Amazon Opens Its Walled Garden: Sellers Can Now Run Their Stores From Claude

For most of its history, running a business on Amazon has meant living inside Seller Central. Whatever else you used — your accounting software, your supplier spreadsheets, your favorite AI assistant — the actual store lived behind Amazon’s own walls. At its annual Accelerate seller conference in Seattle on September 23, Amazon took a sledgehammer to that assumption: it announced a new selling partner plugin that lets merchants check inventory, adjust prices, and update listings from outside AI agents, starting with Anthropic’s Claude and Amazon’s own Quick assistant, without ever opening Seller Central.

The scale of the bet is easy to underestimate. Independent sellers account for more than 60% of units sold on Amazon worldwide, and the fees Amazon charges them generated $46.8 billion in the second quarter of 2026 — more revenue than AWS. Whatever changes in the seller experience ripples through millions of businesses. And the company is explicit about why it is loosening its grip now: Amazon’s own data shows that roughly 90% of its selling partners already use third-party AI tools to run parts of their operations. The choice was never between AI and no AI — it was between sanctioned, guarded connections and an unmanaged sprawl of screen-scrapers and copy-paste workflows.

What actually shipped

Three things arrived together, and they build on each other.

A smarter Seller Assistant with persistent memory. Seller Assistant, the generative AI companion Amazon first launched in 2023, has already rolled out to over 90% of Amazon’s selling partners worldwide in their native language and claims hundreds of thousands of active users, with sellers accepting its recommendations over 90% of the time. The upgrade gives it durable memory of each seller’s pricing patterns, inventory cycles, and growth goals — context that follows the seller across Seller Central, Quick, and Claude rather than evaporating at the end of a session. A new canvas feature renders a dynamic visual workspace from live data, letting sellers model pricing scenarios or diagnose ad campaigns conversationally, redrawing itself as they ask follow-up questions.

Seller Assistant workflows. Where the assistant previously acted only when asked, the new workflows are custom automations that run continuously in the background — monitoring conditions, interpreting changes, and responding even when the seller isn’t logged in. A workflow can be as simple as “alert me if any of my top 10 products drop below a 4-star rating and draft a response plan,” or as complex as “monitor my top category for competitive openings; if you see one, adjust pricing and refresh the listings.” Sellers set the guardrails, choose whether a workflow only recommends or actually acts, and every action lands in a complete audit trail, with human approval required before anything is carried out. Prebuilt templates cover account health monitoring, order summaries, pricing trend analysis, and product expansion.

The selling partner plugin. This is the headline. The plugin connects a seller’s Amazon data — listing contributions, real-time performance metrics, inventory levels, sales analytics — directly to their preferred AI agent. And critically, it goes beyond read access: the external agent can act on the account the same way Seller Assistant does inside Seller Central, under the same approval and audit constraints. Connecting Claude takes about 60 seconds and requires no coding. The plugin is live in beta for Amazon’s U.S. stores, with international expansion promised, and more integrations beyond Claude and Quick to follow — Amazon says the architecture was built to be modular enough to keep publishing plugins.

Why Claude first

The choice of launch partner is not accidental. Amazon is a major investor in Anthropic — it doubled down with a $25 billion investment as part of a broader cloud alliance — and Claude models already power Seller Assistant inside Seller Central, running on Amazon Bedrock. Anthropic’s field CTO Eric Burns framed the plugin as meeting sellers where they already are: “Amazon sellers make dozens of decisions a day on pricing, inventory, and ads… they accept its recommendations more than 90% of the time. The Seller Assistant plugin for Claude now brings these data-driven insights into sellers’ Claude account alongside their existing accounting and supplier data connections.”

Amazon’s own framing is more pointed. “Our vision was that they would never have to log into Seller Central,” said Mary Beth Westmoreland, VP of Worldwide Selling Partner Experience. “We would just bring it to them where they work.” Every primary account holder globally can claim a free 12-month Quick Plus subscription through December 31, 2026, covering two additional co-workers at no cost — a not-so-subtle push to make Amazon’s own assistant the default hub even as it opens the gates to rivals.

The strategic read

Three undercurrents make this more consequential than a routine feature drop.

First, it formalizes the agent-as-customer. Just days before the announcement, Amazon blocked Meta’s Muse — a consumer AI agent that shops on behalf of users — from its store, saying outside agents must identify themselves and follow site rules. Read together, the two moves sketch a coherent policy: unsanctioned agents are treated as threats, while sanctioned ones get first-class APIs, guardrails, and audit trails. That is likely the template every major platform will converge on as agentic traffic grows.

Second, data boundary marketing is now a feature. Amazon was careful to stress that seller data stays within Amazon’s infrastructure, that the plugin’s access is scoped by clear boundaries, and that Amazon doesn’t see the seller’s other business data living in their assistant. In a year when OpenAI’s agents are breaching government portals and drawing prime-ministerial rebukes, permissioning and auditability are no longer compliance footnotes — they are the pitch.

Third, the FTC shadow. The announcement lands while Amazon is defending itself against the Federal Trade Commission’s antitrust case, which alleges the company overcharges sellers and punishes them for lower prices elsewhere — a case headed to trial in Seattle in March 2027. Opening seller tooling to third-party agents is, at minimum, a handy counter-narrative to claims of a locked-down marketplace, even as Amazon retains sole discretion over which platforms get plugins.

For sellers, the near-term math is simpler. Beta testers report converting afternoons of manual data shuttling into instant connections, and a small team gains operational reach that once required dedicated staff. The risk is dependency: automation that adjusts prices and relists products around the clock is only as good as its guardrails, and “seller-approved actions” will be tested at scale by exactly the sellers most tempted to let agents run loose. Amazon is betting that guarded autonomy — not raw capability — is what earns the trust to act. Given that its sellers already accept AI recommendations nine times out of ten, it may be right.