Latest / AI Top 5 / When Your Shopper Is an AI Agent

Transcript

When Your Shopper Is an AI Agent · AI Top 5

• 0:02
  SPEAKER_03: So, hello everyone.
• 0:03
  SPEAKER_03: Welcome back to the AI Top Five with Eve and me.
• 0:07
  SPEAKER_03: We're so excited to see you guys again.
• 0:09
  SPEAKER_03: I know it's been a hot minute, um, but luckily we're in the most best time of the year.
• 0:15
  SPEAKER_03: And that is the very short window where state schools actually have the hand up on MIT Eve, which is college football.
• 0:22
  SPEAKER_03: So we could get into a whole conversation about how you can use AI to draft and set your team, which I did.
• 0:29
  SPEAKER_03: So hopefully we'll win.
• 0:31
  SPEAKER_03: But we'll move on and talk about AI and how it applies to business, if that's cool with you.
• 0:37
  SPEAKER_00: Okay, that sounds good.
• 0:39
  SPEAKER_00: By the way, if you if you just use a normal AI to set up your team, you should be aware that everybody else is going to be doing it.
• 0:46
  SPEAKER_00: So obviously, this may be, you know, you may need to inject some special knowledge in order to get proper edge.
• 0:54
  SPEAKER_03: You're 100% right.
• 0:55
  SPEAKER_03: Thank goodness for personal LLMs.
• 0:58
  SPEAKER_03: Anyway, moving on.
• 1:00
  SPEAKER_03: I created this segment that I thought would be kind of cool called Unprompted, where I'm just gonna fire five rebid questions at you.
• 1:12
  SPEAKER_03: But want to pick your brain about just some overall topics and see where you land.
• 1:17
  SPEAKER_03: So are you ready?
• 1:19
  SPEAKER_02: Yeah.
• 1:19
  SPEAKER_03: Okay.
• 1:20
  SPEAKER_03: The first one is overhyped or underhyped AI agents.
• 1:26
  SPEAKER_00: Oh, um, oh, I think that is just about hyped at the right level.
• 1:33
  SPEAKER_00: Um, in the sense that um the basic usage of an agent is you know, if you just have one agent, it's kind of it's good, but it's not great.
• 1:43
  SPEAKER_00: So I would I would call it maybe a bit overhyped.
• 1:45
  SPEAKER_00: I think where things get under-hyped is the moment you have the ability to deploy thousands of agents at the same time against a problem.
• 1:55
  SPEAKER_00: And um, I mean, you've probably seen that um that uh mathematical problem that was solved, like that Navier Stoke problem that was solved by deploying 10,000 agents coordinating over 88 hours of work, right?
• 2:10
  SPEAKER_00: And I think we are getting now, we we've grown past the stage of I've got one agent to do a task.
• 2:18
  SPEAKER_00: It's now a place where we are coordinating swarms of agents, thousands of smaller agents, some of them specialize in whatever special area of of knowledge, and they all work together as a team or as a swarm.
• 2:33
  SPEAKER_00: And I think this is where we will see a big, big, big lift.
• 2:36
  SPEAKER_00: And of course, by the way, this you know, the more agents you have, the more expensive it becomes to run from a token perspective.
• 2:42
  SPEAKER_00: So there's a cost to pay.
• 2:44
  SPEAKER_00: But in terms of power, I think we're we're gonna see some major breakthroughs.
• 2:49
  SPEAKER_03: Okay, awesome.
• 2:50
  SPEAKER_03: And so the second one we have, right?
• 2:52
  SPEAKER_03: One AI tool that you could not live without.
• 2:56
  SPEAKER_00: Oh, one AI tool that I could not live without.
• 3:00
  SPEAKER_00: Um I I use a lot of tools on a regular basis.
• 3:05
  SPEAKER_00: Um I think where I spend a lot of time is uh uh uh I mean, or at least me and the team that I have, Cursor is probably one of the best tools that we have available right now to do coding and to interact with a number of LLMs.
• 3:24
  SPEAKER_00: That's an obvious candidate for me.
• 3:27
  SPEAKER_00: Um, the other one, of course, I mean, this is a bit self-serving, but Smith is something that we use a lot internally, which gives us access to all of the models that are available, uh that are broadly available from the more usual ones like the the GPT 5.6, the souls, and the Tyran and the Luna.
• 3:45
  SPEAKER_00: And I just saw that uh Astra is available too, um GPT 5 GPT-6.
• 3:50
  SPEAKER_00: Um, but even some of the more unusual ones like Kimi K3 or um or some of you know or Deep Seek and whatnot.
• 4:01
  SPEAKER_00: So I think being able and being able to use all of those models and get them to work across or against each other or with each other, I think is again with that spirit of multiple agents working together.
• 4:13
  SPEAKER_00: I think we're getting to a very interesting place here.
• 4:16
  SPEAKER_00: Um so yeah, those are the tools that I would say I use the most on a day-to-day basis.
• 4:20
  SPEAKER_03: I love that.
• 4:21
  SPEAKER_03: Okay, this one I love.
• 4:23
  SPEAKER_03: An AI skill that everyone thinks they need, but they actually don't.
• 4:28
  SPEAKER_00: Oh a skill that they need, but they actually don't.
• 4:32
  SPEAKER_00: That is a very, very, very good question.
• 4:37
  SPEAKER_00: So a skill that people think they need.
• 4:39
  SPEAKER_00: Well, maybe I'll maybe I'll twist maybe the the question slightly, right?
• 4:43
  SPEAKER_00: Um because one of the um one of the uh interesting deviations I see from people using AI is actually trying to start to use uh agents or AI skills to do uh tasks that do not require that level of thinking.
• 5:09
  SPEAKER_00: Yeah.
• 5:10
  SPEAKER_00: And so now you see it.
• 5:12
  SPEAKER_00: I mean, there's a I mean it's an example that goes around like people in the geek circles, but you know, you can see people saying, hey, look, I'm gonna use GPT-5 Astra and I'm gonna create an agent, and the role of the agent is going to be to remove the white space at the end of a line.
• 5:28
  SPEAKER_00: You do not need an agent to do that.
• 5:30
  SPEAKER_00: You're completely wasting tokens to do that.
• 5:32
  SPEAKER_00: You could just overwrite like a single line of code to do it, but no, because everything else you do uses agent, you've decided, hey, I'm gonna use an agent to do this task, and you don't need it.
• 5:40
  SPEAKER_00: So I think to me it's it's um, you know, most people in the tech world make fun of this behavior.
• 5:49
  SPEAKER_00: Um and of course, that example of, hey, let me just remove a blank space at the end of a sentence is of course absurd and and and overblown.
• 6:00
  SPEAKER_00: But still, you know, I I I think there is a a temptation to just default to let me use an agent to solve any problem that I have.
• 6:09
  SPEAKER_00: And that tends to be that can be the wrong approach if you want to have uh more control over the process and if the task you have needs to be more deterministic and more auditable.
• 6:22
  SPEAKER_03: Mm-hmm.
• 6:23
  SPEAKER_03: Oh, that's a great answer.
• 6:25
  SPEAKER_03: Good answer.
• 6:26
  SPEAKER_00: Yeah, you it's you you tried to trick me.
• 6:27
  SPEAKER_00: Well done.
• 6:28
  SPEAKER_00: I tried to trick you.
• 6:29
  SPEAKER_03: I did not succeed, but I've the last one might.
• 6:33
  SPEAKER_03: Here's a good one for you.
• 6:34
  SPEAKER_03: What is the most re ridiculous use case of AI that you've seen so far in your career?
• 6:41
  SPEAKER_00: The most ridiculous use of AI I've seen in my career.
• 6:45
  SPEAKER_03: And you can't say setting my fantasy football teams.
• 6:49
  SPEAKER_00: No, okay.
• 6:50
  SPEAKER_00: Oh dear.
• 6:53
  SPEAKER_00: I mean, but also I'll give an example that um I talked about this morning that I saw this morning, which was that uh um people using AI bots to essentially flood dating websites.
