I wasn't at Accelerate this year. Life, scheduling, the eternal optimism that I'll "catch the replays" (which I actually did this time, so points for follow-through). I've spent the last few days reading every announcement, watching every session, and cross-referencing the press releases with what Amazon actually said on stage versus what they quietly buried in documentation links nobody clicks. So here's your roundup of everything that matters from Seattle, minus the free tote bag and the networking anxiety.

Amazon announced a Selling Partner plugin that connects your account data directly to an outside AI assistant, launching first through Quick, Amazon's own general-purpose AI assistant for work, and in beta through Anthropic's Claude. The pitch is that you can manage listings, check performance, adjust pricing and monitor inventory without ever opening the portal.

That is a genuinely unusual thing for a company to say about its own core product. Imagine Shopify announcing that, ideally, you'd never visit their dashboard. It has the same energy as a restaurant saying "honestly, you'd be better off eating at home" — technically generous, strategically fascinating.

Amazon framed it as meeting sellers where they already work, noting that around ninety per cent of selling partners already use some form of third-party AI to run parts of their operation. Rather than fight that behaviour (which, knowing Amazon, was presumably option A through D before someone suggested option E), they're formalising it.

The setup reportedly takes sixty seconds, requires no coding, and lets you choose which data categories the plugin can access. Actions require your approval, and everything gets logged. It's in beta for US stores, with international markets to follow.

The convenience is real. The subtext is also real. Amazon decides which AI platforms make the whitelist (currently its own assistant and a model from a company it has invested billions in), your data stays on Amazon's infrastructure, and the "freedom" is extended on terms Amazon sets. It's the velvet-rope version of openness — you can leave the building, as long as you stay in the approved courtyard. (We've all been to that kind of party.)

Three upgrades to Seller Assistant (the AI system running on Anthropic's Claude models via Amazon Bedrock) were announced, and together they change it from a tool you poke when you remember to one that operates while you sleep.

First, memory. Seller Assistant now retains a persistent record of your pricing patterns, inventory cycles and growth goals across sessions, and that memory travels with you into Quick and Claude. No more re-explaining your business every time you open a chat, which honestly felt like breaking up with someone and then having to re-introduce yourself on every date. Each recommendation arrives with the reasoning attached, which is the mechanism that lets you evaluate it rather than just nod along (a distinction worth holding onto, because "trust me" is not an acceptable business strategy from a trillion-dollar company).

Second, workflows — Amazon's term for always-on automations. Until now, the assistant only acted when asked. Workflows run continuously in the background, monitoring conditions and responding even when you're logged out. You describe what you want in plain language and set the guardrails. Amazon's examples ranged from alerting you when a top product drops below four stars (sensible, useful, no notes) to automatically adjusting pricing and refreshing listings based on competitive openings (which is either liberating or the kind of thing that makes you check your account at 6 AM "just in case," depending on how much you trust a platform's AI to optimise pricing on a platform the AI's parent company also controls).

Third, Canvas, a visual workspace that generates from live data and redraws as you ask follow-up questions — discount modelling, category benchmarking, campaign diagnostics. Worth noting: Canvas was actually announced back in March for US and UK. Accelerate is the expansion showcase, not the debut, a distinction Amazon presented with all the clarity of someone mumbling "I already told you this" into their sleeve.

Amazon's own (unaudited) adoption figures: Seller Assistant has reached more than ninety per cent of selling partners worldwide, has hundreds of thousands of active users, and sellers accept its recommendations somewhere between "nearly" and "more than" ninety per cent of the time. Setting up a new product, which used to mean filling in more than twenty fields, now requires fewer than ten.

Several ad announcements matter more than the shiny AI demos, because they touch how customers actually find products now.

Your existing Sponsored Products and Sponsored Brands campaigns now appear inside conversations with Alexa, Amazon's AI shopping assistant — no new campaigns to build or budgets to set. Amazon also introduced Ads Data Manager, a free tool that lets you bring first-party data (email lists, loyalty data) into Amazon Ads to bid more aggressively on known high-intent shoppers.

