For most of digital advertising's history, media buying has meant learning how to operate increasingly complicated software. A buyer logged into an advertising platform, selected a campaign objective, chose a geography, built an audience, picked placements, entered bids and budgets, uploaded creative, configured attribution and then spent the following days adjusting dozens of settings as performance data arrived.
Automation gradually took over parts of that process. Platforms began optimizing bids, expanding audiences and testing creative combinations, but the basic relationship between the advertiser and the advertising system remained largely the same. The buyer operated the interface, and the interface told the machine what to do.
Key takeaways
- On September 29, 2026, Amazon Ads introduced Ads Agent: advertisers can create, analyze and optimize campaigns through natural-language conversation.
- Media buying is moving from configuration to delegation. The advertiser describes the outcome, the system works out the execution.
- The hard part becomes choosing what the machine optimizes for. A system fed only CPL will find cheap leads, not profitable customers.
- Agents need structured information about supply. A listing that says only "finance traffic available" gives them nothing to compare.
Amazon Ads Agent and the new interface
Amazon's conversational capabilities are in closed beta for eligible advertisers, with new capabilities rolling out over the coming months, according to its availability announcement.
That relationship is beginning to change. On September 29, Amazon Ads introduced Ads Agent, an agentic AI system that allows advertisers to build and manage advertising through natural-language interaction. The announcement is significant not because Amazon has added another AI feature to an advertising product, but because it illustrates where the interface for media buying is heading. Instead of configuring every part of a campaign individually, advertisers are increasingly being asked to describe the outcome they want and allow an AI system to work out more of the execution.
The difference becomes obvious when comparing how a campaign might have been constructed only a few years ago with the direction platforms are now taking. A performance marketer looking for refinancing customers in Texas might traditionally select the state, construct an audience, choose inventory, define bidding parameters, set a CPA target, build campaign structures and monitor individual placements. In an agentic environment, the starting instruction could eventually be much closer to the actual business objective:
Find high-intent consumers interested in refinancing in Texas and keep the cost per acquired customer below $80.
That sounds like a relatively small interface change. In practice, it transfers a considerable amount of decision-making from the buyer to the advertising system.
Amazon says Ads Agent can help advertisers create campaigns from a conversation, conduct research, generate audience and creative recommendations, analyze performance and make optimizations. It is being integrated across Amazon's advertising products rather than being confined to a single sponsored-ad format, which means the same conversational layer can increasingly sit above different types of inventory and campaign objectives. The wider set of announcements is collected in Amazon's unBoxed 2026 news.
From configuration to delegation
The broader significance is that advertising platforms have spent years accumulating exactly the capabilities required to make this possible. They already have automated bidding systems capable of changing bids in real time, machine-learning models that predict conversion probability, tools that generate and adapt creative, audience-expansion systems, attribution models and enormous amounts of historical campaign data. Agentic AI adds a new layer above those systems: an interface capable of interpreting what the advertiser is trying to achieve and coordinating the tools underneath it.
Media buying is therefore moving closer to a model that resembles delegation rather than configuration.
This transition is not limited to Amazon. Google is moving Search advertisers toward AI Max, where the platform can expand beyond the advertiser's original keyword list, adapt advertising copy and select relevant landing pages (we covered it in Keywords Are Losing Control). Other major advertising platforms have been introducing increasingly automated campaign types in which targeting, bidding, creative selection and placement decisions are handled by machine-learning systems.
The direction across the industry is remarkably consistent. Advertisers provide fewer instructions about exactly how a campaign should operate and more information about the business outcome they want the platform to produce.
What expertise looks like now
That changes what expertise looks like in media buying.
A skilled buyer has traditionally been partly defined by mastery of the platform itself. Knowing how to structure campaigns, manipulate bidding strategies, segment audiences, find overlooked placements and interpret hundreds of settings created a meaningful advantage. Agencies built teams around people who understood these systems better than the advertisers using them directly.
If the interface becomes conversational, some of that operational advantage inevitably becomes less valuable. Knowing where a particular setting lives inside a dashboard matters less when an advertiser can describe the requirement in ordinary language and allow an agent to configure the campaign.
The job does not disappear, but it moves upward.
The difficult part becomes deciding what the machine should optimize for.
The wrong objective, perfectly optimized
That is more complicated than it sounds because advertising platforms are very good at optimizing the objective they are given, even when that objective is a poor representation of the advertiser's actual business.
- An insurance company might ask an AI system to minimize cost per lead. If cheap leads come from users who rarely answer the phone or never purchase a policy, the system can successfully improve the campaign metric while making the business worse.
- A lender optimizing toward completed forms can encounter the same problem if those applications have low approval rates.
- A casino advertiser can generate inexpensive registrations that produce very few depositing players.
The more autonomy advertising systems receive, the more important it becomes to provide them with signals that represent genuine commercial value.
This is why the shift toward agentic media buying is closely connected to another major change in performance advertising: platforms are increasingly being fed deeper conversion data. Instead of optimizing only toward a click or form submission, advertisers can send information about qualified leads, approved applications, purchases, deposits, subscriptions and customer value.