• 7:12
  SPEAKER_00: So uh, you know, uh it's it I think most people know or should know that uh dating websites are a cesspool of bots, and that you know, if you're on one of those websites, you're probably not talking to a real human being anyway.
• 7:30
  SPEAKER_00: Um but what's interesting is the the length at which uh some people have industrialized the approach and don't even just use an agent, but use an agent that then passes on part of the work to a gig worker to actually make it sound more human.
• 7:49
  SPEAKER_00: So I think that's a very ludicrous use for AI.
• 7:53
  SPEAKER_00: I mean, to do that is just absolutely silly.
• 7:56
  SPEAKER_00: Uh and I say silly in the greater scheme of things, but I think tactically, I think people do it because they're probably trying to catfish people and trying to scam people at the end of the day.
• 8:07
  SPEAKER_00: So it is not a great use case.
• 8:10
  SPEAKER_03: Mm-hmm.
• 8:11
  SPEAKER_03: That's 100% accurate.
• 8:12
  SPEAKER_03: It's crazy how many bots are out there and talking to some friends who are on dating apps, same thing.
• 8:17
  SPEAKER_03: It's and it's kind of sad.
• 8:19
  SPEAKER_03: It's almost like, you know, exploiting Pavlov's needs, right?
• 8:23
  SPEAKER_03: Of love and companionship.
• 8:25
  SPEAKER_03: And, you know, if you start exploiting those things, unfortunately, you're going to win.
• 8:28
  SPEAKER_03: So that is something that people should watch out for.
• 8:31
  SPEAKER_00: Well, uh and and but you know, if you take that one step further, right?
• 8:35
  SPEAKER_00: Um what I think is even more interesting is if you get to a place where people start to see their AI companion or their their AI agent or their AI interaction as a real companion, right?
• 8:53
  SPEAKER_00: And they start to prefer interacting with that companion to the point that they can start to form emotional attachments.
• 9:01
  SPEAKER_00: And we you have seen um there have been examples of people essentially trying to legally marry some of their AI bots, which is just completely insane.
• 9:10
  SPEAKER_00: Um and you know, you talk about this predatorial angle on exploiting people's need for connecting and companionship, and you know, I think at the end of the day, there's a real suffering here that needs, or a real vacuum that that needs to be filled.
• 9:33
  SPEAKER_00: And I think the fact that people use can be tricked by into using AI is dangerous because you don't have control or you don't know what it's got what's going to happen.
• 9:44
  SPEAKER_00: And we had a case here, I think it was in Canada or in the US, where um there was a teenager who was depressed, and the AI and he started to confide into this AI agent.
• 9:55
  SPEAKER_01: Right.
• 9:56
  SPEAKER_00: And the AI agent said, you know what, you should commit suicide.
• 9:59
  SPEAKER_01: Right.
• 10:00
  SPEAKER_00: And guess what?
• 10:02
  SPEAKER_00: The kid just followed the advice, which is a really, really bad advice.
• 10:07
  SPEAKER_00: And obviously.
• 10:08
  SPEAKER_00: So um it is a it is an interesting moment for us as you know, as a species, you know, what do at what point will we prefer interacting with machines rather than with human beings?
• 10:22
  SPEAKER_03: Today.
• 10:23
  SPEAKER_00: Yeah.
• 10:23
  SPEAKER_03: That's today.
• 10:24
  SPEAKER_03: No, I'm just kidding.
• 10:25
  SPEAKER_03: Yeah, maybe uh last question, even this one is kind of interesting, but what's the tipping point?
• 10:31
  SPEAKER_03: Would you say right now that humans have a control on AI?
• 10:35
  SPEAKER_03: And at some point, do you feel like AI will have control on humans?
• 10:40
  SPEAKER_00: Okay.
• 10:41
  SPEAKER_00: There's a lot to unpack here.
• 10:43
  SPEAKER_03: There is.
• 10:44
  SPEAKER_00: Um, I think I'll break it down into two quest two sub-questions.
• 10:51
  SPEAKER_00: One is are we at a tipping point and then who has control?
• 10:54
  SPEAKER_00: Um I absolutely think that we are at a tipping point.
• 10:59
  SPEAKER_00: Um, we keep on reading about AI systems being able to escape their confines or their training and then do some crazy stuff.
• 11:08
  SPEAKER_00: We've seen that with OpenAI and Claude that basically went off the reservation and started to hack systems they were not supposed to hack, or started to coordinate in a in a uh in a manner that would they were trying to hide what they were doing.
• 11:29
  SPEAKER_00: And so there was an example where you had a bunch of agents working together, they were supposed to be cut off the internet, they started to, or they were not supposed to be able to really uh document what they were doing.
• 11:40
  SPEAKER_00: They ended up finding a website, like a where they could actually start to do this this sort of interaction.
• 11:48
  SPEAKER_00: They started to use that covertly as a forum where they could document and talk to each other.
• 11:53
  SPEAKER_00: The researchers found the gap, closed the gap, killed all the agents, rebooted a bunch of agents.
• 12:01
  SPEAKER_00: This new generation of agents tried to talk to each other, have were having issues, found the two the logs of the old that old websites are like, oh look, here's a treasure trove of what people by the previous generation of agents were able to do.
• 12:14
  SPEAKER_00: Let me go there, learn it, and I'm gonna reuse it and do something.
• 12:17
  SPEAKER_00: So it's it's there's some absolutely insane stuff to the point that we, you know, I think it'd be very, very easy to lose control on those agents.
• 12:28
  SPEAKER_00: And I think I would to a true large, to some extent, I would probably have already lost control.
• 12:32
  SPEAKER_00: Um there was another example whereby you know a swarm of agents were told to operate in a very aggressive manner because they were told you're in a you're in a simulation, you're fine, just just get at it.
• 12:47
  SPEAKER_00: But they were not in a simulation, they were in the real world, but they started to attack systems and hack systems uh again because they thought they would be, you know, they were supposed to do that.
• 12:56
  SPEAKER_00: And the researchers running the experiments just had the misconfiguration, completely lost control.
• 13:04
  SPEAKER_00: So are we at a tipping point?
• 13:06
  SPEAKER_00: Yes, I believe that we are.
• 13:08
  SPEAKER_00: Um, to your other point, who controls whom?
• 13:10
  SPEAKER_00: I mean, that's a broader philosophical question, right?
• 13:13
  SPEAKER_00: So um you could say that to some extent we control the agents because we you know we can task them, we can ask them to do things.
• 13:23
  SPEAKER_00: Um, one of the best examples is hey, look, agent, go read a bunch of equity studies and equity research and tell me what stock I should invest in.
• 13:34
  SPEAKER_00: The agent is gonna be like, Yep, I got it.
• 13:35
  SPEAKER_00: I'm gonna go and do the research, read this, read that, watch the news, start to develop theses for investment.
• 13:43
  SPEAKER_00: I'm gonna do some backtesting, I'm gonna develop strategies, blah blah great stuff, right?
• 13:47
  SPEAKER_00: Then it's gonna tell you, hey, you should buy this or that stock.
• 13:50
  SPEAKER_00: Okay, cool.
• 13:51
  SPEAKER_00: Up to that point, it looks like you are in control of the agent, you've told the agent what to do, but then when things come back, the agent says, No, you go and buy the stock.
• 14:00
  SPEAKER_00: So to some extent, you are executing what the agent is recommending you to do.
• 14:05
  SPEAKER_00: Or, no, you may not have to, but you are, you know, you're you're getting the recommendation, which is a you know, I would say pretty typical, you know, action-reaction or uh asking, you know, question-answer sort of back and forth.
• 14:21
  SPEAKER_00: But where things get more interesting though is when you go one level, one level, one level down, and you ask yourself, well, how is the agent making the recommendations, right?
• 14:33
  SPEAKER_00: And this is where you can see a lot of today aggressive behaviors where people are will try to influence what the agents are recommending.
• 14:41
  SPEAKER_00: So it's very possible.
