Who Owns Intelligence? Eddie Drake on AI, Intellectual Property, Data Clouds, and Why Trust Will Decide Enterprise AI

Watch on Youtube

Listen on Spotify

For the past two years, the AI conversation has centered on one question: Which model is best?

But as AI models become increasingly commoditized, the real competitive advantage may lie elsewhere.

In this episode of Signal & Noise, hosts Rio Longacre and Brett House sit down with Eddie Drake, Global Head of Marketing, AI & Industry Go-to-Market at Snowflake, to explore why proprietary data, governance, and enterprise context are becoming critical to the future of AI.

As businesses integrate AI across their organizations, new questions are emerging. Who owns the intelligence created by AI? How can companies protect institutional knowledge and competitive insights? And what happens when the context that makes a brand unique becomes valuable intellectual property?

Drawing on his research into AI governance, Eddie explains the rise of the Marketing Context Layer and why organizations need to rethink how they manage data, evaluate AI vendors, and protect their competitive advantage.

Rio and Brett also explore the evolution of the Data Cloud, Customer Data Platforms, modular architectures, AI agents, token economics, and the changing enterprise technology stack.

Topics include:

• Why proprietary data is becoming a major AI advantage

• The hidden IP risks of generative AI

• The rise of the Marketing Context Layer

• Why governance can help companies move faster with AI

• Protecting competitive intelligence

• The evolution of Data Clouds and CDPs

• AI agents and modular enterprise architectures

• Token economics and open-source models

• Why trust and control will matter in enterprise AI

If you enjoyed this conversation, subscribe to Signal & Noise for more discussions with the leaders shaping AI, marketing, advertising, and enterprise technology.

Read the full transcript below.

---

Brett (00:00.921)

Hey everybody, welcome back to Signal and Noise. This is Brett House with my co-host Rio Longacre. And today's guest, who we're thrilled to have an old buddy, industry colleague, Eddie Drake, who is industry principal, marketing, advertising, and experience, for regulated brands. That's a mouthful. Regulated industries, exactly, like financial services, things like that, at Snowflake. and I knew Eddie from his days at Bank of America.

Rio Longacre (00:06.446)

Yeah, they're

Rio Longacre (00:19.32)

Regulated industries, right?

Eddie (00:21.236)

That's

Brett (00:30.503)

where you were the SVP of marketing data strategy and transformation. I'm getting these titles right. I'm nailing them Eddie. Yeah. Very regulated industries. And and Eddie was always one of the few people that I saw especially from regulated industries, brands in general have have struggles with this, outside of the financial services and insurance companies, of actually going out and speaking in public and and offering thought leadership and points of view. Maybe not directly tied to the brand, but

Rio Longacre (00:36.418)

Well, a regulated company, right?

Eddie (00:37.758)

So, that's all.

Brett (00:59.755)

You somehow got that ability at Bank of America. You just asked enough. We were the gear that squeaked somebody had to oil Eddie Drake, otherwise otherwise you're gonna walk. But but so so he's been a a you know a close sort of colleague. You you you were certainly involved with a lot of what we did at New Star, from a public thought leadership and subject matter expertise standpoint, you know, events like Brave New Worlds, I think you might have even been on my old podcast, No Hype.

Rio Longacre (01:05.793)

Ha ha.

Eddie (01:23.656)

Yeah. Yeah, yeah, yeah.

Brett (01:27.057)

so so very big client and great to have you on the show. And you know, I think you know, you're unique because in in in having met a lot of people across the industry, because you've really lived in this sort of intersection of technology, data strategy, some analytics and data science thrown in there and marketing, and and you know, you're kind of a jack of all trades, master of none, but you've always had deep, deep knowledge on how these things are built, architected, and work.

And I think that's why that's why we always keep inviting you to all of our thought leadership events. So you could offer some intel. So thanks for joining Signal and Noise. Yeah, and we've got my and my intro is not done yet, though. We've got more, right? We've got more for you, Eddie. Right? He's you're you're a Boston boy, which I didn't realize. I didn't realize. Brace yourself. He went to Babson College, I was a B U guy, so th you know, I didn't realize that you're also a UMass Eisenberg School of Management.

Rio Longacre (02:04.334)

Yeah, good to see you Eddie, for sure.

Eddie (02:06.664)

Yeah, it would

Eddie (02:10.474)

boy. boy.

Rio Longacre (02:11.918)

Prace yourself, Eddie.

Brett (02:23.545)

graduate studies, right? The the MBA, I'm assuming.

Eddie (02:25.941)

Little little time in there little time in there. Yes sprinkled Boston through and through

Brett (02:29.41)

Yeah, I'm you I'm a UMass Amherst guy.

Rio Longacre (02:32.162)

Funny, Brett, actually all three of us went to school in the great state of Massachusetts, right? I I'm a New Yorker, so I technically I guess I I studied it in Belly of the Beast, right? If we're gonna talk I know we're probably gonna go into sports at some point during this conversation today, but that that is funny.

Brett (02:35.938)

Yeah, real

Brett (02:45.6)

Yeah, and you were a UMass Boston guy, right? So so we all, you know, certainly my family still lives in the area. I've been in the New York area for twenty-six years at this point. But you know, loved your recent article, which we'll talk a lot about today. If you haven't read it, the audience, it's on snowflake.com. It's it's called Why Marketers Need to Own Their AI context layer, which is a tactical top title that I don't think sort of sheds light on really what this is truly about, which is

Rio Longacre (02:48.11)

That's right. Yep.

Brett (03:13.518)

really around AI governance, trust, enterprise context, brands IP ownership and brands being able to protect their IP from being sort of taken from the AI and the companies, the platforms that they're working with.

Rio Longacre (03:17.804)

And IP ownership too, right?

Rio Longacre (03:28.248)

Certainly by the frontier models, right, who are using d to train their future models.

Brett (03:33.719)

Yeah. So so yeah, so great article. that you know, that was a a a a huge question that I think we wanna talk about today is IP ownership and the fear of sort of data governance and the topics of sovereign AI and you know, do I really have control of the information that I'm sharing with either the platforms or the AI services that I'm working with? Or is it going to escape the confines and suddenly you know, I'm gonna be disintermediated or my advertising, my marketing, my strategy is gonna look a lot less

unique and differentiated and kind of homogenized, you know, across all of my competitive set. That's a huge topic and a huge threat. So thrilled to have you on the show. you know and I wanted to say one thing, the f the the best quote from that article was you said I'm a I'm a lifelong Bostonian and have watched avidly as the city of champions raised banner after banner to the rafters, a total of 13 sporting championships in this situation alone. So take that New York Yankees.

Rio Longacre (04:33.269)

trade that for one next championship guys. Just saying

Brett (04:33.344)

Brett (04:38.144)

Yeah. The Dicks the Dicks are the heart and soul of of New York City. So so so yeah, so you

Rio Longacre (04:43.437)

That was a long time coming, that that championship and it that felt good to watch, so.

Brett (04:47.586)

Yeah, it really did. It really did for all of us living there. So so I th I loved your sports analogy. We'll get into that and how it tied into how you're talking about this fundamental problem. Like how, you know, you think of of sports leagues and how they compete and they're not giving away their IP. They need any strategic ad advantage and edge that they have. And and that's kind of analogous to what you're seeing in sort of companies, big brands giving away IP unknowingly and contractually in some cases. So super interesting topic. Eddie

Thanks sorry for the long winded intro, but I think you deserve it. You're awesome. Thanks for joining the show. And let let the audience know kind of like a little of like you know, where you Yeah, what did I not miss in that that short novelette about about Eddie?

Rio Longacre (05:25.769)

What what did House not miss in that long winded intro?

