding Workflow by Workflow

Amazon lets you set each workflow to surface recommendations or to take actions you approve, so the choice is yours to make one workflow at a time. The stakes differ by task. A wrong recommendation costs you the minute it takes to dismiss it. An approved price cut on your best seller costs margin on every unit sold until someone notices it went through.

Where Qore Fits

Qore runs the same kind of workflow against what Amazon can't reach: your own definitions, applied across Amazon, Walmart, Google, Shopify, and TikTok, and across every account in your roster. You describe the review once, and Qore runs it on a schedule. Actions wait for your approval by default, and you can let a check act on its own once you trust it. Q, the assistant inside Qore, helps you turn a review you do by hand into a workflow your whole team can run.

Qore can't see inside Amazon's assistant or override what it decides, and it doesn't replace Seller Central. If you run one Amazon account and mostly want rating alerts and order summaries, Amazon's workflows cover that without us. Qore earns its place when the standard has to hold across channels, across accounts, and against costs Amazon doesn't see.

Conclusion

Seller Assistant workflows are a meaningful addition to Seller Central, and they confirm the direction workflow tools in this category have been heading. Use them where Amazon sees everything that matters: rating alerts, account health, and order summaries in a single account. Keep pricing and restock actions at recommendations until you've confirmed the guardrails hold your margin, and get answers to the open questions before a workflow becomes part of how you run the business. For anything that has to hold across channels, accounts, or costs outside Seller Central, keep the standard somewhere you own. For the wider map of what's worth automating on Amazon, see our guide to Amazon automation.