The ‘is a hot dog a sandwich’ problem in AI advertising

Summary

---

The ‘Is a Hot Dog a Sandwich’ Problem in AI Advertising

The Digiday Podcast
Published: November 18, 2025
Host: Kamiko McCoy & Tim Peterson
Guests: Seb Joseph (Executive Editor of News, Digiday), Ronan Shields (Senior Ad Tech Reporter, Digiday)

Episode Overview

This episode unpacks the explosive evolution of AI agents within digital advertising, focusing on emerging standards designed to bring order to a rapidly automating, increasingly complex programmatic ad ecosystem. The hosts and guests dive into the practical, ethical, and technological challenges of integrating AI agents—from arbitrage and agency transparency issues to new protocols (Ad Context Protocol—ADCP and Agentic RTB Framework), all while expressing skepticism and hope in equal measure (“Is a hot dog a sandwich?” becomes the metaphor for the ambiguities and debates embedded in these new technologies).

---

Key Discussion Points & Insights

1. Ad Agency Arbitrage & Transparency

• The episode starts by exploring a recent eyebrow-raising move from Paramount, who revealed that agencies like Publicis and IPG are not only handling their media buying but also selling their ad inventory (01:00–05:00).
  • This sparked industry concerns about arbitrage: potential kickbacks, savings guarantees, and whether agencies push inventory from their media-owner clients onto other advertiser clients.
  • Tim: “This Paramount example seemed to, I think why it stood out so much is because it seemed to initially confirm that. Now again, the companies have walked that back and said, that's not going on here.” (06:19)

Notable Moment:

• Kamiko: “I feel like there was this Whitney Houston meme... somebody on that earnings call who whipped their head around to say, ‘Did everybody else just hear what I just heard?’” (03:49)

2. Rise of AI Agents in Ad Operations

• Major tech companies (Amazon, Google, Meta) are rolling out new AI ‘ad agents’—essentially chatbots or tools for campaign planning and optimization (09:41–12:00).
  • These agents promise to speed up routine tasks (ad buying, analytics, etc.), but questions remain about control and oversight.
  • Meta’s Advantage+ tool, for example, auto-generates ad creatives, sometimes with questionable results because of a default setting many didn’t know about.

Notable Quote:

• Kamiko: “If you don't also have your human standing by with a leash and a collar now, you end up with weird AI generated ads... AI sheen happening here.” (12:18)
• Tim: “One of the challenges with these AI tools is it makes it super easy for people to be really lazy.” (13:14)

3. Guardrails, Prompt Engineering, and the Future of AdOps Skills

• The need for guardrails is a recurring theme—ensuring AI agents don’t go rogue requires new skills (like ‘prompt engineering’) and diligent oversight:
  • Everyone in marketing may need a basic ability to “prompt engineer,” similar to everyone once needing basic typing skills (14:42).
  • Human oversight remains vital: “The humans are going to need to stay in the loop.” (14:31)

---

Industry Frameworks: ADCP & Agentic RTB

4. What Is the Ad Context Protocol (ADCP)?

• Explained by Ronan Shields: “It is an open source standard designed... to bring more accountability and structure into Programmatic Trading.” (24:09)
  • Acts as a common language or set of rules for AI agents to communicate across the programmatic ad supply chain (27:46).
  • Developed by industry players like Yahoo, PubMatic, Scope3, Optable, Swivel, and Triton Digital.
  • Metaphors abound: ADCP is described as pipes, a standardized language, or even “the slang” of advertising AI agents.
• The ‘Hot Dog’ Metaphor:
  • Tim: “...that’s similar to me saying to an AI agent, make me a sandwich... it could come back to me with a hot dog. In which case then we're getting into the conversation of is a sandwich a hot dog?” (29:00)

Key Viewpoints:

• Seb: “Better to think of it as a language, a standardized, common language for AI agents...” (27:46).
• Concerns around data quality: “If the data driving these… workflows is kind of weak, then the protocol is just going to move the mess differently.” (32:21)

5. Agentic RTB Framework (ARTF) & Speed Concerns

• Tim explains the IAB Tech Lab’s ARTF, which sets a specification for enabling AI agents to participate in the Programmatic Bid Stream (15:43).
  • Designed to address the latency problem: traditional real-time bidding takes ~400–600 ms, which is much faster than current AI chatbot response times.
  • Solution: “containerization”—moving AI agent code closer to servers processing ad auctions to accelerate responses (16:36–18:08).
  • Potentially could speed up RTB by up to 80% rather than slow it down.

6. Is the Industry Ready? Adoption, Fragmentation, and Data Cleanliness

• Much debate over whether these protocols are arriving too early given messy, incomplete, or inaccurate data at many publishers (33:05–34:36).
  • Many major players (Amazon, the “walled gardens”) haven’t signed on, risking fragmentation and standards wars.
  • “It’s almost just as interesting who isn’t backing ADCP as it is who is.” — Seb (34:36).

Memorable Analogy:

• Tim: “It feels like AI agents are kind of like the flying cars of advertising… Maybe a monorail system [is safer].” (51:49)

---

Challenges and Cautions

• Brand safety: There’s industry worry that early mistakes will lead to outsized brand-safety blunders as “AI agents learn on the job” (31:48–33:05).
• Lack of inclusivity: Protocols are being steered mostly by vendors, with little input so far from advertisers, publishers, and broadcasters—mirroring early mistakes in the last era of programmatic standards (50:28).

