CNN builds in-house agent infrastructure as it prepares for AI-driven media trading

“In terms of having it as a full trading model and a new revenue generator, I would say Q1,” Karmali told Digiday. “In Q4 we’d like to be ready for people who want to test budgets, so that when we want to begin to trade against it fully, it’s already tried and tested,” he added. 

Naturally, this timeline isn’t set in stone — it could shift as agentic protocols and the broader landscape continue to develop and evolve, which will affect the timings, he said.

CNN’s plan pairs in-house development of agentic infrastructure with third-party tech and trading vendors, including DSPs, though he wouldn’t give specific partner names. “We’re building out our own infrastructure to see how that can be traded. That’s being led out of the U.S., but internationally we could adopt that quite easily,” he said. 

It will also work closely with the IAB Tech Lab and its agentic roadmap framework to ensure the scheme and protocols it’s rolling out are read correctly and translate into the right outcomes on the buy side. It has been scoping for all protocols available, including AdCP.

The idea is that if CNN implements this soon, even without revenue flowing, the system can learn how the agents communicate and operate. That’s a contrast from what publishers have faced traditionally with OpenRTB or other programmatic ad tech, often needing to wait for buy-side adoption of new tech for it to warrant their own investment, per Karmali. 

“This is a learning protocol, which means that if we implement it now and there’s no revenue being traded, when it does start trading, it’ll be super efficient. It’ll kind of leapfrog that learning step,” he said. 

Agentic protocols are the rules and standards that let AI agents interact with each other and with systems autonomously. Karmali said the publisher’s main focus has been on the agent-to-agent communication, figuring out how the buy and sell-side agents can request information from each other, negotiate terms on pricing and usage, and delegate tasks. Ensuring those are consistent is paramount, stressed Karmali. 

Other publishers are also exploring both in-house and third-party partnerships with tech platforms to explore sell-side agent development. News Corp is developing its internal agentic infrastructure for its titles, which include The Sun tabloid. Meanwhile, the buy side is also running tests. Independent media agency Butler/Till recently concluded the first test of a programmatic media-buying agent on a campaign for brewer Geloso Beverage Group. The experimental run successfully cut intermediary fees by over 80% and reduced CPMs while hitting industry benchmarks on fraud and inventory standards, according to the agency and its test partner PubMatic.

For now, there are more examples to point to on the buy side. Scott Messer, principal and founder of Messer Media, said while several tests are underway, having fully-fledged agentic media trading is still some way off. Most sell-side examples still are heavy on agentic campaign setup, rather than agentic trading. In the early stages, he expects the bulk of cases to be reservation setups of small buys across more publishers – guaranteed, pre-arranged deals between buyers and publishers, as opposed to open, real-time auction buying, or private marketplaces. 

For CNN, the focus is on ensuring the agent-to-agent communication is consistent between buyers and sellers. “We want to have conversations with the buy side to ensure that the context protocols they’re using are consistent with what we’re using, so both our agents can trade efficiently,” said Karmali. “So we’re always open to talking to any partner on the marketing side who wants to whiteboard this and see what’s the preferred way of doing it.”

Publishers can build their own sales agents and run them alongside external ones, creating a model of “agentic abundance,” where multiple agents operate in parallel, according to Andrew Mole, CEO and founder of agentic advertising platform PubX, which is integrated with thousands of publishers, he said. While publisher-specific agents sell only their inventory, broader platform agents aggregate supply and negotiate with buyers across multiple publishers.

“Part of what we’re building outside of the protocols is tooling for publishers so they can have control,” said Mole. “So they see who’s buying, and can go and speak to those advertisers directly and say: ‘hey, you’ve spent quite a lot through the PubX sales agent, why don’t we do a big marketing event together, or something more in-depth like that,’” he added. 

The current aim is to test and learn as much as possible, work closely with the buy side on alignment, incorporating it as “another string to the trading bow,” said Karmali. Initially, agent-based trading will likely be most effective at handling lower-funnel performance-driven activity, like conversions, direct response, or bottom-of-funnel demand, because it’s highly data-driven and optimizable in real time. Then it has the potential to move up the funnel (areas like branding and discovery) once payments and commerce capabilities are integrated, per Karmali. 

“We’ve got an opportunity here to kind of start from scratch, start a new way of trading, which we don’t get often. So I think just having that consistency across both sides [buy and sell-side agents] is super important,” he added.