DJ Patil | An Update From the Front Lines of AI - A Perspective From Spock on the Bridge

0:00The CMO Confidential Podcast is a proud member of the I Hear Everything Podcast Network. Looking to launch or scale your podcast? I Hear Everything delivers podcast production, growth, and monetization solutions that transform your words into profit. Ready to give your brand a voice? Then visit IHearEverything.com. Welcome to CMO Confidential, the podcast that takes you inside the drama, decisions, and choices that go with being the head of marketing. Hosted by five-time CMO Mike Linton. Welcome marketers, advertisers, and those who love them to Chief Marketing Officer Confidential. CMO Confidential is a program that takes you inside the drama, the decisions, and the politics that go with being the head of marketing at any company in what is one of the most scrutinized jobs in the executive suite.

0:54I'm Mike Linton, the former CMO of Best Buy, eBay, Farmers Insurance, and Ancestry.com here today with my guest, DJ Patel. Today's topic, an update from the front lines of AI, a perspective from Spock on the bridge. Now, DJ held leadership positions at eBay and LinkedIn and served as the chief data scientist for the United States of America. He's a senior fellow at Berkeley and also an advisor and investor. Full disclosure, we worked together at eBay where DJ was, and I really mean this in Star Trek-ing terms, the Spock on the bridge for the company. This is his third time on the show, and today we are going to dive into AI from that perspective, and he's going to do his best Star Trek impersonation.

1:48Welcome back, DJ. Hey, thanks for having me back. I feel like now I get to be, do I get to call a regular? Is that like how it works? You can, I think you can if you want. Yeah, that's totally good. All right. First question, DJ, where are we really in the AI adoption curve? Yeah. So I think the best way to call it is you ever like make a cake or do anything and it's got crazy lumps and you're just like, we're lumpy. It's like lumpy in a ridiculous way. There's certain areas where AI is, it's knocking it out of the park. And there's other areas where you're like, eh, kind of there. The teams that are adopting it look very different.

2:38There's different areas. There's all sorts of things. Maybe here's the best way to think about it. So the first time you use one of these AI systems, you're like, wow, oh my gosh, this is amazing. Look at this. And then like the third, fourth time, fifth time, you're doing the same exact thing. You're like, could it be a little bit better? Could it just actually be a little more helpful? You want this thing to actually work for you? That's kind of where we are on this. But then you flip it around and you kind of look at self-driving cars and automation on that side. and you go, wow, Waymo keeps expanding its footprint.

3:16You're seeing lots of great stuff happening there. You're seeing the kind of big warehousing, like Amazon-style warehousing, take advantage of robotics and AI. And that's working wonders too. And then in national security, you're seeing it really work well in battlefield conditions like in Ukraine. Well, I want to go back to the Ukraine thing in a minute. it, but you described this as cake in its batter form, where you've poured it into the, correct me if I'm wrong in this, you poured it into your baking dish, and you're not sure what kind of cake it is yet. Is that a fair way to look at it? Yeah, I think that is.

3:58One of the things that I think is really essential is, and here's the framing that I use right now is we're in this kind of very early phase of the technology and the applications where the people who are what we might call native the the ai native people they haven't actually graduated yet they're like sophomores in college and the way to think about this is you know mike you and i are not ai or or we're not going to be ai native because we're not that age we weren't even mobile native we weren't cloud native you're killing me here dj but well we weren't we weren't even desktop native right like you know like we went to labs and those things but the thing that i think is essential for people to know out there is that you have to go you we had to be fluent we had to learn how to be compute fluent we had to learn how to be mobile fluent cloud fluent and right now what you're seeing is you have a lot of teams that are working to be ai fluent and then there's going to be a wave as those people come in and those that new guard that's coming in those are going to be the ai native people and many of the ways same ways you know like you know kind of kids people have kids like they like you know my kids at least they they they take my phone they're like you're too slow at this and i'm like hey did you know i i actually built that I made that.