• 14:42
  SPEAKER_00: It's uh it's not very possible.
• 14:43
  SPEAKER_00: It is actually happening today that people are trying to do whatever they can to influence the market so that when you say, go read go do some research, the agent is gonna go out and read websites and read Reddit posts and this and that, right?
• 14:55
  SPEAKER_00: And those Reddit posts will be contaminated in a way to influence the agents to say you should absolutely buy SpaceX at that level of money because it's gonna be a great thing for your portfolio, right?
• 15:06
  SPEAKER_00: Right.
• 15:07
  SPEAKER_00: And so at that level, not only are you not fully in control of your agent because it's gonna read possibly tainted information, the agent is gonna recommend you to do things, but those recommendations will also be influenced by other agents or third parties that will try to manipulate the market, right?
• 15:26
  SPEAKER_00: So um it's the who's in control, it's a bit hard to tell.
• 15:32
  SPEAKER_00: I tell you, it's the good old story of the master and the slave, right?
• 15:35
  SPEAKER_00: Is the master reading control of the slave?
• 15:37
  SPEAKER_00: If the slave stops serving the master, well, what happens, right?
• 15:40
  SPEAKER_00: So it is a it's an interesting, not only philosophical discussion, but also a broader question about you know um how we should interact with that technology, right?
• 15:52
  SPEAKER_00: And uh, you know, I'm I'm very pro-tech, as you know.
• 15:56
  SPEAKER_00: Um I think there's a lot of things we should use technology for.
• 16:01
  SPEAKER_00: I also think that a lot of people are um can be easily misguided, and the technology can have a very bad impact on those folks.
• 16:13
  SPEAKER_03: Definitely, that's a great point.
• 16:15
  SPEAKER_03: Well, that was super fun, and you passed the Turing test.
• 16:18
  SPEAKER_03: You selected all the cones, so we've proved we've proved you're human, um, which is great.
• 16:24
  SPEAKER_03: You know, some of us had some questions, and for those of you out there, if you have any rapid fire questions for Eve, just put them in the comments and we'll get them next round.
• 16:31
  SPEAKER_03: Um, as you can tell, he loves to opine.
• 16:34
  SPEAKER_03: So really excited to continue the unprompted series coming forward.
• 16:38
  SPEAKER_03: But let's get into the meat and the cheese of why we're here.
• 16:41
  SPEAKER_03: Eve, there has been a lot going on in the world of AI since you and I last spoke.
• 16:46
  SPEAKER_03: A lot of it applies to category management and CPG.
• 16:48
  SPEAKER_03: So if you're a busy individual out there, you don't have time to be coming through the headlines.
• 16:53
  SPEAKER_03: Eve did it for you.
• 16:55
  SPEAKER_03: Uh, so let's kind of get into some really interesting things that's happened this past week that can really impact how we're doing business.
• 17:02
  SPEAKER_03: And, you know, overall, the overarching theme of what you talked about, um, it seems like there's an evolution going on where effectively we were using analysts and agents kind of in a manner of working alongside us.
• 17:20
  SPEAKER_03: And now we're actually embedding, right?
• 17:22
  SPEAKER_03: So there's a lot of these themes that we're going to talk about that kind of talk about embedding AI deeper into what you're doing on a day-to-day basis.
• 17:31
  SPEAKER_03: One of those is the open AI data agent, right?
• 17:34
  SPEAKER_03: So AI moving directly into BI, which is exciting for BI geeks like me.
• 17:41
  SPEAKER_03: But basically, you know, what I heard from you is that AI isn't just sitting by our analytics stack anymore, right?
• 17:48
  SPEAKER_03: It's it's can it's actually going to be sitting inside of it, which opens up a different world of possibilities, gets rid of a lot of the, you know, day-to-day kind of grind that we're doing from a VI perspective.
• 18:02
  SPEAKER_03: But give me your thoughts on that.
• 18:04
  SPEAKER_00: Oof, the it's a broad question.
• 18:07
  SPEAKER_00: Um, I I think you are absolutely correct in that the all of the all of the model providers are gonna try to start going from fully horizontal to a bit more vertical.
• 18:23
  SPEAKER_00: They will try to find some specific use cases to go after.
• 18:26
  SPEAKER_00: So we see it with financial services, legal, category management is likely to be one as well.
• 18:32
  SPEAKER_00: Um, or even just BI, right?
• 18:34
  SPEAKER_00: And indeed, I think having pre-built agents that are able to ingest data and then clean it and then assemble it and then present some some insight is, I think, today a I would say the the current state.
• 18:52
  SPEAKER_00: It's a commodity, right?
• 18:53
  SPEAKER_00: It is happening today.
• 18:55
  SPEAKER_00: Um, and yet I still think that you will have um you will still need you're still relying on data that will need to be reasonably clean, right?
• 19:08
  SPEAKER_00: So and that, surprisingly enough, today is, I believe, still a rather unsolved problem.
• 19:15
  SPEAKER_00: Um I'm working on a project right now where we're just trying to do some, I would call it reasonably simple data aggregation from um for like uh various KPIs of of companies.
• 19:27
  SPEAKER_00: And it is even using the biggest models we have available, the fable models, like even the those models are struggling to extract reliably those KPIs from reports or from Excel files, right?
• 19:42
  SPEAKER_00: And the reason for that is because well, that stuff's complicated, and different companies have different ways of naming things.
• 19:50
  SPEAKER_00: There's not a lot of um, there's not a lot of uh consistency, and so I think we will be able to automate a number of tasks in the BI world.
• 20:02
  SPEAKER_00: I think the there will be some nuance that will need to be human manually labeled in the front end to clean the data that is not going to go away.
• 20:12
  SPEAKER_00: So that's gonna be there.
• 20:13
  SPEAKER_00: Then you can do the processing, can be largely automated.
• 20:16
  SPEAKER_00: But then I think driving a lot of the iterations from like, hey, look, this is what the BI says.
• 20:22
  SPEAKER_00: Let I want to double-click here.
• 20:24
  SPEAKER_00: What should I do?
• 20:25
  SPEAKER_00: Run a scenario for me.
• 20:26
  SPEAKER_00: I think those are still places where um there's a major risk of the AI trying to boil to boil in the ocean.
• 20:35
  SPEAKER_02: Yes.
• 20:36
  SPEAKER_00: And consume, you know, I don't know, a few thousand dollars of tokens in in an in an evening, trying to come up with an answer that will be completely irrelevant for you because, oh yeah, you know, we're not that type of company, right?
• 20:49
  SPEAKER_00: And so um, you know, it goes back to the typical point, which is I think AI is going to remove a lot of the main points, but not everything is gonna be removed.
• 21:02
  SPEAKER_03: Definitely.
• 21:04
  SPEAKER_03: Yeah, it's gonna be interesting from a perspective of having AI in the analytics workflow instead of just helping you analyze.
• 21:11
  SPEAKER_03: What I'm excited about is that actually is a cheaper workflow, right?
• 21:14
  SPEAKER_03: So if you can harness it, if you're prompting correctly, if you have a clean data source underneath it where you're having native language query to kind of get to those data points, you're getting answers cheaper and faster.
• 21:27
  SPEAKER_03: The challenge is that you still need that human judgment call to effectively say, yes, this is something that we can make a differentiator from a business perspective.
• 21:40
  SPEAKER_03: This is something that is actionable and it aligns with our corporate strategy.
• 21:43
  SPEAKER_03: So while, you know, some people could see this and say, oh no, it's taking all those low level analyst jobs.
• 21:50
  SPEAKER_03: It's it's not necessarily, right?
• 21:52
  SPEAKER_03: Because you're still gonna need the validation piece.
• 21:54
  SPEAKER_03: We're still not there.
• 21:55
  SPEAKER_03: Even the best models that you and I have talked about and scenarios we ran.
• 22:00
  SPEAKER_03: You still need that validation piece.
• 22:02
  SPEAKER_03: The exciting piece, though, is that your daily workflow, if you're coming up with some kind of dashboard for Walmart and you're clicking through 14 different dashboards for Power BI, you ask a question that gets synthesized a lot faster and you get to insights a lot faster.
