Amazon hands sellers a Claude plugin so they never open Seller Central

Share
Amazon hands sellers a Claude plugin so they never open Seller Central

Amazon spent a year treating outside AI agents as trespassers. At its Accelerate seller conference on Wednesday it opened the door itself — and handed the keys to Anthropic.

Amazon launched a Selling Partner plugin that lets third-party sellers run their business through an AI assistant instead of the Seller Central portal. Sellers connect either Anthropic's Claude or Amazon's own Quick assistant to their seller account in about 60 seconds, no coding, and from there the assistant can pull listings, inventory levels, sales analytics and performance metrics, then act on them — checking stock, adjusting prices, updating listings — the same way Seller Assistant does inside Seller Central. Amazon's vice president of Worldwide Selling Partner Experience Mary Beth Westmoreland told GeekWire the goal plainly: "Our vision was that they would never have to log into Seller Central. We would just bring it to them where they work." The plugin is in beta for US sellers, with international expansion to follow.

The controls are the part worth reading closely, because this is an assistant with write access to a business. Amazon says sellers choose which data types the plugin can reach, must explicitly approve each action before it executes, can revoke access, and get a full audit trail — and that it does not see other business data inside a seller's assistant. Sellers can also run it in recommend-only mode. Anthropic's own connector listing confirms the beta and the review-before-action flow. Whether a seller trusts that boundary is a separate question from whether it exists, and Amazon has answered the first one by making approval the default rather than an option.

The timing is the tell. Amazon blocked Meta's Muse agent from shopping its site days earlier on the principle that outside agents must identify themselves and follow a site's rules; GeekWire and The Next Web both frame Wednesday's announcement against that block. Amazon's newsroom post does not mention Muse. But the asymmetry is real and it is not about rules, it is about who is allowed to be the agent: Anthropic is a major Amazon investee whose Claude already powers Seller Assistant, and Amazon Quick is Amazon's own. Amazon is not opening its storefront to agents — it is opening its seller tooling to the two agents it controls or co-owns, and publishing a plugin architecture so it can pick the rest later. Westmoreland said as much: "We've built this in a way that should be modular enough for us to continue to publish plugins."

Amazon also upgraded Seller Assistant itself with persistent memory of a seller's pricing patterns, inventory cycles and growth goals, carried across Seller Central, Quick or Claude, plus a canvas that models scenarios and compares pricing strategies on live data. Every seller on the platform gets a free 12-month Amazon Quick Plus subscription through December 31, with two co-workers included at no extra cost — a distribution move as much as a discount. Amazon says about 90% of its selling partners already use third-party AI tools; Anthropic field CTO Eric Burns says sellers accept Seller Assistant recommendations more than 90% of the time. Independent sellers are more than 60% of units sold on Amazon worldwide, and seller fees reached $46.8 billion in the second quarter, so the upside of moving even a slice of that portal traffic into an assistant Amazon can meter is not small.


Separately, Meta put a price on its best-selling AI hardware. Meta announced the third-generation Ray-Ban Meta glasses at Connect, starting at $449 — up from $379 for the second generation — and shipping now. The spec sheet is a modest refresh: up to nine hours of battery, an hour more than Gen 2; a sixth microphone in a six-mic array that Meta claims cuts more than 90% of background noise; a new programmable action button that can be set in the Meta AI app for one-touch Meta AI or live translation; and three frame styles — a new Aviator, a new cat-eye Zena, and the carried-over Wayfarer — across 27 color and lens combinations, with a limited-edition 90th-anniversary Aviator. A special-edition Aviator in Milky Warm Beige runs $579.

The Verge called the changes "fairly minor refinements," and that is the honest read — the genuine new products at Connect were the camera-free Audio glasses and the $1,299 VR Glasses we covered Wednesday — Meta's VR Glasses land at $1,299 with no headset at all. What matters is the price direction. Meta's glasses are the category's volume leader, and the company is raising the entry point while the category absorbs a privacy backlash — The Guardian notes the frames are frequently dubbed "perv glasses" or "spy glasses," and Wired notes Meta spent months being "pummeled" over camera-enabled glasses. A $70 increase on a refresh this incremental tests whether Meta's brand carries the category or the category's reputation drags Meta.

What to watch: whether Amazon publishes a plugin for an agent it does not own or invest in — that is the real test of the "modular" line — and whether the Ray-Ban Meta price bump holds once holiday sales data lands.

Amazon is opening its seller tools to AI agents it controls while blocking ones it doesn't. Is that platform governance or just picking winners? Tell us in the comments.

Read more

Brain organoids classified Braille at 83% — in a trio, not alone

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon turned its always-listening Bee line into a two-device family on Thursday, and NVIDIA opened a 2.8-trillion-token Indic corpus recipe. The connecting thread: both are bets that the next AI money is in memory — yours, or your model's. Amazon put two new always-listening devices on sale, and both now lead to a subscription. Bee Frontier, the next-generation wristband that also clips to clothing, ships now at $74.99 in Onyx, Lavender and Sage — up from the $50 predecessor — with seven-day

Perplexity open-sources its agent sandbox after 118 failed escapes

Perplexity open-sources its agent sandbox after 118 failed escapes

Perplexity handed its agent containment layer to the open-source world, and Washington's AI task force stopped reading briefings and started working. Perplexity open-sourced SPACE on Thursday — the sandbox it built to keep its own AI agents from getting out — and says nine frontier models failed to break out of it across 118 attempts. Each agent runs inside a small virtual machine with its own operating system, with credentials and network access managed outside the sandbox so an agent can re