Foresight Warning:

• Kamiko: “It seems like guardrails, safety and things like this have taken a backseat when it comes to AI. The focus is on efficiencies, curbing costs, faster—not necessarily better...” (46:14)

---

Timestamps for Key Segments

• 00:10–05:09 – Arbitrage, agency kickbacks, and the Paramount Publicis/IPG deal
• 09:41–14:31 – Rise of AI ad agents; pitfalls with Meta’s Advantage+; the blessing/curse of automation
• 15:43–18:08 – Agentic RTB Framework explained (containerization and speed)
• 24:09–30:28 – Deep dive: What does ADCP actually do? (plumbing/language metaphors)
• 33:05–37:05 – Adoption challenges: Brand safety, data quality, and why the “walled gardens” might not play ball
• 39:07–42:51 – Standards “dialects”: MCP, ADCP, and whether mega-platforms will standardize or fragment the ecosystem
• 44:51–47:34 – Human oversight, optimism vs. pessimism, and guardrails in an agentic future

---

Notable Quotes & Soundbites

• Tim (on AI agency laziness):
  “One of the challenges with these AI tools is it makes it super easy for people to be really lazy.” (13:14)

• Kamiko (on loss of brand control):
  “If you don't also have your human standing by with a leash and a collar now, you end up with weird AI generated ads.” (12:18)

• Ronan (on protocols):
  “It is an open source standard designed... to bring more accountability and structure into Programmatic Trading…” (24:09)

• Seb (on future risks):
  “If the data driving these… workflows is kind of weak, then the protocol is just going to move the mess kind of differently.” (32:21)

• Tim (on standards wars):
  “If Google and Amazon come in with their own pigeon, that wrecks things a little bit.” (39:14)

---

Conclusion & Takeaways

• The rapid rollout of AI agents in advertising is already exposing old wounds around transparency, control, and data cleanliness.
• There’s cautious hope new protocols like ADCP and ARTF can introduce order and guardrails, but only if major platforms and all key stakeholders, not just vendors, agree to participate.
• Much of the industry still feels unprepared for this transition—echoing the “is a hot dog a sandwich?” debate, foundational definitions and standards remain unsettled.
• The next 12–18 months will be critical in determining whether AI agents can deliver both speed and safety—or simply automate old problems in new ways.

---

For more insights and a ringside seat at the digital advertising transformation, check out the upcoming Digital Programmatic Marketing Summit and keep an eye on Digiday’s explainer content for ongoing updates.

---

Transcript

[A]

Foreign.

[B]

Welcome to the Digiday Podcast, a show about the business of media and marketing. I'm Kamiko McCoy, senior marketing reporter here at Digiday.

[A]

And I'm Tim Peterson, executive editor of Video and Audio Digital Media. What's up, Kimiko?

[B]

How's it going? Tim? Welcome back.

[A]

Welcome back. Yeah, another day we were off from the podcast last week. We were not off from work. And no kidding, because, geez, there's a lot of news that's happened in the past week, in the past two weeks since our last episode, but particularly in the past week.

[B]

Yeah, we've got AI agents, we've got AI frameworks, we've got arbitrage and probably a couple other A words in there somewhere.

[A]

Yep. And then our featured segment, we have Seb Joseph, our executive editor of news, and Ronan Shields, senior ad tech reporter at Digiday, on to talk about more AI stuff. Another a word ADD cp, ADD context Protocol, which is a new proposed standard for AI agents to be able to interact with existing advertising tools such as DSPs, SSPs, AD exchanges, other non acronym words. So yeah, it's a very AI agent filled ad centric episode, which I guess increasingly that's what our episodes are.

[B]

Yeah, I was just gonna say. But the good thing is for, you know, in this roundtable discussion that we have later on in this episode, for anybody's brain that just went into like a busy signal, the dial up tone, like mine did when we first had this conversation, there are a lot of metaphors to help you kind of wrap your head around what the hell is going on.

[A]

And a bit of a conversation about like, is a hot dog a sandwich? Though we don't quite answer that. For what it's worth, I believe a hot dog is sandwich. I'm willing to have that conversation with anyone who would like to engage with me. But Kawika, we got a lot else to discuss. For example, ad arbitrage, to what extent that's alive and well, and even taking on a new form. So Paramount Skydance had its earnings call last week, its quarterly earnings call, and during the earnings call, Paramount executives, specifically Jeff Shell drops this mention of like, yeah, hey, we just hired publicist and IPG to handle our marketing. So we're clients of them now. And you know what's great about that? They're also serving as ad sellers for us too. In which everyone's antenna kind of immediately shot up. Ad Age reported on this of just what questions does this raise? Is this agencies getting kickbacks?

[B]

Yeah.

[A]

So Seb will have a story that will be up on the DJ site on Tuesday about this. And Paramount is saying these are savings guarantees on Paramount's own media buys through Publicist and ipg. It's not revenue commitments tied to Publicist and IPG going to their other clients and saying, hey, you, you want to buy on cbs, you want to buy on Paramount? Plus we got inventory. We can make that happen. Let's make that happen. Let's make a deal here. But this has been like one of the questions around agencies that are representing media companies and big tech platforms of like if Google is a client, if Meta, if Amazon, if Disney, if NBCU are a client, is part of that deal for the agency to also sell that media company or platform's inventory to its other clients.

[B]

Yeah, the thing here is like one, I'm surprised that I feel like there's this Whitney Houston meme where at some award show, right, Somebody says something on stage and she gets up in her chair and starts looking around and be like, did everybody else just also hear what I just heard? And I imagine there was somebody on that earnings call who whipped their head around to say, did everybody else just hear what I just heard? Because what happened is like the arbitrage and the idea that they kind of said the quiet part out loud. No.