Eddie (05:31.231)

No, look, you pulled on so many threads there, Brett. I mean, I'd say this, and I smile and joke because, to speak publicly from a conservative brand, even from a platform, it has to be something valuable, meaningful. And I think back to my days on the brand side, almost 20 years, again, you kind of hit.

the high points of what I covered there, but in the last decade, there was so much focus on privacy. In fact, that's where I started a lot of my research was on how do modern brands actually thrive and protect in the cases and instances of privacy. In the AI era, and perhaps even more critical to marketers, it's now we certainly still need to protect privacy, but how is it that

from a competitive differentiation, from an intellectual property, from like, what makes my brand my brand? Really critical and meaningful in the future of brand strategy. It's interesting here because I'll talk a little bit about my research later, but every marketer when they hear governance is like, compliance tax, don't talk to me about slowing down my campaigns.

Rio Longacre (06:51.015)

Governance can mean different things to different people too, right?

Eddie (06:54.75)

For sure. And I think that's the point I really want to hit home. And I toyed with, we call this AI governance? Do we call this AI strategy? It's like, all right, I'm going to let all that be aside. But as a CMO, I now own and have accountability for the AI that my teams use. I've got to make sure that that protects what differentiates me as a brand. I've got to ensure that I can go fast. But I still have a seat belt there or a break.

to hit this thing well, what I've done to protect in this hard fought differentiation, I've got to make sure is not only consistent on all the places where AI is going to shape and shift the future of the brand, but not make my competitors better. That's a big part of the thesis and a big part of why I chose to look at this so deeply in this seat now.

Rio Longacre (07:50.04)

So, Eddie, we're really happy to have you on. We as Brett mentioned, we have wanted to do this for a while. And I I do like the fact that Snowflake is a little more open with letting its people do things like this. I Aaron Foxworth, he's appeared on a number of one of these. If Aaron, Aaron, if you're listening, we'd love to have you on at some point soon. I think I believe she's your boss over Snowflake, right? Yeah, she yeah. Yeah, she's great. I've known her for a number of years. the timing of this is really good in a lot of ways. The last thing since the Chat GPT moment, what is it's been almost been four years, but or

Eddie (08:07.028)

Cheers,

She is great. She is really great.

Rio Longacre (08:20.471)

Which is kinda hard to believe, right? But since then it w the conversations have always been on which is the best model. is this new model how much better it or faster it? How much like deeper does it think, or what benchmarks does it blow past relative to the previous models or relative to the competitors? That's really been the c where the conversation has that's been most of the conversation the past few years, right? But I think that has fundamentally shifted, especially over the past few months. There's been more of a focus, I think, and

Brett, I know that I think you agree, right? More on the data itself as the models themselves become a little more commoditized, right? Or a little har harder to discern what are the advantages of one over the other, right? So the real differentiator when you when you boil it down is the underlying data itself, specifically that proprietary enterprise data. Everyone's trained in everything that's been out before, right? People are, I mean, even you read about the LLM is looking for old rare books to try to to try to take and scan and

And and use to train. But the new model, specifically the enterprise data, right? That's become critically important, especially as we look to build that application layer on top of these large language models, right? You need that specifically vertical information from these enterprises in order to do that to build really, really good applications that work and actually solve use cases that are relevant to businesses themselves, right? So that's where the conversation shifted shifted. And I and within that, how do we

protect intellectual property or IP, how do we govern AI responsibly? And how do we build systems that both people as well as organizations can trust as well as rely on? I think that's where the conversation's going. And that's why Eddie, when we saw that article you you published a couple of weeks ago that I know you wanted to talk about, that's when we just say, okay, this is the timing is good. Let's make this happen. So I guess first question then would be like looking at the governance versus space

But we'll first maybe taking a step back. Why do you think that why do you think we've shifted? Why do you think the conversation has shifted so fundamentally to the data and to data governance and and and IP and you met Brett you mentioned sovereignty as well.

Eddie (10:19.23)

Yeah.

Eddie (10:29.448)

Yeah, it's a good question. There's a couple of things in there. Maybe I'll touch quickly on two years ago, we were talking all about the model. They were profound. They were unharnessed. They were something that everyone needed to get after. And like any good competition, we've seen races neck and neck. We've also seen differentiation and distinguishing, where I see the data becoming more and more important plays out in a couple of ways. And I'll keep this

or obvious reasons in the marketing context. But if we had asked the question a year ago, who's the best model? Everyone is going to raise their hand and have a vote. I won't venture mine, but we'll just keep that here. The same thing happens today if you ask a question, but it may or may not be the same answer. How a brand uses that AI, though, is likely very similar. Their competitive strategy is going to evolve. Their brand priorities and goals are going to evolve.

Brett (11:07.595)

Have an opinion, yeah.

Eddie (11:29.48)

that context as the industry is calling it shouldn't end. In fact, it should multiply and get smarter on itself as brands use AI more and more to build content and creative, to build journeys, to build messaging strategies, even to evaluate performance. The way I talk about my brand, to find my metrics, my KPIs, my outcomes, even the vernacular. Yeah, the vernacular we use as a brand. mean, I'll pick on Bank of America as an example.

Brett (11:51.223)

My audiences.

Eddie (11:59.271)

If I said teammate to a large language model, that could be confusing unless I've defined like, well, teammate is actually how we talk about our employees. So there are considerations again with this big C context that have become really important so that I cut down hallucination, I have predictability, I have consistency across those, I'm using AI. But the thing that I'm seeing actually come through more and more here is that

depending on where you sit in the marketing organization, you may have a different need or ask. You may say, hey, this model is really great for creative development, generation, versioning. This other model is awesome for workflow, for deep research, for competitive intelligence. I'm gonna use social scraping with a different model, having silos of different information about who we are as a brand, engaging and interacting with those brands.

Believe it or not, that's governance. That is how we stay consistent from a brand voice perspective. As we see these models continue to get better and better or do different things, want the ability to be versatile and to actually engage with what is best for the job that's being done. As simple as that.

Brett (13:16.055)

So so you said in your article, and and and maybe this is one of the bigger points, you know, with that in mind, that the context layer is sort of emerging as as the most strategic asset a brand can own. You know, it used to be the data, like we've got first party data, we've got enriched third and second party data, we've created this sort of data moat that is our our strategic advantage. Now it seems like you're pointing at the fact that there's a real risk of that moat, which leads to strategic differentiation and all the other things that you discussed, right, around

Eddie (13:23.667)

Yeah.

Brett (13:45.971)

everything you do from a brand and marketing perspective, you want to create a mode around and keep it unique from from from your competition. You seem to be suggesting that that's now starting to leak out and it's leaking out because of of how brands end up signing contracts with with you know and you read one of these contracts to me about what the the vendor was be was asking to do.

Eddie (14:08.116)

it

Brett (14:12.347)

and it was, you know, to to you to leverage some of this information for developing other services. Can you go into that? Because I think that's a really interesting idea, generally.

Eddie (14:19.432)

Yeah.

Eddie (14:23.688)

Yeah, for sure. So at the risk of not bringing the whole audience with us, real quick on what context is, because I think there's a number of definitions that are floating around out there. again, operational rules, policies, business metric definition, like tribal knowledge about what makes your brand a brand, like codified in a way that AI, machines, agents can read that.

Rio Longacre (14:35.436)

Yeah, agreed.

Eddie (14:47.55)

Like I like to think back 20 years ago, everyone was saying like, we have an integrated marketing communications plan. That was a big deal in a business school setting. It was like, hey, here's the campaign brief. Here's the engagement playbook. But like those PDFs mean arguably very little in the state of AI. Right. So I've got all this IP, this knowledge, like these charters. So now this is what in a, in an AI setting, we are

hotifying so a machine can use and read. It's things like net margin, churn rate, et cetera. Great. So this is all uniquely valuable information to a brand. As we think about it now in the context of some of the research, what I said was like, hey, there is a battle frontier here. Those that are providing services to brands, whether in the ad tech space, the martech space, the model space,

In order for these firms to be valuable, how are they taking a brand's thesis goals, objectives and making them better? In some cases, it's purely that. It's a brand has their context, they bring an AI solution to it. We get something like better campaign optimization, the ability to execute journeys, agents that are gonna help with content and creative.