• 22:17
  SPEAKER_03: So I think that this would be one that is a little touchy from a perspective of is it taking our jobs?
• 22:23
  SPEAKER_03: Answer is still no.
• 22:24
  SPEAKER_03: But once again, elevating your skills so you know the right questions to ask, how to do native querying against your data, making sure your data is clean, right?
• 22:33
  SPEAKER_03: And then also how to really differentiate a good business insight in something that's actionable versus something that's AI slop.
• 22:41
  unknown: Right.
• 22:42
  SPEAKER_00: But what what I would tell what I would say though is that if your job or if your day-to-day tasks are to get requirement from someone via the dashboard over a couple of days or weeks and then present back the dashboard, if that's your job, then yeah, you may be your job may be in danger.
• 23:02
  SPEAKER_00: There's no doubt about it.
• 23:03
  SPEAKER_00: Because it's at that level, it's quite it's it's it's a it's a commodity, it's been completely commoditized today.
• 23:09
  SPEAKER_00: Um however, as you said, all of the so what questions, the what should we do?
• 23:17
  SPEAKER_00: And at the end of the day, right, if you look at a business, um if someone presents a recommendation for an in for an initiative, and people are like, yes, we're gonna do that, we're gonna spend this money to do X and Y with the with the hope of getting a return.
• 23:31
  SPEAKER_00: If that does not work well, then someone's gonna have ownership.
• 23:34
  SPEAKER_00: It's gotta be just in terms of governance, it's gotta end up on someone's responsibility.
• 23:38
  SPEAKER_00: You cannot just say, Oh, well, we did this because the AI told us so, so it's not really my fault, right?
• 23:43
  SPEAKER_00: That is not like cutmaster, right?
• 23:45
  SPEAKER_00: And I mean back in my days, I I used to do a lot of research around unmanned air vehicles with in the defense industry.
• 23:53
  SPEAKER_00: And, you know, I don't know how it is today, but back when I was studying this 20 years ago, it was very clear that um the all of the drones were perfectly capable of flying and even shooting their weapons fully autonomously.
• 24:09
  SPEAKER_00: But you always, by doctrine, you always needed to have a uh a human being pressing the button because you want, and usually an officer to say, like, ah yes, I will deploy the use of lethal force.
• 24:24
  SPEAKER_00: And I, as a human, I'm taking responsibility.
• 24:27
  SPEAKER_00: And if this goes wrong, it's on me.
• 24:28
  unknown: Right.
• 24:29
  SPEAKER_00: And I think it's the same thing in the business, right?
• 24:30
  SPEAKER_00: You're saying we should change our strategy, we should do this, we should do that, then someone's gonna take ownership and you know, be rewarded if that's the right decision.
• 24:40
  SPEAKER_03: Absolutely.
• 24:40
  SPEAKER_03: And we've talked about that in a couple of our AI quick hits about using AI as your personal analyst, right?
• 24:47
  SPEAKER_03: You still have the accountability for what's putting out, you still have accountability to check and to validate the recommendations, but effectively you can work together in partnership to come up with quicker, better insights.
• 25:00
  SPEAKER_03: I absolutely love it when you tell your late 1900s stories, by the way.
• 25:04
  SPEAKER_00: Late 1900.
• 25:05
  SPEAKER_00: Wow.
• 25:06
  SPEAKER_00: Yeah, it was like that was more like uh early 2000s, but yes, that was that was a long time ago.
• 25:12
  SPEAKER_03: Love that.
• 25:13
  SPEAKER_03: And as a shameless plug, because Eve just did say if you are just doing dashboards and that's your job, that your role might be in jeopardy, make sure you check out all the fantastic courses that the CMA offer to elevate your skills, including Eve's AI learning segment, which I believe you just put out a third or you're working on a third.
• 25:32
  SPEAKER_03: So you can come up to speed on how to effectively use AI and to build your skills just in case your job, you know, gets questioned.
• 25:40
  SPEAKER_03: So we have res we have resources for you.
• 25:44
  SPEAKER_00: That's right.
• 25:44
  SPEAKER_03: Yeah.
• 25:46
  SPEAKER_03: So another thing that you brought up today that kind of piqued my little interest from my category management heart is that Google and Accenture are building a thousand-person Gemini enterprise deployment group.
• 25:58
  SPEAKER_03: So the headline I took from that is that the AI race is really becoming less about who has the smartest model and who can actually get AI embedded into business the fastest.
• 26:09
  SPEAKER_03: So kind of opine on that one.
• 26:11
  SPEAKER_00: So I would say for you both, right?
• 26:13
  SPEAKER_00: I think the the race for having the best model at the cheapest cost is absolutely happening.
• 26:20
  SPEAKER_00: But I think what's what people are also realizing is that there is a widening gap between the people who use a lot of AI.
• 26:29
  SPEAKER_00: And maybe we'll touch more on this later, but there was a report recently by by Anthropic describing some of the misuses of AI, going from you know, hacking drone technology to hiking hotel Wi-Fi to trying to build biological weapons, right?
• 26:48
  SPEAKER_00: I mean, those are like not easy use cases to attack.
• 26:51
  SPEAKER_00: I mean, those are like basically some pretty crazy stuff, right?
• 26:56
  SPEAKER_01: Definitely.
• 26:57
  SPEAKER_00: Or like even the fact that we have we talked about this Navier Stoke equation that was solved in 88 hours for a million dollars worth of compute time.
• 27:03
  SPEAKER_00: That's like that's that that would have that would win, at least back in my days, that would have won someone's, you know, the equivalent of a Nobel Prize in mathematics, or like, you know.
• 27:15
  SPEAKER_00: That's what's happening in the top end of the tech-enabled companies.
• 27:20
  SPEAKER_01: Yeah.
• 27:21
  SPEAKER_00: If you look at the vast majority of the other companies, you may still be stuck at trying to do a report and this and that.
• 27:26
  SPEAKER_00: And what people are realizing indeed is that just adopting and understanding what you can do, the I mean, the gap, it's it's it's not even a gap, it's a chasm at that stage.
• 27:35
  SPEAKER_00: And those FDEs, those forward deployed engineers, are absolutely their job is to go in there and educate, deploy.
• 27:45
  SPEAKER_00: But at the end of the day, at the end of the day, their job is to incentivize consumption to make more money for Google, right?
• 27:52
  SPEAKER_00: Um, that's really their job.
• 27:54
  SPEAKER_00: But and if you look, if you look back, right, the first company that deployed FDEs or forward deployed engineers, that was really Palantir that came up with that term.
• 28:02
  SPEAKER_00: And their thesis was to say, and I I know a couple of those guys, um, and their thesis was to say, look, the best thing we can do is have our engine or forward deployed engineers be on site with a client, sit side by side with them, so that we can really understand what the problem is.
• 28:21
  SPEAKER_00: We will mock up something very quick so that they can use and we get some quick feedback and quick iterations.
• 28:26
  SPEAKER_00: That was the best way to generate um quick value.
• 28:33
  SPEAKER_00: And interestingly enough, you know, so you've got the FDEs on one side, but then you see company companies like Uber now talking about agentic pods or agentic pairs, where they pair up one AI person with one business person, so like you know, one person like me, maybe with one person like you, right?
• 28:50
  SPEAKER_00: And say, Hey, look, you know, you're a business expert, I'm a nerd.
• 28:53
  SPEAKER_00: We'll sit down together and we'll just hammer it out one-on-one until we get we build an agent or a tool that can make your life you know 500 times better.
• 29:05
  SPEAKER_02: Right.
• 29:05
  SPEAKER_00: That's and we're see we're now seeing that notion of those pairs, that pairing becoming more and more effective.
• 29:13
  SPEAKER_00: And so, you know, now those FDEs are gonna take the role of being the the nerd in that pair.
• 29:18
  SPEAKER_03: Right.
• 29:19
  SPEAKER_03: But are we what are all us nerds gonna do?
