By Amazon is letting independent sellers run parts of their Amazon business from inside the AI assistants they already use. The company announced a new Selling Partner plugin that connects Amazon seller data and actions to Anthropic’s Claude and Amazon’s own Quick assistant, with more integrations to come.
The shift is significant because it moves Amazon’s seller ecosystem from a closed portal — Seller Central — to an open-agent model where sellers can check inventory, review performance, and update prices and listings through a conversational AI, without manually clicking through multiple Seller Central pages.
Amazon unveiled the changes at its annual Accelerate seller conference in Seattle on September 23, 2026, alongside upgrades to its built-in Seller Assistant and new background automation workflows.
Amazon is opening its seller tools to AI assistants
For years, sellers have managed their Amazon presence inside Seller Central, the web portal used to manage listings, inventory, pricing, advertising, compliance and performance metrics. It is powerful but complex, especially for small teams juggling multiple sales channels.
According to Amazon, about 90% of selling partners already use some form of third-party AI to run parts of their business, whether for listing copy, demand forecasting, or accounting.
The new approach flips the interface. Instead of forcing sellers to bring their AI work into Seller Central, Amazon is bringing Seller Central intelligence to where sellers already work.
The vehicle for that is the Amazon Selling Partner plugin. The plugin is currently available in beta for sellers in Amazon’s U.S. stores, with international expansion planned. It is launching first with two assistants: Amazon Quick and Anthropic’s Claude.
Amazon says connecting Claude takes about 60 seconds and requires no coding. As reported by GeekWire’s coverage of the announcement, the vision is that sellers would never have to log into Seller Central for routine tasks. As Mary Beth Westmoreland, Amazon’s VP of Worldwide Selling Partner Experience, put it, Amazon wants to bring the tools to sellers where they work.
What can Claude actually do for Amazon sellers?
Claude is Anthropic’s AI assistant, known for long-context reasoning and its use in work tasks like analysis, document creation, and coding. It is not an e-commerce platform itself, but Amazon’s plugin gives it a secure bridge into a seller’s Amazon business.
Once connected, Claude can pull in four core types of Amazon data:
- listing information and contributions
- real-time inventory levels
- sales analytics
- performance metrics and account health signals
More importantly, it can act on that data the same way Seller Assistant does inside Seller Central. In practice, that means a seller can ask in plain language:
- “Check inventory for my top 10 ASINs and flag anything below 14 days of cover.”
- “Why did sales dip for this listing last week?”
- “Draft a more competitive title and bullets based on recent search trends.”
- “Adjust pricing for this product to stay within 2% of the lowest competitive offer.”
Claude analyzes the data, provides reasoning, and proposes an action. The action only goes live after seller review and approval. Amazon says every action is logged with a complete audit trail, so sellers can see what was changed, when, and why.
This is not full autonomy. It is assisted execution with guardrails. Amazon emphasizes that sellers choose what data the plugin can access and whether workflows should only recommend or also prepare actions for approval.
How the new Selling Partner plugin works
The plugin is built on Amazon Bedrock, Amazon’s foundation model service, and uses Anthropic’s Claude models combined with Amazon’s own Nova models. Seller Assistant itself runs on this stack.
Technically, the plugin does not give Claude or Quick direct, unrestricted access to Seller Central. It exposes specific Selling Partner APIs through a controlled boundary. Amazon says seller data stays within Amazon’s infrastructure and is not shared outside Amazon except as needed to fulfill the seller’s request through the approved agent.
When a seller connects Claude, Amazon does not see the rest of the seller’s business data that might be connected in Claude — such as accounting software, supplier sheets, or Shopify data. The connection is one-way from an Amazon perspective: Amazon provides the Amazon commerce context, Claude can combine it with other business context the seller already has in Claude.
For sellers, the practical difference is speed. One early beta tester, Atlas Bar founder James Oliver, said in Amazon’s announcement that what previously took an afternoon of manually exporting data and feeding it into AI tools now happens instantly because the data is built in.