[A]

Yeah, exactly. Which again, they've been, you know, walking it. But Paramount has walked it back. And Seb reached out to Publicist for comment and they backed up what Paramount said. Not on their earnings call, after their earnings call of just like, it's just media savings guarantees on Paramount's own media buys. It's not us selling, committing, you know, revenue guarantees or commitments to Paramount to sell their inventory. He also reached out to IBG and they just referred him to what Paramount had to say. So hard to say now we've had.

[B]

These conversations behind closed doors. No, because like even at our town halls and things like that, that's kind of been like the hush hush. You've been managing those spaces a lot longer than I have. So kind of what's been the, like the back history of the rumblings about arbitrage and the concerns there that kind of got us to this point?

[A]

I mean, the concerns have been there for a long time, but it's also been, what was it, 2018, 2017 when the ANA had their transparency report that really put numbers on all of this and drove a lot of attention to it. But yeah, we've had town halls at our summits where agencies talk about this kind of arbitrage that's still Going on is understood to still be going on. And so it's not even necessarily like quiet part out loud. This has been out loud for a minute. Business Insider reported on former WPP media executive Richard Foster filing a lawsuit against wpp, alleging that he was fired for getting vocal about WPP Media, formerly Group M, having a whole agency kickback program. So this has been going on kind of seems to be continuing to go on based on, like the conversations we have at the town halls. The newer wrinkle is there's been speculation of like, if you have an agency holding company representing a big media company or tech platform, to what extent is that agency holding company pushing that media company or tech platform's inventory on its other clients? This Paramount example seemed to, I think why it stood out so much is because it seemed to initially confirm that. Now again, the companies have walked that back and said, that's not going on here. But it's definitely going to further fan the flames of like, okay, even if this isn't going on, maybe in this one specific example, it's still probably going on elsewhere. Just because ultimately this is business, this is capitalism, people are going to chase money.

[B]

I was just about to say. And then like we, we've talked about this before too and had like Michael Bergey on the podcast to talk about this with us. Agency holding companies have been playing catch up for a while here, looking for new revenue streams and ways to kind of keep business going. And I wonder how much of this kind of ties into needing to continue to ensure dollars are there and that they're getting paid.

[A]

Yeah, because there could be like pretty clear incentives of.

[B]

Correct.

[A]

Okay, again, Paramount and publicists have walked back this stuff. But let's use this as the hypothetical example. It would make sense for publicist for IPG to be like, cool Paramount, we want your money. So we're going to like get you as a client and we'll do whatever we need to do to like keep you as a client, including help you with ad sales. Especially if that means we can get discounts on Paramount inventory to then take just to our other clients, because then that's going to help us keep those other clients. Because if the next time Paramount has or CBS has the super bowl and if publicist clients can get a discount or get like super bowl spot at a discount compared to what clients of other agency holding companies are going to do, that's going to be helpful for publicist, that's going to be helpful for ipg. So the incentives are very clear and obvious here. Which is why you would expect this kind of thing is going on. Although no one would ever want to admit it because if you're a CPG brand, any other. Well, this could be going on with retail media networks too, given that everyone has some inventory.

[B]

Absolutely.

[A]

But you don't want to be thinking every advertiser cares about is my money going towards the media? Am I being as cost efficient as possible? This kind of arrangement would raise questions of like, how cost efficient is this? I don't want to just be patting other people's pockets. I want my money to be leading to product sales or, you know, whatever transactions I need. Because my cfo, my CEO is on my back to prove that our dollars are working.

[B]

Absolutely. Now if arbitrage is not the answer, maybe ad agents are. No, that's a good transition.

[A]

So yeah. So last week Amazon and Google both announced their ads agents. Amazon's is called the Amazon Ads Agent. Google actually announced two. One is the Google Ads Advisor and the other is the Google Analytics Advisor. Both of these are just like AI chatbots and both of them are basically from everything I was able to read up on. You know how Amazon and Google have self serve buying tools. It seems to just be like taking those and saying, well now instead of like clicking buttons and typing in text, you can just talk with the chatbot. And it's all it seems to really be is the AI chatbot version of a self serve buying tool. Self serve analytics tool also in Google's case.

[B]

But it also speaks to like the ambitions that these tech companies have of one, you know, becoming one stop shops leveraging AI tools. Right. I do wonder if there's going to be like a premium placed on these type of things because it's labeled AI but like the push for, for, for AI agents and like agentic AI to like, I don't know, have you talking to a talking head and say hey, like an assistant, essentially do this, do this, do this. And it just happens because you know the chatbot is set up that way as opposed to you clicking around seems to be because Meta's got tools like this too.

[A]

Meta's got tools like this which are drawing some questions and some feedback from folks. So as Business Insider again reported on this of like Meta's Advantage plus program, which is its AI ad creation platform product that brands including True Classic Karuna, which is a European shoe brand and E bike company Lectric all found out that Meta was just creating weird ad creative for their campaigns. I guess because there's some toggle in Advantage plus where if you don't disable this, it's basically saying meta, if you don't like our creative and you want to create your own AI generated ad creative that maybe can perform better than ours, just go ahead and do that. These brands, it seems, weren't aware of that toggle existing. I don't know how many advertisers were aware of that toggle existing. I imagine a ton of them have toggled this off since this story broke in Business Insider last week.

[B]

That's something, right? And we talk about this later on in the conversation with 7Ron is just like, how much, how much of the reins do you handle hand over to like, AI agents into AI? Because like, on the one hand, I won't take away that it's made a lot of improvements in terms of like, getting things done faster and cheaper. No comment on better? Not yet. But you also run to this idea of like, if you don't also have your human standing by with a leash and a collar now, you end up with weird AI generated ads. And what one brand marketer that I talked to at one point called the AI sheen happening here.