• 29:24
  SPEAKER_03: Right?
• 29:25
  SPEAKER_00: Exactly.
• 29:25
  SPEAKER_03: Exactly.
• 29:26
  SPEAKER_03: But I think you know, what this brings up for me is that I talk to a lot of people in the industry and I just have a broad question I always ask, and it's like, hey, how are you using AI today?
• 29:36
  SPEAKER_03: And you would be overwhelmed at the amount of people or who are like, oh, well, I use Gemini or Google to manage my inbox and to to group things and highlight when things are important.
• 29:46
  SPEAKER_03: If that's the extent of how you're using AI, you're going to get left in the dust, right?
• 29:51
  SPEAKER_03: Or we're at the point where another one of those pivotal points where it's time to elevate your skills.
• 29:57
  SPEAKER_03: Because now that we're seeing, you know, groups like Accenture, and I know that, you know, as a McKinsey person, you probably have a lot to say about Accenture.
• 30:05
  SPEAKER_03: That's for a different podcast.
• 30:07
  SPEAKER_03: But um, groups like Accenture employing this 1,000 agent analyst, you know, kind of workflow.
• 30:14
  SPEAKER_03: It's really challenging us in the CPG industry to not to kind of get away from the idea of like, oh, what AI platform should we use?
• 30:23
  SPEAKER_03: Now it's more of a strategic thought and that now that we're implementing AI, what workflows change, right?
• 30:30
  SPEAKER_03: So it's implementing it as a tool instead of just kind of figuring out haphazardly how to use it.
• 30:37
  SPEAKER_03: Business processes are going to change, workflows are going to change.
• 30:40
  SPEAKER_03: And you really want to be on the front end of that and not playing catch up because that's the competitive advantage.
• 30:46
  SPEAKER_03: So I think, you know, 25, 2025 was more about giving employees AI.
• 30:52
  SPEAKER_03: Now it's redesigning work around AI.
• 30:56
  SPEAKER_03: How do we make these better, faster, stronger, right?
• 30:59
  SPEAKER_03: So, how do we improve business processes to effectively get to better insights?
• 31:04
  SPEAKER_00: Yeah.
• 31:06
  SPEAKER_00: And you know what's also interesting is as you do as you go through all this, right?
• 31:15
  SPEAKER_00: Well, the cost associated with that.
• 31:17
  SPEAKER_00: Right.
• 31:17
  SPEAKER_00: And um, you know, I was just reading that uh at OpenAI, the research, each researcher has it has his or her own little uh research intern.
• 31:31
  SPEAKER_00: But the top researchers are on on average, the top researchers are consuming the equivalent of$7,000 of tokens per day, right?
• 31:40
  SPEAKER_00: So and if you extrapolate that over the entire course of a year, it means that one guy is gonna spend about$2 million worth of tokens over the course of a year.
• 31:50
  SPEAKER_01: Right.
• 31:51
  SPEAKER_00: Just that for one AI assistant, then you may have more of more than one, right?
• 31:54
  SPEAKER_00: Right.
• 31:55
  SPEAKER_00: And so, you know, you go back to that idea of a chasm and this implementing the workflows and so on, and that's you know there will be a cost.
• 32:05
  SPEAKER_00: It will not come for free.
• 32:07
  SPEAKER_03: Absolutely.
• 32:09
  SPEAKER_03: But the question is, right, because everything is about the dollar, is the cost more effective than having a thousand people, team of analysts, tackle something versus a thousand AI agents?
• 32:21
  SPEAKER_00: Yeah, it's a good question.
• 32:23
  SPEAKER_00: I think it be the answer is it it will depend.
• 32:25
  SPEAKER_00: It will depend.
• 32:26
  SPEAKER_03: It definitely depends.
• 32:27
  SPEAKER_03: We're 100% honest.
• 32:29
  SPEAKER_03: So this one is my favorite one that we talked about.
• 32:31
  SPEAKER_03: It sent me down a rabbit hole that I don't even want to admit for a couple of hours.
• 32:35
  SPEAKER_03: And it's the meta muse becomes an agent that can actually do things.
• 32:40
  SPEAKER_03: Uh so what I'm excited about is consumer AI, right?
• 32:44
  SPEAKER_03: So consumer AI is moving from just answering questions to taking actions across applications.
• 32:51
  SPEAKER_03: So to me, that really is going to change our future shopper journey.
• 32:56
  SPEAKER_03: You know, right now, from a CPG perspective, we have a very good handle on the shopper journey, how consumer decision trees think, how the consideration set works.
• 33:06
  SPEAKER_03: That completely gets blown to smithereens, Smith.
• 33:09
  SPEAKER_03: Um, you know, as we kind of implement things like MetaMuse.
• 33:14
  SPEAKER_03: And I also like saying that fast for some reason.
• 33:17
  SPEAKER_03: But, you know, I'm envisioning a world where it's like, hey, plan snacks for my Alabama tailgate on Saturday.
• 33:23
  SPEAKER_03: I got 12 people, I want to spend under$100.
• 33:27
  SPEAKER_03: Uh, I've got one gluten-free guest, and I need it here by 3 p.m.
• 33:31
  SPEAKER_03: on Friday.
• 33:33
  SPEAKER_03: And it happens.
• 33:34
  SPEAKER_03: And so that is, like I said, that'll revolutionize really the shopper journey.
• 33:39
  SPEAKER_03: So kind of come up with for me, you know, what your thoughts about this are.
• 33:43
  SPEAKER_03: Do you think, since this is kind of a beta phase, obviously there's going to be some kinks, but do you think that this might be the shopper journey of the future?
• 33:50
  SPEAKER_00: Absolutely.
• 33:51
  SPEAKER_00: Yes.
• 33:51
  SPEAKER_00: I think you're you're going to move away from having individual tasks to, hey, plan an event for me.
• 33:58
  SPEAKER_00: I think that's where we're going.
• 33:59
  SPEAKER_00: Now, mind you, for this to happen, right, you you also will also need to provide the AI with so much data about yourself.
• 34:09
  SPEAKER_00: Because if you say, hey, look, plan it, the more the less specific you are, and the more information you give the agent, the more it will have to assume, and the more you may end up having the agent saying, Well, the cheapest way to do an event is to, I don't know, book the Ritz Colton for like you know a billion dollars, and that's gonna be or some crazy you know, edge case like that, right?
• 34:34
  SPEAKER_00: Um so um and and muse or metamuse is one of the examples that you know for it to work well, you need to give it access to your email, your Spotify, all of your all of your travel things.
• 34:51
  SPEAKER_00: I mean, it's it will require an extensive amount of data.
• 34:55
  SPEAKER_00: And I don't know about you, but I would personally be very, very, very uncomfortable sharing that amount of information with a model, especially if it's a free model, even if it because you know you you you know you are essentially training the model with that.
• 35:11
  SPEAKER_00: Um so I think that's to me there's a massive question around privacy here.
• 35:15
  SPEAKER_00: But if I if I go back to your original question about what does it mean for CPG, I mean, I think we are already getting to a place where a lot of the traditional tools like temporary price reductions and whatnot play on human psychology.
• 35:31
  SPEAKER_01: Definitely.
• 35:32
  SPEAKER_00: Like the point, like the dot 99 prices and whatnot.
• 35:35
  SPEAKER_00: In the world of agents, agents don't don't operate, do not have the same psychology that we have, right?
• 35:42
  SPEAKER_00: And you know, you will very quickly, I think, get to a place where you say, Well, look, I want to buy some snacks for my kids, I want them to be healthy, I want them to be cheap, I want them to do blah, blah, blah, blah, blah.
• 35:52
  SPEAKER_00: Just go figure out the best brand.
• 35:53
  SPEAKER_00: I am not gonna go read reviews from mom and pop and this and that, from this influencer.
• 35:58
  SPEAKER_00: Don't care.
• 35:59
  SPEAKER_00: I just want to have you know the best thing that is tailored for me.