Sponsored prompts — conversation starters on product pages — were reported to drive a six per cent lift in conversion for Sponsored Brands. New auto-optimisation for keyword and product targeting was said to produce a first sale forty per cent faster in early results.

Read together, this is Amazon's ad machinery being re-plumbed for conversational shopping. The question worth asking isn't whether your ads still work, but where they now appear and how much of the discovery moment Amazon is quietly moving inside its own assistant. If you're not thinking about how your products show up in a chat interface rather than a search grid, this is the nudge.

The most materially useful announcement for many sellers had nothing to do with AI (which tells you something about how much AI we actually needed versus how much we got).

Amazon is extending its Account Health Assurance programme — which covers sellers who maintain an account health rating above 250 — with listing-level protections. Most listing enforcements will now carry a grace period rather than an immediate takedown, and in most cases a listing stays active while an appeal is under review. Pricing enforcement is being softened too — you'll get a warning before a pricing change triggers a deactivation.

Alongside this sits the Seller Challenge, an enhanced-review mechanism for sellers who believe an enforcement decision was wrong. You get three challenges every 180 days, with a quoted turnaround of twenty-four to forty-eight hours.

The caveats: Seller Challenge has been rolling out in beta, not to every eligible account at once. A listing that's already deactivated stays down while a challenge is reviewed. Amazon has published nothing on success rates or the standards its reviewers apply. The protections are due to be live in the US by year-end and worldwide in 2027.

For anyone who has lost a listing overnight to an automated flag and then spent three days trying to reach a person who could explain why — this is the announcement with the clearest immediate value. It's also, read charitably, an admission that the automated enforcement systems get things wrong often enough that a formal appeals process became necessary rather than, say, a thoughtful gesture of goodwill. (If you've ever had a listing vanish on a Friday evening with no explanation and a support ticket that auto-closes, you already knew this. But it's nice to see Amazon catch up.)

An event framed around seller autonomy lands differently when you remember the commercial reality around it.

Seller fees brought in $46.8 billion in Q2 alone — more than AWS generated in the same period. Amazon's advertising business hit $19.8 billion, up twenty-six per cent year on year. Those fees sit at the centre of the FTC's antitrust case, which goes to trial in March 2027 in Seattle, in which the regulator has argued that advertising has become functionally mandatory and that Amazon's combined fees take close to half of many sellers' revenue.

There's a second piece of context worth holding. Days before Accelerate, Amazon blocked Meta's Muse — an AI agent that shops on behalf of consumers — from its store. The stated reason was that outside agents must identify themselves and follow the rules, which is a perfectly reasonable policy that also happens to be very convenient timing. Set next to the plugin, the pattern is tidy: Amazon is opening the seller side to outside AI agents while policing the buyer side. It decides which platforms the plugin supports, and those platforms (so far) run on models from a company Amazon has invested heavily in.

Every account holder is being offered a free year of Quick. Strip away the demonstrations and that generous onboarding gift, and a harder question sits underneath the week. If every seller gets handed the same assistant, trained on the same platform data, and roughly nine in ten act on its recommendations, then the advantage any single seller gets from it is temporary by definition. Once everyone is optimising pricing, listings and inventory with the same tool against the same signals, the efficiency stops being an edge and becomes the floor.

The tools are useful. Some of them — the account health protections especially — are overdue. But the question worth sitting with is where the value of all this productivity actually lands. It could accrue to sellers as recovered time and healthier margins. It could be competed away as identical optimisation cancels itself out across a category. Or it could be absorbed upward, into a fee structure that already draws more than AWS, as the cost of competing on a platform that just made competing easier for everyone at once.

Amazon spent the week answering how the tools work. It didn't answer that.

✒️ Stori AI - Create and publish branded content effortlessly. Reach your audience across multiple channels and make a lasting impression.

📊 CB Insights Analyst - Your always on AI driven research analyst.

📹 MeetGeek - Automagically record videos, transcribe, summarize, and share insights from every meeting to any tool.

👾 HeyGen - Create personalized outreach videos at scale.

😍It was great, all I needed

🙂Good, not great

🙄It sucked