Latest / AI Top 5 / Always-On Agents, AI Shoppers & the Death of the Digital Shelf

Transcript

Always-On Agents, AI Shoppers & the Death of the Digital Shelf · AI Top 5

0:04
SPEAKER_02: Welcome back to the AI Top 5.
0:06
SPEAKER_02: If you're watching this, we did not get a nuclear winner, just like Eve suggested that we would by AI taking over the world.
0:12
SPEAKER_02: So we live to see another day.
0:13
SPEAKER_02: Super excited about that.
0:15
SPEAKER_02: Eve, how have you been?
0:16
SPEAKER_01: Been good.
0:17
SPEAKER_01: Been good.
0:18
SPEAKER_01: Uh excited and still a bit terrified every week that passes with the news, but yes, mostly good.
0:25
SPEAKER_02: I totally get that.
0:26
SPEAKER_02: And we're gonna dive into the news right after we do our favorite segment with Eve called Unprompted.
0:31
SPEAKER_02: So we've come up with a list of questions that Eve has not seen and he has to answer truthfully.
0:36
SPEAKER_02: For those of you who are just joining us, Eve has a pedigree that looks like it's been written by AI.
0:42
SPEAKER_02: So he has a PhD from MIT.
0:45
SPEAKER_02: He worked for McKenzie and their Quantum Black program, if I'm correct, right?
0:50
SPEAKER_01: Yeah.
0:51
SPEAKER_02: Okay, perfect.
0:52
SPEAKER_02: And while at Quantum Black, he works specifically on some projects for Formula One racing.
0:57
SPEAKER_02: So this one's a really important one, Eve, and it's gonna cause you to do some thinking if that's okay.
1:02
SPEAKER_02: So there is a cultural promo phenomenon called the Kim Kardashian effect, where if you go back and look at famous athletes that Kim Kardashian has dated, their careers get significantly worse once they've dated and broken up.
1:19
SPEAKER_02: Are you concerned about Lewis Hamilton at all?
1:21
SPEAKER_02: And have you done any AI predictions on if this might affect his career?
1:26
SPEAKER_01: Wow.
1:27
SPEAKER_01: Wait, is he dating someone?
1:28
SPEAKER_01: I don't even know.
1:31
SPEAKER_01: Oh, I I I I follow F1, but mostly what happens on the track, not off the track.
1:38
SPEAKER_01: Wait, really?
1:40
SPEAKER_02: He's really yeah, he's really dating Kim Kardashian.
1:43
SPEAKER_02: That's the thing.
1:43
SPEAKER_01: Oh wow, okay.
1:44
SPEAKER_01: Um so that is very unprompted.
1:46
SPEAKER_01: I I did not know.
1:48
SPEAKER_01: I did not know.
1:49
SPEAKER_02: Um breaking news.
1:51
SPEAKER_01: Breaking wow, okay.
1:52
SPEAKER_01: Well, yeah, I wish him luck.
1:54
SPEAKER_01: He needs all the luck indeed he can get at uh with his uh at Ferrari right now.
1:59
SPEAKER_01: So yeah.
2:00
SPEAKER_02: Yeah.
2:01
SPEAKER_02: That's hilarious.
2:02
SPEAKER_02: No, so I read that.
2:03
SPEAKER_02: I was it was a weird thing where someone did some kind of statistical correlation.
2:08
SPEAKER_02: And after I think Odell Beckham Jr., Reggie Bush, and there was one more dated and broke up with Kim Kardashian, they're calling it the curse because their career got they got injured or they got significantly um worse when it comes to their athletic ability.
2:24
SPEAKER_02: So we wish Lewis Hamilton the best, but that gives us the intel that Eve is deep on news for AI, just for you, not pop culture updates and insights.
2:35
SPEAKER_02: So cool.
2:36
SPEAKER_02: Let's try a second one.
2:38
SPEAKER_02: This one doesn't require pop culture knowledge, so I think you've got this one, okay?
2:42
SPEAKER_00: Okay.
2:43
SPEAKER_02: Recently, OpenAI, aka Chat GPT, opened up a module that allows you to connect your financial institutions to it so it can give you some information on spending, subscriptions, ETC.