An AI buying system that understands only CPL will look for cheap leads. A system that receives reliable information about which leads became profitable customers has a much better chance of finding more people who resemble them. We looked at why cheap leads are often the expensive ones in The $50 Insurance Lead Is Becoming a Data Supply Chain.
Data is the new advantage
For sophisticated advertisers, the competitive advantage may therefore move away from operating the advertising platform and toward building the data infrastructure around it.
Two companies could use exactly the same AI buying agent and still achieve dramatically different results. One feeds the system superficial conversion events and asks for cheaper traffic. The other provides high-quality first-party data, downstream conversion signals, margins, customer lifetime value and clear constraints around where and how customers should be acquired.
The interface may be the same, but the quality of the instructions and data is not.
Supply has to be described for machines
There is another consequence that receives less attention: agentic buying changes how advertising supply needs to be described.
Human media buyers are surprisingly good at dealing with incomplete information. They can speak to a publisher, read a media kit, understand a niche website, ask questions about traffic sources and make a judgment about whether an opportunity is worth testing.
AI agents need information that can be interpreted systematically.
If an advertising opportunity is described only as "finance traffic available," there is very little for an automated buyer to evaluate. If the same opportunity includes the following, the agent has something it can compare with other opportunities:
- geography and audience characteristics;
- acquisition source and device distribution;
- available volume and pricing model;
- historical performance and traffic restrictions;
- information about the publisher.
This could make structured information about traffic supply considerably more valuable.
Imagine an advertiser asking an AI buying agent to find 10,000 high-intent users interested in personal loans in Spain, avoid incentivized traffic, prioritize direct publishers and keep qualified-lead acquisition below €60. To execute that instruction outside a single closed advertising platform, the agent would need access to a market in which available supply is discoverable and described consistently enough to evaluate.
What this means for traffic marketplaces
That is where the development becomes particularly relevant to marketplaces such as Traffberry.
Today, the primary user of a traffic marketplace is a person. An advertiser searches for available opportunities, looks at publishers and decides whom to contact. As agentic buying develops, the user of the same marketplace could increasingly be software acting on behalf of that advertiser, a shift we explored in The Media Buyer Is Becoming an AI Agent.
That does not require Traffberry to become an automated media-buying platform. Discovery itself can become machine-assisted. The important requirement is that the supply available on the marketplace contains enough structured information for both humans and AI systems to understand what is being offered, the same principle behind our ad network profiles, where every fact is linked to its source.
The same applies to publishers. In a market dominated by human relationships, a publisher can explain the value of its traffic during a conversation with a buyer. In an agent-driven market, more of that value needs to be visible in the description of the opportunity itself. The publisher that provides clear information about audience, source, GEO, quality and commercial terms is easier for an automated system to evaluate than one offering little more than a traffic volume and price. A detailed listing is where that starts.
This could eventually change how traffic is discovered across the wider advertising market. Much of media supply remains fragmented between major platforms, independent publishers, affiliate networks, agencies, direct relationships and private deals. Human buyers navigate that fragmentation through experience and relationships. AI agents may instead search across available sources, compare opportunities and recommend combinations of supply that meet a particular objective.
The advertiser's interaction with that market would look very different. Instead of opening several dashboards, contacting networks and manually comparing media plans, the buyer could describe a requirement and review the strategy assembled in response.
The limits, and who the agent works for
There are obvious limits to how far this can go. Media buying involves brand safety, regulation, commercial relationships and strategic judgments that cannot simply be reduced to an optimization target. An AI system may also make poor decisions when the underlying data is incomplete, biased or misleading. Giving a machine more autonomy does not eliminate the need for oversight; in many cases, it makes understanding the system's decisions more important.
There is also a commercial tension for the platforms themselves. Amazon, Google and other large advertising companies are building agents inside environments where they also own or control much of the inventory being recommended. An agent may be excellent at optimizing within that ecosystem, but advertisers will still have reasons to compare opportunities across platforms and independent media sources rather than delegating their entire acquisition strategy to a single seller of advertising.
That distinction may become one of the more interesting questions in the next phase of ad tech. The first generation of advertising agents is being built by the platforms that sell the inventory. The next generation may increasingly work for the buyer.
If that happens, the media-buying interface could change much more dramatically than the current wave of AI assistants suggests. Advertisers would no longer need to begin with the question of which platform to open or which campaign type to select. They could begin with the business problem they are trying to solve and allow software to search for the appropriate supply.
Learning the language of the buyer
Amazon Ads Agent is one step in that direction. The larger change is already visible across the industry: the language advertisers use to communicate with advertising systems is moving closer to the language they use to describe their businesses.
For twenty years, media buyers learned how to speak the language of advertising platforms. The next generation of advertising platforms is learning how to speak the language of media buyers.
Sources
- Amazon Ads: Introducing Amazon Ads Agent, your AI-powered ad platform (September 29, 2026)
- Amazon Ads: unBoxed 2026 news and announcements
- Google Ads Help: How AI Max for Search campaigns works