The open web is moving beyond identity-based systems built around audience data to context systems that understand who a person is, where they are, what surrounds them and what’s happening right now.
We already know what this next stage of open web ad tech will look like.
At my daughter’s recent dance studio gala, a friend quipped to me, “I’ll never retire in this country.” He wasn’t bitter – just joking the way parents do after two hours of sequined choreography, raffles and a DJ jumping between 2000s pop and Disney songs. Then he added, “We’re heading to Italy this summer, and I’m going to look for a place to retire.”
On the way home, while scrolling through Instagram, I started seeing ads for travel to Italy: cheap flights, coastal getaways, Tuscan villas.
My phone’s apps weren’t listening to me. They didn’t need to. They knew I’d spent several hours near someone who had been searching and engaging with topics related to “retire in Italy.” They inferred proximity, interest and timing – context surfaced by models trained on millions of similar moments and applied at the speed of the moment itself.
The advertising industry has traditionally defined context narrowly, simply as the audience that is activated through a given channel. But modern-day tools compile a much richer definition of media activation.
The ad itself, a person’s physical location, the time of day and the social environment around them can now shape how ads are ranked and delivered. The challenge is that these signals are too complex and constantly changing to fit neatly into traditional planning and targeting rules.
That’s where modern context systems come in. Instead of relying on rigid audience segments, they can interpret a real-world moment in real time, combining layers of information to better predict what someone might actually care about right now.
Audience systems vs. context systems
For the last decade, digital advertising has been powered by audience systems: identity graphs, clean rooms and data-collaboration platforms built around deterministic IDs. The major data hubs put themselves at the center of that world, allowing brands and data providers to collaborate on deterministic first-party data without moving it.
That approach was a breakthrough for privacy and interoperability. But it also locked the ecosystem into a static view of what people bought and how they behaved. It is the past tense of intent.
The next generation of AI-backed advertising infrastructure is shifting toward something more dynamic: understanding the present moment. Where is the person physically? What are they reading? What is the creative saying? What time of day is it? Who are they near? What is happening in the real world? What is the probability they will engage with this ad vs. that ad? What is the probability the ad will have long-term influence?
Audience systems explain who. Context systems explain what matters right now.
What containerized bidding actually unlocks
The ad tech industry has recently introduced the first containerized bidding platforms. In simple terms, these systems allow ad-buying software to run directly inside the same infrastructure where ad auctions take place.
On the surface, this architectural shift compresses the latency of ad auctions and optimizes decision-making around which platforms to send queries to and when. It also reduces the costly tax associated with taking data out of the cloud.
But, strategically, it represents something bigger: a shift toward real-time AI-powered advertising decisions happening inside the auction itself. Instead of simply deciding whether to buy an impression, these systems can process contextual signals, run lightweight AI models and evaluate relevance in the moment. The auction becomes less about static targeting and more about live interpretation.
The IAB Tech Lab’s Agentic Real Time Framework (ARTF) is designed to support this model by building containerization into real-time bidding systems. Host platforms run third-party “agent services” as co-located containers within their own infrastructure, communicating over gRPC and Protocol Buffers, mutating the bidstream through a standardized API that exposes only the data each task requires. ARTF’s goal is to reduce bid request/response time by roughly 80%, from several hundred milliseconds to around one hundred.
The complementary spec is Agentic Audiences (formerly the User Context Protocol, donated to the IAB Tech Lab by LiveRamp). Where ARTF defines the container, Agentic Audiences defines the payload: dense vector embeddings that compress identity signals, contextual signals and reinforcement signals. This framework is meant to capture a user’s real-time response to advertising in a format fast enough for in-loop inference.
ARTF is the box. Agentic Audiences is what flows through it. That distinction matters because containerization read in isolation looks like plumbing. Read together with an embedding exchange, it looks like the substrate for a different kind of advertising.
When a DSP’s decisioning logic is containerized inside the supply environment, the traditional limits of the open web fall away. Today, container execution windows run in under five milliseconds. With deliberate colocation and peering, that environment can be expanded toward one hundred. The compression of transit time and the expansion of processing time create a new budget for machine thinking.
Given that thinking time, the bidder doesn’t just decide whether to buy an ad; it can also process the contextual signals flowing through the system, run deep models inside the decision loop and make the auction itself a forum for contextual inference.
To be clear, this is not an LLM reasoning inside the auction. That type of computation will not fit in the window. It is single-pass inference over rich embeddings: The slow, agentic thinking happens upstream, and the auction runs what it compiled.
This is the same hardware-and-software-integrated strategy that allowed Google and Meta to build structural performance advantages by keeping data, compute and decisioning in the same place. Bringing intelligence and execution side by side is how the open web finally behaves like it was intended to.
The result is the kind of relevance that feels like your phone is listening to you without anyone actually needing to listen.
The strategic gravity of colocation
What clean rooms did for first-party data, containerized bidding is now doing for real-time advertising infrastructure.
Modern clean rooms created centralized environments where companies could safely collaborate on data. Joining those networks became necessary.
Containerized exchange environments are building a similar dynamic for real-time media. By running closer to the auction, platforms gain faster access to supply, lower operating costs and more opportunities to improve their models using real-time contextual signals.
From listening to understanding
When I saw those Italy travel ads, no one had been listening to my conversation. They had simply connected proximity, intent and timing into a relevant moment.
That is where the industry has to go: from listening to understanding, from static targeting to contextual inference.
The future of advertising will not favor those who know the most about audiences; it will favor those who also understand the moment and can act on it instantly.
Containerized bidding provides the environment where that can happen. Embedding interoperability provides the language that makes it an exchange. Secure data collaboration provides the memory that makes it meaningful.
Together, they mark the transition from the audience economy to the context economy, where advertising stops chasing people and starts delivering the moment the user intuitively knows they want.
“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Follow Proof in data and AdExchanger on LinkedIn.
For more articles featuring Nathan Woodman, click here.
