Seb Joseph (2:59)

Yeah, yeah, definitely.

Tim Peterson (3:01)

That said, like interns, as someone who was once an intern and had no idea what the hell I was doing. AI agents similarly don't know what they're doing, but they're also overly confident in a way that I feel like most interns probably aren't. So there's a lot of guardrails and you need kind of like guardrails and a leash when it comes to these AI agents. But so that was a lot of conversation that we had during the during DPMs. And we'll get into that again later in this episode. In this interview with Christopher Francia from Attention Arc about the case against AI agents and programmatic ad buying. But then we got back from DPMS and all hell. So I've started thinking there's this weird pattern that's emerged where this two week period, the week before Thanksgiving in the US and the week after Thanksgiving in the US and is kind of like the death rattle of the year when it comes to marketing and media news. Like you think, okay, we're heading into the holidays, things are going to quiet down. I'm going to be able to get the rest of my work done for the year and then have a nice break and recover for the next year. And in the past, we've had this be roughly the period of time in which Disney says, hey, Bob Iger's coming back as CEO, in which OpenAI's board says, Sam Altman, get the hell out of here. And then Sam Altman decides, no, actually I'm not going to get the hell out of here. You're going to get out of here.

Tim Peterson (4:38)

Roughly. I think it's a bit looser timeline. But when Omnicom says, hey, we're going to buy Interpublic Group and become the biggest agency holding company. And now this year in which Netflix says, we would like to acquire Warner Brothers, not all of Warner Brothers Discovery, but kind of the most valuable part. So Warner Brothers has kind of been up for sale since the summer when Paramount made, I guess, an unsolicited approach after Skydance acquired Paramount. Paramount new owner David Ellison, son of Larry Ellison, Oracle's founder, said, you know what? I like this shopping spree. I'm going to keep it going. Warner Brothers Discovery. You're planning to split up the company next year. One side, one company that's the studios and streaming business, one company that's going to be the cable TV network business. What if instead you just come to me and become part of Paramount? Warner's Brothers Discovery has said, I think it's three different times now. No, thank you. But this is an interesting Idea of us just selling. So, okay, we'll take bids. Three companies have bid on Warner Bros. Discovery, Paramount, Netflix and Comcast. And the second round of bids were due last Thursday. And within hours it turns out that like, oh, Warner revisit. Discovery doesn't need to do much due diligence on these bids. Netflix offered enough money. We're just going to sell to Netflix to the point where Netflix then announces the next morning, hey, we're acquiring Warner Brothers Discovery, everybody. We're going to pay $83 billion including debt, but we're just interested in picking up the film and TV studio. So Warner Brothers Studios and the streaming business, meaning hbo. Max, this happened so quickly. It's going to take a long time before this deal closes, but this happened so quickly. Seb, were you expecting all of this to transpire as quickly as it did?

Seb Joseph (6:56)

No, is the kind of short answer right to your point. These things sort of normally play out.

Seb Joseph (7:06)

In a protracted fashion, but then I think when you kind of dig into it, this seems like the most fitting of kind of ways that it would play out given.

Seb Joseph (7:17)

It was pretty clear that Netflix were going to make some sort of moves despite, you know, the.

Seb Joseph (7:25)

The, the, the claims otherwise made by kind of both, the, both CEOs.

Seb Joseph (7:31)

You know, once Paramount kind of made it clear that they were, they were going to sort of try and do a deal quickly and had the back the sort of Trump administration, it kind of felt like there was no other kind of option but for Netflix to sort of make a move. I know it's kind of a lot has been made as to kind of why Netflix would kind of do this given their sort of build versus kind of via approach to sort of. Yeah, you know, kind of the, the market writ large, but.

Seb Joseph (8:04)

Makes sense, right? Like, because otherwise what do you, do you do you see ground to the kind of the emerging sort of rival that already has sort of stakes in, you know, kind of all manner of media businesses, including sort of TikTok. Like what. Why would Netflix not make them particularly given that they have been, you know, at least sort of looking at other sort of parties to acquire right from Electronic Arts right the way through to sort of Disney. So it's a long winded way of me saying like, no, I'm not surprised once I started to sort of look into kind of this, this more and start to understand the rationale behind what Netflix is really sort of trying to do here.

Tim Peterson (8:44)

Yeah, because it feels like, and I'm far from the first person to make this point, I'm maybe the like 3563rd person to make this point since Friday, but that.

Tim Peterson (9:00)

It seems like a lot of the motivation for Netflix wouldn't even necessarily have to be a Disney, an NBCU, a Paramount, but a YouTube, where YouTube is the dominant streaming service on TV screens. And a lot of that dominance comes from it having a big library. Now, granted, that library is not the same as Warner Brothers library. Warner Brothers has Harry Potter, has dc. It has all these premium titles, but to today's kids, Mr. Beast is premium. And so that said, this is also an audience that is interested in franchises like Harry Potter, especially HBO has the Harry Potter TV series that's going to be coming out. It has DC Studios, which is in a resurgent period with James Gunn now overseeing and the success of the Superman movie from this summer. All of which then makes me think of Netflix acquiring Warner Bros. Discovery is kind of the Omnicom IPG of TV streaming. This seems purely so in the way that Omnicom IPG seems so much about.