2:57
SPEAKER_02: As an AI expert, would you do this?
3:01
SPEAKER_01: The short answer is hell no.
3:04
SPEAKER_01: Absolutely not.
3:05
SPEAKER_01: Um well, I mean, let me let me slightly take a slightly more nuanced answer.
3:10
SPEAKER_01: Um if I'm using a um a consumer level Chat GPT, I would absolutely not connect it because again, their data privacy um statements is that if you're using one of those, you know, personal consumer level, 20 bucks a month, even you know, like 100 bucks a month for like more than that for the pro versions, um any data that you upload can be used for training.
3:44
SPEAKER_01: And um I personally am not super keen on providing my banking data to OpenAI.
3:49
SPEAKER_01: I think they have enough data.
3:50
SPEAKER_01: Um now, to be fair, using it with a more secure model, like a enterprised version, enterprise grade GPT, I think would not be would be conceivable, but on a personal consumer level, absolutely not.
4:05
SPEAKER_01: I strongly just do not recommend you do that.
4:08
SPEAKER_02: Okay, perfect.
4:09
SPEAKER_02: That was the one thing that a lot of people reached out and asked a question about is hey, this is a really cool feature.
4:15
SPEAKER_02: Is it safe to do?
4:16
SPEAKER_02: So at this term, it looks like a red card from Eve, not even a proceed with caution, but unless you have an enterprise model, right?
4:24
SPEAKER_02: But once again, and and there has been some question, Eve, about there is a feature in Chat GPT for personal where you can toggle a button that says, you know, don't use my information for learning.
4:36
SPEAKER_02: I feel like that gives people a little bit of a false sense of security.
4:40
SPEAKER_02: Do you feel like that's accurate?
4:42
SPEAKER_01: Well, here's the thing that buttons may say, yeah, don't use my data for learning, but here's the reality.
4:49
SPEAKER_01: If you actually read deeper into the the fine print, the moment you provide feedback to the agent, like a thumbs up, thumbs down, I like the answer, or do not do not like the answer, all of a sudden that that bypasses all of those restrictions because it means, oh yeah, I gave feedback to the to to to the model, therefore the model needs to be able to learn from the feedback, therefore it needs to retain the data, therefore it will retain the conversation and possibly the data for up to five years.
5:19
SPEAKER_01: Again, I personally am not super keen on that.
5:22
SPEAKER_01: Some people may be completely comfortable, but hey, look, to each his own.
5:26
SPEAKER_02: Totally understood.
5:28
SPEAKER_02: And the last one that we're gonna hit you with today, since I rattled you with that first one, this one's gonna be an easy one.
5:33
SPEAKER_02: What ridiculous everyday thing would you outsource to AI if it was completely private, you know, nobody would know that AI was doing it for you.
5:44
SPEAKER_02: What would you do?
5:45
SPEAKER_01: Oh, wow.
5:46
SPEAKER_01: Um that's a very good question.
5:51
SPEAKER_01: You know, I think still shopping is still is a one of those things that still takes a lot of time, like trying to find the best item for what I need at the best price at the best moment, right?
6:05
SPEAKER_01: And you know, there's a bit of gaming, like you know, it's always good to find the best possible deal, it's kind of fun, but at the same time, can it but it can become quite confusing to do I buy this, do I buy that?
6:16
SPEAKER_01: And I think having AI do that for me and just provide to me a digested view that I can just choose from, I think is something that would be really good.
6:26
SPEAKER_01: And I think it's getting there, right?
6:27
SPEAKER_01: I think we're getting close to having to have this this capability.
6:31
SPEAKER_01: But again, uh to do it in a private and secure way, I think there may be some ways to go.
6:38
SPEAKER_02: Absolutely.
6:39
SPEAKER_02: And that's one of the concerns that a lot of category managers have is as more people who feel exactly like you, you know, shopping is not uh it'd be interesting to actually do some analysis work on how many people actually enjoy shopping, right?