In other cases, again, a lot of the aspiration to do this research came from seeing what I saw in kind of the privacy and the identity in some of these spaces historically. There are providers that are saying, hey, we actually want to own a context layer and we want to bring anonymous or aggregate or contextual detail from the brand.

into our context layer actually train models, products, and services with you and your competitors' data.

Rio Longacre (16:45.334)

So Eddie, looking at that question, digging in a little bit, one thing you brought up in the article, which I thought was an interesting point, was about marketers spending a long time, to your point, creating all of this IP, whether it's brand, brand standards, creative briefs, process documentation, all things that we we use as marketers to run marketing organizations, right? On one hand, you need it to train these models to be better and more tuned to your needs and more tuned to the way your organization and your team operates. But on the other,

And you're potentially risking giving away that I that IP, and that's really what it is, that could in theory not only train the models to be better, but then be used by your competitors. what like how do you draw the line between the two? Is it is easy to do, difficult to do?

Eddie (17:31.123)

Yeah, it's evolving every day. The idea though, fundamentally, and in some cases this is easier said than done, certainly we see this as a model of success for many, is to bring AI to the data instead of putting your data into AI. As simple as that sounds, it's not wildly profound, but the idea that you bring the intelligence

Brett (17:50.837)

Yeah.

Eddie (18:00.181)

to what makes you protected and unique is ultimately the guidance here.

Brett (18:07.444)

Yeah, and and did you w now your research, how many what did this how many what was the the the the sample that you used? I mean how many how many companies did you interview on the vendor side, on the brand side? What kind of information did you gather?

Eddie (18:16.221)

Yes.

Yeah, 98, roughly 98 of those that are across AdTech and MarTech looking at identity, CDP, CEP, programmatic, the list kind of goes on and on there, but everyone that you would think of as the usual suspect of being able to enrich, provide value, execute a campaign, measure against it, right, in the realm of what is marketing and advertising technology.

I also looked closely at the LLMs who, know, coincidentally actually are pretty good when it comes to saying, hey, we're not going to take and train any, you know, client or customer data in our models, right, and how those are used or applied. The bigger challenge that the models have though is saying, hey, if I've got, you know, this engagement where I'm gonna...

use a chat GPT model, this engagement, I'm to use a Gemini model or a cloud model, making sure that that context is consistent. Even if it's protected by policy, being able to have a consistent brand voice, or set a brand considerations across those. Whereas on the MarTech and AdTech space, we're more seeing these providers are saying, how can we understand and learn the patterns of those customers that

engage and work on our platforms to actually make those products and services better for the whole that are part of that.

Brett (19:50.327)

Yeah, it's sort of it's sort of a shared it's a shared vendor model, right? And it's and it's those it's those pat that pattern recognition that can be applied to multiple sort of use cases within different companies. And I've actually seen this sort of in real life with clients where you know you can take something that applies to retail and apply it to airline because it's a similar pattern. It's just different products and services, it's butts and seats versus, you know, cash register, point of sale purchases, right?

So that type of pattern recognition though is should be differentiated at the brand level, right? They're they're these are patterns that they're using for, you know, customer acquisition, customer lifetime value, inventory management potentially or or or product design, all these sorts of things. That's IP, right? I mean, would you define that as

Eddie (20:36.252)

without a note. Yeah, I mean, like here's a pretty benign example. So we've seen send time optimization. We've seen subject line testing, right? Understanding across everyone that's using an email platform, what's the best time to send an email to get an open rate or to minimize unsubscribes or to encourage interactivity. The reality is that can either be trained on one client's data.

Rio Longacre (20:37.248)

Yeah.

Eddie (21:03.788)

or that can be trained across a category or the entire leveraging set. Most brands don't stop and think, hey, what is actually the source of this data or this metric? And the privacy policies, the AI terms, which is what the foundation of the research was based on, is actually where you find a lot of information about how clean or what rights you're relinquishing when it comes to your vendors, your model providers, your partners using your data.

At the end of the day, it is a direct pipe back to that sports analogy saying, hey, I've taken all my scouting reports. I've taken my playbook. I've taken all my strategy. Am I just uploading that to a place where all the other teams in a national league can get better based on my intelligence? Or am I really protecting that with care so that, again, from a differentiation perspective, I'm OK?

Brett (21:55.669)

Yeah.

Brett (21:59.112)

And it's yeah, and it's important to understand that it's you're not it's not necessarily it could be fully anonymized or pseudonymized data, right? But it's the patterns and that that it go across a bunch of different you know, in and no s no front office in the sports industry would do that in their right mind because you know, i i it suddenly you'd have a bunch of teams on you know, they they they wouldn't have the competitive edge and I think that's the

Eddie (22:04.51)

For sure.

Rio Longacre (22:20.383)

Well, but looking at the patterns themselves, the regulations, you mentioned privacy, you mentioned rules, I would I would categorize all those things under the word you used earlier, Eddie, which was governance. Looking at governance, I know marketers are under enormous pressure now to move quicker, especially with AI. I think that's given them the ability to move quicker, but it's also put a lot of pressure in in terms of elevated expectations for it should not take three months to have a new campaign out.

Turning around a banner should not take two weeks anymore. These things should be immediate. People see how quick they are. But there's the balance, right? Okay, you need to move quicker, but then if you add governance, it always throws sand in the gears, right? It does inevitably slow slow things down. How are you working with clients or what are you hearing about that balancing act between moving fast yet adding governance, which will slow things to a certain extent?

Eddie (22:49.758)

without a doubt.

Eddie (23:13.01)

Yeah. So there's a couple of things here. I'd say there is an investment of time upfront that pays dividends tremendously and exponentially in the back end. So we talk about context layer in this paper or around that definition, but it's really the pattern of how I'm going to connect my information and leverage these AI models, which to your point, Rio, could come and go or be interchanged, et cetera, et cetera.

I'm not saying that we stop shopping for models. I'm saying we put a one-time governed harness in place so that anytime I want to use AI, interchange a model, come up with a new partner, I'm bringing it to a base that's consistent. I think about this a lot like in an F1 analogy, right? So F1 fastest race cars in the world.

You realize and appreciate though for people that have driven these cars, it's actually having a break there that makes them win the race. They have the ability to know that they can go fast because they have something that slows them down, stops them, is surgical in the place before they can accelerate again. And that it's not a perfect parallel, but the idea here is you'd have to let your foot off the accelerator, come to a slow down, turn the corner and then go, as opposed to just saying,

Hey, I've got a tool in my toolkit that lets me go fast. I got a tool in my toolkit that lets me move and be dynamic, right? And then I'm off to the races again. I think the same thing from an AI strategy perspective plays out. I set the thing once I get rules in place. I do that as far left as possible. You got it. You got it. Right.

Brett (24:44.893)

Yeah.

Brett (24:50.877)

Yeah. Short term pain for long term gain, right? Right. And what but what is the scope and scale? Like from these ninety eight companies vendors that you talk to, what is the scope and scale of of like the contractual issues that that sort of give away data that brands don't want to give away? And are are brands hyper aware of this and becoming skittish because they're looking at the contractual language and they're like, I am not gonna dive into this pool, right? Because, you know

Eddie (24:59.625)

Yeah.

Brett (25:18.355)

you know, I might be disintermediated down the down the down the road or whatever. I mean, what is the scale of the of the problem?