Seller approval and data permissions
Permissions and approval are central to how Amazon is framing this rollout, particularly as sellers worry about AI making costly pricing mistakes.
Sellers control:
- What data is accessible: They can choose which categories of data the plugin can reach, such as listings versus sales analytics.
- What mode the agent operates in: Recommend only, or prepare actions for approval.
- Whether to approve: Even when a workflow is set to take action, Amazon says the action waits for seller approval before execution.
- Audit and rollback: Every action is logged. Sellers can review the reasoning behind a recommendation and revert if needed.
Amazon says external connections do not change how it protects seller data, and that plugin interactions are protected by clear boundaries, human approval, and audit trails. It has not claimed that Anthropic never processes Amazon data — the models run on Bedrock — but says data safety and trustworthiness are built into the design and that data is not shared outside Amazon’s infrastructure without seller permission.
Amazon Quick and the upgraded Seller Assistant
Alongside Claude, Amazon is launching the plugin first in Amazon Quick, its own AI assistant for work. Quick is a general-purpose agent that creates documents and spreadsheets, conducts research, and connects to email, calendar, and files. Amazon describes it as working across phone, web, and desktop with continuous conversations.
For sellers, Quick becomes more powerful because it can combine Amazon selling data with the rest of their operations — financials, marketing, manufacturing — to give a fuller picture. Sellers can also build specialized agents inside Quick, such as a pricing agent, a listing agent, or a restock agent.
The underlying Seller Assistant is also getting a major upgrade:
- Persistent memory: Seller Assistant now remembers pricing patterns, inventory cycles, and growth goals across sessions. That memory follows sellers whether they are in Seller Central, Quick, or Claude.
- Workflows that work around the clock: Previously, Seller Assistant acted when asked. Now sellers can create automations that continuously monitor conditions, such as “Alert me if any top 10 products drop below 4 stars and draft a response plan” or “Monitor competitive openings and refresh listings.” Sellers describe needs in plain language and set guardrails.
- Canvas: A new dynamic visual workspace where sellers can model pricing scenarios, compare ad campaign performance, and interact with live data through conversation. The canvas redraws as they ask follow-ups.
Amazon says Seller Assistant has rolled out to over 90% of its selling partners worldwide in their native language and has hundreds of thousands of active users, with recommendations accepted over 90% of the time. The company built it with AWS and Anthropic, harnessing foundation models and 25 years of Amazon commerce expertise.
To accelerate adoption, Amazon is offering every primary account holder worldwide a free 12-month subscription to Amazon Quick Plus, available through December 31, 2026, with two additional co-workers covered at no extra cost.
Why Amazon is moving toward agentic AI
To understand the announcement, it helps to separate two types of AI that have emerged in e-commerce.
Traditional chatbot: User asks a question → AI answers from its knowledge. Example: “What does stranded inventory mean?”
Agentic AI: User asks for an outcome → AI interprets the request, accesses relevant tools and data, reasons across domains, proposes or prepares an action, waits for approval, and reports the result. Example: “Find my stranded inventory, explain why it’s stranded, and draft a plan to fix it.”
Seller Assistant was already moving in the agentic direction inside Seller Central. The plugin extends that agentic capability outside Seller Central.
Why now? Independent sellers account for more than 60% of units sold on Amazon worldwide. Fees Amazon charges them generated $46.8 billion in the second quarter of 2026 alone, more than AWS, as noted in GeekWire’s reporting. Seller efficiency directly impacts Amazon’s marketplace health, advertising revenue — $19.8 billion that quarter — and its defense in the FTC antitrust case set for trial in March 2027, which alleges Amazon overcharges sellers. Amazon has denied those allegations.
From a product perspective, if sellers are already using external AI tools, Amazon has two choices: try to pull them back into Seller Central, or meet them where they are and remain the system of record. Amazon is choosing the second, while keeping control over which agents are officially supported and how they access data.