5:23They're like, shows what you know. You can make the technology, but it doesn't mean you know how to use the technology, is what they constantly remind me. And it's a good lesson, because I think what we see right now in this lumpiness is people who are adverse to even becoming AI fluent. And we were talking about this a little bit before this, you see this in classroom some professors are like no ai allowed at all and others are like use it like you are mike like if you're teaching they're like use it all you want yolo can't help it it's not going to help you get the answer but it can help you think and create the platform exactly exactly i i think this thing i have to go back because i can't let the how ai is affecting the battlefield between Ukraine and Russia just as a sidebar.

6:17Tell us what's going on over there. Yeah. So you look at, take a look at the, the, the autonomous weapon systems. These are, think of these as drones. And these are what are referred to as kinetic strike capabilities, where you have a drone, it kind of flies out. Maybe it, it goes and hits something. Maybe it tries to deploy a payload, shoot something, drop something, those type of things. Recently, you've heard about it maybe you heard about it in the news around uh uh the the russian fleet the tankers the ghost fleet kind of that's been moving oil secretly those have been being taken out by by sea drones well when you kind of launch these things you know some of them may have some human assisted type technology but they're more and more say oh that's the target let me figure out how to how to do it because um if if you're if you're connected to it as a human from somewhere somebody can intercept that feed hijack it and you know change everything up to you know counter it's a counter attack and so you're here what you want to say is like look that's the target you've got it yeah go do your terminator thing exactly and so you're seeing and and it's it's it's it's less about tech innovation it's about forced necessity of survival and and this technology as it's kind of moved along and and we've seen this over time is like you can use these huge models like you know the open ai the anthropics the googles etc or you can use a lot of times these small models these these miniature models many of which are referred to as open source models these are the things that you hear out of china deep seek or uh llama which is uh has been facebook's uh version so i want to flip this over now to the all the circular investing going on you know you got the chip makers you got the models you got uh i feel like you need a backdrop with a flow chart i you know if i could i would maybe next show and maybe next when you come in you know, later this year.

8:30How do, how do you think about that? You've got all this energy needs, you've got stuff. I think you've got at least seven big players in this. How are you thinking about this and how is everyone going to eventually make this payout or won't they all do it? I mean, one thing is just to recap what a crazy, I don't know, six months it's been in terms of flip-flopping of whose model is the best between open AI, Anthropic, and even Google's Gemini. In the last three months, we had the Math Olympiad, this famous exam that high school students take. The year before this last Olympiad, only four out of the six problems had been solved by AI systems and it took them a long time computationally to do it four uh five out of the six this year were solved uh so that qualifies you as a gold medal so these these models effectively got a gold medal a number of students did get six out of six we're expecting all six out of six to be solved next year yeah but we'll also but what's happened in this small short time period is one of the hardest math proofs that's out there, the Erdos problem, one of his famous conjectures, was just solved by one of the models out there.

9:58And so it's introduced a whole new way of thinking about solving some of the hardest math problems that have ever been created. And then you kind of go, okay, that's this high level big stuff, right? And then you kind of go, okay, well, we've got vibe coding that's happened, we now have all these other kind of like ways that people are starting to figure this out and so i might separate i think everyone is really focused on the big giant players and that there's right to do so because of the volume of of of the amount of dollars that are going into these things and you're right there's like you can't keep track it looks like if those are remember reading all of her twist you're like wait who's related to who and what's happening kind of feels like that like and you're just like wait are they investors are they not are they frenemies now like exactly he's like game of thrones and very very ask uh and then like the talent it keeps moving you're like wait that person like alex was over at scale now he's at meta and you're like like everyone's kind of moving around uh um and apple's making moves everyone's making moves so a lot's taking there what i find more interesting is the applications that people are starting to really leverage for these these technologies so specifically like we're starting to see where people are able to say hey this is actually adding lots of value on the legal front uh inside inside the corporation we're actually able to use a product might be harvey might be one of the many others that are out there we're actually able to do this you take a look at figma which is on the design side something that you know i've been really fortunate to be involved in for a long time um you take a look at figma and now you can do a lot of the design components by you know just talking to the system and saying what you want you are you going to get something perfect no but boy remember the days when it took you like weeks and weeks to talk to the designer so dj there's The question beneath all that is, we can talk about Amari's law, but also there's all this thing like, will this pay out or not?

