Following the Agentic Dollars

AI continues to dominate the conversation in the advertising industry. It’s difficult to make it through a newsletter, earnings call, podcast, or casual industry discussion without the topic eventually turning to AI—or its newer, buzzier cousin: agentic AI.

But for all the excitement, the answer to one question remains surprisingly unclear: Where do so-called “agentic” advertising dollars actually come from?

How does that revenue move from advertiser to publisher? Should it be thought of as a replacement for direct revenue? Is it simply programmatic demand arriving through a new interface? Or does it represent a genuinely distinct revenue category?

As of today, I think there are two primary ways agentic dollars can realistically flow from buyers to publishers: Direct IO Workflow Automation and Agentic via Programmatic.

The diagrams below map those two paths.

Both pathways begin with the same fundamental shift: Automation expands from execution into planning, discovery, negotiation, and decision-making.

That intent might originate in natural language through an LLM interface like Claude or ChatGPT, but it could just as easily come from a planning tool, structured campaign brief, rules-based workflow, or another software system. What matters is that campaign objectives are translated into structured instructions that software can act on autonomously.

The real innovation is not that AI can generate recommendations. It’s that workflows historically dependent on planners, account teams, traffickers, and buyers could increasingly be automated end-to-end: campaign planning, supply discovery, negotiation, activation, optimization, and eventually ongoing performance management.

In a more mature future state, buyer and seller agents could communicate continuously, adjusting campaigns autonomously to achieve advertiser outcomes.

But if we’re going to talk about “agentic revenue” as though it’s a meaningful category, we need a clearer definition.

My working definition is: If autonomous software, acting on advertiser intent, can discover, evaluate, and transact media opportunities across one or more supply partners with limited human intervention, I’d consider that agentic revenue.

I call this a working definition because the category itself might prove temporary as agentic workflows are woven into existing technology.

However, I do believe it’s useful to categorize now to assign value to any work required to capture novel opportunities arising from this technological shift.

Before diving deeper, it’s worth defining a few more terms.

Quick Definitions

AdCP (Ad Context Protocol): An open protocol designed to let AI agents buy and sell advertising through standardized interfaces.

Buyer Agent: Autonomous software acting on behalf of a buyer to discover supply, evaluate options, negotiate, and execute purchases.

Seller Agent: Autonomous software acting on behalf of publishers or sell-side platforms to interpret buyer requests, package supply, negotiate terms, fulfill campaigns, and report results.

LLM: A potential conversational entry point for expressing campaign intent—not necessarily the system executing the transaction itself.

Path 1: Direct IO Workflow Automation

The first path represents the more operationally disruptive model.

In this scenario, a publisher—or its ad server provider—develops a seller agent capable of interpreting buyer requests, packaging relevant inventory, negotiating commercial terms, and accepting buys from buyer agents through AdCP or a similar integration layer.

A buyer or upstream planning system defines campaign goals (maybe with the help of an LLM). That intent is translated into a structured media brief, which is passed to a buyer agent capable of negotiating directly with seller systems.

The seller agent then operationalizes that campaign using existing publisher infrastructure: creating campaigns, configuring line items, applying targeting, ingesting creatives, setting pricing, and managing flight dates through ad server APIs.

At the end of the process, the campaign still runs directly within the publisher’s ad server.

No traditional DSP execution. No programmatic transaction path. Which could eliminate some intermediary transaction fees.

From a publisher’s perspective, this looks functionally similar to a direct insertion order workflow—but with much of the planning, negotiation, and trafficking automation abstracted away.

But while this might seem simple on the surface, there is a long list of existing operational processes that sit between the intent to purchase inventory and an ad actually running. A few examples:

• Approval workflows

• Makegoods

• Trafficking dependencies

• Ad ops QA

• Legal/compliance

• Creative specs validation

• Billing/reconciliation

• Sales channel conflict rules

Someone has to figure out how to design, govern, and operationalize agents to handle the heavy lifting behind each of these tasks and more — which could make this workflow seem more aspirational than short-term reality.

This may not be a fundamentally new monetization model, but it does represent a materially different operational model for direct demand.

Path 2: Agentic via Programmatic

The second path begins similarly but diverges at execution.

A buyer or upstream planning system defines campaign intent (maybe with the help of an LLM). A buyer agent evaluates opportunities and coordinates with supply-side programmatic systems. But instead of activating the campaign directly within a publisher ad server, execution occurs through a programmatic buying platform.

Rather than campaigns being created directly within a publisher ad server, they are created within a programmatic buying platform and executed through familiar programmatic infrastructure (like OpenRTB).