[A]

What do they mean by AI Shane.

[B]

Where it clearly looks like AI, you know, when you're watching an AI Little too smooth. Chatbot created something talk. Yeah, exactly. Exactly. Where the hair looks more like a cotton pad than it does this person has Deal.

[A]

What's that about? It's like Photoshop on steroids. I get that. Yeah, yeah, I like your like leash and collar analogy. It seems like one of the issues that like this meta example raises is and people need to make sure they have a firm grip on that. Because I think one of the, one of the challenges or issues with these AI tools is it makes it super easy for people to be really lazy. In this case, it just seemed like people may have been, oh, cool, I'm using Meta's advantage plus AI's taking care of things. I'm just going to type in my budget or whatever my KPI is and sit back and it's going to take care of everything and didn't go through and look at every single toggle that that was toggled on, every single selection, every single option that was enabled and that if they had just looked at those things, maybe they would have recognized, oh, I maybe want to disable this one. And so, yeah, as we talk about with Seb and Ronin, they're going to need to be guardrails. They're also going to need to be people enforcing those guardrails. The humans are going to need to stay in the loop.

[B]

Yeah, Prompt engineers and things like that. Every week I say this, I should have chosen a lucrative career here and this would have been it. Prompt engineer, that would have been it behind it.

[A]

Because I feel like prompt engineer is one of those roles where it's like come into question how long that's going to be like a single standalone.

[B]

Oh, flash in the pan.

[A]

Well, just because of the idea of like, everyone needs to be a prompt engineer in a way. When you have all of these companies demanding that employees start using these AI tools, figuring out how to be adopting them in that case, then everyone has to Prompt engineer just becomes a skill that everyone's required to have. In a similar way of back in the day when people would have been asked, how many words per minute can you type?

[B]

That's a real throwback there.

[A]

But speaking of guardrails, another kind of guardrail that got introduced last week is IAB. Tech Lab unveiled the AgentIC RTB framework ARTF, a new acronym for us. Kimiko, love that.

[B]

What is it? What is this lovely acronym that we're now being introduced to?

[A]

This is a specification for enabling AI agents to participate in the Programmatic Bid stream. The main thing that it looks to tackle is, and there was actually agency exec I was talking to about this last week. So right now, when it comes to the Programmatic Ad Auction, like we have this term real time bidding, because this stuff happens super fast, near real time. It's like Seb wrote a WTF on the Agentic RTB framework And he mentions that the typical Programmatic Ad option ad auction happens within 400 to 600 milliseconds. So super quick, half a second. Have you ever used an AI chatbot, a chatgpt, a Gemini, what have you, and had it respond within half a second?

[B]

Absolutely not.

[A]

No. And so with AI agents being brought into the programmatic mix, the question it raises is how long is that going to delay things? If every time you go on a website and it is loading ads, if it has to be calling GPT, Gemini, Claude, whatever other large language model powered AI agents, how much time is that going to add up before an ad gets loaded? Or you say, I don't got it this time, I'm leaving, I'm not sticking around for these ads. And so what the Agentic RTB framework aims to do is create, to institute a process that's called containerization to speed things up. Basically what it is is companies will take their AI agents or like the code powering These AI agents put it in a box and ship it off to the various servers that are handling the ad auctions or otherwise are like necessary parts of the RTB process. Because if the AI agents can be effectively on site in these servers, then they're not going to take as long to have to respond. That's one way in which IAB Tech Lab says, actually this could even speed up the RTB process by as much as 80% from that 400 milliseconds. Not only would having AI agents in the fold with agentic RTB framework not add more time to it, it would actually take time away.

[B]

The key theme here, and this goes in even to our conversation later on about like AI agents is like, continuity is such a big theme here to ensure that these processes, like the pipes have to continue flowing for things to continue to work as quickly as inefficiently as AI promises. Which is, like I said, something that we get into here later at the next dpms, you guys are going to have a working group breakfast about this, no?

[A]

Yeah, yeah. This is going to be really exciting. So at the Digital Programmatic Marketing Summit, which is right after Thanksgiving in New Orleans, a city that I've only been to once, but I loved my time there. I'm really looking forward to being in New Orleans again. So December 1st through 3rd, we will have an exclusive working group session with IAB Tech Lab in which an IAB Tech Lab executive will be presenting the AgentIQ RTB framework, going over it, and then Seb and I will be asking questions about it. But we're also going to open it up where any of the attendees in the room we're going to be setting aside, I think close to an hour for this, for an open Q and A, so that anyone on site is going to be able to ask questions about the agentic RTB framework, also provide their own feedback on the agentic RTB framework, which I expect there will be. And IAB Tech Lab is very much open and looking for that feedback because this framework is in a public comment period until mid January. And so one of the great things of being able to have them come to dpms is this will be a forum for that feedback. So I'm really looking forward to that. I know seb's really looking forward to it. I even think IB Tech Labs really looking forward to it. So it should be a good time. So if anyone listening hasn't registered to come to New Orleans for DPMS and is looking for a reason to be on BEIGNETS and Bourbon Street. We got the exclusive breakfast on the agentic RTB framework that'll be happening during the event.

[B]

Well, hopefully many, many juicy scoops come out of that and some answers to questions. Continuing on the AI beat and wrapping this up, we've got Seb Joseph and we've got Ronan Shields who are going to be leading us to this discussion of ad cp, ad context, protocol and kind of the programmatic ad market and how it is adapting to the era of AI agents. Like I mentioned, the big theme here to me is continuity to make sure that every the stars are aligned to ensure that the AI agents are working the way that you need to. And also kind of the caution tape that needs to be placed up around these places to make sure, like I said earlier, your leash and collar works.

