Amazon Opened the Seller Back Office to Outside Agents, and Kept the Buyer-Side Door Shut

At Amazon Accelerate on September 23, Amazon opened its Seller Central APIs to outside AI agents, launching a US beta plugin that lets sellers manage inventory, prices, listings and analytics through Anthropic's Claude or Amazon's own Quick assistant without logging into Seller Central. Mary Beth Westmoreland, vice president of Worldwide Selling Partner Experience, framed the goal in the announcement: the vision was that sellers would never have to log into Seller Central, and that Amazon would bring the capability to where they work. The plugin connects a seller's listing contributions, real-time performance metrics, inventory levels and sales analytics to their chosen AI agent, and goes beyond data access — the AI can act on the business the same way Seller Assistant does inside Seller Central. The Claude plugin connects via OAuth and covers 11 skills including sales-drop diagnosis, FBA stockout risk checks and suppressed-listing fixes, is built on Amazon's official Selling Partner agentic toolkit, signs in through the normal Seller Central permission screen, and the assistant can only use tools the seller's Seller Central role allows, drafting changes for approval. On governance, sellers grant access through a new Manage Agents page in Seller Central, choose which data types the agent can read, and must approve every action before it is carried out, with every plugin interaction protected by access boundaries, human approval and complete audit trails. Seller Assistant itself gained persistent memory, always-on workflows and the new plugin, plus a Canvas visualization layer; Amazon says it now reaches over 90% of selling partners worldwide in their native language, with hundreds of thousands of active users accepting recommendations over 90% of the time. On availability, it is a US beta requiring a Professional selling plan, with some sellers waitlisted; every primary selling account holder worldwide gets a free 12-month Amazon Quick Plus subscription through December 31, 2026, including two additional co-workers. On the stack, the plugin runs on Amazon Bedrock combining Amazon Nova models with Claude for reasoning and execution, and Amazon is a major investor in Anthropic. Two caveats: advertising data is explicitly excluded from the current beta, and once an outside assistant is connected, the data is handled under that assistant's terms and the seller is responsible for anything the agent does on their account. Amazon also kept its consumer storefront closed to outside agents, days after blocking Meta's consumer shopping agent Muse from its store.

The Door That Opened Is the Seller Side; the One That Stayed Shut Is the Buyer Side

What deserves attention here is not what was opened but what was kept closed at the same time. The seller back office is now open to outside agents. The consumer storefront is not — and this comes days after Amazon blocked Meta's consumer shopping agent Muse from its store. **The same company gave opposite answers to two kinds of agents in the same week.** The dividing line is not technical, it is whose attention is at stake. Sellers are its supply side, and pushing tools to where sellers already work raises supply efficiency while turning Amazon's data and execution into a dependency inside someone else's agent. Buyers are its demand side, and if purchase decisions happen inside someone else's agent, Amazon degrades from "the app a consumer opens" into a back-end supplier. So the accurate description of this opening is: **willing to give up the interface, unwilling to give up the entry point to demand.** In its own words, sellers will "never have to log into Seller Central" — the interface is being deliberately surrendered, while the data layer and the execution layer underneath remain Amazon's, and arguably more secure than before, since other agents now have to call its toolkit. For anyone building a platform, that is a directly transferable judgment: **separate your interface from your position, and work out which one is actually the moat.** The former is what you can afford to give away.

The Permission Model Is the Part Most Worth Copying

Strategy aside, this has concrete value for anyone actually deploying agents, because it lays out a relatively complete authorization model — the piece most teams handle worst when wiring up their own. Four layers: **One, a dedicated grant surface.** Sellers authorize through a new Manage Agents page in Seller Central. Agents are not mixed in with ordinary API keys; they have their own management view. **Two, data-type selection.** The seller picks which data the agent may read, rather than all-or-nothing. **Three, inherited rather than newly minted permissions.** The assistant can only use tools the seller's own Seller Central role already allows — an agent does not get extra reach for being an agent. This one matters most: it caps an agent's authority at the account that granted it, with no lateral escalation. **Four, human approval on actions.** Changes are drafted for approval, every action must be approved, and there are complete audit trails. Writing about Plugin4Shell on September 21, this site made the point that once your agent supports auto-updating third-party extensions, your trust boundary stops being "what did I install" and becomes "who did each link in this chain decide to trust." Amazon's model answers that directly — **it separates grant, scope, approval and record into four things handled individually.** When building your own agent integration, those four layers work as a checklist.

But Two Boundaries Leave the Risk With You

Detailed governance does not mean risk went away. Two sentences have to be read carefully. **First, advertising data is explicitly excluded from the current beta.** So the plugin can tell you a listing is suppressed but cannot answer why your advertising cost spiked. For most sellers, advertising is the second largest variable after inventory, which means this "automated back office" currently has a hole in it — and the hole is where the money goes. **Second, once an outside assistant is connected, the data is handled under that assistant's terms, and you are responsible for whatever the agent does on your account.** That draws the line cleanly: Amazon supplies boundaries and audit, and does not backstop an outside agent's behavior. **Audit trails let you determine afterward who did what; they do not prevent it beforehand** — the only preventive gate is the human approval step, so do not be quick to switch it off.

Read the "90% Acceptance Rate" on Its Terms

Amazon says Seller Assistant reaches over 90% of selling partners worldwide in their native language, has hundreds of thousands of active users, and that **recommendations are accepted over 90% of the time.** Ninety percent sounds high, but it measures how many of the recommendations that were offered got accepted, not how accurate the AI's judgment is. A system that only speaks up when confidence is high will naturally show a high acceptance rate; the number mostly says **the threshold for offering a recommendation is set conservatively**, not that the model understands the business better than the seller. That is not a criticism — for a system executing actions across hundreds of thousands of seller accounts, conservative is the correct design. But do not read the figure as a capability metric. One commercial detail to record: every primary account holder worldwide gets a free 12-month Amazon Quick Plus subscription through December 31, 2026, including two co-worker seats. Free for a year and then billed is the standard move for making an agent part of a daily workflow — a year from now, these sellers' operating habits will have grown around the tool.

via: Amazon's announcement, GeekWire, PYMNTS's analysis, API Evangelist's read