Tim Peterson (10:23)

Accumulating more data, because advertising is becoming more and more of a data business where the winners are, to a large part contingent on who has the most data and could do the most with that data. This is about intellectual property, about the film and TV library, which is data in its own kind of way. And Netflix even sent an email out to subscribers. I think I got this on Friday or Saturday.

Tim Peterson (10:54)

Promoting the deal, like, hey, we're buying Warner versus Discovery, everyone. Which is a strong move. I don't remember the last time a company that sent this email to its customers so quickly after announcing that it had agreed to acquire a company. I feel like usually companies wait until the deal closes. This feels like Netflix kind of preempting that by like, if everyone accepts this as a reality, then it's going to be inevitable that it'll close. But. But in that email that Netflix sent to subscribers saying, hey, we're buying Warner Brothers Discovery or we're buying Warner Brothers.

Tim Peterson (11:29)

It particularly highlighted the intellectual property that would be added to the Netflix library.

Seb Joseph (11:37)

Listen, like, Netflix has probably kind of wasted a small sort of fortune trying to kind of build its own sort of kind of IP or at least kind of find the next. The next kind of big, big one, you know, and it's kind of spent large sums acquiring some existing kind of IP that hasn't necessarily sort of panned out like, or not necessarily panned out, hasn't necessarily kind of yielded kind of big payoffs yet. Like the Roald Dahl sort of library. It kind of bought. I think you sort of. You've nailed it. And I think that that seems to be the kind of broader sort of consensus here, that this is about kind of almost one foot in the past with the sort of extensive, rich kind of TV library that they have kind of bought, but then with one eye kind of on the future, with the kind of abundance of IP potential.

Seb Joseph (12:30)

That they kind of have. I think to that I'd be quite surprised if Netflix turns its back on sort of kind of theaters, like, because that is a very lucrative business still. Right. When you sort of land on the right ip. Why. Why would you blow that up? Like, you know, maybe it will kind of reduce the slate. But, you know, I think despite some of the cynicism that's kind of met this kind of deal from, you know, the industry, from Hollywood, I. I can't imagine them sort of put in the next Batman film.

Tim Peterson (13:06)

Yeah.

Seb Joseph (13:07)

Straight to, you know, like, these are $1 billion, so 100 billion, $1 billion franchises. Right. I can't. I just can't see that you don't spend that amount of money if you're just going to need to see the whole. The whole thing. To your point about Omnicom, I think the parallels are sort of interesting. Although.

Seb Joseph (13:26)

I would say the Netflix one is for me, at least, maybe more interesting, even though they kind of both. Not the most.

Seb Joseph (13:37)

Interesting of Thesis is right investments or Thesis is. I think the Netflix one is probably the one. I'm a bit more intrigued about the Omnicom one. Look, it's big enough to shake agencies, but I don't think it's big enough to change who actually runs advertising. Right. The platform still do and they will always do.

Tim Peterson (13:57)

Yeah. Yeah. And I mean to your point around.

Tim Peterson (14:02)

The theatrical model with Netflix.

Seb Joseph (14:05)

Yeah.

Tim Peterson (14:06)

I don't know. I kind of expect Netflix to. Not immediately, certainly with not like if this deal goes through. And obviously that's a huge if because, you know, there's the question of David Ellison and his father, Larry Ellison. So the Paramount contingent, who are the spurned suitors in this case? They're tight with U.S. president Donald Trump. Could they try to get Trump to have his Department of Justice sue to block the deal? Like how the Department of Justice sued to try to block AT&Z's acquisition of Time Warner, which eventually became WarnerMedia and then became Warner Bros. Discovery, because this company can never have a consistent name, despite how long it's been around. So there's the question of is the US Government going to try to scuttle this deal you alluded to. It seems like a lot of folks in Hollywood do not like the idea of this deal. Jane Fonda for one, she doesn't like it.

Tim Peterson (15:15)

The Writers Guild of America also doesn't like it. Both have voiced opposition, A lot of people have voiced opposition. And then Paramount still not willing to let this go through. So this morning we're recording this. On Monday, December 8, CNBC reported that Paramount is offering Warner Brothers Discovery shareholders $108 billion in cash and a hostile takeover bid. Now that's a lot of money, but I actually wonder if that's not as good a deal because Paramount's trying to buy all of Warner Brothers Discovery, including the cable TV networks that are going to be spun off, which include cnn, because Netflix is not buying CNN as part of this deal. And so maybe that business is only worth, what is that, $25 billion. Yeah, but I wonder if like shareholders might think actually like if we get Netflix is 83 billion and whatever we stand to make from spinning off the cable TV business, especially if that were to get acquired by Versant or you know, rolled up some other way, maybe we stand to make more money. All to say, this is far from over. As.

Tim Peterson (16:27)

Eager as Netflix was to let subscribers know this is happening, it's like a year and a half, two years from actually happening. But when it does happen, if it does happen, it's really interesting to think what are going to be the impacts of this. I could see Netflix at some point saying.

Tim Peterson (16:49)

Batman, we're gonna put it on Netflix same day and date as it goes in theaters. I they're doing so much to try to drive the appointment viewing that I know but like yeah.

Tim Peterson (17:03)

They will lose money on that. But Netflix is a company that like.