[A]

Yeah, yeah. I think my big takeaway from this conversation was just the importance of there being guardrails in place as, as more and more of the advertising work gets handed over to these AI agents and adcp. The Agents Regard framework seem like two examples of systems for guardrails that the industry is starting to put in place. But we'll see to what extent the industry adopts it because adoption is a big question that SEB raises, particularly with adcp, given those who are currently supporting it and those who haven't come out in support of it just yet.

[B]

Absolutely. Well, with no further ado, let's get to it.

[A]

Sup, Ronan, welcome back to the show.

[C]

Thank you.

[D]

Good to be here.

[A]

Kimiko and I are very glad to have you here because we're trying to make sense of what the hell is going on when it comes to AI agents and programmatic advertising right now. So this very much. Oh, oh. Throughout the conversation. We'll just put that as a caveat on this whole conversation. Start of the year 2025 was kind of proclaimed as the year of the AI agent. It was the year when there was expectation that AI agents were just going to pervade all of technology, all of business. That's kind of been true. Agencies like WPP have talked about all the AI agents that they've rolled out. Ad tech companies are rolling out AI agents to. The problem with these AI agents is they need a way to talk not only to each other, but also to be able to use the existing tools and technologies that exist, like be able to talk to DSPs, to SSPs, to AD exchanges, to everything in the programmatic supply chain. And within the past month, ish, we've seen two now big developments that seem to create this communication layer for AI agents with respect to programmatic advertising. First was the creation of the Ad Context Protocol and then the second was IAB Tech Labs introduction of the Agentic RTB Framework. Both of these are terrible names for things. I wish it was something that was much clearer. Also, we don't have much detail at this point over about the Agentic RTB framework. So we're like the. Oh, we're going to use that as an asterisk to couch a lot of what we have to say about the Agentic RTP framework. The Add Context protocol though. Ronan, Seb, you both have done a bunch of reporting on this. There's also been a lot said and posted about ADCP online. So everyone's kind of had more time to kind of understand what, what this is. Ronan, you wrote the piece explaining for Digiday what ADCP is like kind of base level. What exactly is adcp?

[C]

Okay, so Ad Context Protocol or adcp. I'm just kind of repeating that and putting those two next to each other just because I don't know, maybe it's going to be an acronym that we're all going to have to familiarize yourselves with at least over the next year. So just trying to get that into the lexicon. It is an open source standard designed, well, I mean from the parties behind it, designed to bring more accountability and structure into Programmatic Trading, I guess, as it would have been. There are those. No asterisks. Who would say this is the death of Programmatic Trading? I don't know if I'd say that. I mean, stuff's happening, it's just maybe a different technology doing it.

[A]

Because ADCP doesn't actually get involved in the execution of programmatic buying and selling. It's more planning, strategy, gathering information to inform deals. But when it comes to the actual real time bidding component, that seems to be where ADCP goes. Hands off.

[C]

Yeah, yeah, well, I was going to say so. So maybe I'll just start with. Well, continue by saying some of the parties that were behind it and then maybe that would give listeners a bit more of an idea as to where it's coming from and what. Well, who, who's behind it and what's driving at. So some of the parties behind it. Sorry, I have them listed here. And please, if I forget any names, can we put a caveat at the end? Because I'm sure this would happen. So the parties behind it include Yahoo, PubMatic Scope, 3Optable, Swivel and Triton Digital. So obviously all those companies are Quite high profile Scope 3 CEO Brian O', Kelly, the Godfather of ad tech. On the eve of the announcement at Prebid Summit, he gave an interesting theatrical display of just how insane the programmatic ad ecosystem is of just, you know, one person. He got people from the audience up on stage and one started to be like the buyers. And then they were basically just running around with all these like random bits of information. And it looked quite ludicrous and it was quite a good visual representation of how ludicrous and chaotic the ecosystem is. His point is that AD CP will or can in theory bring a bit more order to it, to the process, particularly as we become, become more automated through these AI agents. So it is a framework, it is a proposal framework, open source, by the way, for these agents to do business together. That would be one way of putting it.

[B]

I feel like at the beginning of this conversation I had my wits about me. Now it's like, remember the.

[A]

I don't know if you guys like three minutes.

[B]

You guys remember the dial tone that used to happen in the 90s? That's what's happening in my head right now.

[D]

Sorry, pal.

[B]

Okay, so how I am interpreting this in, you know, God bless out there, who, who is listening in kind of the same capacity that I am, it. It seems like pipes here, right, that are kind of being Connected by your scope 3, by your Yahoo and some of the other players in this space. Is that the best way to kind of describe it here?

[D]

Better to think of it as a language, a standardized, common language for AI agents to communicate with each other across the ad supply chain. Like that would be like, you know, the real bare bones sort of explanation of it. Obviously there's a lot of nuance, but in terms of how you think about at CP as a, in terms of what it's designed to do it is to give a agency a standardized way of talking to each other. Essentially.

[B]

Kind of like how emails talk to one another or just, just, just the.

[D]

Standard, like just, just a dictionary. Right. That they can all reference in order to be able to figure out what each other, you know, act accordingly.

