Amazon has run its advertising business through the same basic furniture since 2012. Advertisers logged in, built campaigns, set bids, pulled reports, and if they were large enough, graduated to the separate world of the demand-side platform for programmatic display, video and audio. At unBoxed this year, Amazon announced that the Ads Console is being retired and replaced by a single agent-powered platform called Amazon Ads Agent.

That sentence deserves a moment. Last year's unBoxed consolidated the stack, merging sponsored ads and the DSP into one campaign manager. This year's goes further and changes its nature entirely. The dashboard—the thing an advertiser opened and operated by hand—is being replaced by an agent the advertiser instructs in plain language and then supervises. Amazon is rebuilding the engine on two fronts simultaneously: how advertisers buy, and where their ads actually appear to shoppers.

Amazon's framing for all of this is simplification, and the complexity of running campaigns across formats and consoles is genuinely real. But a rebuild around a single agent does something else at the same time, and it's the thing worth watching. It removes the surfaces where advertisers could see—and contest—what they were actually buying. (More on that in a minute.)

Amazon Ads Agent brings sponsored ads and the DSP into one connected experience with a single conversational interface for planning, building and optimising. Underneath it sit three campaign types.

Sponsored ads continue as before. Full-Funnel Campaigns let an advertiser set one budget, add products, and hand targeting, creative generation and pacing to Amazon's AI across both sponsored and display, video and audio inventory. The third type—display, video and audio—comes in two flavours: a plus version that hands optimisation to Amazon against an objective you set, and an advanced version that keeps the full programmatic toolkit and manual control, now with an AI assistant sitting alongside it like that colleague who keeps suggesting you "just let the algorithm handle it." The conversational experience is in expanded beta, with a US launch signalled for January.

The pain point Amazon cites is concrete, and most operators will recognise it immediately. The company says the majority of traders feel overwhelmed by campaign complexity and that teams lose more than ten hours a week to manual campaign management. (Ten hours a week. That's an entire working day spent clicking through tabs, which makes you wonder what those tabs were doing for us in the first place.)

The early performance figures Amazon put forward are, as ever, its own and unaudited, but they point at the pitch: beta testers of Full-Funnel Campaigns reportedly saw materially higher long-term ROAS and a lower cost to acquire new-to-brand customers, with named test brands reporting a double-digit reduction in the cost of a product-page view.

What this means for advertisers depends heavily on size, and the design admits as much. One algorithm managing an entire budget with no segment-level control is a fast, guided way in for a smaller advertiser who wants full-funnel media without building each campaign from scratch. For an enterprise team running a complex, multi-market portfolio, the same arrangement looks rather different, because the segment-level visibility that governance depends on is exactly what the simplest mode does not offer. The trade at the heart of the buy-side rebuild is control for automation, and the right answer is not the same for a brand spending a few thousand a month as for one spending millions across markets.

The buy-side changes would be significant on their own. The demand-side change is the more structural one, because it alters where an ad even is.

Shoppers are increasingly researching and buying through Alexa for Shopping rather than through a results page. The new format, Branded Conversations, is built for that behaviour. It begins from a sponsored brands ad and turns a single question-and-answer exchange into a multi-turn dialogue, with the assistant drawing its answers from the brand's own catalogue, product guides, compatibility details, imagery and video. An optional brief lets the advertiser steer what the assistant emphasises. It charges only for the first prompt click, and Amazon has pointed to earlier conversational placements converting markedly more often. General availability is expected in the US in January.

It is worth naming what is actually being sold here, because it is not what advertising has historically sold. A display ad buys attention. A search ad buys intent. Both leave the actual decision to the shopper. An ad inside a conversation buys something closer to a seat within the relationship the assistant already has with that shopper—an assistant that holds their purchase history, their preferences, and has earned a measure of persuasive authority through months of ordinary, non-commercial use.

That is a more valuable position and a substantially more opaque one. And it's the direction Amazon is now building its highest-intent inventory toward. (If the phrase "opaque advertising position built inside a trusted relationship" doesn't make the regulatory lawyers sit up slightly straighter, nothing will.)

The practical consequence for advertisers is that content operations become media. A conversation is only as good as the material feeding it, so keeping product guides, compatibility information and imagery current stops being a back-office chore and becomes the raw material an AI draws from in front of an actual buyer. Visibility is no longer a ranking position to win on a page. It's the quality and completeness of what a brand has given the assistant to say on its behalf.