Tim Peterson (17:10)

Has, it.

Tim Peterson (17:14)

Has always been willing to spend more money than it necessarily is taking in if it sees the opportunity to make that money in the long run. Like it over invested so much in programming as it was building up its originals business. But then it got to a point, I think it was like five, six years ago, where its operating income was in the positive to the point where it was almost in the positive and there was no looking back. It had crossed the Rubicon like it was, I don't even know if it was like a necessarily a bet that paid off as so much as like a smart investment. So Netflix plays the long game also. It's acquiring IP as we're talking about and this is the kind of IP that you do a ton of co marketing deals around it similar to like what we just saw with NBCUniversal do with Wicked and so and Netflix. So there was an agency exec I was talking to earlier this year who said Netflix has gotten much more involved in, like, brand integration, sponsorship deals in shows. Previously it had, like, left that to the production companies. And it's also really strongly packaging those kinds of things into broader media buys or attaching broader media buys to these kinds of sponsorship integrations. If you're able to. Like, I don't know how you would integrate a brand like Coca Cola or Nike into the Harry Potter franchise, but I'm sure Netflix will find a way to make that work financially. And similarly, like, if there's a Batman movie, if there is a way to have, like, brands subsidize much of that.

Seb Joseph (19:00)

How you're closer to this than I'll ever be. But how, like, does that kind of work for them, though, when it comes to being a place to attract the best talent, the best directors? Doesn't that just kind of strengthen the position of their rivals who are more committed, at least for now anyway, to kind of cinema?

Tim Peterson (19:24)

It depends. Like a Christopher Nolan. Absolutely not. But he's already left Warner and gone to Universal. Sure.

Seb Joseph (19:33)

But, like, even the up and coming ones, right? Like, I just, I wonder if, because everything is going so far in the sort of streaming appointment to sort of you kind of direction, if that doesn't then sort of at least kind of spark some sort of reversion to, you know, those kind of big screen moments. Right. Like, I get it. Like, the days of, like, you know, all of those blockbusters kind of making money are pretty much done and dusted, at least for the foreseeable future. But, you know, I still want to see the Batman on the kind of big screen or, you know, Me and the Kids or Zootropoulos on.

Tim Peterson (20:17)

On.

Seb Joseph (20:18)

On the big screen the other day. Like, there are just certain things, right? Like, if you like film, like, you want to experience it a certain type of way. Way, not everything, right.

Tim Peterson (20:28)

But.

Seb Joseph (20:32)

Yeah, definitely.

Tim Peterson (20:32)

And I could see Netflix doing day and date where it's like, okay, Batman movie, New Batman movies coming out will debut that in theaters, but it's also going to be available on Netflix that day. Because one of the things this does is it would give Netflix ownership of the IP that's most important to theater owners, which would put Netflix in. In an even stronger bargaining position with theaters. But you'd raise a really good point of, like, how are the filmmakers gonna feel about that? And we've already seen a case study where Greta Gerwig is doing the Chronicles of Narnia movie for Netflix. She reportedly wanted that movie in IMAX theaters and her and her team made a strong case for that to the point where Netflix acquiesced. And so that movie's gonna be in theaters even though it will also be on Netflix or eventually on Netflix. But I could see that becoming more rarefied air that that's not going to be the case for necessarily everyone and that I'm sure Netflix definitely just came up with the monthly active viewer stat. Right. The metric, like this completely made up metric that's based entirely on proprietary data. As far as I know. I could see Netflix coming up with some stat that like makes the case to filmmakers on like why they're better off letting a movie just be released directly on Netflix and just have like reserve the theater experience for tentpole stuff.

Seb Joseph (22:13)

Yeah. And to your point, the market is so concentrated from a buyer perspective is not like talent have a lot of places to, to sort of go now. Anyway. So.

Tim Peterson (22:24)

Yeah, yeah. And one thing I'm interested too with this deal is whether this is Netflix playing the sports rights long game. So Netflix, like with everything else, like with making a big acquisition, like with getting into advertising with sports, had said no, we're not doing sports until it started doing live sports. But it's also been very choosy with which sports it's doing. It's not. It didn't try to like buy out Thursday Night Football rights from the NFL. It just created these new Christmas Day games. And so it doesn't want to be tied up in like the current economics or the current like sports deal model. But I wonder if Netflix now having Warner Brothers Discovery would and the scale that'll bring if it folds HBO Max into Netflix and if having Batman, Harry Potter, etc. Adds to Netflix's subscriber base, if that then gives Netflix the sale to be a more attractive place for the next round of NFL rights deals, which may be coming up sooner than we might expect. So currently the deals have an opt out option for the league to opt out as early as 2029. Now Amazon, NBCU would need to agree to that. The Disney deals are until 2030, so I don't know why any of them, Fox, Paramount would say like, yeah, absolutely, let's do this. You can go shop again for some more because this is the most important programming. But if the NFL were to make that somehow worth its while, if Netflix would then be in a position to be a really attractive bidder for that and would want to bid for big packages.

Seb Joseph (24:24)

Yeah.

Tim Peterson (24:25)

I have no idea. No one has any idea.

Christopher Francia (24:27)

No.