[A]

Yeah, like I think I. So as I've been trying to think about it and make sense of it in my own mind, I've been thinking about it. Like the idea everyone seems to have when it comes to AI agents is I'm going to be able to tell an AI agent in this context, hey, go create a programmatic guaranteed deal for me with this publisher. But if I'm as simple as that with my prompted AI agent AI agents can be like, bet I got this, I'll take care of it. But that's similar to me saying to an AI agent, make me a sandwich. If I'm that like, broad with it, the AI agent gets to then like, figure out what kind of bread it's going to use. It's going to figure out, do I want turkey, do I want ham, do I want bologna? And that assumes an AI agent has a solid understanding of what a sandwich is. It could know that a sandwich includes bread. It could come back to me with a hot dog. In which case then we're getting into the conversation of is a sandwich a hot dog? Yes or no? I believe it is. We can table that conversation for something else. But basically for an advertiser and agency to be able to have an AI agent handle parts of the advertising workflows, they're going to need to be rules and guardrails. And so to Seb's point, what ADCP does is creates the language for those rules so that if I say AI agent make me a sandwich, it understands, okay, I'm going to need to have some sort of bread, some sort of filling for that. And a hot dog is or is not a sandwich. For this person who's asking does that. That's my interpretation of adcp, Ronan. Seb, does that check out or is that like completely off base?

[D]

No, no, no. You noted.

[B]

I do think that makes sense. The dial tone in my head has gone away, however.

[A]

Hooray.

[B]

Now that brings more questions of like, if it's meant, to Ronan's point, to like simplify a very muddied programmatic process, like, how viable is that? Because to your point, if we're talking about like a hot dog versus a sandwich and like AI agents have to figure that out, how viable is this in streamlining processes?

[A]

I think it's going to get messy before it gets cleaned up. But I think it's also, it feels necessary because it feels like everyone's just assuming AI agents are going to need to be part of workflows. And there's definitely a lot of inefficiency here of things that can be handled by AI agents. But it feels like AI agents are going to need to be trained probably even more than you would have, like a junior media planner who needs to get trained. Like, you know, anytime we talk to kind of anyone in the advertising industry and they always reference the 25 year old media planner who's just like the root of all problems, AI agents are going to be like a five year Old media planner.

[C]

Yeah, I just, just personally wondering about all the, you know, brand safety uproar that is almost certainly bound to emanate from the, in the early stages. Not going to say it's going to be endemic or blah, blah, blah, but you know, this brand safety is always the one that or the brand safety mess ups, they always deliver and they'll always be around. So yeah, maybe that'll be more for us to report on year or two down the line.

[D]

I think it's an interesting point, like just kind of threaded the needle to both of those things. It's, you know, if the data driving these, you know, kind of, these workflows is kind of weak, then the protocol is just going to move the mess kind of differently.

[B]

Right?

[D]

Rubbish in, rubbish out. I think that came up at the kind of last summit we did when you've got publishers actively really thinking about the substance of their kind of data and trying to get it into real kind of ship shape in order to be able to power these agents.

[A]

You.

[D]

Know, and I think unless you know that that data quality concern is prioritized up top, then all of the issues, those long standing issues we've all been writing about for years, right, from opaque fees to fuzzy supply chains, fraud risk, they're just gonna, they won't vanish because of this protocol. They're just gonna sort of play out in a new form.

[A]

Which kind of raises the question of is it too early for something like an adcp? Because yes, seb, as you were mentioning, we were at the Digiday Publishing Summit in Europe recently and all of the conversation was around AI there. But so much of the conversation came back to publishers feeling like I can't really have AI do a lot of stuff for me because AI needs really clean data and we don't have enough clean data. And even seb, was it Immediate Media when you spoke with them on stage, who said like 2026, their big priority was cleaning up their data?

[D]

Yeah, that was it. It was their kind of, you know, revenue operations kind of lead. And he was, you know, he said kind of that, yeah, their biggest priority kind of next year at least as far as his team is concerned, is cleaning up their data, you know, in order to be able to kind of feed it into the agents that they're kind of building to your point about like the, whether it's too soon for a protocol race, I think we'll see how these things kind of shake out.

[A]

Right.

[D]

I do think though, you know, it is something probably to be commended that the Industry is trying to anticipate disruption before it hits a kind of almost like a, a preemptive open RTB moment. Right. Everyone knows agentic workflows are coming, but nobody wants a repeat of the 2010s, right? Where, you know, kind of infrastructure ossified under one company's ecosystem, essentially. You know, I do think there are some kind of caveats. The fact that it already seems like there could be overlap between some of these protocols, you know, the data point that we've kind of just spoken about, I think it's, it's almost just as interesting who isn't backing ADCP as it is who is. You know, when you look at the companies that are missing no index, no OpenX, no big DSP really.

[A]

So no Amazon.

[D]

No Amazon. And that's actually, that's a really good point. You know, we, we were speaking to them on Monday, I think, and we asked them and they were, you know, know, as you'd expect, pretty reticent to kind of talk about it. But there was no, there was no even. There was no real kind of hint that it would be something that we would genuinely consider either. So we'll see.

[A]

Which I think is interesting because like, to Kimiko's point around, like, is that cp, the pipes, it feels like it's in a way the like train tracks or the road system for an agentic advertising landscape. And I mean, thinking about the railroad barons of yore, there's a lot of power if you're in charge of that. And so good on scope three and PubMatic and Yahoo and the others who are involved in adcp and especially in making it open source. But you would think at Google, Google particularly, who has a chokehold on digital advertising, would be a little reticent to support anything that would cede some of that power, at least without courts getting involved, regulators getting involved.

[B]

But also if, if it has to be standardized for it to like work, right? If you don't have all these players on board simultaneously in an agreeance, then again, you get back to this point, it's like, is it too soon? Does it muddy the process? Does it solve the things that it's.