Eddie (25:25.456)

Yeah, mean, so it'll show up in a couple of these places that we talked about in privacy policies, in AI terms and things of that nature, which isn't always the same as the marketing material that's on the headline of the website. What to look for here is the, hey, we're going to have a brand kind of hereby grant royalty free, unlimited use.

Brett (25:39.881)

Yeah.

Eddie (25:52.533)

consideration in the form of customer data that's going to help our services become better for you and potentially others that are using this platform. In a nutshell, Brett, I've seen it probably written in 15 different forms of variation of that same very thing. Back on the brand side, I saw the same thing play out in the identity space.

hey, you've got email addresses, you've got postal addresses, you've got phone numbers from your client base. We're gonna use these for verification, right? We're also going to use them to enrich our graph in the position of identity. What's different here is, as you guys pointed out, the privacy is a checkbox. We've accounted for that. I think we as an industry have done a lot better here. Now it becomes, hey, how do I plug that whole of accidental

strategic intelligence leakage around my patterns, pathways to conversion that are naturally observed and inherent in this data exhaust. Even if it's anonymous, right again, protecting privacy and security, but in this case, it's data that's moving in a way that's actually very detrimental to the brand and what makes them unique or differentiated.

Rio Longacre (27:11.263)

So Eddie, on the subject of identity, I think that actually pivots us quite well into another topic we wanted to talk about, which was the cloud data warehouse and then becoming what people are now referring to as the data cloud, right? I know that Snowflake was the I think really coined the phrase cloud data warehouse for marketers, right? again great market share. And a number of places I've worked over my career was the was the cloud data warehouse that was used by marketing as the, you know, for to really be the

Eddie (27:23.892)

Sure. Sure.

Rio Longacre (27:39.871)

the central kind of nucleus around which all of the marketing data flowed. So Snowflake's done an amazing job there. Love your thoughts on the evolution of the space, the new term, the data cloud, like how is Snowflake looking at that? And what implications does that have for, let's say, best practices and enterprise, like data and marketture supporting marketing and advertising.

Eddie (28:07.55)

Yeah, there's a couple of interesting factors here. Obviously not all of them Rio have to do directly with AI, but we are certainly seeing and I have since the last two and a half years seen an inflection, is the warehouse is moving from a storage layer to actually an ecosystem or a platform layer. There is a lot here from a brand perspective that's important.

openness, neutrality, protection, the idea of giving greater empowerment by bringing solutions, models, what have you to the data. Not too dissimilar from the idea that we're talking about, like, hey, I have a safe place to keep my data, to leverage AI without the risk of, again, let's say leakage of IP.

What is interesting here, I think, is as the places where, again, the brand's neutral ground kind of dry up or shrink and narrow, it's giving the opportunity for flexibility, for choice, for almost the brand to be kind of the referee about how and where they do business, which has been different to the case in place as when your data moves or goes into

let's say another solution, sometimes you're beholden to that platform or that product roadmap or strategy that also has your data alongside with it. And it's not just.

Brett (29:37.685)

Yeah, and there's a there's a lot less con there's a lot less control of where that data goes and how it's being used once it's in a th a third party environment. Where you know, there's contractual agr agre you know, obligations, but but certainly you lose a little bit of visibility control versus having it centralized in your own own and oper not an owner it doesn't even have to be owned and operated, it could be rented like from Snowflake central repository. Does that solve the problem of I mean if so if so if people are are working within

Rio Longacre (29:37.694)

Yeah. W one thing I think is it interesting, yeah.

Brett (30:05.682)

They've got all their data housed in one place, whether it's a Databricks, a Snowflake, an Azure. it's, you know, th th there's application layers being built, native applications being built into these environments, by vendors that are headless, right? And so they're forced to work within this environment. Does that really tie to this whole notion of sort of sovereign AI and and sort of zero copy data and zero trust data?

Eddie (30:30.416)

It gets you a lot closer. I'll say that. There is innovation that's happening in all places. In this pattern, we like to see and believe that there's a lot more good than not. But you've got to imagine that everyone that's playing in this game has set of considerations where they would like to engineer recurring revenue. They'd like to create stickiness in their offerings.

I think they're going to become increasingly easier to see though. I think it's the point, Brett. And as brands get more flexibility, control, choice, see their data as actually having gravitational pull for those in the ecosystem and being able to audit or interchange those that are serving a purpose or doing a task very well today that may be replaced or moved out.

if they don't meet the paradigm of excellence that the brand wants to work to. But I think one of the biggest disruptive factors there is if the data doesn't have to move and there's far less cost and time in the mix, your optionality when it comes to choice, it gives you a lot more flexibility. And that's really what's prompting, I think, so much disruption and so much focus on this pattern now.

Rio Longacre (31:53.428)

Yeah. It's interesting, Eddie, with this pattern. I mean, I've worked with a few clients on this recently. Some of the implications are actually kind of interesting to think through. Like for example, if the if the data cloud becomes your system of record for customer data, right, then what does your CDP do? Well, it's probably you're probably doing zero copy, right? And that's why I think high touch, for example, has done pretty well in recent years because it's just, you know, it's just really a zero copy way to just activate data, right? And push it to different activation channels.

Right. And then so you're really looking at a redefinition of the CDP. And I think because of that, certain things that the CDPs did a few years ago, identity resolution, which you brought up maybe being the most important, is kinda up for grabs now. Like where does it sit? Where does it sit for for known data, for CRM data? As well as where does it sit for unknown, maybe even cookie data, which you may be

you know, or or data from let's say paid channels, which which you may be collecting a lot of, and it's you're probably you would never put that in a C D P 'cause it's it's totally unstructured, right? You're probably gonna be put it in your into your data cloud or cloud data warehouse. So I think rethinking some of these assumptions about where do things like identity set are coming up a lot, thoughts on that.

Eddie (33:01.48)

Yeah, it's definitely, I mean, you hit the nail on the head. It's a disruptive moment. And we saw a lot and heard a lot around composability. I think that's been one of those terms that had really strong resonance for the idea of how I'm going to assemble an architecture, my partners around my data. Increasingly, and this may just be me, but I'm thinking about this more as modularity than necessarily composability. Because the other thing AI is

Brett (33:28.969)

Yeah.

Rio Longacre (33:29.32)

Better term.

Eddie (33:30.65)

is doing is opening up the ability to fill in the blanks and create custom code, custom applications. We're certainly seeing that. We still see a tremendous amount of domain specialized applications from the partner community. Hightouch, a great example that, where I have a certain set of features, functions that I want to achieve in my stack. Partners that do that exceptionally well should play and fill into that role.

Brett (33:39.081)

Yeah.

Eddie (34:00.501)

There are plenty though that say, hey, I from a brand side have this unique consideration in how I do my business. It could be a process. It could be, you know, something that's well protected in terms of how they build propensity models or, you know, journey prediction models. I've got to modularize where I plug that in. And I want to do that in the least disjointed in the smoothest way possible when it comes to my future quote unquote staff.

Brett (34:27.293)

Yeah.

Eddie (34:30.152)

Right, so it's.

Brett (34:30.161)

And w and with as with as little lock in as possible, right? You don't want to have to go to another platform to buy a whole bunch of other capabilities when you only need that one discrete thing and you don't and you want it to be plugged into your stack, into your data infrastructure, versus having to go somewhere else to do it, right? To keep it unified and connected, right?

Eddie (34:33.694)

For sure.

Eddie (34:46.494)

Yeah.

Rio Longacre (34:49.747)

Yeah. Yeah, and by the way, I think modular is a much better term than composable when describing this type of pattern and this type of architecture ready. So I r I really like that a lot in some of the work you've done. So kudos to you there.

Eddie (34:50.375)

Absolutely.

Brett (34:54.824)

Yeah.

