Andrew Mole, the CEO and co-founder of UK-based agentic AI ad startup pubX, has a theory about why agentic ad tech has a tendency to hallucinate.
In his view, it’s not the LLM’s fault. Hallucinations happen when a large language model queries a demand-side platform that doesn’t know how to talk to an agent. Legacy DSPs just weren’t designed for it, he says.
So Mole built one that is.
PubX, which was founded in 2020, started out on the sell side helping publishers price and optimize their inventory. This year, it launched buyer agents to serve brands and agencies so it can handle both sides of a transaction.
But for brands to hand their media budgets over to an agent, they need to trust what it’s doing.
So, on Tuesday, pubX announced its acquisition of Compliant, a media quality and governance startup co-founded by former marketers, among them Jamie Barnard, who spent more than 15 years at Unilever, including as its general counsel for global marketing, media and ecommerce.
On top of that, pubX also announced a $5 million Series A led by Chicago Ventures to fund its expansion into the US, build out the demand side of its business and integrate Compliant’s data integrity signals and transparency metrics into its platform.
Mole declined to share the Compliant deal price, but one can extrapolate based on the raise.
Agent, meet agent
PubX’s buyer and seller agents transact directly with one another without routing through a DSP or SSP.
A brand or agency feeds a brief into the buyer agent, then the agent queries publisher seller agents, analyzes their inventory and audiences, drops what doesn’t meet the brief’s criteria – while also explaining why – and executes the buy.
There’s a live log that lets buyers watch exactly what an agent is doing at every given step.
“You can pop out a window and watch the rationale,” Mole said. “For example, ‘I’ve just analyzed 56 audiences from this publisher and I’ve dropped 30 of them, because they don’t need the criteria in the brief.’”
The level of autonomy is calibrated by the client. A brand can require approval at every stage or just let the agents do their thing. “As trust is earned, autonomy grows,” Mole said.
The infrastructure runs on Ad Context Protocol (AdCP), which is an open standard for agent-to-agent advertising developed through AgenticAdvertising.org, a consortium of ad tech companies of which pubX is a founding member.
The IAB Tech Lab is building its own framework, called ARTF, and pubX is a member there, too. But Mole is candid about his preference for AdCP, because it’s “committed to building and shipping code.”
The Tech Lab’s approach, he argued, risks being shaped by the very companies that have the most to lose from a real reset.
“We’re part of the Tech Lab because we think we can maybe change things from within,” Mole said, “but we favor AdCP because it doesn’t have any of that baggage; it’s a completely fresh start.”
The quality question
Compliant’s job, meanwhile, will be to make sure that pubX’s agents are buying the right stuff.
It produces data integrity and CTV quality indexes that distinguish high-quality, premium inventory from impressions that shouldn’t be bought at all. Inside pubX, those signals will be baked into an agent’s decision-making process and also help with optimization.
Think of it as a prebid filter of sorts.
“The biggest barrier to adoption of agentic tools will be any uncertainty over whether they can be trusted to make the right decisions,” said Barnard, who will become pubX’s president of global demand growth. “This gives CMOs the assurance that quality and trust are built into the fabric of it.”
And they need all the assurance they can get on the open web. Call it PTSD. Although, to be fair, buyers, despite their handwringing, are often complicit in the proliferation of crappy inventory.
“Waste is the worst kept secret in Adland,” Barnard said. “A lot of senior leaders in media are quite happy to tolerate buying low-value, low-quality media as a cost of doing business.”
The innovator’s dilemma
At the same time, legacy DSPs are trapped by their own business models.
It’s not as if they can tell shareholders they’re going to go dark for a year to rebuild from scratch so, instead, they bolt new agentic interfaces onto old pipes, Mole said. It’s the classic innovator’s dilemma.
But the economics of running agents on legacy infrastructure will eventually force the issue, Mole said.
“Silicon Valley is letting us all eat for free right now,” he said. “But as soon as the cost of tokens has to be priced in, the cost of running an ad is going to go up.”
But there’s a bigger question lurking here: the future of the open web itself. On one side, AI is shrinking referral traffic and pulling attention into closed systems; on the other side, companies like pubX are trying to optimize a market that many think is dying.
That’s a little reductive, though, Barnard said.
One of the reasons so much media spend goes to the walled gardens is because their tools are so easy to use. “Agentic is going to bring that level of simplicity to the open web.”
Not that it’ll happen overnight. There will be a “liminal period,” Barnard said, during which DSPs and SSPs continue as they are, adding some agentic tools as leaner, newer players gain ground.
“The writing is on the wall for the old programmatic stack,” Barnard said.