6:51
SPEAKER_02: And shopping is a broad umbrella, you know, whether it's closed, that's a different story than grocery shopping, which is more habitual, feels a little bit more like a chore.
7:00
SPEAKER_02: A lot of people still getting delivery, right?
7:02
SPEAKER_02: But you still have to manually go in, select your items, go through that process.
7:07
SPEAKER_02: So as we implement AI more into shopping tools, how do, as category managers, we react to AI doing shopping?
7:16
SPEAKER_02: What does that look like?
7:17
SPEAKER_02: And I think that's a really interesting topic we kind of broached on what, two weeks ago.
7:22
SPEAKER_02: Um, that definitely deserves a little bit more thought/slash diving into the mechanics of how that could work.
7:30
SPEAKER_01: Yep.
7:31
SPEAKER_01: Yep.
7:31
SPEAKER_01: And I think it's also, and then maybe maybe I'm jumping ahead a bit here, but if you look at some of the latest news with dots and muse that came out uh recently, um I think that the direction of travel is very clear.
7:46
SPEAKER_01: And I I think that you're already seeing some some of the platforms take very different stances with respect to those um AI bots.
7:56
SPEAKER_02: Exactly.
7:57
SPEAKER_02: And if you are scared about AI taking over the world and doing horrible things and being able to replace you, please read the story about how Muse botched completely a sale on marketplace for someone, had a bunch of people showed up to their apartment like at five o'clock in the morning or super late.
8:14
SPEAKER_02: So we still are a long ways away from AI being perfect in that space, which is great news for us.
8:21
SPEAKER_02: Okay.
8:22
SPEAKER_02: So, but you hit on something that was really interesting, Eve, and that's the AI dots.
8:27
SPEAKER_02: So AI going from answering my question to own this work.
8:31
SPEAKER_02: Talk to us a little bit about what we learned about this this week.
8:35
SPEAKER_01: Yeah, so I think this is a this is just the next a natural evolution from um very simple reactive chatbots to having um always on agents for you.
8:52
SPEAKER_01: So if you go back back in the early days, you know, you'd have GPT and you'd answer, you'd you'd pose a question, you'd get a pretty bland answer, and you couldn't really understand why it would give you the answer.
9:04
SPEAKER_01: Then we had the next evolution was chain of thought.
9:07
SPEAKER_01: So the the agent was able to extract to tell you why it was doing and why it was thinking what it was thinking, right?
9:14
SPEAKER_01: Little dots, and you say, like, oh, I'm doing this, I'm doing that, right?
9:17
SPEAKER_01: Uh, which was very helpful.
9:19
SPEAKER_01: Um, next level up was then to start to have some of those agents that were doing small to medium tasks.
9:26
SPEAKER_01: Hey, go research this topic and come back to me.
9:29
SPEAKER_01: It would go for like an hour or two, come back.
9:31
SPEAKER_01: Next evolution was, hey, group of agents, go and figure this out for me.
9:35
SPEAKER_01: Again, a couple of hours, but then you may have five, six agents.
9:39
SPEAKER_01: Hey, those agents are gonna research those websites, those agents are gonna read those papers, those agents are gonna do blah, blah, blah, right?
9:46
SPEAKER_01: Um, and I think now we're getting to the point where it's a lot more of a, hey, you are my always on agent.
9:54
SPEAKER_01: I need you to monitor my inbox, I need you to always look at an ongoing process, my calendar, my inbox, my shopping habits, my deliveries, my my bank, if you're into this, right?
10:06
SPEAKER_01: And then, you know, warn me when there's something that I should be aware of.
10:12
SPEAKER_01: Right.
10:12
SPEAKER_01: And so at this stage, it's not a here's a task, come back when you're ready.
10:17
SPEAKER_01: It's you know, just give me the uh the UAV, the drone, the ISR that's going to monitor and loiter on target for like days and month on end, right?
10:28
SPEAKER_01: Now, obviously, doing that can be super helpful.
10:32
SPEAKER_01: Um and at the same time, doing that also means that you're gonna consume a lot more tokens, right?
10:38
SPEAKER_02: Definitely.
10:39
SPEAKER_01: You know, always on agents don't come for free.