Seb Joseph (24:28)

But I think you're right. So much of it is incumbent on like the rights holders, right, to, to make it more appealing to the platforms because to your point, it's been something they've dabbled with for a while, but the, the idea of sort of having to kind of manage rights across so many different markets just is not appealing to those sorts of platforms. So, you know, I think you've seen, you know, the NFL in the States, the, the, you know, you wait for the Champions League over here, start to sort of really be more cognizant of that in recent.

Seb Joseph (25:04)

You know, kind of negotiation sort of rounds and start to think about, well, how can we start to carve out, partition out, create new sort of packages that bring those platforms, the Netflix of the world, the Amazons of the world, to the, to the table.

Seb Joseph (25:18)

Yeah, I'm surprised FIFA didn't try and do that around the World cup like next year. I thought that given everything that sort of happened, I thought that might have been something on their radar.

Seb Joseph (25:31)

At least in sort of par. Right. Maybe like the opening game or the kind of final, but doesn't seem to be.

Tim Peterson (25:38)

Yeah, yeah, we'll see the next time those rights are up. So obviously Netflix, still plenty to see, a lot to unpack as we've been doing. Before we get to our featured segment, I just wanted to touch on some other big news of the week. Now, the Netflix Warner stuff kind of overshadows everything, but interesting development on Friday. So just Davies, our senior media editor, reported on this. Meta is going to start paying publishers again. In this case, it's going to pay publishers to license their content to train its AI model. So CNN people, USA Today are among the publishers that have signed these deals with Meta. It's unclear what the compensation model. It's even more unclear like how long the compensation model is going to exist because Meta has a history of saying, hey, publishers, we're open for business. You can make money here. And then within a matter of years saying, okay, we changed our mind. Just like Meta seems to be changing its mind on the Metaverse, That's a whole other conversation. But, Seb, did it surprise you that Meta has started signing AI content licensing deals with publishers?

Seb Joseph (26:50)

Funny enough, no, because you know, our amazing sort of reporter, media reporter Sarah Gargoni, she'd kind of picked up on this a while back, right, because they were at some meetings with the IAB sort of tech lab gathering where there were lots of sort of publishers and there were reps from sort of Facebook trying to get a handle on publishers want from these types of kind of deals. So it's, you know, kind of we knew that there was something in the works, but to your earlier point, weren't kind of really sure as to how serious kind of Meta was because they kind of, they talk a lot when it comes to sort of publishers and don't always sort of deliver on it. But, you know, it kind of makes sense. Right. Given the implications for kind of Llama and just the kind of broader competition that we're seeing across all of the platforms. Right. From kind of Google to sort of Microsoft as well.

Tim Peterson (27:48)

Yeah, because Microsoft's another interesting one. Adweek reported last week, Microsoft's hired Bloomberg COO Julia Bazer to oversee its AI News product. Microsoft's been signing content licensing deals with publishers, including Conde Nast, who is, I think, the newest publisher to do a deal with Microsoft. On the other end of the spectrum, New York Times has now sued Perplexity for copyright violation. New York Times has an ongoing lawsuit against OpenAI. Chicago Tribune has also filed a lawsuit against Perplexity. So it seems like at the same time as there's more competition and more opportunity for publishers to be making money from these AI platforms that also seems to be empowering publishers in a way to go after the AI platforms that maybe undercutting their revenues by using their content without deals in place and violating copyright. And then like one of the big questions with this is OpenAI's done deal with publishers, but how does it scale those deals? One option would be to figure out some sort of advertising revenue share program, but that would require OpenAI ChatGPT to have an advertising product in the first place. There's probably going to be one at some point, but it seems like it's going to take a longer time until that happens. Because before all hell broke loose with Netflix and Warner Brothers.

Tim Peterson (29:26)

OpenAI CEO Sam Altman declared in an internal memo reported on by the Information Code Red. And that as part of this Code Red, which is in response to Google introducing Gemini 3, which is like a really powerful model and kind of raised the question of like, is Gemini 3 better than ChatGPT at this point? In response to that, OpenAI is putting everything into ChatGPT and deprioritizing work on other things such as an advertising product. And it's this Pulse product that's like kind of like a daily assistant recap kind of thing. So God, there's so much going on on the AI publisher side. Of things.

Seb Joseph (30:08)

On that point. Tim, how important do you think next year is going to be for OpenAI? Because I kind of wonder, the longer they don't have ads, does that work against them in the sense of like, the longer they leave it, the more pushback there will be from people who have grown accustomed to not being there. But then also the longer they leave it, the more time Google has to, you know, kind of launch it. So like scale its own sort of alternative and lock in the kind of market which we know, you know, it's very. Well, it's very adept at like that kind of first move advantage, you know, open Air and subsequently Chat GPT kind of had. It feels like it doesn't feel like an advantage anymore.

Tim Peterson (30:49)

Yes, and yes.

Tim Peterson (30:52)

Because like, so we saw this play out last week where so OpenAI did this deal with like Target, among other companies, where like Target has a ChatGPT app that people can then use to shop from Target. But within the ChatGPT interface, it seems like some people got exposed to this or like OpenAI maybe promoted it in the interface to people and people thought that that was an ad. It's all like very fuzzy, but that's the best sense I can make of it since there are reports on tweets and other kinds of posts of people being like, what the hell? Why is ChatGPT trying to get me a shop at Target? This is an ad. This sucks.