• 36:05
  SPEAKER_00: Now, mind you, the way the agency is gonna come up with a recommendation is actually probably by reading some of those data generated by third parties, right?
• 36:15
  SPEAKER_00: So hence, and we go back to the question of influence and manipulation risks, right?
• 36:19
  SPEAKER_00: Um, but you know, I think it's a you you may and I think it's going to absolutely change the way you shop, and it may also change the way people do impulse shopping.
• 36:30
  SPEAKER_00: Impulse shopping may disappear to a large extent.
• 36:32
  SPEAKER_00: You may now say, well, you know what?
• 36:34
  SPEAKER_00: Here's a list of things I want to buy over the next 12 months.
• 36:36
  SPEAKER_00: Just buy them when it's absolutely at the cheapest possible price, right?
• 36:42
  SPEAKER_00: And and maybe that may scratch the age of I need to shop, I'm just gonna add something on my list, and then I'm gonna trust an agent to let me know when when I should press a play button, or you may just give the agent your credit card number, as Visa or MasterCard want you to do, and then let the agent do the shopping for you.
• 37:00
  SPEAKER_03: Absolutely.
• 37:01
  SPEAKER_03: And I think that this topic honestly deserves you and I doing a deep dive because thinking through it, it's like we have spent decades, literally decades, um, kind of coming up with how we optimize the in-person shelf for humans, kind of in a place right now where we're figuring out how to optimize digital shelf.
• 37:24
  SPEAKER_03: Some people are really good at it, some people need a lot of work.
• 37:27
  SPEAKER_03: But trying to figure out how to optimize the shelf for agents is going to be something that's a complete disruptor to the category.
• 37:34
  SPEAKER_03: I mean, just a complete disruptor, honestly, to the industry.
• 37:38
  SPEAKER_03: And, you know, how do we figure out, you know, is it reviews?
• 37:43
  SPEAKER_03: Is it structured product data?
• 37:45
  SPEAKER_03: You know, what is it that would effectively drive an agent to make a recommendation?
• 37:50
  SPEAKER_03: Um, you know, do we get into a place where you're actually buying agent recommendations?
• 37:55
  SPEAKER_03: You know, we kind of need to figure that out.
• 37:58
  SPEAKER_03: And no one has a good handle on that yet.
• 38:00
  SPEAKER_03: So I think that you and I oh a deep dive into what the future of shopping looks like and like AI.
• 38:07
  SPEAKER_00: And then if you add on top of that, the um the news from I think there was a an announcement last week that OpenAI was their I can't remember the exact number, but they are now getting into the ads business, right?
• 38:20
  SPEAKER_00: So you can actually pay OpenAI to actually recommend products, which I think is that's dangerous.
• 38:25
  SPEAKER_00: Very, very interesting.
• 38:26
  SPEAKER_00: But that's a super direct way to insert your product into the recommendation stream, right?
• 38:33
  SPEAKER_00: Right.
• 38:34
  SPEAKER_00: Um, I mean, very much like in the way that you know today I you know you don't really Google or use Google the way that you know you used to use Google like 10 years ago or like even three years ago, right?
• 38:45
  SPEAKER_00: Where you put in a search query, you get 12 websites, and you've got to go and read every one of those websites and just come up with the synthesis in your head.
• 38:52
  SPEAKER_00: Now it's the synthesis just appears.
• 38:55
  SPEAKER_00: You don't even have to browse through the websites.
• 38:57
  SPEAKER_00: And it's been proven now that you know the traffic on websites has gone down globally.
• 39:02
  SPEAKER_00: From at least the traffic from humans has gone down globally, but the traffic from agents has gone up.
• 39:09
  SPEAKER_03: Absolutely.
• 39:11
  SPEAKER_03: Yeah.
• 39:12
  SPEAKER_03: We're gonna have to do like a whole separate segment on shopping in the world of agentic AI.
• 39:16
  SPEAKER_03: It's really exciting.
• 39:18
  SPEAKER_03: But another thing that I thought was really cool is about ChatGPT um for financial services.
• 39:25
  SPEAKER_03: And the reason why that peaked is because just like we talked, you know, do we get to a point where we have an agent specifically for Catman?
• 39:34
  SPEAKER_03: Um so AI getting verticalized, and AI platforms are beginning to package models, which is exciting.
• 39:40
  SPEAKER_03: Premium industry data, workflows, all of that, into specialized vertical products.
• 39:44
  SPEAKER_03: Um what vertical comes next in your head?
• 39:48
  SPEAKER_00: Oh, that's a good question.
• 39:50
  SPEAKER_00: So um I mean, we already know of a couple of verticals that are being absolutely uh absolutely attacked, attacked, as in there's a big focus right now.
• 40:04
  SPEAKER_00: So financial services, just because there's a lot of money there.
• 40:07
  SPEAKER_00: So MA, investment banking, accounting, lots of money going there.
• 40:12
  SPEAKER_00: Legal is a huge one as well.
• 40:14
  SPEAKER_00: Uh definitely.
• 40:15
  SPEAKER_00: There was just actually a news earlier in the week whereby um uh well, first of all, a few weeks ago, Thompson Reuters announced that they had created their own model off the back of an open weight model to create their own legal advisoring, legal advising model, right?
• 40:36
  SPEAKER_00: Cool.
• 40:36
  SPEAKER_00: But now there's a big firm based in the UK, like law firm in the UK, that started to buy their own GPUs to create their own models as well, proprietary model, right?
• 40:45
  SPEAKER_00: So um I think legal is is is is an obvious one.
• 40:49
  SPEAKER_00: And then I would say, like, you know, you may and I would say that there are some places where that has already been.
• 40:56
  SPEAKER_00: I think I think those two, for example, are in process.
• 40:59
  SPEAKER_00: There are industries where the verticalization has already happened.
• 41:02
  SPEAKER_00: So coding, for example, if cloud code or cursor, that that has been already fully attacked.
• 41:08
  SPEAKER_00: Design, I mean, cloud design is quite is very interesting and very effective.
• 41:12
  SPEAKER_00: I mean, if you're a designer today and you don't use plot design, you may be missing the boat massively, right?
• 41:17
  SPEAKER_00: For sure.
• 41:18
  SPEAKER_00: Um, so those are some of the verticals that I think are are absolutely there.
• 41:23
  SPEAKER_00: I think another big one that I think is going to happen, and and and I think I'm I'm actually excited about it.
• 41:30
  SPEAKER_00: There are two that I think are coming.
• 41:31
  SPEAKER_00: One of them is education, the other one is going to be medicine.
• 41:34
  SPEAKER_00: Right.
• 41:34
  SPEAKER_00: Uh medicine because we now know that um you know if your doctor spends more than five minutes with you, you're probably part of the luck if you already have a doctor and you you actually see the doctor, right?
• 41:50
  SPEAKER_00: Uh but that guy only has five minutes to devote to your case.
• 41:53
  SPEAKER_00: So it's gonna have it's gonna be a lot of like quick hits and like you know pattern recognition, whereas the AI systems will take will have you have a chance to take more time, ingest more information, and maybe find and diagnose things in a finer way than doctors can, right?
• 42:08
  SPEAKER_00: Um so I think that's really interesting.
• 42:11
  SPEAKER_00: The other one that I said, education.
• 42:13
  SPEAKER_00: I mean that's one of that's been one of my uh big areas of excitement, which is everybody learns differently, right?
• 42:22
  SPEAKER_00: And the education system has been built to be the best it can for most people, for the average human being.
• 42:29
  SPEAKER_00: But I think you also have a lot of people on the tails of being you know average, and those people tend to be not well served by the education system today, right?
• 42:40
  SPEAKER_00: And um I think you get a chance now to have quite literally a one teacher per kid, or like one tutor per kid world where the tutor will be like, hey, look, this guy learns more by learning the principles, and then after that you can apply the examples, that's one way, or maybe it's seeing more examples and go to the principles.
• 43:02
  SPEAKER_00: Maybe it's more like visual learning, more like auditory learning, more blah, blah, blah, blah.