10:41
SPEAKER_01: There's a lot of reasoning, there's a lot of that stuff happening.
10:45
SPEAKER_01: And so if you are a provider of AI and you're able to charge on a consumption basis, then it's great for you.
10:53
SPEAKER_01: It's absolutely amazing because you you will multiply your consumption tenfold.
10:57
SPEAKER_01: I mean, if you again if you go back, the consumption went about tenfold between the models with the simple models to the models with chain of thought.
11:05
SPEAKER_01: That's about a 10x increase in in token consumption.
11:09
SPEAKER_01: And then after that, the moment you had agents, it's it was another like 10 or 100x consumption.
11:14
SPEAKER_01: And I think with the always-on, then you can always you can sort of see the pattern of more and more consumption.
11:19
SPEAKER_01: So I think Muse and Dots are probably making the Nvidia's of the world extremely, extremely happy right now.
11:26
SPEAKER_01: Because that is absolutely driving more and more demand.
11:30
SPEAKER_02: Yeah.
11:30
SPEAKER_02: And this seems like it's the latest, you know, AI race, if you will, as far as always-on agents.
11:35
SPEAKER_02: You know, we're seeing a lot of information come out.
11:38
SPEAKER_02: At this point, who do you think is winning?
11:41
SPEAKER_02: And what do you think the next evolution of always on is going to look like?
11:45
SPEAKER_02: Because really, this is great for, like you said, those repetitive, multi-step, time-consuming tasks that we use category managers do every single day.
11:54
SPEAKER_02: Pull this, you know, deck and refresh this, pull this POS every Monday.
11:58
SPEAKER_02: So it will help to get rid of that monotonous kind of time-consuming work, but at what cost, to your point?
12:06
SPEAKER_02: You know, that eats up a lot of credits and tokens.
12:08
SPEAKER_02: So talk a little bit about that.
12:10
SPEAKER_01: So it's a very good question.
12:11
SPEAKER_01: So, look, I think Muse has been a resounding success for all the reservations I have about Muse and the data privacy.
12:19
SPEAKER_01: It's been extremely popular.
12:21
SPEAKER_01: And I mean, I I don't know about you, but I'm getting ads from Muse on every all the time, right?
12:26
SPEAKER_01: There.
12:27
SPEAKER_01: And even like we talked about F1.
12:29
SPEAKER_01: I think there was an F1 ad with uh George Russell, the empty Mercedes driver pitching Muse to the world, right?
12:37
SPEAKER_01: Um, so I think I would differentiate between agents that are consumer level and agents for work.
12:44
SPEAKER_01: I think in the consumer level, I think you know Meta legitimately has a good shot, right?
12:50
SPEAKER_01: They're they're they were first on the market with this, they've got good branding, good marketing, they have a lot of information about you.
12:57
SPEAKER_01: Like from everything you look at on marketplace to the friends to the ads to the you know, the moment you spend more than a quarter of a second scrolling on your on your on your Facebook or your meta thing, you know, it will know, so it will know what you like, right?
13:12
SPEAKER_01: So it knows a lot about you.
13:14
SPEAKER_01: So I think that's one angle.
13:18
SPEAKER_01: But I think the consumer version is important but not not huge.
13:22
SPEAKER_01: The biggest question is going to be in the businesses, right?
13:25
SPEAKER_01: When you start to have those agents embedded in the workflows.
13:28
SPEAKER_01: And this is where I think Meta just does not have a chance.
13:31
SPEAKER_01: They just do not have the right to win because they're not a business platform, just fundamentally.
13:37
SPEAKER_01: And this is where I see so OpenAI is making the play, right?
13:41
SPEAKER_01: Right now, they're they're first.
13:43
SPEAKER_01: Uh something tells me that um cloud is gonna come next, but you know, I would actually I would actually think that companies that are fully embedded already in the fabric of organizations, like maybe companies closer to the to the Googles of the world, um, may actually be in better positions to do this.
14:04
SPEAKER_01: Now, mind you, if you start to turn on those agents and you're locked in with one specific provider, then you're locked in, as I said.
