SEATTLE — For millions of Amazon sellers who have spent the last two years running parts of their operations through outside AI tools anyway, Tuesday’s announcement at Amazon Accelerate felt less like a revolution than a formalization. Amazon is officially opening Seller Central to third-party AI agents, starting with Claude, in a move that reveals as much about what the company intends to protect as what it has decided to share.
The new “Selling Partner plugin” lets sellers manage inventory, adjust prices, update listings, pull analytics, and handle customer messages through Anthropic’s Claude without opening Seller Central at all. Connecting takes roughly 60 seconds, with no code required. Amazon says seller data stays entirely within its own infrastructure throughout, never shared externally. The plugin is built on Amazon Bedrock, the company’s managed AI infrastructure layer, combining Amazon Nova models with Claude for reasoning and execution. 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. Amazon’s announcement notes that Seller Assistant now carries persistent memory of each seller’s pricing patterns, inventory cycles, and growth goals, memory that follows sellers across Seller Central, Amazon Quick, and Claude instead of resetting between sessions.
Amazon’s own Quick assistant is receiving the same capability, along with a new Canvas feature that pulls live seller data into a visual workspace for financial planning and inventory analysis. Quick Plus subscriptions are free for primary account holders through December 31, 2026. The beta is US-only for now, with international expansion coming in weeks.
What Amazon is not doing is as significant as what it is. The consumer-facing storefront, including the product listings, the checkout flow, and the discovery layer Amazon has defended aggressively against outside AI agents, stays closed. Courts have already started ruling on the edges of this: earlier this year, a federal appeals court cleared Perplexity’s Comet AI to shop Amazon without the company’s permission, opening a legal corridor Amazon continues to contest. Tuesday’s announcement draws a deliberate internal boundary. AI agents are welcome to help sellers run their operations more efficiently inside Amazon’s system. They are not welcome to make purchase decisions on behalf of Amazon’s customers.
That line is not incidental. Amazon’s margin lives at the moment of consumer discovery and purchase, not during a seller’s overnight inventory review. Letting agents manage the back office improves seller productivity without surrendering the customer relationship, the element Amazon has treated as non-negotiable throughout the broader debate over agentic commerce. The Selling Partner plugin enforces this boundary structurally. Its permissions architecture ends at the seller’s operational boundary. Nothing in the pipeline touches the storefront.
The choice of Anthropic as the launch partner is less a product endorsement than an extension of a structural alignment that already runs to the infrastructure layer. Amazon has committed over $13 billion in investment to Anthropic, making AWS its primary cloud training partner. Claude models run across a wide range of enterprise AWS deployments, as TechCrunch reported in April when the latest tranche was announced. The Selling Partner plugin allowlisted Anthropic’s client on September 9th, roughly two weeks before the public announcement at Accelerate. Whether other AI providers can join that allowlist, and on what terms, Amazon has not said.
The update also adds agentic supply chain capabilities running through Amazon’s own models: new inbound planning tools that optimize inventory routing, and aged inventory alerts with automatic recommendations. These run through Amazon’s systems rather than external agents, but they sit within the same strategic direction. The idea is that Seller Central’s interface layer is increasingly a conversation, not a dashboard.
The question that remains unanswered is what happens after December 2026, when the free Quick Plus period ends, and which providers follow Anthropic into the plugin program. Amazon has not described the criteria for joining the allowlist, nor announced pricing for what comes after the trial. There is a version of this where access to Amazon’s seller data becomes a competitive lever for a small number of AI companies the marketplace chooses to certify, concentrating the seller toolchain inside approved systems. Whether this becomes a walled ecosystem or an open standard will determine how much leverage sellers retain over their own operations. The broader debate over who controls AI agents operating across commercial infrastructure has not been settled by this announcement.
What is clear is the line Amazon drew. The company that spent two years contesting outside agents at its storefront has now opened the side entrance, into the seller’s machine room, one sanctioned action at a time.