That raises an important classification question for publishers: Is this actually agentic revenue, or simply programmatic demand arriving through a new interaction model?

The answer likely depends on context.

Is this net new spend that would not have entered the ecosystem otherwise? Does supporting it require publishers to expose new metadata, rethink packaging strategies, or modify integrations? Or is this simply workflow automation layered on top of existing infrastructure?

Regardless, if buyer behavior increasingly shifts toward autonomous execution, publishers might need to rethink how they merchandise inventory—even if the underlying pipes remain unchanged.

Are We Reinventing Programmatic?

This is where the conversation gets more interesting.

The most ambitious vision for agentic advertising imagines thousands of buyer agents continuously negotiating, activating, optimizing, and revising campaigns with publisher seller agents in real time.

That might happen eventually.

But the infrastructure requirements are non-trivial.

Even if communication standards like AdCP mature, agentic execution raises operational questions that the industry would still need to address.

How are campaigns reviewed? Who approves creatives? How are standards communicated? Do publishers expose planning and forecasting data? How are implementation failures handled? What happens when autonomous decisions break something?

The interesting twist is that many of these problems have already been solved for programmatic.

We already have standardized creative metadata schemas, creative review workflows, deal syncing, automated error communication, and established reporting pipelines.

Which raises an uncomfortable but worthwhile question:

Do we need to build something fundamentally new, or can the infrastructure we already have evolve to support an agentic future?

Ad platforms already exist to connect buyers and sellers, manage operational complexity at scale, process enormous volumes of metadata, and provide the execution, reporting, and financial infrastructure that keeps the ecosystem functioning.

In many ways, the underlying operational challenges of agentic advertising are not new. Automated activation, reporting, error handling, creative review, deal orchestration, and supply-side coordination are exactly the kinds of problems programmatic infrastructure was built to solve.

That suggests the path forward might be less about inventing something entirely new and more about evolving the infrastructure we’ve already spent decades building.

Of course, operational readiness does not automatically mean strategic alignment.

A more agentic ecosystem could require publishers to expose enough structured supply intelligence for autonomous buyers to make meaningful decisions. That creates an important tension: Greater machine-readable transparency could improve automated buying efficiency, but it could also accelerate commoditization pressure for sellers that benefit from differentiated packaging, controlled access, or human-led merchandising.

And publishers are not the only intermediaries facing change.

Agencies could also see parts of their value chain compressed.

Agentic advertising could partially disintermediate agency workflows that primarily exist to translate advertiser intent into media execution: vendor discovery, packaging comparisons, operational coordination, trafficking, and routine optimization. Those are exactly the kinds of repetitive workflows autonomous software is well-positioned to compress.

Direct workflow automation could reduce the need for some traditional intermediated workflows, but agencies themselves could remain important orchestrators of those systems.

But that does not necessarily make agencies less relevant. Strategic planning, cross-platform orchestration, and accountability could become even more important in a world where autonomous systems increasingly make execution decisions on a brand’s behalf.

The question might be less whether agencies remain relevant, and more which parts of the agency value chain survive automation.

But advertising platforms and agencies do demand their piece of the pie in the form of platform fees, service fees, and transaction economics.

A potential benefit of the direct workflow automation path is that buyers and sellers could reduce some of the fees associated with traditional intermediated execution while still preserving many of the benefits of automation.

Buyers could potentially direct more spend toward media rather than intermediary fees in a world where buyers and sellers manage their own agents. But realistically, they might both need tech providers and partners to manage agentic workflows, for the same reason publishers and advertisers choose not to build their own ad tech stacks.

Will existing platforms ultimately justify their fees by becoming the operational backbone for agentic execution? Time will tell.

In the meantime, we will need to ask if “agentic revenue” is actually a meaningful economic category, or just a behavioral lens applied to existing revenue streams. Even if this proves temporary as a reporting construct, I believe it’s useful in the near term to look at an agentic revenue category purely for planning how to support this evolving buying path.

Agentic revenue might not be a durable accounting category. But it might be a useful strategic classification during this transition, because it highlights changing buyer workflows and infrastructure requirements.

None of this means agentic advertising is overhyped.

Interface shifts matter. Reducing operational complexity matters. Buyer behavior changes matter.

But if much of the underlying execution infrastructure remains recognizable, then the real transformation could be less about inventing an entirely new ad economy and more about changing how that infrastructure is accessed and operated.

Activate with Marketecture Media at Cannes Lions

Marketecture Media is back at Cannes, and this year, we’re doing even more!

From our signature hosted dinners to our Hot Yachts series, 1:1 interviews, and our always-on content studio, we create the content and gatherings where the advertising industry comes together.

Want to work with us? Click the button below.