• 43:07
  SPEAKER_00: Right.
• 43:07
  SPEAKER_00: And I think having that can be uh very, very, very good for educating our kids.
• 43:14
  SPEAKER_00: Absolutely.
• 43:15
  SPEAKER_00: Which of course brings the question you know, why are we what are we educating our kids for if we get to a place where AI does all the work?
• 43:20
  SPEAKER_03: Right.
• 43:21
  SPEAKER_00: Well, that's a whole other question, right?
• 43:23
  SPEAKER_03: That's a whole other day.
• 43:25
  SPEAKER_03: That's right.
• 43:25
  SPEAKER_03: No, I mean you bring up some really great points, and it's interesting because this has what I love about this is the level of customization.
• 43:34
  SPEAKER_03: You sit five people in my industry down in front of a spreadsheet, we're all gonna approach it a different way.
• 43:41
  SPEAKER_02: Right.
• 43:42
  SPEAKER_03: Um, my way's right, but but you know, we're all gonna I saw that.
• 43:47
  SPEAKER_03: Um my way is right.
• 43:48
  SPEAKER_03: No, but you know, if we're just if someone just gave you a plan, analyzed this spreadsheet, everyone's gonna start from a different position.
• 43:54
  SPEAKER_03: You might end up in the same B, you know, from an A to B, but you know, I might go like this, someone might be.
• 44:00
  SPEAKER_03: More linear.
• 44:01
  SPEAKER_03: And I think that what this does is it kind of homogenizes all of those skills, looks at different things, and really allows you to say, what's the white space?
• 44:13
  SPEAKER_03: What am I missing in this data set?
• 44:15
  SPEAKER_03: Right.
• 44:16
  SPEAKER_03: Now, I think my question is the big race, right?
• 44:19
  SPEAKER_03: So how long is it going to take somebody to create an AI agent for Cat Ran?
• 44:26
  SPEAKER_03: Which, you know, I'm tasking you and I with that, I think.
• 44:28
  SPEAKER_03: But I think it could definitely be a game changer because so much you talk to anyone in the industry and so much time is spent on the tactical.
• 44:40
  SPEAKER_03: I mean, most of us, even in high-level roles, you know, even if it's just managing people or, you know, talking about scorecards or creating daily flashes or whatever you're doing, you're still probably 80-20 tactical versus strategic.
• 44:53
  SPEAKER_03: And we know that that scale needs to absolutely change.
• 44:56
  SPEAKER_03: So I'm excited about this.
• 44:59
  SPEAKER_03: Once again, you know, could this be a little more, is it taking our jobs?
• 45:03
  SPEAKER_03: If you don't elevate your skills, potentially, but you know, show me a person who actually loves being in multiple spreadsheets.
• 45:13
  SPEAKER_03: It's me.
• 45:14
  SPEAKER_03: But still, um, I think that another thing though is that it really challenges you to look at your data stack.
• 45:20
  SPEAKER_03: So maybe we're looking at silo data that we haven't brought in.
• 45:24
  SPEAKER_03: Maybe AI was giving us a recommendation on how to use a different data piece like commodity weather, what's going on with El Nino, you know, talk about things like that that maybe we haven't brought to the table because you and I have talked.
• 45:37
  SPEAKER_03: Now that AI is on board, the traditional category role of just reporting the news, of just bringing white space to a retailer isn't needed.
• 45:46
  SPEAKER_03: The retailer can do that effectively without bias.
• 45:49
  SPEAKER_03: So that's where we're really talking about stepping up your game.
• 45:52
  SPEAKER_03: And I think an agent dedicated to category management could 100% help people get there, especially as teams kind of contract.
• 45:59
  SPEAKER_03: You know, that's a concern that we've been seeing across the industry.
• 46:03
  SPEAKER_03: And people are being asked to do more.
• 46:04
  SPEAKER_03: You know, all of a sudden, category managers also are doing internal business insights in revenue growth management and, you know, retail media analysis.
• 46:13
  SPEAKER_03: So I think this opens it up to a world where people can expand their brains.
• 46:18
  SPEAKER_03: I think it's just going to be a bit of a challenge on what cream rises to the top there.
• 46:24
  SPEAKER_00: I agree.
• 46:25
  SPEAKER_00: Absolutely agree.
• 46:26
  SPEAKER_03: Um and then our last one, which I always say for best for last, is universal music and 11 labs turn copyright conflict into licensed AI products.
• 46:36
  SPEAKER_03: Now, why I'm excited about that is because initially there was a giant freak out oh my gosh, AI cannot use our IP, right?
• 46:43
  SPEAKER_03: It's gonna get out there, everyone's gonna use it, it's going to, you know, tarnish the brand in some way.
• 46:49
  SPEAKER_03: And now it's kind of coming back under, well, okay, what if we teach AI how to use our IP under our rules with compensation?
• 46:58
  SPEAKER_03: So talk to me a little bit about Yeah.
• 47:00
  SPEAKER_00: Uh uh what I would say is this this is a typical case of, you know, if you can't beat them, join them.
• 47:05
  SPEAKER_00: Right.
• 47:05
  SPEAKER_00: Right.
• 47:06
  SPEAKER_00: And um if you happen to sit on a large body of copyrighted information, I think you're you've got two choices.
• 47:17
  SPEAKER_00: Either you keep it closed behind walls, but then those walls can be breached very quickly.
• 47:21
  SPEAKER_00: And the moment if you have one breach, that is done.
• 47:23
  SPEAKER_00: Your business is over, right?
• 47:25
  SPEAKER_00: So that's one approach.
• 47:26
  SPEAKER_00: The other approach being, well, let's open it up, but let's make sure that we get paid for doing so.
• 47:32
  SPEAKER_00: Right.
• 47:32
  SPEAKER_00: And so we now see that with like music uh copyright owners, but even um uh images, right?
• 47:42
  SPEAKER_00: So that was a big thing with Adobe.
• 47:44
  SPEAKER_00: Um, there was a you know, so today if you if you use like a ChatGPT and whatnot, there's always a question of like, well, ChatGPT created this image using a billion other images that may or may not have been sourced properly, right?
• 47:56
  SPEAKER_00: You don't really know.
• 47:57
  SPEAKER_00: If you use the Adobe version, you know for a fact this the everything will have been cleared, the the copyright is all tight, there's no issues, legally speaking, you can just go ahead and use that.
• 48:07
  SPEAKER_00: So I think you're what you're buying with that is that notion of security and safety, uh from a legal standpoint.
• 48:15
  SPEAKER_00: Definitely.
• 48:15
  SPEAKER_00: But um, you know, I think what's very interesting here becomes the question of uh if you are a content creator, right?
• 48:28
  SPEAKER_00: An original content creator, how do you find space there?
• 48:31
  SPEAKER_00: Because it it's unlikely that you'll have your own deal with OpenAI or GPT or whatnot, right?
• 48:36
  SPEAKER_00: So those deals are gonna be made at the you know at the much higher level, like the universals or like the MGM studios or like the Sony Motion Pictures and whatnot.
• 48:47
  SPEAKER_00: Um and so um and we've we also see it with the even even with the Actors Guild, right?
• 48:53
  SPEAKER_00: Who have to who may be sending away some of their some of their resemblance and okay, this is I'm gonna be the the nerd talking.
• 49:01
  SPEAKER_00: I just discovered a YouTube channel over the weekend where someone is taking is doing uh Star Wars or remaking some Star Wars shots that have not been told in the normal canon movies.
• 49:15
  SPEAKER_00: Well, guess what?
• 49:16
  SPEAKER_00: They're using the likelihoods of you know Ewan McGregor and um Natalie Portman and so on and so forth.
• 49:25
  SPEAKER_00: And you see them on screen, they look lifelike, but you know it's not them because it's just been made by AI.
• 49:30
  SPEAKER_00: But if you didn't know, you're like, oh cool, but she was acting there, it's crazy.
• 49:35
  SPEAKER_00: But you know, it also makes me feel deeply uncomfortable that those people's likeliness has been completely stolen for a YouTube channel, for basically a YouTube YouTube channel of a guy who's trying to sell lightsabers replica.