12:06One day, Wall Street's up. The next day, Wall Street's in this flop sweat panic that we've overinvested. How, you know, Amari's law says you underestimate it now and you overestimate the short term and underestimate the long term. How should people think about this in terms of ROI and these investments paying off? And, you know, you're an investor. How should they be thinking about it? So, you know, I think that public markets are a super confusing way to look at the world. And like the valuation of these things and what is the actual like the actual ROI on this. I think there's absolutely going to be a small set of massive, big companies that are able to take advantage of AI and disproportionately leverage them.

12:59I think that's going to be the case. You know, if you look at Google and the infrastructure, they have all those things. NVIDIA, the chip designs, all this stuff. Amazon. They're sitting not just on data. You can think of there's like a Maslow's hierarchy. It's like in life, you have food, clean, water, shelter, and all. up the same way exists for AI power. You need power. You need data. You need water, um, to basically build data centers. You need all that. Keep the data centers cool. Yeah. Right. But then you need to center. So drink, but they do stay cool. They just stay cool. And then you kind of go up the stack.

13:36And one of the things that you people don't realize is like, you need the model. You need lots more data that start to come in and you need user feedback loops. You need user feedback loops to keep iterating this, creating different data, interaction points, all of those things. That's the one to watch is like where the feedback loops. So, but here's the way I would encourage everyone out there to think about this. How are, like at the balance sheet level, at the board level, are we spending pennies, nickels, dimes, or dollars? and when do we start seeing the savings so right now on the balance sheet for a corporation you're spending pennies and you'll be spending pennies relatively through 2026 you'll start making and my my very dangerous things to do make forecasts but here we are kicking out 2026 you know the beginning is i think we start to make pennies towards the end of 2026 maybe into 2027 and Now you're talking like the average corporate user, the big corporate.

14:45Like, so you're still only like on your, on your revenue lines, you're not seeing big dollars through 2026. You're seeing pennies. You're still seeing like, Hey, look, I saved, you know, if you're a United healthcare, you're like, look, I saved billions of dollars, but you're like, what's billions on this, on this kind of, or something. Yeah. Yeah. You're like, maybe we'll call that nickels. Maybe we'll call that a dime. You start to really start seeing, and your spend moves from pennies to maybe a nickel in there towards the end of 2026. You really start to see the impacts in 2027. Why does it take so long?

15:27It's not just only about the technology getting there. The cultural component of this is one of the big... We did a whole show on this that said that is the biggest blocker. Yeah. With the CEO of Typeface saying it's culture and it's also and I want to flip over. There's all the hype around it. And also you have a lot of boards probably going, oh, my God, what are we doing in AI? What are we doing in AI? And a lot of our user or a lot of people that have come on our show have said, look, the board shouldn't be doing that. it should be sorting out use cases and getting traction here. It shouldn't.

16:05And also leaders can't delegate this. Tell us what you think about that and then weave the culture story into it. Yeah. So let me give you first a framework why I think this is, why this is, it sort of epitomizes how bad things are going to be. So I think of these things in time scales. So, So, you know, you think about product releases. A major company has one big product launch every year. Like you get your new iPhone every year. You hear about the launch and you're like, wow, okay, this is what I get. 1x kind of for a major company. The lesser companies, it's like 0.2x, maybe every, or 0.5x every other year.

16:48So startups typically release three times a year. So they're 3x. So 1x versus 3x puts you on exponential. AI companies are launching 7 to 10x, 7 to 10x major releases a year. So that puts them on this crazy scale. Culture changes roughly once every 10 years for a company, if you're lucky, if you're able to transform it. So that's 0.1x. Policy changes every 20 to 30 years. So that's a 0.02x. So you've got a 0.02x versus a 10x in technology change. So out of that, there's three things that can happen. You can have a hard landing, you can have a bumpy landing, soft landing, or you can leverage it on opportunities.

17:42And so to your point, Mike, I think what we see is executives think of this like traditional software. Let's make a big investment. Let's go do this stuff. It should be lots of experimentation, lots of play to allow your teams to learn and codify with this technology. Allow them to become AI fluent, to leverage it, figure out what are those things work and then go. The infrastructure for these bets, the infrastructure, the raw infrastructure to allow all this stuff to work is super in the infancy. And it's likely to only get hardened and codified over a year. So you better be like, and Mike, you and I live this, like we were trying to do all this stuff with data and you're like, well, that data, you know, that data warehouse style technology doesn't work.