[D]

Here to solve or just kind of creates more fragmentation? Like the Wall Gardens are going to do what the Wall Gardens do and build their own systems and then everyone else is going to have to figure it out on their own altogether? I think we will see that, right? We haven't even started to kind of talk about how OpenAI is going to kind of move into this space, but I'm pretty sure agents are going to be a big part of whatever they're cooking up on their ad business side.

[A]

Yeah, and I mean, we're already kind of seeing some of that with OpenAI, with the news that they had recently regarding apps in ChatGPT. And those are going to be built on this model context protocol, which is kind of like the underlying layer to everything we're talking about. Sontext protocol is built on model context protocol, mcp, the number of acronyms that we have in this industry. But basically. So that's kind of the fundamental layer to all of this. So we have an explainer that Sarah Guaglioni wrote up on Digiday. We also have an explainer video to go along with that around mcp. The idea of MCP is it creates the communication layer for AI agents. ADCP is a flavor of that. It's basically tailoring this to advertising systems because it's. I guess we could think of ADCP as like a dialect. We all, you know, the group of us speak English, but I'm in California. We have certain slang that we use. Kameko, you're in the south, you have certain slang. Rana and Seb, you're from Ireland and England, you have entirely different slang in terms of words that you use where in your communities you know what you're talking about. Ultimately we're all speaking English, but we need to like have that slang. Ad CP is kind of the slang for AI agents that are going to be used for advertising purposes.

[C]

Okay, so like the Creole of ad tech.

[A]

But if Google and Amazon come in with their own pigeon, that wrecks things a little bit.

[C]

Yeah, well, one thing I would say actually in some of the wording that the entities behind ADCP put out there, they did make note that there's going to be a. Hold on one second. I have it here in my notes. It'll be governed by an independent nonprofit working group purposely to avoid dominance by any single vendor. Now, if we cast our mind back to the Google antitrust trial, some of the evidence that came out in the findings was prebid, which is another sort of non profit entity that's involved with some of the tech standards that are being used by parties out there in programmatic trading. They wanted to take certain standards to Ivy Tech Lab and Google effectively vetoed that when it was with the iab. So I think, you know, that speaks to that point that these people are trying to make themselves out to be this purest of the pure. And we're very, you know, no one Single party is going to dominate this. Like there are perceptions, There are perceptions that entities like IAB are dominated by Google at all. I'm not saying they are. I'm just saying, you know, most people listen to this will be aware people perception. There is, there is this perception as such. I'm not saying they are, but I just find.

[A]

Very diplomatic of you.

[C]

The wording is such a. Yeah, I come from a conflict resolution society, but yeah, I think it was. I think it's kind of interesting. It just seems to be awareness and to that point that you bring up, it's like. And SEB as well. The rubber will really start to hit the road whenever we find out what those big household name companies like a Google or an OpenAI. Yeah, it'll be interesting to see what they put to the market.

[A]

Yeah.

[B]

Which takes us back to the walled gardens conversation. Maybe we'll get another acronym out of it. Who knows?

[A]

Even then, even with adcp, because it's not actually handling the execution, it's handling things like if an advertiser wants to have an AI agent go out to different data providers to find an audience for it to add to a buy or to set up some sort of PMP deal with a publisher or to handle a negotiation, identity resolution, things like that. Ari Poparo had blog posts that he published recently where he weighed in on ADCP and kind of the issue that he took with it. He had a lot of positive things to say about it. But one of his criticisms was the AI agents are going to be vulnerable to getting bad data effectively because if they're going out to data providers or to publishers and saying, hey, do you have this audience? Or I'm looking to buy this many impressions at this price, the publisher, the data provider could just give the agent the answer that it's looking for. It's going to miss some of that negotiation that humans are really good at. Kind of the discerning part of this strategy, which I thought was a really interesting perspective on adcp.

[B]

Yeah, that is. And then. Yeah, because like one of the big critiques of like AI in general is its capability to like reason versus just kind of like regurgitate. And I could imagine something like that showing up here as well. Especially do you guys point earlier where publishers are still working on getting like clean data to be able to supply these things?

[D]

I guess it depends also on like how quickly the market wants to get to that point. You know, I think at the moment the agents that are in play, they're automating a lot of the workflow stuff. But there is a, there is still a kind of very acute human kind of component. I'd like to think that the industry will be cognizant of that. You know, as these agents kind of grow and you know, the kind of data provenance sort of aspect becomes just much more of a priority because. Yeah, Tim, you're right. Otherwise, you know, it just becomes a real slippery slope. Right. When you're just letting these agents. Yeah. Just kind of run away with all of the kind of decision making that, that goes into sort of trading ads, but without kind of the clear kind of guardrails that you need and also the expertise that, you know, humans have kind of built years and years and years sort of kind of accumulating. So I, I'd like to think that, you know, there will be some guardrails in place that mean that the way the amount that agencies are given, the amount that agents are given, what to do is sort of controlled or limited to some capacity and sort of almost kind of handled over sort of gradually in order to prevent some of that stuff. Because I read that post from re2 and thought similarly.

[A]

So I always appreciate the, the optimism.

[B]

I was just about to say these last couple of conversations, you've really been bringing, bringing the silver lining, man.

[D]

Like, I'm with you free. So like, you know, someone's got to kind of balance it out. Going all Luke Skywalker rated.

[A]

Because I agree with your point. I worry though of like, there are going to be the people out there who are just like, cool, I can have my AI agent handle my work for me, but I'm going to do that. I'm going to go for a surf and I'll be back in three hours and then it's just AI agents run amok at that point.