Tim Peterson (31:37)

My hunch is that this is just chatgpt app that Target made and OpenAI just maybe overzealously promoting its use because it's a big shopping season. But people didn't like that. And that wasn't even like an actual ad, as far as I can tell. At the same time, Google has said it's going to bring ads to AI mode. Search Engine Land, I think, has even reported on potentially there being ads already being tested in AI mode that look kind of just like standard ads. So Google definitely has the leg up. Google already has search dollars and display dollars, at least until we see what happens with this ad tech antitrust case and what the remedies will need to be. But.

Tim Peterson (32:28)

Google has a head start. It seems to be keeping a head start. And yeah, I think that's going to be a tough look for OpenAI to try to get in on that money because inertia is real in advertising and also like anthropic's going kind of the other direction where instead of as much as it has consumer facing products, it's really going after enterprise and businesses. And so that Kind of leaves OpenAI and chatgpt in the middle in this weird position where it's the has the strongest brand in AI when it comes to like AI chatbots people can use, but it doesn't have the beachhead that it once had.

Seb Joseph (33:15)

Okay, yeah, yeah. I wonder like.

Seb Joseph (33:21)

Given how much money that sort of company is burning through, given it will now face a Google that is fully focused on winning right now, that the sort of trajectory of both of these antitrust cases seems to kind of be pretty much kind of sorted. That's a scary prospect.

Seb Joseph (33:41)

And I kind of just wonder like how we'll be talking about ChatGPT OpenAI this time next year. So here's a long time in sort of any industry, but particularly kind of media and advertising.

Tim Peterson (33:54)

Yeah, no, I mean, how are we going to be talking about it? Because you and I, as well as Kimiko and then our managing editor Sarah Jordy will also, I'm sure be talking about that as we do our year in review and then our 2026 preview episode. So obviously there's a lot that's gone on in the past week. This has become a super sized news roundup, but we had some super sized news to cover. Now we're going to go to my conversation with Christopher Francia, who's the director of project development and client performance at attentionarc. This was a live recording that we recorded during last week's DJ Programmatic Marketing Summit in which Chris makes the case against AI agents and programmatic ad buying. Seb, thanks so much for joining me.

Seb Joseph (34:38)

Thanks for having me.

Tim Peterson (34:40)

All right, now here's the conversation with Chris.

Interviewer (34:54)

So the case against AI agents. Are you all at attention arc just like anti AI agents or do you see there being a role for them and maybe just a tight leash?

Christopher Francia (35:05)

I think there's a role.

Christopher Francia (35:08)

For AI in general in a lot of different avenues. But I think when it comes to programmatic activation, we're not trusting that to large language model based agents.

Interviewer (35:22)

Should we define what AI agents are?

Christopher Francia (35:24)

We probably should because a lot of people throw out different terminology. So to the average person, AI today is ChatGPT. 10 years ago it was DeepMind from Google. 20 years ago was Deep Blue from IBM. It has evolved and every new advancement creates a new meaning. So this is probably the largest where most people in the world now understand AI to mean ChatGPT. Most of them even know AI was anything before that. So I think when that's the definition of AI today when you say agents, people are going to associate it with ChatGPT with Gemini, the large language models that exist. So when we talk about agents, and I talk about agents, I'm talking about the common person is thinking of other people in this industry and in the space will say agents or any computer program that's using a protocol to talk to another computer program through a protocol. I mean, by that definition, all the technology in the world is an agent. It's kind of a little ridiculous in my opinion, because that's not definition that's ever really been true. So that's what we mean. So we don't use, we don't tell ChatGPT, go and set up this campaign and activate this campaign and select the best targeting options you think would deliver for this. For a lot of logical reasons, most, mostly it's going to hallucinate. Mostly there's not really a good way for it to do that without there's an error that gets triggered because some condition wasn't known ahead of time. It's just too many variables for an AI to do. AIs are really good at very specific, narrowly tailored tasks. The more complicated a task becomes, the harder it is for it to figure out what it's supposed to be doing.

Interviewer (37:09)

And so it feels like two.

Tim Peterson (37:13)

Representative examples.

Interviewer (37:14)

But also like big examples of AI agents existing in digital advertising, not necessarily programmatic would be Meta's Advantage plus and Google's performance Max. Would you agree that those are AI agents? Because effectively you're saying here's my budget, here's my KPI go.

Christopher Francia (37:31)

So what they're doing there is. I know with Google they're not doing large language models under the hood for that interpretation. What they're doing is taking some key data points and plugging it into a machine learning model that's going to be deterministic and figuring out what to buy. I assume Facebook is doing something similar because they can't afford a hallucination, right? Because they can't afford it to screw up. So they're, they're making it feel to us like it's a chat agent and things like that. But they're probably faking that on the front end and in the back end they're using a.

Christopher Francia (38:06)

Network and you could do that with many different tools. We utilize that to some degree because they're two very separate systems and the way they think, one is very easy to audit and control and the other one is not. So I imagine anything that's touching spending is mimicking at best if it appears to look like that with the typing. With the exception of creative work, Creative work, you definitely do see agent based stuff where it's like build this and do this and it will seek out different software to build you a beautiful creative. And I think that's where it's been the most disruptive is in creative workflows and agencies have vastly sped up and the resources you need to do iterations of creatives have now reduced, which has made it for us to do more, take on more clients. It's been a really great boon to the industry.