18:30Got to rip it out, put in a new one and you do it again and again to stay competitive. These companies are not ready for that level. But this says you have to really have a board, in particular a board in the C-suite, that is cognizant of everything you just said, or they will either pass this down to someone to do it or demand someone do it that isn't them. if your executive team doesn't have carved out time to be doing the ai playing with the ai themselves and they are just delegating it they're going to mess up this is that facebook moment where they you know mark we know we did this at linkedin too is like if you tell the exec team you're like you can't use like you can't use the product on the laptop you have to be only mobile mobile.

19:24If you don't do that, you're out. Like we can't have you on this team. The board has to be this way too. A lot of the board things that people groups that I interact with, and I see the same thing we saw during the data science era, the same thing we saw during the web era, 1.0 to 2.0. If they're not using the tech, they don't get it. They don't get it, but they're getting all their information out of either business posts that they're reading or talking to their friends, which is not going to be enough. You know, the, all the consulting companies are saying they can help you with this. How do you feel about that?

20:06I'm going to get myself in trouble. Well, we'd love that on the show, you know, be tasteful, but you know, because what you just said kind of goes counter to saying, I'm going to have a consultant come in and tell me how to do this because if I don't know how to do it myself, I'm not going to be able actually to lead it. Yeah. So tell me about the consulting companies and then, you know, how you feel about it. And then we can talk about if you still have to hire one, what do you look for? So here's what at least I'm seeing out there is you see a lot of people who are trying to do the traditional handed to a bunch of consultants, management consultants who historically have been really effective and really helpful to the companies.

20:48and doing this. They're not here because the stuff has to be integrated. It's like changing your DNA. You can't change your DNA as a company by, you know, we saw this with mobile where suddenly they're like, oh, we should have a mobile team. And you're like, no, everyone has to have mobile DNA. You don't, there's not like a team that you hand it over to. The same thing as like on marketing teams, like if you have this separate, well, let's call on the social media team. You're like uh well this is probably going to go bad real fast so so we see that the other thing is because the tech is so changing so quickly what does it mean to be a consultant who understands and utilizes technology the shift that we're seeing is more what people refer to as a forward deployed engineer or somebody from the vendor so for example you work with palantir you work with open ai you work with anthropic pick fill in your name of sort of cutting edge technology shop you know you spend so much they are going to give you the team they are going to give you their team of super you know talented young spirited people that before you were like oh right you would be a you'd be at like one of these big five shops like kind of thing they look the same they act the same but they are They're way more savvy on the technology.

22:11And because they are from the DNA of the company, they're not like a system integrator. They have a direct line. They're directly connected in. And they're like, hey, here's how you use our technology. We see this all the time in healthcare right now where you work with these players and then you get the management consultant comes in and they're learning too. They're figuring it out also. The tech is so early. Well, and also, if it's evolving, they're not going to leave anything behind when they finally go. So tell us about, you know, there's all the job cuts, Amazon, Matt, Goldman Sachs, there's so many job cuts.

22:52How should people feel about that? And is this just the beginning of the job cuts? All right. I'm going to get myself in trouble again, but hot takes. So we'll see how much what people come at me with. So my take is the following. At the junior levels, the pullback on hiring that we're hearing about, like kids aren't getting internships, new job hiring is not there, all of those type things. There's been a lot of talk of, oh this is ai it is ai for like the amazons and other places that have invested in this for decades already like you know they're from year 15 plus on this journey they're they those ones you are seeing some of the advantages for everybody else it's just economic pullback it's a bigger sign of economic weakness and everyone's like hey we're not it's not in our hiring plan Talk to lots of companies out there and they're just saying, hey, we're not going to take interns this year because we're not planning to hire.

24:01We're going to stay flat or go slightly down next year because we just overhired our efficiencies. For the later stage employees, call it 10 years out or something like that to more, that's just streamlining with the hope that AI is going to take those efficiencies out there. The place where we see AI having big impacts right now is more on just straight up automation and a lot of things. A place that is counter to this is healthcare. So in healthcare, and this has been driven by the last few quarters of job growth, the job growth in the United States is driven by healthcare. That's back office work.