[D]

I just have to have some faith that the last decade plus years haven't been for nothing. The number of missteps and faux pas that this industry has repeatedly done has to amount for something, right? Surely there has to be some learnings that can be applied for what is arguably the biggest transition the industry is sort of going to make. And I just like to think that there's a lot more discernment that goes into this next phase given how existential it could be, not just for companies, but the yard industry writ large.

[B]

I am almost inclined to believe because I don't know where in my reporting there. It seems like guardrails, safety and things like this have taken a backseat when it comes to AI. The focus is on efficiencies curbing costs faster, not necessarily better, but cheaper. And so my fear is not to be in opposition, but the fear is that it will be when either dollars show up missing, spent unwell. Right. Or if there is a hallucination or something like that, or God forbid, a lawsuit, that's when the thought starts being like, how do we put some guardrails around this?

[D]

But who lose this out? Then can we go, you know, like if the industry does that, then more money just goes into the wall gardens where, you know, at least on paper it looks like it's working. So like that that benefits no one but, you know, the wall garden. So if the, to be fair, you know, what if the industry is that stupid to, to, to repeat those, you know, mistakes, you know, then it, then it deserves it in some respects. But I'd like to think that because of everything that has happened, there has to be a little bit more thought into how it sticks the landing this time around.

[B]

Yeah.

[D]

Because otherwise that money is, that money is only going to go one place.

[B]

Yeah, yeah. What did George Bush say? Fool me once, fool me twice, can't get fooled again.

[D]

Yeah. What's the effect?

[A]

For sure, It is a real concern though, because I think all of this is moving so quickly, but it also kind of has to be moving quickly. And obviously there's a lot of incentives for things to be moving quickly, quickly. When you have a publicist in its most recent quarterly earnings call saying, hey, our revenue forecast is looking real good because we got advertisers wanting AI things from us and so we're going to give them AI things. And so there's pretty clear financial incentives at this point for all of these companies to be implementing AI agents to reduce headcount, shitty as that is, but then also because there's a lot of demand, it's the shiny new thing. And the risk, there is a lack of guardrails, a lack of security in coding, there's this principle of least privilege where basically you don't give an app any sort of technology any more power than it absolutely needs to do the task that you need to do. And that's kind of where it feels like things like ADCP and potentially this agentic RTB framework come in is putting very clear guardrails on what these AI agents are capable of. But all this is going to require as many participants in the industry to come together to actually inform these things. And so hopefully the cautionary tale of the past 15 years, and especially that initial 5 to 10 year phase after OpenRTB was introduced, tells people like, oh, actually everyone needs to get involved with this public comment period that IB Tech Lab is going to be having for the agentic RTB framework that goes until January 15th of 2026. People need to really comment and be very public with those comments about that because it feels like once these initial versions get baked in, yes, they can be updated, but updates take a lot of time. I mean, with OpenRTB, it takes multiple years before people actually update to the latest versions of that. And so it's that first impression thing of like, the foundation is being set right now and anyone who isn't involved in laying that foundation is kind of gonna get run over or at least it's gonna have to like, deal with whatever gets laid. Seb, you have a comp Sci background. This principle of least privilege idea, does it apply here?

[D]

Yeah, yeah, I think you sort of nailed it. Like, I don't. It's hard to sort of add any more to it. I think it really does. I do think it's. There's just so much to try and be out in front of. So I think that, you know, that will continue to sort of be really tricky, particularly as things continue to evolve at a clip. It's interesting that the stakeholders that are amassing around all of these standards, I don't know about you, but advertisers and publishers and broadcasters, they don't seem to be involved in any yet. Like, you know, so you'd like to think that at some point they will have to be otherwise, again, like we, we go back to, you know, what was happening a decade ago, right, with some of those sort of issues. So I would like to see maybe even more urgency, but also kind of more thought that goes into it. And these, these, these, these protocols, these guard was just being a lot more inclusive about the resident because it does feel a little bit like, you know, they're, they're very specific to a contingent of the market that cynically stands to benefit the most from all of this stuff.

[A]

Yeah, yeah. I keep coming back to Kameka, what you said around the pipes analogy and then the idea of roads and railways and all of that. And it feels like AI agents are kind of like the flying cars of advertising and. Jesus Christ. You needed. Well, can't really speak for the FAA at this point, but you need some sort of regulator, like an FAA for that, but also an AI agent. If it has a flying car, it can go up, down, sideways, backwards, all over the place. You would want tracks for that. I don't know that the industry is at all ready for full on flying cars when it comes to AI agents. Maybe a monorail system.

[B]

I think that is a fantastic idea. Part of this conversation has been exploratory and the other part is like a warning.

[A]

I mean, it'll be interesting. So, Seb, you and I will be at the JJ Programmatic Marketing Summit in New Orleans early December. It feels like fantastic timing to get a bunch of programmatic experts from brands and agencies in the room to talk about what's going on with AI agents, which will be the focus of that event.

[B]

Yeah. Something good to chat about over beignets.

[A]

Absolutely. Always. Always. I'll talk to anyone about anything over beignets. All right. Well, with that, Zab, Ronan, always appreciate you guys coming on to help Kimiko and I wrap our heads around what the hell's going on in it Programmatic these days. Yeah.

[D]

Thanks for having us. Really enjoyed it.

[B]

No problem. Well, that brings us to the end of this episode of the Digiday Podcast. Thank you to everyone for listening. And please don't forget to share this episode with someone who you think would enjoy it. You can even rate us and leave us a comment on Apple Podcasts. We'll be back next week with another episode of the Digiday Podcast. Thank you so much for joining us.