Interviewer (38:53)

And is that instructive?

Tim Peterson (38:55)

Because it feels like right now the.

Interviewer (38:56)

Biggest case against AI agents in programmatic advertising is no one can agree on what an AI agent is, but the creative side of things. So we have AI generated creative now, which seems like the next leap from dynamic creative optimization of you have different assets, whether it be fonts, actual text colors, actual graphic assets, and dynamic creative optimization is we're just going to package these up in different ways based on what the audiences or whatever other instructions are given. But now we have AI generated creative. What's the parallel there that you would apply to programmatic buying?

Christopher Francia (39:38)

So with AI generated creative versus dynamic, I think the difference is when it comes to programmatic and executing these things, is AI.

Christopher Francia (39:50)

Trying to. If you're talking about AI modifying creatives in real time in the midstream, that's going to be a huge no, no for a lot of brands. No. Right. Because again, hallucination, it just takes one mistake for that brand's 100 year legacy to have a huge crisis. And we had a company pitch us this and I said very plainly, it's like, it sounds like a really novel idea. No brand on earth is going that has like a reputation. It's going to risk that because they have no control. And that's one thing. If you work in the agencies, brands, you know, especially larger brands, really want to control their image and their reputation and where they're running and what their creatives say. So taking that out sounds like a great idea. We'll make a custom ad. But that generative real time adjustments is very dangerous until you can have full control, which we're just not there yet.

Interviewer (40:38)

And I imagine that's similar to having an AI agent actually be the one managing bitstream as opposed to doing planning.

Christopher Francia (40:47)

Correct. And I think when you talk about where the bitstream comes into play, when you make a change in DV360 or the trade Desk, you're not touching the bitstream, what you're doing is touching Trade Desk's internal system. And then at some point trade Desk updates their logic. So when the bitstream comes in, it then starts using its bidder taking that information. So you don't actually touch the bitstream director, you're touching another system which is then going to talk to a bidder. So when they're talking about AI in the bitstream, they're mostly talking about AI talking to demand side platforms, but they can't talk to the bitstream. And the reason is there is physics. The midstream is a very small window of time to communicate with it. I think a DSP has maybe up to 100 milliseconds. So your fastest LLM can even get close to that.

Tim Peterson (41:32)

I don't know.

Interviewer (41:32)

I asked Gemini this morning. Gemini was just like a really optimized one can be 10 milliseconds. Now this may be hallucination.

Christopher Francia (41:38)

It's a little bit hallucination because you have to think of it's not just speed but volume. Okay. So you have to have it take that. But also doing 100,000amillion quests per second. So when you take that it becomes a physics problem where there may be a way to scale it. But the cost of doing that, to make it that fast is not going to back out ever in terms of an roi.

Interviewer (42:00)

But is this where things like the Agentic RTB framework, where so much of that, as much as Agentic is in the name, is really about speeding up the programmatic process. Or even like Amazon with RTB fabric, which is a similar idea where there's like a bit of co location going on there. But it's about like hey, ad tech companies just like put your logic up in the cloud in AWS and we're going to put it all very close together. Similar data centers are the same data center and so when they need to talk to each other, that data is going to transfer a lot quicker.

Christopher Francia (42:36)

Correct, Correct. That's a lot of what that is all really designed to do. And we're a little behind the curve in our industry. When you look at the app ecosystems, your iPhone is when people want to deploy an app, they give it to Apple. Apple co loads down your computer in a container and that's your application and then you can use it and it needs an update. Just update your one app. They don't have to update all Macaus to do it. We've been kind of operating the old way where people are being manually plugged in with separate APIs for each. So the Agentic RTP framework is really there to standardize plugins doing whatever they need to do and recommending they colo them, let them install them as containerized apps on the servers that are running the bidder or the multiple servers running the bidder. So you reduce latency. So if you do that, then you're dealing with only internal latencies. You're maybe gaining 80% efficiency from just making an outside call. So that's really what that's for. That's a huge value. I think that's actually a really good protocol for that. But those containers are not going to be an LLM that's at the current state of the industry. It's just not fast enough. And you couldn't containerize the size of a container for an LLM would be insane. They're just not doing that.

Interviewer (43:47)

Even something like Gemini, like the light version or the diode version of the.

Christopher Francia (43:51)

LLMs, I mean if you do that, you still. It's the architecture underneath it and all the requirements it has and things like that, it's not going to be very fast and it's going to be quite large. And then if it's really a small model, it's not going to be very accurate, going to have a lot more hallucinations because it's on a smaller subset of data.

Interviewer (44:06)

Could you take the Gemini Pro or the. What was last week, Opus 4.5 from Claude. Could you just take that, have like a single model and put and containerize that within the same data center as.

Tim Peterson (44:22)

Your viyant container, your trade desk container?

Christopher Francia (44:25)

You could. It's about. If you're talking about the bitstream though, they only get maybe 15 to 25 milliseconds to respond. 50 milliseconds if we're generous.

Tim Peterson (44:34)

So still not enough.

