At Accelerate on September 23, Amazon gave Seller Assistant the ability to watch your account while you’re logged out, and it opened your seller data to outside AI assistants for the first time, starting with Amazon Quick and Anthropic’s Claude. Amazon also announced new agents for aged inventory and inbound planning the next day.
Most coverage stopped at the announcement. This post covers what an operator actually needs to know: what each piece does, where your data goes, what it can’t do yet, and the order we’d switch things on.
What Amazon announced
- Memory. In Amazon’s words, Seller Assistant “now carries persistent memory of each seller’s pricing patterns, inventory cycles, and growth goals.”
- Workflows. Automations you describe in plain language that keep running “even when the seller isn’t logged in to Seller Central.” You choose whether a workflow only recommends or takes action, and actions are reviewed and approved before they run. Every action is logged with a full audit trail. One of Amazon’s example workflows: alert me if any of my top 10 products drop below a 4-star rating and draft a response plan.
- Canvas. A visual workspace built on live data, for modeling scenarios, comparing pricing strategies and diagnosing campaign performance.
- The selling partner plugin. It connects your listings, real-time performance metrics, inventory levels and sales analytics to the AI assistant you already use. It launched in Amazon Quick and, in beta, in Claude. Amazon says setup takes about 60 seconds with no code, and it’s available in beta to sellers in Amazon’s U.S. stores.
- Supply chain agents. New agents for aged inventory and inbound planning that send alerts, explain their reasoning and show the data behind it. Amazon says they’re rolling out over the coming weeks and months.
Amazon also said Seller Assistant now reaches more than 90% of its selling partners and that sellers accept its recommendations more than 90% of the time. Every primary account holder can get a free 12-month Amazon Quick Plus subscription by signing up by December 31, 2026.
For context, Seller Assistant started life as Project Amelia at Accelerate 2024. At Accelerate 2025, Amazon gave it the ability to plan and take actions with a seller’s permission. This year’s update is the first that lets it keep working on its own schedule, and the first that lets a tool outside Amazon do the same.
How the Claude and Quick plugin works
The plugin is an MCP connector, the standard way AI assistants connect to outside tools. Amazon published the skills behind it as an open-source toolkit on GitHub. When you connect, you go through the normal Seller Central sign-in and consent screen, and the assistant can only use tools your Seller Central role allows.
The toolkit’s 11 skills show what it’s built for today:
| Area | Skills |
|---|---|
| Sales | Seller analytics |
| Inventory | Stockout prevention, FBA inbound management |
| Listings | Listing troubleshooter, listing issues, buyability, searchability, compliance checks |
| Account | Inviting secondary users, preparing Seller Support cases |
| Developers | SP-API knowledge |
It also ships three shortcuts: a stockout check, a listing fix and a sales-drop diagnosis. The toolkit states that write actions are always drafted for your approval before anything changes.
Where your data goes
This is the part most sellers will skip, and it’s the part worth reading twice.
Seller Assistant inside Seller Central runs on Amazon’s infrastructure. The plugin is different. Amazon’s toolkit is explicit that data handled by your AI assistant is “subject to that assistant’s own terms and privacy policy.” Amazon’s side doesn’t send your data anywhere except the assistant you connect, and it doesn’t keep your account identifiers beyond the session. Once your sales figures are in a Claude or Quick conversation, though, that assistant’s data policy applies.
That’s not a reason to avoid it. It’s a reason to decide deliberately:
- Connect the plugin from an account on a business plan whose data terms your company has actually reviewed, not a personal login.
- Use a Seller Central user with the smallest role that does the job. If the connection only needs to read inventory and sales, don’t connect it as the primary account holder.
- Add the connection to your access list. If you haven’t reviewed who can reach your account lately, our Amazon account access audit walks through it.
What it can’t do yet
- Advertising isn’t in the toolkit. None of the 11 published skills cover Sponsored Products, Sponsored Brands or DSP. Canvas can help diagnose campaigns inside Seller Assistant, but your bids and budgets still live in the ad console. Our take on using AI for Amazon ads still applies.
- The plugin is U.S. only, and in beta. Amazon says international expansion will follow.
- Compliance checks are advisory. The toolkit’s compliance skill flags risks before you list. It doesn’t clear a product for you.
- It can be confidently wrong. A 90% acceptance rate also means some recommendations get rejected. Agents that act at scale can repeat a bad call across a whole catalog, which is why the approval step matters.
The order we’d switch things on
- Start in recommend-only mode. Run every workflow as alerts for the first two to three weeks and compare its calls with what your team would have done.
- Begin with watch-and-alert jobs. Review drops on top products, stockout risk on your top sellers and aged inventory approaching storage fees. These are low risk and high value, and they match what the new supply-chain agents are built for.
- Keep pricing on a leash. Amazon’s own example workflow adjusts prices when it spots a competitive opening. Before you allow that, set a floor and a ceiling, and block changes in the two weeks before major events. Shoppers can now see a full year of price history, which we covered in what Amazon’s 365-day price history means for sellers.
- Read the audit trail weekly. Put it on the same calendar as your account health review, and assign one person to own it.
- Connect Claude or Quick last. Once the in-house workflows are trusted, add the plugin for the questions that cross data sources, like pairing inventory with your own margin model. If you also sell on Walmart, TikTok Shop or Shopify, Amazon’s new multichannel selling tools bring those orders into the same Seller Central view.
What this means for your team
An assistant that watches your account around the clock handles more of the checking. It doesn’t set strategy, and it doesn’t take responsibility for what it changes. Someone still decides which workflows run, where the price floors sit and what gets approved. The brands that get the most from this will treat it like a new team member, with a job description, limits and a weekly review.
That review is part of how our Amazon account management team runs accounts, alongside inventory planning that the new agents now support. If you want help deciding what to automate first, book a free Amazon audit.
Sources: Amazon, “Amazon gives sellers an even smarter Seller Assistant and a new plugin for Amazon Quick and Anthropic’s Claude” (September 23, 2026); Amazon, “Amazon announces new tools and services to help sellers go global from day one” (September 24, 2026); Amazon, Selling Partner Agentic Toolkit on GitHub; GeekWire coverage of Accelerate 2026. The plugin is in beta, so features and availability may change.