Brett (35:02.449)

Yep. Totally. Composability's got g gotten some baggage lately, recently. And I've I've heard some pushback from clients about using that term because people have associations with what it means. But you know, I'm another thing I'm gonna steal from you, Eddie, modularity. I think that's a good way. Headless is another term that they talk about. Headless headless applications or whatever.

Eddie (35:17.364)

dip.

Rio Longacre (35:21.201)

Yeah, that that's coming that's coming back. It's you know, that making a rebound now. but I got another trump we did want to talk about, Eddie, which was very prominent in not only this article, but a lot of work you've done is is the the concept of a context layer. I mean, I've read, I know that you've been big on this, very important. love if you can maybe define that for listeners. And then I know you argued that marketers should own it, so maybe give a little context about why you think the argument is even needs to be made.

Eddie (35:25.074)

It is.

Eddie (35:34.526)

Hmm

Eddie (35:49.897)

Yeah, so again, context layer here is that which can be codified, actually written down in a way that can be reused across AI or truthfully even conventional marketing. But we're talking about operational rules, things like frequency caps or management. We're talking about policies, the types of customers that should get a certain offer. We may have a rule that says,

If we've got somebody on our site browsing high margin luxury goods, like we do not, you know, pop up a 20 % discount code. That's all knowledge that is institutionalized within marketing. That's part of the experience that has historically been in documents and campaign strategies, right? That now actually can get encoded in a layer that any AI that's connecting to that context layer.

can be smarter. it's things like governance. Again, it's things like tribal knowledge, vernacular. Importantly, in the analytic community, it's, how do we define cost per acquisition? How do we define churn? What are the things that we as a brand may adopt and embrace as standard terms like audience or campaign, but may also have, if we call the subsets or the subsegments pods,

We want to have that in a way that, again, AI or any of the solution that connects to our data can use. I think that from a brand position, it says, hey, this is completely a brand-centric exercise. This is something that lives, breathes, gets smarter as marketers use AI as our business continues. The inverse side of that is saying,

Hey, I've got a CDP, an ESP, a CEP, or any other three letter acronym. We get all the way into a black box media bitter. The idea here is that those companies are actually collecting information from across potentially their client base when it comes to usage, telemetry, execution of actual client campaigns, perhaps even across

Rio Longacre (37:46.449)

And there are many, yeah.

Eddie (38:09.096)

their clients, is again, where, you know, kind of this relative danger comes in. But that's something that from a marketer's perspective doesn't port well, you know, from channel to channel or campaign to campaign. It's something that actually just trains, you know, the the engine of the execution solution to get better. And I want to know.

Rio Longacre (38:32.593)

Eddie, I think a lot sorry to interrupt, but I I think over time there's always been this tension between marketing and IT about who owns customer data, right? Is is it owned by the CIO? Is it owned by marketing? I think that's CDPs gave more power to marketers over time. And and I think the the pushback to that was you have too much power, you're screw you're creating risk and compliance risk speci specifically, or data leakage, or you name it. And IT, especially with the rise of composable, gradually took some of that power back. So

Seemed that your argument was more marketers should own this context layer. It's too critically important. Thoughts on that?

Brett (39:07.215)

It's point of differentiation for for it it seems like your primary

Eddie (39:12.55)

Yeah, I will say this. I think it's a huge resolution moment for the IT offices and for the marketing offices. And when I say a brand should own it, I would include IT inside of the brand. Like this is really where the CIO, the CDAO, the CMO come together and say, hey, it's our job to institutionalize and protect this knowledge. It's our job to go out, acquire, deepen, retain, grow the business.

we together can assemble this and rightfully the only ones who should assemble and own it as opposed to outsourcing or renting it by a place that it may take advantage of it into how it helps, let's say, power or train our competitors. And I think that's the point. It's not a CMO or a CTO. It's a, what does the brand do collectively and the functions that make it up and support it?

relative to, let's say, those that stand to benefit from the data that aren't part of the brand themselves.

Rio Longacre (40:14.842)

Eddie, actually like that answer a lot. Instead of saying marketers should own it, what you're saying is it's a continuation of what we've kind of already solved for with marketing data, right? Instead of no marketing's gonna own it, no, see see like IT is gonna own it. We've kind of come to this okay, this detente almost. Okay, well, actually we both like we like the steward of it's maybe gonna be IT, but marketing's the users of it. We need to get together to to manage and run this together and have the right systems in place so we

Eddie (40:37.331)

You got it.

Rio Longacre (40:42.738)

So we're not gonna run afoul of compliance issues or privacy issues. I think that's kind of where we come to the customer. That's actually a very logical answer for like this kind of context layer and like where it sits and who runs it and who owns it.

Brett (40:54.096)

Yeah. yeah, no, no. So it seems like it's it seems like you're arguing 'cause you know, I go back to to Scott Brinker a lot and the composable canvas and sort of data is the new operating layer and and it almost seems as if the ar you're making an argument that data's less commoditized or more commoditized, I should say, at this point. Well you you've collected a mask first party data. It's really w the context layer, how do you how do you translate this stuff into brand

Eddie (40:54.388)

Yeah, it's so fascinating. Go ahead, Brett.

Eddie (41:02.205)

Yes.

Rio Longacre (41:18.28)

Well, if the models are commoditized, right, the the the data is like is your IP, right? Like like the context layer becomes the way that you could activate and build on top of it and do it in a way that's compliant with what your brand wants, right? Is that fair to say?

Eddie (41:23.55)

That's it.

Eddie (41:31.528)

That's precisely it. Yeah. I mean, that's what makes you different from your competitor, right? Is what's unique to you. What are the patterns you see? How do you define your profitability, your margin, your conversion, your campaigns? That's far from commodity. That actually is probably.

Brett (41:48.401)

Yeah, and so and that's and that's a combination of all those things. It's the data operating layer, it's the context layer that interprets that and governs that, but that as a whole is fundamentally your IP. It's how you make decisions on on every step of the way.

Rio Longacre (41:58.601)

Yeah. And this is where you would deploy your your agents, right? I mean, sorry to interrupt you, House. Like this is you think about it, like, and I always thought this argument about like, Salesforce is gonna have agents, like ultimately I do think most big brands, right, who are going to invest in a context layer. And if you own a big brand with a lot of IP, good lord, you better be doing it, right? If you're going to be doing that, you will want your agents to be trained on that data.

Eddie (41:58.651)

Absolutely.

Rio Longacre (42:23.362)

using that context layer and all of the rules and r and regulations and and ways of working and markdown files, whatever else you fed it, right? I I personally think that's where you'd want your agents running. You wouldn't want to be using other people's a agents. You might be doing handoffs, right, between your agents and let's say agents run by different by different partners or by different vendors or by different customers, maybe even. But I do think if you're a big organization, that's where you're gonna want to deploy your agents and you won't be using a platform's agents. I don't really think so.

Thoughts on that?

Eddie (42:54.26)

No, think that's right. That's certainly what we're seeing, Rio. There are certainly cases to be made for things that are repetitive and things that can be automated that aren't truly a reflection of who you are from a differentiation perspective of a brand. Yeah, there's an idea out there for pre-built agents that serves well and serves efficiently. That said, to the point of the whole conversation,

those things that differentiate you, those things that should become smarter, learn, act autonomously, or very close to the humans in your organization with that training around your data and your context, you absolutely want to build those. And that's, again, no longer a challenging thing now, that's becoming easier and easier.

Brett (43:40.272)

You wanna you wanna build and gate those and prevent those things from being shared with anybody, vendor, you know, of of any type, right? That's that's the key here. And contractually that's gotta be built in.

Rio Longacre (43:48.553)

Yeah. Well this is yeah, I mean if you look at his bread in that model in that model everything goes most tasks goes headless, right? And then like you have brands deploying their agents and then using these headless tools to actually in interface with these databases and these and these processes sitting underneath in order to accomplish things. I think that's kind of what we're seeing.

