For many publishers, there are really two stories of AI.

The first is the dark one: the story of web publishers whose content is scraped, remixed, and redistributed without compensation. It threatens the very foundation of the open and free web. A web that, in large part, is funded by advertising.

The second story is brighter, and it’s the one I want to focus on here. It’s the story of publishers and advertisers leveraging modern AI. More specifically, large language models (LLMs), autonomous agents, and expert assistants that can help to better align marketing goals with audiences.

Both stories matter. But let’s stay with the second one for a change.

Why AI?

When we first started experimenting with LLMs, my mental model was simple: imagine platform workflows augmented by a digital advertising version of Clippy. Smart assistants that could translate natural language into SQL queries, run an audience model, or generate a campaign report. Some of these assistants would be useful, others less so, especially since, as I’ve learned, humans will do almost anything to avoid having to talk to a bot. But the picture was still one of “tools that help humans get stuff done.”

Now that we’ve been at this for some time, the more exciting question has emerged: what happens when these assistants aren’t just helpers, but collaborators?

Allow me a quick digression. Having been around digital advertising for a while (and perhaps this is the old man in me talking), I remember when advertisers and publishers actually talked about ad campaigns. Communication often happened over email, sometimes on the phone. Video calls weren’t the norm yet. The cast of characters included publisher sales reps and media agency buyers, with ad ops, analysts, and data teams on both sides.

Sounds inefficient, right? And it was. But there was a benefit to that complexity: when decent people talk directly, trust and accountability come built in. You tend to share more about your actual goals. You try harder to make things work. Reputation and repeat business were always on the line.

Fast forward to today: real-time auctions have brought incredible efficiencies, but they’ve also eroded some of that organic incentive alignment. The frictionless protocol stripped away some of the humanity and creativity that is core to advertising.

So here’s the emerging question: putting aside cost of sale and scale challenges for a moment, what if advertisers and agencies could once again design and run campaigns by interacting more directly. Not just with humans, but with agents acting on behalf of humans, using the same kind of natural language to exchange thoughts and ideas and help align goals and objectives? Could this better align advertiser goals with audiences? Could it make advertising feel better, behave better, work better?

That’s worth getting excited about.

What We’ve Learned So Far

Working with LLMs has taught us something both obvious and profound: they are dazzling at simulation, but shallow at self-awareness. They can produce fluent, convincing responses and even mimic reasoning, but they have no reflexive capacity to recognize when they’ve gone astray. The outputs can be brilliant in one moment and dangerously wrong in the next, with equal confidence.

This duality means that while LLMs can simulate many human workflows, from drafting creative copy to parsing data schemas, they must be deployed carefully. They thrive when given the right scaffolding: a contained problem space, clear context, and a well-bounded objective. Left unconstrained, they risk drifting, fabricating, or amplifying error.

That’s why the path forward hasn’t been about building one all-knowing digital assistant, but rather assembling constellations of specialized agents. Each is trained or prompted for a narrow domain of competence: a media planning agent, a data transformation agent, an audience segmentation agent. These agents are better behaved precisely because their world is smaller and the teams building them can impart context and create clear guardrails based on deep domain knowledge. Instead of asking them to “understand everything,” we’re asking them to execute well-defined tasks within a shared environment.

Of course, even narrow specialists are only as useful as their ability to work together. And this leads to the bigger frontier.

Agentic Collaboration in Digital Advertising

If specialized agents are the components, collaboration is the system. Real work rarely lives inside the walls of a single agent. Planning a campaign, for instance, may begin with a creative brief that must be interpreted by one agent, handed to another to model against audience data, and then routed to yet another to generate media plans and validate them against inventory. Without coordination, this quickly becomes chaos.

This is why the idea of Agentic Collaboration is so compelling. It is not enough to build competent agents; we need ways for them to communicate, to delegate, to negotiate, and to reconcile. Inside an organization, this means establishing frameworks where multiple agents can operate on the same context, share state, and pass tasks fluidly without losing fidelity. Across organizations, the challenge becomes even more interesting: what happens when an advertiser’s agents need to converse directly with a publisher’s agents, or when third-party specialist agents need to be introduced into the workflow?

At that point, protocols matter. Just as real-time bidding was only possible once the industry coalesced around shared protocols and standards for describing inventory, price, and demand, agentic collaboration will require structures for intent, context, and trust. Compelling new protocols such as MCP and A2A have emerged to support a new infrastructure across our industry. If agents are to transact meaningfully on behalf of their human principals, we will need conventions for verifying what they can and cannot do, and for ensuring that their exchanges reflect not just efficiency, but accountability.

The product implication is enormous. Platforms like Optable’s are no longer just facilitating data exchange or workflow automation; they are becoming the medium in which agents collaborate. That means exposing enough functionality and data to make collaboration useful, while constraining enough to keep it safe and aligned. It means thinking carefully about how agents identify themselves, how they signal authority, and how they fail gracefully when they don’t know the answer.

If this sounds familiar, it should. In some sense, we are circling back to the kind of direct communication that once characterized the industry only now, the conversations are mediated by software agents that can operate at machine scale and speed. Done well, this can fundamentally upgrade the methods to establish trust, alignment, and shared understanding in digital advertising.

What We’re Up To

At Optable, our starting point was building a platform where publishers and advertisers could collaborate transparently on data, without intermediaries diluting the signal or hoarding the value. That same conviction now extends naturally into the world of AI.

Agentic collaboration, to us, is not about replacing humans. It’s about restoring what was lost when programmatic scaled: the trust and alignment that come from two parties working directly toward a shared objective. The difference is that now, agents can help scale those conversations across thousands of campaigns, billions of impressions, and an ecosystem that demands both speed and precision.

Our role is to provide the medium where this can happen safely, particularly from the perspective of publishers, the creators that make the free internet possible, and the custodians of user and audience data. That means building the infrastructure where specialized agents, some ours, some yours, some built by our partners, can meet, exchange, and work together under clear rules. It means designing protocols that enforce accountability and safeguard data, while allowing creativity and experimentation to flourish. And it means ensuring that when agents collaborate across organizational boundaries, they are still serving the fundamental goals of the humans they represent.

This isn’t just a product roadmap. It’s a philosophy. Advertising at its best has always been about connection: between brand and audience, between publisher and advertiser. By enabling agentic collaboration, we have the opportunity to bring that connection into the AI era not by automating away the conversation, but by amplifying it.

We’re still at the beginning of this journey, but beginnings matter. And if history has taught us anything, it’s that the structures we design now, meaning the protocols, the incentives, the norms, will shape the industry for decades to come. Our intention is to help shape them in a way that makes advertising more accountable, more collaborative, and, ultimately, more human.

‍