Ad agencies have an overwhelming amount of buying tools to choose from. And they’re increasingly relying on AI to focus just on what will help them meet their clients’ needs.
Between general-purpose and specialized DSPs, social platforms, CTV environments and direct publisher deals, agencies have to stitch campaigns together across a variety of fragmented systems. Given this fragmentation, ad tech vendors see an opportunity to create a unified ad-buying platform built on agentic AI.
Kargo, an SSP that specializes in high-impact ad formats for mobile, desktop and CTV, is the latest platform to launch a new agentic buying interface. It announced last Tuesday the release of Project Kera, a chat-based planning and buying environment that feels similar to consumer AI tools, but that’s tuned for media buyers. The solution is currently in closed beta, with Hershey, travel media network Navigator and independent ad agency Wpromote testing full campaigns.
Kargo’s play is to create a unified agentic buying interface that will “aggregate all biddable and non-biddable inventory into a single platform,” its CEO, Harry Kargman, told AdExchanger.
But Wpromote is approaching Project Kera less as a standalone tool and more as part of a broader effort to build its own AI-powered media operating system that sits on top of all the other various platforms.
“Technology and agents are really about enabling marketers to be more strategic in decision-making and orchestration across systems,” said Skyler McGill, head of programmatic and video at Wpromote. “It’s less about just moving faster and more about creating a step change in how we operate.”
Streamlining creative iteration
Wpromote has been testing Project Kera for about four weeks. According to McGill, the early testing phase has centered on enhancing the agency’s existing workflows. The biggest impact so far, he said, has been in speeding up the process of going from a human-created campaign brief to a programmatically activated campaign.
To start using Project Kera, buyers log in to its chatbot interface and describe a goal, such as growing brand awareness among specific audiences. They can upload brand assets, including videos, images and copy, and tell the platform where they’d like these assets to appear across streaming TV, social feeds and premium news sites.
The platform then generates a plan with suggested creative uses (such as recommending pause ads, large display units or social videos) and a channel mix with the budget split across Meta, TikTok, Pinterest, connected TV and the open web.
Once the buyer reviews the proposed plan and hits go, the platform starts pushing ads live through Kargo’s SSP integrations.
“What’s really exciting is the ability to operationalize a campaign brief and build creativity at scale, then optimize in real time,” McGill said. “That’s a big shift from the manual processes we’ve had.”
That shift has already allowed Wpromote to test and iterate on campaigns at a much faster pace and at greater scale than before, more akin to a social platform’s testing workflow, McGill added. The agency has so far used Project Kera to test different high-impact CTV formats, such as pause ads and immersive overlays, as well as bespoke display units.
Concrete performance results are still a few steps away. McGill said it’s too soon to see measurable outcomes from these test campaigns. But Wpromote anticipates that Project Kera could drive meaningful lift once the agency’s workflows and prompts are more fully refined.
Building the OS, not buying it
While Kargo envisions a unified buying environment, agencies like Wpromote are more likely to treat Project Kera as one piece in their arsenal of AI tools.
For instance, Wpromote built an internal AI tool, Polaris IQ, which handles bid management, budget allocation, media planning and measurement. The agency’s goal is to create its own centralized decisioning layer that sits above DSPs, SSPs and other buying tools.
That approach also explains how Wpromote is testing the market. Rather than betting on one platform, it’s piloting multiple agentic partners, McGill said, including Kargo and PubMatic on the SSP side, while watching DSP-led efforts from companies like Yahoo and Amazon.
So far, much of Wpromote’s experimentation has leaned toward SSP-provided solutions. Those platforms have moved faster to bundle inventory, creative formats and activation into unified environments, said McGill, making them a more practical entry point for agentic workflows. DSPs are catching up, he added, but for now, SSPs offer a clearer testing ground.
Long term, McGill expects a more interoperable model.
“It really comes down to where you can drive the best outcomes and reach the right audiences,” he said. “You’ll have a lot of optionality, and the systems will need to connect.”
From beta to buy-in
Still, speed and automation aren’t enough to secure a place in the stack. Wpromote is also evaluating Project Kera against familiar agency criteria: performance, efficiency and control.
On the performance side, the agency is looking for measurable lift across both brand and performance metrics, McGill said, whether in the form of improved reach and attention or increases in conversions and return on investment.
Operational gains matter, too, he said. Faster campaign setup, reduced manual work and better real-time insights could all justify deeper adoption.
But ensuring strong testing guardrails is just as critical as tracking performance gains, McGill said.
As agentic systems take on more responsibility, agencies are wary of too much automation and black-box decision-making. Data quality and transparency remain top concerns, especially when campaigns are being shaped through prompts rather than traditional workflows.
“We’re being really mindful of overreliance on automation,” McGill said. “You still need the right data inputs, governance and QA processes in place to understand what’s actually happening.”
That caution reflects a broader industry reality.
Agentic platforms promise to unify planning and execution, but agencies aren’t ready to hand over control. Instead, they’re building systems that let them decide when and how to plug those tools in.