Christopher Francia (44:35)

It's still not enough time for a large language model. It's enough time for a machine learning model that's been tuned and it's deterministic or whatever, like a scoring model. Because you can do a very fast prediction that way. You don't need as much on the back end. It's been trained. But large language models have to do parsing the transfer input. They then have to compute that, do a mapping to a very large graph and then output the closest result in a way that's spit out line by line or whatever it is. So it's just not fast enough yet. It will be one day. I'm sure they'll figure out ways to make it faster. We're just not there yet.

Interviewer (45:09)

Yeah, because even if you bring in user context protocol where the whole idea there is instead of having to deal with A bunch of text to describe an audience. We're going to translate that into these embeddings, which is.

Tim Peterson (45:23)

Seems similar.

Interviewer (45:24)

The idea is similar to like binary code of like, computers don't need a bunch of text to be able to understand things, they can understand ones and zeros. Embeddings are like a version of that. A very complicated version of that.

Christopher Francia (45:35)

Yeah, it's what the base of machine learning is. Tokenized embeddings, vector databasing and things like that. It makes it quicker for lookups. We actually use vector database and lookups for our audience discovery platforms. So we tokenize all the possible audiences out there and then we can do natural language lookups using a sentence transformer, which is a very technical way of saying you type it in and we can infer what it is and return things back. We do it very quickly because we're not using an LLM to process that. We're just doing. We call it elastasearch in the industry is what it is just a tokenized vector database. They've been around quite a while and it's a very good use case for that.

Interviewer (46:11)

Okay, and is that effectively what user context protocol is or do you see differences?

Christopher Francia (46:15)

I think everything there is. All these protocols are trying to create universal search standards because every single company has a different way of plugging in. So they're all trying to create a level playing field of standards. So if we could all just learn one thing.

Christopher Francia (46:28)

The issue is they're all trying to solve a problem that doesn't yet exist. Right. We don't have agents demanding to talk to other computers right now. I mean, we can ask everyone in the room. I don't think the majority here are using agentic AI daily to buy media all over the place.

Interviewer (46:45)

I'm curious, the show of hands, is anyone using agentic AI for your ad bias?

Christopher Francia (46:50)

Yeah, I think that's telling. Not a single hand.

Interviewer (46:52)

Not a single hand.

Christopher Francia (46:54)

So there you go. It's just not. It's a problem that doesn't exist yet. Whereas when they Compare it to OpenRTB. OpenRTB was developed because the industry was at a crisis. We had. Everything else was either done through these weird custom APIs or it's done through tags. And everyone had different macros, everything had different requirements. Anyone who used AOL back in the day, the AOL one platform ripped their hair out so many times just trying to optimize and figure all the bugs out there.

Christopher Francia (47:19)

It was a mess. And OpenRTB was there to solve a real problem. We had just too many different ways of doing it and the whole system was collapsing under its own complexity. OpenRTB was to standardize that. And the key there is the big boys were the ones the Google and they weren't exactly an early adopter but they were on board with developing it and they help, you know, these larger companies help push it forward. And I think with these other protocols you don't have the largest companies backing them right now.

Tim Peterson (47:48)

Right.

Interviewer (47:49)

Like Google, Amazon, trade desk.

Tim Peterson (47:51)

They haven't come out in support of ad contacts protocol.

Christopher Francia (47:53)

No, no they haven't. And when you look at under the hood there's a lot of logic to that. It's a business choice. There's nothing wrong with trying to standardize communication layers between systems. Where the issue becomes is they want to make it. I think in their pitches you could be able to have an agent shift budget between the DSP to find the right flow. Well that's not good for Google's business model. It's definitely not good for trade desk business model and it's not good for Amazon's business model because they don't want to be losing money to the other be compared to each other that way. It's really good for a mid tier DSP who's just looking for any chance to prove itself. So I don't imagine there's going to be a heavy adoption from the big players in the space.

Christopher Francia (48:33)

But I mean I could be proven wrong. Well no, I've seen other protocols go this way and I think buyers JSON or demand JSON that was pushed years ago, that was dead on arrival because the buyers, the big platform said not a chance, we're not going to support that. And that was the end of it. So well meaning, good intentioned but I do think it has a very big hurdle to cross before it really changes.

Interviewer (48:56)

Anything because it seems like there's just kind of a fragmentation period going on right now.

Christopher Francia (49:02)

Oh yeah.

Interviewer (49:02)

Because not just like what we're talking about here, but even again Amazon with RTB fabric, I don't know what the incentive is for Google to be like yeah, bet we're going to also support things living within aws let alone for a trade desk to be like yeah, cool, I'm going to definitely support the technology owned by one of my biggest rivals.

Christopher Francia (49:24)

Yeah, it wouldn't do good in an investor conference for sure when they have to speak at quarterly earnings report which is they're all public companies. No, and that's the sad truth in the room is we build to those platforms when we're building at the agency level, the things we develop, we are not building our, we're not spending our resources and our time building to a small API to a smaller player because it's a resource allocation thing. We have to go where most of the budget is. So with Google to get it working with our systems, it's three APIs. That's just DV360 if you want to do CM360 for your ad serving. If you want to do Google Ads, that's another two APIs that you're adding in there. Trade desk is also two APIs. So we've already done the work and it is fragmented but they like it that way. I'd be fine. If there's one protocol, we would all love that as agencies but we don't get to make that decision because they control it. Their world that we live in tech wise and sadly we have to. Whatever they give us is what we get. So even if we wanted to standardize it and we're pressing it, I don't think we have enough power because we've already built the connections. What are we going to do? Turn them off and say we're not going to use it until that's not how it works. Right. We're going to keep using it, we're all going to keep buying it even if we're frustrated. We all have gripes with everybody at some point in the industry with some buying platform, but we still use it.