24:45But I would argue that is not good jobs. and those jobs are likely to disappear because that's the stuff that you hear about that costs a lot like claim denials oh yeah pay passing paperwork all the yachts that nobody likes to do nobody likes to do those jobs like well if they do they're gonna get hired by like uh scrooge and marley's yeah exactly yeah and these are the jobs that i would call stupid boring problems and they're ready, those tech problems are more likely to get evaporated. And that's the form that you're already starting to see disappear elsewhere. And we are seeing these things like on the call centers and other type areas.

25:27The tech is getting decent enough to deploy there. You know, some of the metrics that are there, Brett Taylor's Sierra company already announced that they've surpassed in Q4 of 2025. They've surpassed$100 million in ARR for a company that's been around in a short-term while. But people actually like talking to it. People like working on that system. So, DJ, if I read between the lines there, what I hear you saying is AI may take a lot of jobs in the long run, But a lot of the stuff you're reading about now, it hasn't even started working yet. We're not seeing the real impact yet. The way I would think about a lot of these AI systems today in the corporate world, they're 80 %-ish solutions.

26:16They're good-ish. But they're not ready to really be prime time. Sure, can they solve this ridiculous math problem? Can they do all this amazing stuff? Yes. Can they do real work? Yeah. So they're like autopilot, but they can't fly the plane yet. Exactly. But it is getting there. Right. And it's but like the same way, like you saw this with Waymo, you saw this with other autonomous driving vehicles kind of technologies. It's about a decade for that stuff to really get hardened. So, all right, let's flip this over to investing because we had the two gentlemen that wrote AI first on the show. And they said, look, AI is the biggest prize in the history of capitalism.

27:08How are you investing in the biggest prize in the history of capitalism? So I think that a lot of the big bets are already taken for the big, you know, you kind of think about this as like boulders. and then you've got to put in the smaller rocks and you put in the pebbles and you put in the sand. The big boulders have largely been put in for the big foundation models. I still think there's plenty to be done on more tailored or what we might call small models where they're actually very specific to an application area. And then I think there's a ton that can be done on verticalized solutions. The place where I am really focused on a lot of these things is kind of two-ish, three-ish fold.

27:50One is the analogy I would use, and I know I'm mixing metaphors here, is like, you know, we got this giant muscle, that's the big LLM, but you don't have the ligaments intended. And you can't run without ligaments intended. You have been an analogy machine. We started with cake and now we are to tendons. And in the introduction, we were on the bridge of the enterprise. So no way I use no way I use no AI. So we're working on tendons and ligaments now. Exactly. But like, like the way I think about a lot of times is like, people are like, look at my super cool AI. And I'm like, well, like, you know, like, here's a concrete version.

28:30So you're like, Hey, we're going to have, you know, maybe 20 agents, you know, AI agents doing stuff this year. And I'm like, great. Once you unleash that, then are you going to have 50, 1 ,000, 2 ,000, 5 ,000? Who's going to manage that? What happens when the model changes? How do you do evaluation? How do you think about all these other things? Where are your checkpoints? And all those little things that are required, that's a lot of stuff. You've got to hold the body together focus. Exactly. And you said there were three things. Are there two more? Okay. So that's like the small muscles, ligaments, tendons.

