Earlier this month, I attended the Agentic Advertising Standards & Community launch presentation. I sat back and absorbed the pitch for what could ultimately develop into a new form of ad buying. The presentation is now available on YouTube for all to see and either laud or criticize.

The presentation was compelling enough that I, like many others in the industry, will have to decide whether agentic media buying is worth dedicating time, brainpower, and resources to in the short term.

With so much constantly changing in digital advertising and AI advancements, knowing where to devote your time is more crucial now than ever. Is this the next revolution in digital advertising or a waste of time like the Google Privacy Sandbox? In any case, it is prudent to approach anything that claims to upend digital media buying with some skepticism while remaining open-minded.

What is AdCP?

The presentation made the case for agentic buying via a new Ad Context Protocol. From the new protocol's GitHub:

Ad Context Protocol (AdCP) is an open standard for advertising automation that enables AI assistants to interact with advertising platforms through natural language.

A collection of companies is forming a non-profit entity to govern AdCP and the standards spun out of the initiative. Curiously, the IAB Tech Lab is not involved at all. On the group's website, they clearly outline the goals of AdCP:

Agentic Media Buying

AdCP looks to facilitate agentic advertising or agentic media buying. If you think about media buying today, there are really two paths: direct and programmatic. 

Direct buys are either set up by advertisers on self-serve platforms or by a publisher on their own ad server on behalf of an advertiser. 

Programmatic buying automates this process by using specialized systems (SSPs + DSPs) to connect inventory to demand and facilitate real-time auctions. 

While technically a form of direct buying, you can consider buying anything on the walled gardens as another form of media buying that requires specialized skill sets and API integrations. 

Each form of buying has its pros and cons. 

• Direct buying ensures no unnecessary ad tech taxes, but it requires human communication and can be hard to scale across publishers.

• Programmatic offers scale and automation, but it can lead to unnecessary fees from intermediaries.

• Walled gardens all have their own way of doing things and can require bespoke integrations, which can be burdensome. 

Agentic buying could extract the best parts of each of these buying paths. It could allow advertisers to automate media purchases via natural language powered by large language models. Here is an example prompt from the AdCP website that a buyer could plug into a buyer agent:

A buyer agent could interpret the buyer's natural-language input and interact with various seller agents to accomplish that goal. But what about activation? How do these campaigns actually get set up and run?

Which brings us to perhaps the most interesting portion of the AdCP launch presentation: a demonstration (22:09). The demo shows a buyer agent, powered by Scope3, purchasing inventory from a publisher (LG) via a seller agent, powered by Swivel, and activating a campaign in the SpringServe ad server. 

It's one thing to talk theory, but this demo clearly illustrates what agentic buying is or could be. In the demo, LG has "products" set up in Swivel that let them define a set of inventory available for purchase, including characteristics such as CPM, format, guaranteed vs. non-guaranteed, etc. Think of an ad "product" like a multi-buyer PMP deal in programmatic.

The Scope3 buying agent uses AdCP to interact with the Swivel seller agent to discover relevant ad products based on the buyer's input. The buyer can then ask the Swivel seller agent to execute a buy using one of the listed ad products and to push a proposal to the seller, who will appear in the Swivel UI for a human to review and approve.

Now we arrive at activation. Once the proposal is approved, Swivel then uses its preexisting API integration with SpringServe to create the necessary objects in the ad server to execute the buy (campaign, line item, targeting, etc.).

Do we need agentic buying?

This is the central question we should all be asking. The presentation highlighted what sucks about each current form of media buying. Let's take a look, and I'll provide some thoughts.

Direct

Requires sales teams to sell and humans to traffic campaigns, which limits scale.

True! It is honestly absurd to me that direct buying still exists in 2025. Humans talking to other humans, then pointing and clicking buttons, feels like an anachronism that we should relegate to history.

Walled Gardens / Ads APIs

Bespoke APIs require custom integration, which requires resources and limits scale.

This is pretty much only applicable to the walled gardens like Meta or Google. Maintaining any custom integration is annoying enough; now multiply that by the number of ad APIs you must integrate with. 

My question is: Are we really banking on these walled gardens integrating with AdCP? If they don't build out seller agents to facilitate agentic buying, then we are stuck with the bespoke integrations. 

Programmatic 

1. Difficult to leverage publisher first-party data on buys

2. Less spend goes toward working media 

3. Cannot access the custom capabilities of publishers

The presentation spent more time bashing programmatic than any other form of buying. Coincidence or intentional? Comments on each point:

1. Publishers can layer first-party data on both PMPs and PGs; however, this requires human coordination. 

2. The biggest knock on programmatic. Unnecessary intermediaries in the mix or high fees take away budget that buyers could dedicate to working media. But Swivel, Scope3, and anyone powering an agent will also need to be paid. Are we swapping SSP and DSP fees for Seller and Buyer agent fees?

3. If publishers offer custom units with no defined programmatic standards, then yes, this is an issue. But the most popular ad products gain traction in IAB Tech Lab working groups that develop standards. A current example is the ongoing development of programmatic pause ad standards. But I will admit this process is long and sometimes painful. 

Agentic buying could solve some of these problems, but questions remain. The demo is slick and all, but when put into practice, will agentic buying offer savings in time and resources to warrant adoption? Will more spend ultimately go toward working media, or will it merely be shifted to alternative players?

My initial feeling is that some form of agentic buying should ultimately supplant direct advertising or programmatic guaranteed. If AI agents can automate negotiations and painful setup processes for simple buys that may use publisher first-party data, that is a no-brainer.

But can agentic buying ever really replace programmatic advertising (at least any time soon)? Sure, it has its issues, but programmatic offers a system that is proven to scale, gives advertisers control by letting their buys ultimately live on platforms they control (DSPs), helps them find the lowest price for inventory, and is a proven ecosystem of features, companies, and products.

For now, I view agentic media buying as an experimental alternative to traditional direct buying paths. Something that could eat some direct or minimal programmatic budgets for testing purposes. If it's easy to set up an MVP for both buyers and sellers, why not open an additional demand channel? But the level of difficulty to test is still an open question.

The bull case would be replacing direct in the short term and possibly eating into programmatic in the long term. The bear case is that this new form of buying proves ultimately too cumbersome to use and implement, and does not add enough value to warrant further development. 

Regardless, I'm here for it and will be following along every step of the way.