Several further announcements fill in the picture of an engine that does more of the mechanical work itself. Generative creative tools now sit inside the conversational experience, producing display ads and television-quality video from a brand's existing product assets. Measurement is being pulled into the same natural-language surface, with Amazon saying advertisers can reach close to a thousand metrics by asking for them in plain English and receive a full business review—visualisations, peer benchmarks, the works—in minutes rather than through a dashboard.

For agencies and technology partners, two quieter announcements matter more than they look. Amazon is adding a lighter way for outside AI agents to use its advertising tools, exposing a handful of meta-tools that find and invoke the right capability on demand rather than forcing an agent to hold more than a hundred tools in memory at once. A companion tool turns a plain-language question into a validated query against Amazon Marketing Cloud and is due to be priced at cost when it reaches general availability next year.

The through-line for advertisers is that the specialist skillset is moving up the stack. The value is shifting away from the person who can navigate the console or write the query and toward the person who knows which question is worth asking and what to do with the answer. Writing the query was never the hard part. Knowing which audience, which activation and which measured outcome the answer should lead to—without that becoming its own separate project—is the work that remains, and it is more strategic, not less. (Good news for anyone who's spent years building actual expertise. Less good news for anyone whose entire value proposition was "I know where the buttons are.")

Here is the tension the conference was not built to dwell on. When one agent plans a campaign, builds it, optimises it and reports on its performance—and when a growing share of the inventory it buys sits inside private conversations the advertiser cannot observe—"simpler" also means "less legible." The advertiser sees an objective, a budget and a result. The decisions in between—which audiences were reached, which placements were paid for, how the auction resolved—increasingly happen inside a system the advertiser instructs rather than inspects.

That would be a reasonable trade in a vacuum. It is a sharper question outside one. Amazon's advertising business generated close to $20 billion in a single recent quarter, and it is doing this rebuild while facing a lawsuit from the FTC and a coalition of state attorneys general that alleges, among other things, opaque auction surcharges that advertisers could not see. Amazon disputes the claims, and the case is set for trial next year. The point is not to litigate the suit here. It is that the agentic rebuild concentrates both the controls and the opacity in the same hands, at precisely the moment a regulator is arguing the opacity was already a problem.

For advertisers, that raises the value of the one thing the agent does not provide: independent verification. A second set of eyes on what the machine is actually buying is looking less like a luxury and more like a control function.

A handful of other launches are narrower but worth noting. Sponsored Services extends Amazon's advertising into local service providers for the first time, matching shoppers with qualified providers based on what they've just bought. Review Requests, in open beta for lower-review products, lets advertisers ask verified buyers to leave a review and charges only when a shopper clicks to submit one—a direct lever for newer listings trying to build trust quickly.

The less visible plumbing may matter most over time. Amazon is pushing sales-penetration data straight into DSP targeting, bringing content-level signals into its analytics workflow, and redesigning the data feed so performance data lands faster and more completely. For teams that react inside live event windows, a materially quicker feedback loop changes what is possible to do in the moment, even if it never makes a keynote slide.

The second day of the conference argued that using AI only to do existing work faster and cheaper is the least interesting thing to do with it, and that the real prize is reimagining what the work could be. That's a fair challenge, and it cuts at Amazon's own pitch, which is still largely an efficiency pitch dressed in transformation clothing.

The harder question for advertisers sits underneath it. If every brand runs the same agent, against the same inventory, with the same optimisation handed to the same AI, then the efficiency stops being an edge and becomes the baseline—the way last year's automation already has. What's left to compete on is the material that feeds the conversation and the judgement about which decisions still need a human holding the pen.

And there is a question Amazon didn't answer at all, which is whether "simpler" can survive contact with a trial about transparency. An engine that is easier to use and harder to see into is a reasonable bargain right up until the moment an advertiser needs to know exactly what they paid for, and why.

📖 Clay - Enrich and scrape web data to build hyper-personalized prospecting lists.

📹 InVideo - Create and edit videos from a text prompt for social and marketing.

🌐 Consensus - Search engine that pulls evidence-based answers from scientific papers.

🪶 Arcads - Generate UGC-style video ads from a script using 1,000+ AI actors.

😍It was great, all I needed

🙂Good, not great

🙄It sucked