29:14Then I think there's a lot of space in the application area specifically to something like health tech and healthcare. What is the AI? What could AI do? Right now, everyone's focused on how does the AI help the doctor? Could we have Scribe or all these other things? I think that's nice, but there's a ton more where AI can just be extraordinarily additive, net positive. We have areas of the country where there are no care, like there's no physicians, there's no access to care. So what could we do with AI-assisted technologies to help get you faster care, look over things? And this is the sad thing is, you know, in the era of big data and data science, we have really only done one thing for healthcare is make more money for hospital systems and payers uh the insurance companies we have not delivered the care to people so we have tendons and now we have what i'm going to call it a vertical yeah you know and there's a third thing and then the other vertical sort of area which i would say is like you're going to have things like in government uh um you're gonna have a lot of sense whether it's audits it could be national security but the irs could audit absolutely everybody if it wanted to well i'm like why can't i have like why can't i have ai do my taxes i like that would be like would that be like how much time do you spend on this like why can't it just do it for me that seems like a stupid boring problem like those are the things okay so we've we've gone from cake to tendons i wanted i want to write just before we get to our traditional last question write marketers into this story now yeah what should marketers be doing or people that want a career in marketing be thinking given everything you've just said on the show so number one uh the superpower of marketers in my experience they have one power above all else they come from the liberal arts and the the everybody else around ai and all the technology comes from the hard sciences and they have not done a good job of exposing themselves to the liberal arts that means we've lost human connectivity we've lost touch we've lost the ability to inspire the place where marketers i think have the greatest challenge is they've always kind of said oh i don't do math i don't do tack, like you hand it off.

31:47Not all marketers, but yeah. Not all marketers, right? But like, Mike, you're very far from it. You're like, you should have been a data scientist or if you were, you really were and have been always a data scientist. So, but like, you know, like, what does it mean to really, to use this technology as an asset, as a superpower? So some of the things I would do if I was a marketer, I'd be absolutely making sure I use this technology any chance I could get, even if it makes me 20 % slower. Because like you don't become fluent in language by just sort of being like, I'm going to be fast at it. It sucks.

32:27It's a grind. You have to immerse yourself in it. Number one. Number two is it allows you access to technology and coding in a way that we'd never seen. There's this kind of concept called vibe coding where you tell the, yeah, lovable bold, all these things. Every marketer should be doing this. They should be creating their, their things. And then marketers have to also flex into the design space. They should be using Figma, which I'm biased to, or canvas, you know, Canva or anything else where you're actually you're, you're like, you know, we have this notion of full stack developer where you're like, hey you go from ideation all the way to deployment you build everything you you single superpower person same way you you have to be that as a marketer you have to be the you have to be the cmo all the way to the person writing copy i i think that that you're not going to get budget you will not get budget the new marketing teams are going to look like two to three person shops they're they're not you're not going to have an army and you're not going to be have the budget to go get a massive team of consultants the the best people are going to be doing themselves and you gotta you gotta be social media savvy to all the way all the way through and at the same time the the parts i think that that a lot of marketers that i see now is they haven't taken the lessons mike that like that that people like you and and have learned from the the heart school of hard knocks and the traditional way of doing things.

34:06And if you skip over that stuff, you're just going to repeat history. You're just leaving a ton on the floor. So I hear you saying, marketers, get deep, get involved, do it fast. You heard it here first. This brings us to our traditional last question. Funniest story you can tell on the air and or practical advice we haven't talked about yet. You can pick one or both, but you must pick at least one. Oh, yeah. Well, I'll tell one that's AI related that just kind of shows the awesomeness is, is, is, and, and that sort of goofiness simultaneously. So, uh, uh, Waymo just opened up in our neighborhood.

34:44And so I was like, Oh, we're going on date night. We're going to take a Waymo, uh, um, instead of, cause you, cause you're not going that far, but you're going to get a drink. So you always feel bad with the Uber driver. You're like, I, I gotta beat you. Nothing says romance like a Waymo. Exactly. So we're like, like, so we take the Waymo there. And then I'm like, but it dropped us off 18 minutes from where we should have been dropped off or an 18 minute walk because it didn't understand which side of the train tracks were on. But the saving thing is if you put kind of the customer service, it sends you back to the, it takes you back to the right, the right place.

35:21All right. Well, just as a service to our listeners, that is not dating advice from DJ. That's just a story. So thank you, DJ, and thanks to everyone for listening to CMO Confidential. If you like our content, please like, share, and subscribe. New shows are released every Tuesday, and you can find everything on Apple, YouTube, and Spotify, which include Colonel Mustard in the study with the job spec, what your CFO wants to tell you but won't, the AI application layer, the good, the bad, and the ugly. And of course, DJ's earlier shows, one of which is titled, Is AI Like Taking the Red Pill or the Blue Pill?

36:05Hey, all you marketers, stay safe out there. This is Mike Linton signing off for CMO Confidential.