Brett (44:04.143)

Yeah, I mean we're seeing we're seeing a kind of a bunch of converging trends. I mean you've got sort of I'm gonna call modular instead of headless because I like that modular agentic execution. one thing I want to get your opinion on sort of federated data ecosystems, meaning does data need to be centralized? you know, which some claim is an old model. we we had a we had a good conversation on that topic. Sort of the death of SAS, which I think ties to you know, maybe it's a little hyperbole to say the death of SAS, but it kind of ties to that modularity of

Eddie (44:04.381)

You got it. Absolutely.

Eddie (44:10.462)

Yep, yep. Yep.

Brett (44:33.253)

Being able to plug in natively to a data infrastructure that's owned by the brand and operate like at the use case level as opposed as opposed to you know forcing that data movement out into a platform, right, that lives outside of that. And then and then you have you know kind of frontier model convergence, agentic trading and media buying, and then and then search upheaval, that's a big thing that's that's affecting all of this, right? Like how that's changing consumer behavior and how they find, discover, and interact with brands, right?

so for for let me get let me go to that. So those those are my six. Those are my top six of like forces that are that are collapsing this old sort of data and marketing strategy model. Federated data ecosystems is one I want to sort of leave lean into because you guys play in a place where you're saying, hey, we're centralizing data at the brand level. and then they can they can partner in and have have their vendors come in and and interact with their data natively.

so there's no data movement, you can kind of have all those advantages of a centralized data sort of lake house or repository. But is it is it what's your opinion on on more of a federated system where if you've got headless agents that can go in and plug into all of your key sort of mission critical platforms, the ERP, the CRM, and and we've been talking a bunch about this. does it does it need centralization? Is that required or can can the agents go and operate and be governed?

to get the data that they need from the places, the mission critical places they need to get it and actually provide all the the insight and input that you need.

Eddie (46:07.742)

Yeah, it's a great question, Brett. I actually say the answer is federation is generally okay. It's definitely something that we see when it comes to data. When it comes to context though, that's what we're talking about. There's great value in breaking down what could otherwise be silos. And I won't say that data sprawl is a good thing. Don't get me wrong on that case. There are reasons why you'd want a federated model.

Brett (46:27.151)

Yeah.

Eddie (46:35.796)

There's reasons why you'd want a centralized model. Like I do a lot of work in regulated industries. It's a lot easier to govern how data is used and who's using it when it's in close proximity. But as MCP becomes more proliferant as, you know, kind of the idea of data sharing, data collaboration continue to proliferate, the data itself certainly does not need to be centralized, or at least that's not

Brett (46:47.535)

Yeah.

Eddie (47:05.204)

necessarily the only way. Again, it's definitely the context here and how we're going to apply that data in its essence or its patterns to the task we're trying to accomplish. I think that's really a lot more practical. Back to Rio's point on speed. Hey, I can't wait till I bring every single row, line and column together to run a campaign or come up with a new creative brief.

Brett (47:30.853)

Yeah, yeah.

Eddie (47:32.66)

I think that's the last thing we want to see marketers do. We want to see. For sure, for sure.

Brett (47:36.452)

And oftentimes what they spend a lot of time doing doing when they're putting together their MMM their the markets modeling analyses, right? They're a lot of it is the data wrangling, which we've talked a lot about of bringing data from multiple places, which seems like we've been doing that for fifty, sixty years, right? And one of our old our our guests was like that was the IBM model that they invent that they like data move data from A to B, right? and and then

Eddie (47:57.832)

Yeah, yeah.

Brett (48:03.493)

d derive insight, but that process could take years or months, t you know, in in you know, the it's not really the speed of business.

Rio Longacre (48:08.411)

Well, yeah, Brett, everything in marketing, I always thought that was so crazy. Like you work with these marketing departments that we might have dozens or some even hundreds of people, right? And like there there's so much that's going on between the planning, the execution, the measurement, right? The analysis of it, the optimization of it.

even the creative elements, right? I mean, I you you and a lot of these organizations were typically very process poor, despite having a lot of processes, right? Just things weren't very documented. So I mean I as a consultant, we'd be called in a lot, hey, can you just document these down to level two? Let me be level three, right? And then create instruction manuals to go along with this, codify that in an SOP, which is then agreed upon and then send this to everyone. It just absolute I mean you needed it because otherwise

Brett (48:34.63)

Yeah.

Rio Longacre (48:53.361)

There were so many moving parts, the risk of things breaking down or not going well, or creating even again, privacy, compliance, brand, brand risk was so high. You had to do it. But I almost think like with AI, it's kind of blown that whole thing apart. You really don't need a lot. I mean, you still need process, but you don't need nearly as much. A lot of it can be automated. I mean, that's why these tools like work workfront were created to automate this kind of mess of marketing. I think a lot of that.

Brett (49:09.84)

Yeah.

Rio Longacre (49:21.903)

Is starting to disappear and go away, which will be personally, I think it'll be incredibly liberating for marketers. I'm I already, I mean, for me and my job it already is, but I think for a lot of marketers I'm talking to is letting them focus on the things they want, they want to. But in order to do that right, going back to the context layer and going back to the data, you need the data foundation. That's gotta be in good shape. It's gotta be correct and protected, right? As well as consented, right?

Brett (49:45.777)

Yeah.

Rio Longacre (49:47.365)

And then then that that context layer, it all r really comes down to building that. If you can't build if you don't build that correctly, you don't take the time, I mean you're really coming full circle at it, then none of this will work.

Brett (50:01.423)

I th I think he's in full agreement with you. I was just ho I was hoping that the storm didn't freeze you for a second there. He's couldn't have said it better.

Rio Longacre (50:03.739)

Yeah.

Eddie (50:04.852)

I was anticipating what two through six were on the the Brat House laundry list of everything on top of mine.

Rio Longacre (50:08.281)

Yeah. Yeah.

Brett (50:17.911)

yeah yeah the yeah the

Rio Longacre (50:19.143)

All right. What what was number two house?

Eddie (50:24.851)

Yeah.

Brett (50:24.929)

no, number two is the Federated Data Ecosystems, right? And then we t we we've talked about headlists, we've talked about sort of the the the sort of do you need UIs and sort of the SaaS world, right? Those were those were kind of the top three, right?

Rio Longacre (50:27.141)

Okay.

Eddie (50:27.741)

Rio Longacre (50:38.203)

Yeah. Well, but I do think it's funny that I've seen a lot of pushback on the whole death of SaaS thing recently. which is I mean, SaaS stocks got pounded. I they got pounded not because their revenues really went down. They got pounded because people weren't sure about their future outlook, right? Will they continue to grow at eighteen to thirty percent a year? I mean, when when I mean, no one's suggesting Salesforce will not be here in five years, but I think what people are suggesting is their their model where they sell seats and licenses and

overages for this, like that's gonna be challenged and they may not be able to charge as much. but then again, people have been saying Google search monopoly is gonna be challenged and they grew 30% so far this year. So we're in a weird time where people suspect a lot is changing, but and things are changing. The way people work is changing, but these SaaS companies are not seeing their revenues go down. And we I think people

Like web traffic's definitely being impacted, but I don't think people's time spent in platform has dramatically been impacted or really impacted at all so far. I mean, that's just been my observation. I don't know, Eddie or Brett, if you agree or disagree.

Eddie (51:49.377)

I would say this, there is an increasingly widening maturity curve on this. What is clear to me and maybe something we don't talk enough about as an industry is the idea that everyone is culturally ready for a blinking cursor into ask a question as their only mode of interacting with data as a business user. It's really effective. It has cut a lot of

waste out of our system when it comes to BI dashboards and things like that that don't get used. It is not yet a replacement for where many organizations are in terms of cultural leverage of understanding information. Part of the...