What this means for Amazon sellers
For sellers, the immediate benefit is not that AI can answer questions faster, but that it can interact with business systems and perform actions without constant manual navigation.
Real-world use cases Amazon and early testers cite include:
- Checking stock levels across SKUs without building custom reports
- Identifying underperforming listings and drafting improved copy or images
- Monitoring price competitiveness and adjusting within seller-defined guardrails
- Summarizing sales trends and diagnosing why a campaign underperformed
- Tracking account health alerts and compliance issues before they become suspensions
The memory feature matters because it reduces repetition. A seller who has told Seller Assistant that they prefer to keep 21 days of cover on bestsellers and avoid price wars below a certain margin does not have to repeat that each session.
The risk, which sellers will need to manage, is over-delegation. An AI that adjusts pricing too aggressively can erode margin or trigger repricing wars. That is why Amazon’s emphasis on approval flows, reasoning transparency, and audit trails is more than marketing copy — it is the control layer that determines whether sellers trust the system.
Amazon’s wider approach to AI shopping agents
Amazon’s openness to seller-side agents contrasts with its recent stance toward consumer shopping agents that attempt to act on Amazon without permission.
Just days before this announcement, Amazon blocked Meta’s Muse, an AI agent designed to shop on behalf of consumers, from its store. Amazon said outside agents need to identify themselves and follow the rules of sites they use. It has taken similar steps with other unauthorized shopping agents.
The distinction Amazon is drawing is important:
- Official seller-side plugins: Amazon invites selected AI assistants into its seller tools under defined data boundaries and approval requirements.
- Unofficial consumer shopping agents: Amazon blocks agents that scrape or automate shopping on its consumer storefront without identification or compliance with its terms.
In other words, Amazon is not opening its marketplace to any AI agent. It is controlling which external agents can access its ecosystem and under what conditions. That allows it to preserve trust, security, and its role as the transaction and fulfillment layer while still letting sellers use tools like Claude.
Amazon’s relationship with Anthropic
Claude being first is not coincidental. Amazon is a major investor in Anthropic, with total investment now at $8 billion after additional rounds in 2024 and 2025. Anthropic remains independent, with Google also holding a stake, but its models are deeply embedded in Amazon’s AI products — including its coding assistant Kiro, Alexa shopping features, Quick, and Seller Assistant itself.
Because Seller Assistant already uses Claude models on Bedrock, exposing that same intelligence through Claude’s own interface is a relatively natural extension. Amazon says it built the plugin architecture to be modular, so more third-party tools can be added after Quick and Claude.
What comes next?
The plugin is currently in beta in the U.S. Amazon says international expansion will follow, but has not announced specific dates or markets. More third-party AI platforms beyond Claude and Quick are expected, although Amazon has not named them.
The bigger picture is that e-commerce operations are becoming agent-driven. For years, sellers connected software via APIs and dashboards. The new model is conversational: describe what you want in plain English, let the agent reason across inventory, ads, listings, and financials, and approve actions in one place.
If the model works, it could reduce the overhead that prevents small teams from competing with larger sellers who have dedicated operations staff. A small team could operate with the strategic depth previously available only to larger companies, as Amazon puts it.
The success will depend on reliability, cost, and trust. Sellers will need to see that the agent understands their business context, does not hallucinate inventory or pricing data, and respects guardrails. Amazon will need to prove that opening to external agents does not create new security or data leakage risks.
For now, the rollout is limited and deliberately cautious — U.S. beta, approval-required actions, and audit trails. But it signals where Amazon thinks seller management is headed: less time in portals, more time directing agents that work across the business, with Amazon remaining the infrastructure underneath.
Jalees Ur Rehman is an SEO professional with 13+ years of experience in SEO and Digital Marketing. He helps businesses grow organic traffic and conversions through data-driven digital marketing strategies and practical execution.