Amazon Wants Sellers To Never Log In. Is Your Product Ready?

Last Wednesday, at its Accelerate seller conference in Seattle, Amazon did something that would have sounded strange a few years ago. It announced a plugin that lets independent sellers run their Amazon businesses from outside AI assistants, starting with Amazon Quick and a beta in Anthropic's Claude.

Sellers can check inventory, adjust prices and update listings without opening Seller Central, the dashboard merchants have used for years.

Mary Beth Westmoreland, Amazon's vice president of Worldwide Selling Partner Experience, told GeekWire the goal was simple: "Our vision was that they would never have to log into Seller Central."

The next day, Google said it is testing a Gemini feature that phones businesses on your behalf to book a table, check stock or move an appointment.

Put those two stories together and a pattern appears. More of your users will reach your product through an AI assistant they already use, not through your interface. If you are building software right now, that changes what you should ship next.

Here is what Amazon described in its announcement and what GeekWire added from its interview:

• A selling partner plugin. It connects a seller's listings, inventory levels, sales analytics and performance metrics to their preferred AI assistant. The assistant can act on the account, not just read from it.
• Setup in about 60 seconds, no code. Amazon says connecting Claude takes roughly a minute.
• Seller control. Sellers pick which types of data the plugin can reach and approve each action before it runs. Every interaction has an audit trail.
• Background workflows. Seller Assistant can now monitor conditions and respond on its own, such as alerting a seller if a top product drops below four stars and drafting a response plan.
• Memory. The assistant remembers a seller's pricing patterns, inventory cycles and goals, and that memory follows them across Seller Central, Quick and Claude.

Amazon also shared two adoption numbers. About 90 percent of its selling partners already use third-party AI tools to run their businesses. And sellers accept Seller Assistant's recommendations more than 90 percent of the time.

Amazon earns enormous fees from sellers. GeekWire notes independent sellers account for more than 60 percent of units sold on Amazon, and seller fees brought in $46.8 billion in the second quarter. So why make its own dashboard less necessary?

Because the dashboard was never the point. The point is that sellers keep selling on Amazon. If 90 percent of them already use outside AI tools, they are going to copy and paste Amazon data into those tools anyway. Amazon can either fight that habit or make its business the easiest one to run from wherever the seller works.

It chose the second option, and it kept control of the important parts: which assistants are supported, what data flows out and which actions need approval.

For a startup, the lesson is that your interface is not your moat. Your data, your actions and the trust users place in them are. If a customer would rather manage your product from their assistant, helping them do that safely keeps them. Forcing them back into your dashboard pushes them toward a competitor who makes it easy.

The openness has limits. GeekWire reports that days before Accelerate, Amazon blocked Meta's Muse, an AI agent that shops on behalf of consumers, from its store. Amazon says outside agents need to identify themselves and follow the rules of the sites they use.

So in one week Amazon welcomed agents working for its sellers and blocked an agent working for shoppers. That is not a contradiction. It is a policy: agents are allowed when the platform has approved the integration, can see who is acting and can audit what happened.

If you are building an agent that operates inside someone else's platform, plan for this. Unofficial scraping and screen automation can be shut off overnight. Official integrations, clear identification and respect for the platform's terms are what keep you in business.

If you are building a platform, decide your policy now. Which agents can act on behalf of your users? How do they identify themselves? What can they do without a human approving it? Writing this down early is much easier than cleaning up after an incident.

The Google test pushes the same idea into the physical world. TechCrunch reports that "Call for Me" lets Gemini phone a business to book a restaurant table, check whether an item is in stock, move an appointment or put an item on hold.

It starts small: US Pixel 11 owners who pay for a Gemini subscription and use the beta Phone app. The call comes from the user's own number, and they can follow a live transcript and take over at any time. Google said it is starting slowly because real conversations are nuanced.