Brett (52:22.308)

Yep. Incidence that's

Brett (52:33.806)

Yeah. There are still times when you gotta pull up your Excel spreadsheet and do some analysis because your your conversation with Claude or otherwise, to get that data, or whatever the conversational interface you're using is not quite there, right? There's still a need for doing the doing the math yourself.

Rio Longacre (52:47.803)

You y you know it's funny too, Brett? I yeah, well some things gotta you gotta still use old applications for, right? But but I also think it's interesting too. Like I I find that I'm talking to the interface as much as I'm writing these days. I love it. I mean, I actually think it's funny, I think series just series just the most awful voice to text. It's absolute garbage, right? It's never improved. I don't know how Apple all that money can screw it up so badly, but if you open up the the

Brett (52:54.756)

Yeah.

Eddie (53:03.348)

Same, same.

Brett (53:05.262)

Yeah.

Rio Longacre (53:15.851)

the chat GPT app on an iOS. Good lord, it is incredible. It catches everything. You can I can sometimes talk for 15, 20 minutes. And it's like it's it's a it gets everything. It's just, it's I and then it interprets it and understands. It knows what I like it it even if it misspells things, when it interprets it, it'll catch that and it'll like it'll correct it. It's gotten so damn good. So I almost think that the chat more becomes an actual chat.

as as much, maybe more so than typing, but and but we don't really know what the the future form looks like. And we've talked about that a in this pod Eddie about like not really knowing what it may not be a prompt, maybe a pr a combination of things. But I also think too that like Brett, your point is valid. People are still using you still need to use some of these legacy tools. That hasn't gone away. People are still in dashboards and we're in a transition period. I don't think anyone knows what it looks to and I'll give you one more example before I I I pass over to

I talked to Sarah Robertson, who's the I think she's the CTO at Experian and and they told me she product officer, that's right. Yeah, and she yeah. And she told me, I thought this was fascinating, that they created MCP servers expecting lots of usage. There's been almost none. so like they it's not like you build it, they come. Maybe they will come over time, but as of now they've deplo they're they've raced ahead, built this incredible technology, expecting this agentic revolution of

Brett (54:17.53)

Yep. Chief product officer. Chief product officer. We yeah, we talked to her at Cannes.

Brett (54:29.199)

Yeah.

Rio Longacre (54:40.657)

their partners and customers interfa interacting with their their data products and services agentically hasn't happened yet.

Brett (54:49.22)

Yeah, no, and that's something that I think we've talked a lot about. And it's we we called it the the tr the the trough of disillusionment with AI. in in one of our last podcasts with with Kyle Kyle Chick, right? Is is that there's this sort of b board mandated like we've got to adopt AI and the CIOs have got these incredibly packed schedules, like delivery schedules, right, that might go twelve, fourteen, sixteen months out and they're trying to force AI into into the mix.

And they they you know with with no real strategic direction of of how are we doing this, how are we governing it, they realize it's gonna be a lot of work. And so they just throw up their hands and install Copilot. and just say everybody's got access to an LLM. Yeah, yeah. And and and then you ask a question, hey, c you know, anybody on a on any team, let's say the CEMO, can you can you ask Copilot to tell me, you know, the quarterly revenue numbers or a CAC numbers or LTV numbers?

Rio Longacre (55:32.017)

Then everyone gets disillusioned very quickly, yeah.

Brett (55:46.679)

And the system can't answer those questions. And so it b it begs the question is like that's not a true AI native sort of, you know, where where there's data congruence across the organization where people have, you know, access to kind of common information. That's not what that solution is. And and so there's there's a lot of people sort of kind of put on the brakes and sort of taking it slowly. Are you seeing the same thing on your side of the world, Eddie? For a minute.

Eddie (56:12.692)

I mean, you just gave the commercial, Brett, for why a context layer, right? Why an ability to connect the opportunity to answer those questions is missed in some cases. And so, yes, the answer is yes, we see those symptoms. I think the opportunity and the inflection here is, okay, well, what do I actually need to do in terms of the real work up front?

Rio Longacre (56:16.5)

Ha ha

Brett (56:28.398)

Yeah.

Brett (56:42.885)

Yeah.

Eddie (56:42.932)

to set up my foundation as Rio put it to get there. The tool itself, the model itself isn't going to do it. It's the connectivity of the data and then the ability to get after it.

Brett (56:53.178)

Yeah. It all it all falls back on the on the infrastructure. And the infrastructure is your data layer and your context layer on how you govern and and mo and manage this stuff. And if that's broken, none of this is gonna work. Is

Rio Longacre (57:00.337)

Yeah.

Rio Longacre (57:05.639)

I also think too, like the reliance on these large language models for the most complex and difficult tests for sure. But I mean s people I'm hearing a lot of enterprise clients talk about whether it's running small language models kind of on on the edge, right? For for quicker things. I mean, I think experimentation's starting there, or also maybe even even running some open source things to r to cut some of these incredible, like really high costs for AI, right? As well as to your point about IP protecting IP.

be able to install and run things locally, train them on your own IP and not worry about training the the frontier models so they can let your competitors in on your on your on your your key insights about how you run your business. I can certainly see the the wisdom of doing that. And I I think that'll only continue.

Eddie (57:51.006)

Yeah, another word that he kind of reminded me that's coming up more and more of Rio is tokenomics here.

Brett (57:57.582)

Yeah, yeah. T and whether or not that's a a sound policy or an approach to to pricing strategy, right?

Eddie (58:04.456)

Well, and even like, if I'm going to do some really deep, hard thinking research that I'm going to spin up a bunch of agents and get a whole lot of information. Yeah, that's probably a case that I, as a marketer, want to use the Cadillac version of the model. But there's a case where I'm proofreading, I'm copy cleaning, I'm rewriting or coming up with versioned alternatives for a piece of content.

Brett (58:21.837)

Yeah.

Eddie (58:34.056)

I actually don't need that current model. We have gotten so far in AI that a model that's two, three, four versions old probably can do just as good a job at a dramatic fraction of the cost. So yet another.

Rio Longacre (58:34.897)

You don't need the you yeah, you don't need the current model, no.

Brett (58:48.813)

Yeah. Or an open source model that's cheaper for for really quick, easy tasks. And that's that's something that should be governed and automated within the AI environments that organizations are building internally, right? They should automatically default, right? Just to let me finish this point. They should automatically default to based on the type of task, which which I talked to a ton of I AI companies were implementing one at at high signals that does this, right? It automatically defaults to okay, we we've we've categorized the type of task or the type of request.

Rio Longacre (58:57.339)

Or yeah, or or Brett, you think about it like if you're like

Yeah, it should.

Brett (59:17.965)

And we're gonna use the model that's appropriate for that to reduce for cost efficiencies, right? Tokenization, yeah.

Eddie (59:23.582)

Thank

Rio Longacre (59:24.391)

Yeah. Or also like let's say the if there's sensitive IP, right? If you're a drug company, a pharmaceutical company, you're doing your R and D, there's no way you're gonna want your your research into your new molecules to be used by one of the frontier models, right? I mean, the ch the risk of that being grabbed by a competitor or being recommended to a competitor some ways just I mean that there's a such so much money that goes into these things. You'd never want to do that. So you want to run it locally, right? And then and and

Eddie, to your point too. I and Brett, I like that. You need some kind of switchboard, right? That's gonna be able to decide, okay, this task came in. This is a tough one. We're gonna need the most current model. This one, you know, it's just but just three models ago is gonna be good enough. You know, or the loc the local well, we have the the open source model we have running locally that we train some of their stuff, it's gonna be good enough for maybe seventy eighty percent of all of the requests coming in. We'll cut down our tokenomics, to use your phrase, Eddie, we'll cut down in overall costs.