Christopher Francia (50:46)

So I think that's ADCP's biggest hurdle. I think Agentic RDB will probably actually find adoption because it's mostly a middleware system and there's a lot of fragmentation there that could clean up.

Tim Peterson (50:57)

So it's going to improve and there's.

Interviewer (50:58)

Like a clear need to clean up. There's a clear need eliminate lanes.

Christopher Francia (51:01)

There's more things that need to touch the bid stream that just can't be integrated. There's so many new technologies that people want to test and they just can't even do that. So there's a problem there that needs to be solved. So I think agentic rtb, putting away the agentic stuff really just it's a containerized, the plugin system. That protocol actually has a lot of legs to it. And I do hope it gets adopted more widely because I think it will improve a lot of the tech and lower tech costs which hopefully lower other costs. But the other ones I think are still in search of a problem that hasn't developed yet. Okay.

Interviewer (51:32)

And so I feel like you've made a very clear case against AI agents when it comes to programmatic buying. But what about other parts of the process like on the planning side or the measurement report?

Christopher Francia (51:44)

And this is where AI is really good. AI is good for insights and summarization and helping you. It's a tool helping you ideate and things like that. So we found a lot of success with AI when we're like, hey, take a look at this information here. What are some potential audiences that might be relevant to this? Like an audience searching things, they might not get it all right, but it helps us ideate so it can help us generate ideas. Other ways we utilize it is if you need to do menial tasks like, look, I need to copy this campaign template 10 times and I need to make these three changes in it each time. Can you just do that for me? And they can do it. It might be three, might be 300. So like repetitive tasks, things like that, things that are very clear and binary AI is great at. And that's where it's really starting to come in its own. I think most people here have used probably some form of ChatGPT or Gemini in their daily workflow to solve problems, to help debug something, to ask, advice, an opinion of something. But they're in the driver's seat. They get to control what goes in and goes out and they're hopefully trained well enough that they recognize when it's wrong and they can correct it. And I think that's where it's good.

Interviewer (52:55)

And how do you make sure that they do? Because one of the challenges there can be, you have the 25 year old media planner that everyone always loves to dump on, but that person who I'm sure is grossly underpaid, could be asked to do things, could be pressured to like adopt AI technologies in your workflow because it feels seems like so many agency employees are getting that kind of pressure and they may just think, cool, I'm going to have ChatGPT just knock.

Tim Peterson (53:24)

Out all of it.

Interviewer (53:25)

Or I'm going to use cloud code and I'm going to go into danger mode where I just say cloud code, do, do your thing. I'm going to go get lunch when I, when I get back, this better be done.

Christopher Francia (53:34)

Yeah, I mean, give a human a shortcut, they'll try to take it. I think this is where companies need to evolve to train. And at the end of the day, it's the boss's job to make sure his employees know what they're doing when you hire them to make sure they're fully trained. It's not like people come out of college and they just know media buying like the back of their hand. There's not really a class that teaches that that can keep up with it. It changes too fast.

Interviewer (53:58)

So a lot of I'll do in any of that ascension arc in terms.

Christopher Francia (54:01)

Of training and stuff, we do a lot of that. The tech side and the engineering side is very connected to our media buyers and our activation team. There is always a dialogue and we do lots of trainings. Weekly side conversations. We're always educating each other so everyone knows what's going on. We do not have a push that everyone must use AI everywhere. That's not been a push that I've seen from our company.

Interviewer (54:22)

Is there a nudge? Is there? We recommend our employees.

Christopher Francia (54:28)

What they do is they explore what we teach them, what AI can do, what it's good at. And we don't say don't use it. We say let's find a purpose for it and let's test that purpose and then release it. That way we're a little more careful.

Interviewer (54:39)

Any purposes that have emerged lately?

Christopher Francia (54:41)

Yes, we have an insight driven thing about generating a text based report about a campaign result. So we use LLMs to process that to help build out these, you know, because we have to do you have 10 clients, you might have to do 20 reports a week, you know, depending on the cadence or whatever. So it helps us speed up some of the stuff we used to do copy and pasting and writing it up. So we use it that way. Again, insight driven and things like that. We are experimenting with other stuff, but we're really more into the deterministic ML stuff when it comes to like optimizations and things like that. We have a lot of very advanced stuff there, but when it comes to like your agentic stuff, it's definitely more on the insight analysis and ideation side and research.

Interviewer (55:26)

I think that's a good clear distinction to end the conversation on. Chris, thanks so much. This is really fascinating.

Tim Peterson (55:41)

Thanks for listening to this episode of the Digiday podcast. If you enjoyed it, please leave us a rating and a review on Apple Podcasts, Spotify or wherever you wherever you're listening. Get more from Digiday with our daily newsletter sent out each weekday morning. Visit digiday.comnewsletters to sign up.