If you build software for restaurants, clinics, salons, retailers or any business that takes phone calls, your customers will soon be talking to AI callers. The businesses that do well will be the ones whose hours, inventory and booking rules are accurate and easy for a machine to confirm. That is a product opportunity for anyone serving local businesses.

The same week, OpenAI added plugins to ChatGPT Voice that reach email, calendar and Slack, according to a GTStudios roundup. The assistant layer is filling up fast.

You do not need to rebuild your company. You need to make your core actions safe and reachable from outside your interface. Use this checklist:

1. List your top ten actions. Not features. Actions a user takes to get value, such as "change a price," "issue a refund" or "approve an invoice." These are what an assistant will try to do.

2. Expose them through a documented integration. An API or a plugin that an assistant can call. Keep inputs simple and error messages clear, because a model will read them.

3. Scope permissions tightly. Let users choose what data and actions each connection can reach, the way Amazon lets sellers pick data types.

4. Require approval for anything that costs money or cannot be undone. Amazon's rule is that sellers approve each action. That is a sensible default for an early product.

5. Log everything. An audit trail showing which assistant did what, when and on whose approval. This protects your users and you.

6. Make the assistant identify itself. Know whether a request came from a person in your app or an outside agent. Amazon's Muse decision shows why this matters.

7. Keep context portable. Amazon's memory follows the seller across tools. At minimum, make sure an assistant can pull the recent history it needs to act well.

If users succeed without opening your app, some familiar metrics will look worse even as the business gets better.

Daily active users in your dashboard may fall. Time in app may drop. Neither means your product is failing if users are getting more done through an assistant.

Add metrics that track value wherever it happens:

• Actions completed, split by source: your interface, your API, each assistant.
• Approval rate on agent-proposed actions. Amazon's 90 percent acceptance figure is a useful benchmark for trust.
• Retention of connected accounts versus accounts that never connected an assistant.
• Error and rollback rate on agent actions.

Update your pitch to match. Investors will ask how AI assistants affect your product. "Our actions are reachable from the major assistants, with scoped permissions and audit logs, and connected accounts retain better" is a strong answer. Mapping this out in your Foundra plan, with the metrics you will report each month, keeps the story consistent from your roadmap to your investor updates.

Week 1: Interview five active customers. Ask which AI assistants they use for work and which tasks in your product they would hand to one.

Week 2: Pick the two most requested actions. Build a scoped integration with approval and logging for just those two.

Week 3: Connect it to one assistant your customers already use. Invite ten users into a private beta.

Week 4: Measure actions completed, approval rate and errors. Decide whether to add the next three actions or fix what broke.

Small and safe beats broad and risky. Amazon launched its Claude connection as a beta in one country first, with international rollout to follow. You can afford to start even smaller.

What did Amazon announce on September 23, 2026? A selling partner plugin that lets sellers manage listings, inventory, pricing and analytics from Amazon Quick and, in beta, Anthropic's Claude. It also added background workflows and memory to Seller Assistant.

Does this mean sellers never use Seller Central now? No. Seller Central still exists. Amazon's stated vision is that sellers should not have to log in, because the same capabilities are available in the assistant they already use.

Is the Claude plugin available everywhere? It is in beta for sellers in Amazon's US stores, with international expansion planned.

Why did Amazon block Meta's Muse agent? GeekWire reports Amazon said outside agents need to identify themselves and follow the rules of the sites they use. Amazon still decides which platforms it supports.

What is the first step to make my startup agent-ready? List the ten actions that deliver the most value, then expose the top two through a scoped integration that requires approval and logs every action.

1. Amazon gives sellers an even smarter Seller Assistant and a new plugin for Amazon Quick and Anthropic's Claude, About Amazon, September 23, 2026
2. Amazon opens its seller tools to outside AI agents, starting with Anthropic's Claude, GeekWire, September 23, 2026
3. Google tests letting Gemini call businesses for you, TechCrunch, September 24, 2026
4. 17 Must-Know AI Startup News Stories (Sept 27, 2026), GTStudios