Eddie (01:00:18.526)

Yeah, you got it.

Brett (01:00:20.035)

Yeah. So well this has been a fascinating conversation and and thanks Eddie for you know, I know it's it's it's probably supplements your job, but it's not your day job to do this kind of research. Right. So the article if if if if folks haven't got it go to go check it out on on Snowflake dot com. and you know, certainly interesting to to to hear this kind of stuff and

Rio Longacre (01:00:23.322)

It has

Rio Longacre (01:00:39.43)

Yeah, we'll put it in the show notes. It was it was a cool article. That that's really what spurred this conversation. So thanks for thanks for sharing that and and coming.

Brett (01:00:43.907)

Yeah. And and how br how brands should think about it, you know, their data, their context layer, and also their legal agreements with their vendors, right? And it really does open up the topics of you need to sovereign AI is an important thing. And it's kind of a fancy term, but it's like this should be operating owned and operated within your own environment, right? And and and data movement is super risky for a number of reasons, right? You know, brand differentiation.

IP theft, all this sort of stuff that you need to contractually lock down, right, in the beginning of this process.

Eddie (01:01:20.306)

Yeah, and it's like, hey, we have a lot of folks that are well equipped in the legal space to review these documents here. But there is becoming an increasingly important case to be made that folks in the business that are important for things like competitive differentiation and how we distinguish ourselves from one brand to another should have familiarity that there are exposure points out here, which

let's say don't violate a legal standard, but might really be risky strategically. And I think again, that's the point of advocacy and empowerment that at its core, the article was trying to get after. How do you on the business side as a marketer help understand what's happening behind the scenes in some tech? What's better? What's risky? And how do I, again, from a business strategy perspective, really drive forward

Brett (01:01:51.641)

Yeah.

Yeah.

Brett (01:02:01.294)

Yeah.

Eddie (01:02:16.348)

what I want to do as a marketer in a world that needs differentiation. mean, that's why we exist as marketers.

Brett (01:02:21.827)

Yeah. It's a it's a call for a sort of education, right? And there might be some transformation that certain folks have to go through either to to learn more, to become subject matter experts, to gain more technical expertise, to understand what's happening in the ecosystem, and then how it applies to their business, their category, their industry. yeah, and and oftentimes some of that stuff is missing. They might not have the ecosystem knowledge, they might not be up on the on the latest tech.

Eddie (01:02:26.451)

Yeah, absolutely.

Eddie (01:02:41.737)

You got it.

Brett (01:02:49.453)

You know, they probably understand their category. You understood financial services at Bank of America, but you were one of those people that did extra homework and were w you know, you would go and do extra research and get involved in some of the technical machinery behind this. And it's important. And I think I think marketers that's a demand that marketers have to meet, I think, in in this sort of new age, right? It's

Eddie (01:03:13.052)

Especially when it comes to AI, that's it. The skill set is changing. It's not just, I've got a coding agent, I can vibe code, I can create a campaign with natural language. What are the implications of what this means to the core of my brand strategy? I think you hit it really well, Brett.

Brett (01:03:15.96)

Yeah.

Brett (01:03:30.969)

That's awesome. So I think time to move to quick hits. I know Eddie, you've got a a meeting come up coming up in fifteen minutes. so you you wanna start, Rio?

Rio Longacre (01:03:33.67)

Let's do it.

Eddie (01:03:35.784)

Whoop whoop.

Rio Longacre (01:03:41.67)

All right, cool. let's let's do a non technical one to start out with, Eddie, just for fun. Okay. Do you prefer the mountains or the beach?

Eddie (01:03:50.014)

Ooh, a beach guy. I'm a beach guy. Can't you see me? Long walks, right on the shoreline. Yeah.

Brett (01:03:50.283)

That's gr that's

Rio Longacre (01:03:51.982)

A beach guy. All right.

Rio Longacre (01:03:58.374)

Nice.

Brett (01:03:59.588)

Yeah, so so what's one that that's a good point. I I so I think we all have to answer this question. I think Rhea, are you what are you? You live in the mountains or near the mountains.

Rio Longacre (01:04:05.488)

Funny, I live in the mountains, but I am more of a beach guy. I mean, if I I was telling my we went to Costa Rica, my wife and I first for spring break with with our daughter, and it was like just sitting in a beach, being able to wear shorts, like a shirt like this, or maybe no shirt sometimes. We're I I don't think I wore shoes for a week, right? It was it was I was like drinking like drinks out of coconuts. I was thinking I could live like this the rest of my life. So I'm a beach guy.

Brett (01:04:26.915)

Yeah, it's it's hard. It depends on the mood you're in. I'm gonna say I'm gonna say mountain. I think I'm more of a mountain guy for for winter and summer, but I do love the beach as well. So so back to quick hits. What's one AI governance prac practice every enterprise should implement before deploying AI agents?

Eddie (01:04:26.932)

100%.

Eddie (01:04:44.99)

Define your terms, define what they mean, write them down and put them in a place where any agent that you want to have a real reflection of your strategy can consume. That's it. Be really clear, verbose, get it down in a way that truly reflects your organization and let the agents have at it.

Rio Longacre (01:05:08.568)

Are we in an AI bubble and if so, is it starting to pop?

Eddie (01:05:12.148)

I don't think we're in an AI bubble. I have good reason to think that, no, there is still such early innings from where we are at AI right now. Every day there's becoming more and more tools, strategies, practices, techniques to get after it, as well as opportunities to become more efficient in what we found out.

was available at the early days of AI. We've talked about a number of things in here and I'm going faster than a quick hit, this is a longer than a quick hit, I should say. This is one that I think has tremendous runway for something that is still very, very early for us.

Brett (01:05:53.337)

Yeah. What's the single biggest mistake enterprises, par partly tied to your research, makes about intellectual property in sort of this new AI era?

Eddie (01:06:04.658)

Yeah, it's simple. Don't confuse anonymization with protection. So stripping out names, redacting certain pieces of information, they do one thing for privacy. What they don't do is hide the patterns in training a model that your competitor could also be using.

Rio Longacre (01:06:25.85)

Eddie, I love that. A lot of these privacy laws actually don't like you you are still liable as an organization, even if things are hashed. A lot of people don't realize that. They think, it's enough, right? So totally a hundred percent. Love that. All right. Last one here. Complete the sentence for us. The companies that win the AI era won't necessarily have the smartest models. They'll have

Eddie (01:06:39.316)

For sure, for sure.

Eddie (01:06:53.224)

the deepest and best governed context and a commitment to not give it away.

Brett (01:06:59.215)

There we go. That's that's that's the theme. That's the thesis. You're your your your perfect rap, mic drop. Thanks, Eddie. that was awesome. and for everybody that made it this far, visit us at wdebdeb dot signal and noise.ai. Also subscribe to our newsletter. You'll see it when you come to our site. We're gonna be posting social media. We're we're we're building our subscription base. and the newsletter's got tons of good content editorial, video, podcast, Eddie, and the rest of it. And not not

Rio Longacre (01:07:05.891)

it's a mic drop.

Brett (01:07:28.011)

Eddie of the Iron Maiden fame, but Eddie Drake. and f visit us on TikTok, Instagram. I know I had to I had to drop Iron Maiden at one point or another in one of these episodes. and YouTube, Spotify, Apple Podcasts. We are going everywhere with both short and long form content. So thanks everybody for joining us and thanks Eddie, it was a it was a total pleasure.

Rio Longacre (01:07:33.348)

He waited the whole episode to drop that one, Eddie.

Rio Longacre (01:07:50.214)

Thank you.

Eddie (01:07:51.496)

This was awesome. Thank you guys so much. Love it.