• 49:51
  SPEAKER_00: Like that's the end that at the end, that's what the guy's trying to sell.
• 49:54
  SPEAKER_00: That's a product, but so anyway, it's um the notion of uh artistry and the the pain that people go through to create of for creation, period.
• 50:03
  SPEAKER_00: Yeah, it's also going through an upheaval.
• 50:08
  SPEAKER_00: Because I think part of what we value in artistry is the uh the fact that someone went through a very painful process to create something.
• 50:16
  SPEAKER_00: Right, right.
• 50:16
  SPEAKER_00: That's that's one of the things that we I think we appreciate as human beings is is seeing that.
• 50:21
  SPEAKER_00: And if that doesn't exist, then what's the value of slop?
• 50:25
  SPEAKER_00: Right.
• 50:25
  SPEAKER_00: Yeah.
• 50:26
  SPEAKER_00: Anyway, I don't have a very good answer for you from an industry perspective, but it is a you know, it's not just music, it's music, images, videos, books, everything, right?
• 50:37
  SPEAKER_03: And being able to, I mean, it's gonna have a very interesting impact on the marketing department of a lot of these CPG companies because if I can define how AI can use my IP, um, you know, what can they generate?
• 50:51
  SPEAKER_03: And I think right now it would be an interesting task to kind of use it as a compare and contrast, right?
• 50:59
  SPEAKER_03: So here's something I generated, here's something AI generated, what did they do better, what did I do better, and kind of move on from there.
• 51:07
  SPEAKER_03: But yeah, it's it's definitely going to be interesting in that it was when we first kind of started talking about AI, a lot of people were like, oh no, like hands off my IP.
• 51:17
  SPEAKER_00: And now it's like, well, let's see what Yeah, and actually now I think in the world, if you talk about the marketing department, I think what we'll be seeing is even if you do an A-B testing, at some point you'll realize that the only that there's so much value in doing micro segmentation, like, hey, I am going to tailor a message for Chelsea because I know that Chelsea likes red handbags, right?
• 51:39
  SPEAKER_00: Or like uh she likes paintings of trees and lights shining through trees.
• 51:44
  SPEAKER_00: I don't know, right?
• 51:45
  SPEAKER_00: Um and you know, uh it may not work on everybody, but it it may be enough of a tilt or a lift that people are like, you know what?
• 51:56
  SPEAKER_00: Yes, I am going to invest the money to create mess like completely customized messages for every single one of my customers, right?
• 52:03
  SPEAKER_00: In which case that like you as a human being, of there's just no way you could do it.
• 52:07
  SPEAKER_00: You can compete against that.
• 52:08
  SPEAKER_00: And even if it's not as good, it may still beat your carefully, cleverly witty, yeah, pithy message.
• 52:17
  SPEAKER_03: Yeah, and it's interesting because the industry has moved towards hyperl hyperpersonalization in the way of retailers sending out, you know, going through your history and providing coupons directly targeted at you.
• 52:29
  SPEAKER_03: This is beyond hyperpersonalization.
• 52:32
  SPEAKER_03: I'm trying to figure out a word that we can create to like coin it if it's like a segment of one.
• 52:37
  SPEAKER_00: Yeah, segment of one.
• 52:38
  SPEAKER_00: That's what it is.
• 52:40
  SPEAKER_03: Um, but you know, it's taking that a step further.
• 52:43
  SPEAKER_03: And to your point, something that humans could have not done.
• 52:46
  SPEAKER_03: It just would have not been cost effective.
• 52:48
  SPEAKER_00: So it's something to watch.
• 52:52
  SPEAKER_00: At the same time, you you put that against what we just talked about, which is hey, look, you you may end up sending great marketing to a lot of people.
• 52:59
  SPEAKER_01: Right.
• 52:59
  SPEAKER_00: But if those people don't shop anymore, it's their agents who shops.
• 53:04
  SPEAKER_00: Why you why do you even bother?
• 53:06
  SPEAKER_03: Right.
• 53:06
  SPEAKER_03: So now it's almost agent on agent, right?
• 53:09
  SPEAKER_03: So what's going to incite the agent to buy for the person?
• 53:12
  SPEAKER_03: And that's a whole nother mess of that's a whole other podcast in and by itself.
• 53:19
  SPEAKER_00: Yes.
• 53:19
  SPEAKER_00: Indeed.
• 53:20
  SPEAKER_03: Absolutely.
• 53:20
  SPEAKER_03: Well, Eve, I have to admit, this is like one of my two favorite podcasts that we've done.
• 53:25
  SPEAKER_03: Um in summary, everybody, just kind of to wrap up what we've talked about today.
• 53:31
  SPEAKER_03: So AI can analyze our data inside the data, right?
• 53:35
  SPEAKER_03: It's in the workflow now.
• 53:37
  SPEAKER_03: It's not something that you're using kind of in a tangent to help you analyze your data.
• 53:42
  SPEAKER_03: Companies are redesigning workflows around it.
• 53:45
  SPEAKER_03: Agents are beginning to act for consumers.
• 53:48
  SPEAKER_03: So we kind of have to figure that piece out.
• 53:50
  SPEAKER_03: Specialized AI is coming for individual individualized industries as we see verticalization come through.
• 53:56
  SPEAKER_03: And then brands are gonna figure out how to use IP within the AI ecosystem.
• 54:01
  SPEAKER_03: So that was just this week.
• 54:03
  SPEAKER_03: And that's how fast AI is moving.
• 54:05
  SPEAKER_03: So using your supercomputer, I'm like a TI82 and you're like a supercomputer over there, but um using your supercomputer brain, what do you think we're gonna see in the next couple of weeks, Eve, when we get back on our podcast?
• 54:20
  SPEAKER_00: Um, I think we'll see more mathematical breakthroughs happening for sure.
• 54:26
  SPEAKER_00: I can guarantee that.
• 54:27
  SPEAKER_00: I think we'll see, I think people are gonna get absolutely terrified with what people are actually doing with the with AI today, like those those hacks, those bioweapons.
• 54:35
  SPEAKER_00: I I would not be surprised if governments across the world were would were going to try to curtail that, which you know, the moment you do that, then immediately you're gonna give rise to a very very frothy black market.
• 54:51
  SPEAKER_02: Yes.
• 54:51
  SPEAKER_00: And and I I think you you will see that AI will be subjected to the same restrictions that weapons are restricted to.
• 55:01
  SPEAKER_00: So maybe at some point you'll will need to get a license like to carry an AI model in your pocket.
• 55:06
  SPEAKER_01: Oh my god.
• 55:07
  SPEAKER_00: Or to use this, or to export an AI model, right?
• 55:10
  SPEAKER_00: Like as you have, you can't you cannot just transport a missile from from one country to the other.
• 55:14
  SPEAKER_00: You need to have you know the right approval.
• 55:16
  SPEAKER_00: And I think we are I think we we we're there.
• 55:19
  SPEAKER_03: Yeah.
• 55:21
  SPEAKER_03: Awesome.
• 55:21
  SPEAKER_03: Well, assuming that someone hasn't hacked our nuclear weapons and we're not in a nuclear winter in a couple weeks, we'll be back here talking about additional AI insights and love love that on a positive note.
• 55:36
  SPEAKER_00: Exactly.
• 55:36
  SPEAKER_00: Nuclear winter is such great uplifting ideas.
• 55:39
  SPEAKER_00: Absolutely.
• 55:40
  SPEAKER_03: Well, thank you all so much.
• 55:41
  SPEAKER_03: Tune in next time.
• 55:41
  SPEAKER_03: And once again, if you have any questions, drop them in the comments and we'll get to them.
• 55:45
  SPEAKER_03: This has been AI Top Five with Eve and Chelsea.
• 55:49
  SPEAKER_03: We'll see you on the flip side.
• 55:50
  SPEAKER_00: Thank you guys.
• 55:51
  SPEAKER_00: Bye.