14:12
SPEAKER_01: And this can be a massive, massive operational risk for the for those companies.
14:16
SPEAKER_01: We see every day banks saying, hey, we signed this partnership with that provider, this and that, and great, but then you know what happens if that company goes betty up, you know.
14:27
SPEAKER_01: So um it it is a it is a legitimate point of discussion for a lot of companies.
14:34
SPEAKER_01: And you know, if you look at Microsoft, for example, that's taken a stance of being the platform and then serving other models through this platform, um, and they're fully embedded in companies, obviously.
14:48
SPEAKER_01: I mean, they they may have a very good shot.
14:50
SPEAKER_01: Their latest products have not been, I'm gonna call it, very successful, like Copilot hasn't been massively adopted, Scout not very massively adopted either.
15:02
SPEAKER_01: Right.
15:02
SPEAKER_01: Um, so you know, we have to see.
15:06
SPEAKER_02: Yeah, it's interesting because there's a lot of chatter in the AI world that thinks that in true Microsoft fashion, they launched something to launch something, and then the next version or the next, you know, history of it is gonna be so much more significant where they're gonna actually add a little bit of leadership, like thought leadership to AI.
15:26
SPEAKER_02: What do you think about that?
15:27
SPEAKER_02: Do you think it's gonna be more of a Me Too?
15:29
SPEAKER_02: Or do we are we gonna see like a big innovation come from them?
15:33
SPEAKER_01: Look, I I think I think the the playbook for Microsoft has always been to be to have something on the market reasonably soon after the leaders.
15:42
SPEAKER_01: So take Teams, for example, right?
15:44
SPEAKER_01: Um compared to Zoom, back in the early days, Team was nowhere near So bad.
15:51
SPEAKER_01: Yeah.
15:54
SPEAKER_01: And then all you know, little by little they made it better and better, but also I think it they just made it cheaper, they made it part of the bigger, broader package, right?
16:02
SPEAKER_01: So there was a bundling aspect to it.
16:05
SPEAKER_01: And I could actually see Microsoft following the same playbook saying, hey, you know, um, we're gonna give you and they've they're trying it right now with Copilot.
16:16
SPEAKER_01: It's not as good, but it gives you access to a number of things.
16:20
SPEAKER_01: It's maybe already embedded within within your subscription, right?
16:24
SPEAKER_01: So there's gonna be a price race happening there as well.
16:28
SPEAKER_01: It's if I had to guess, I think they they might be in a good position if they can crack that you know broader mass market angle.
16:40
SPEAKER_02: Right.
16:42
SPEAKER_02: Absolutely.
16:44
SPEAKER_02: I think that one of the things you hit on in your research that was exciting for this week is the A16Z.
16:51
SPEAKER_02: I'm assuming is that the made you say it?
16:52
SPEAKER_02: A16Z?
16:54
SPEAKER_01: Yes, it's Andres and Horovitz, yeah.
16:56
SPEAKER_01: Yeah, the uh the VC firm.
16:58
SPEAKER_02: So super exciting that, you know, essentially essentially a gentic AI is becoming a shopper and really shopping the digital shelf, which is completely revolutionary to what we do from a category management perspective.
17:14
SPEAKER_02: It was a huge race to figure out how to actually optimize the digital shelf.
17:20
SPEAKER_02: There was a lot of Intel, millions and millions of dollars spent trying to understand how people's behaviors and their migration pattern shift when they buy online versus brick and mortar.
17:30
SPEAKER_02: Now we're adding a level of complexity to it, right?
17:33
SPEAKER_02: Because you're effectively trying to add behaviorism to something that isn't sentient.
17:42
SPEAKER_02: So you don't really understand how it shops, you know, how it's making those recommendations.
17:48
SPEAKER_02: Talk a little bit about this because this is really should be top priority, top concern for category management.
17:54
SPEAKER_01: I think there's two angles here.
17:56
SPEAKER_01: Angle number one is indeed, you know, how do we think about like a bot doing shopping for you and what does that mean, right?
18:03
SPEAKER_01: But the other fundamental thing, and I'll get back, I'll get back to that point, but I think the real money mover is that for a lot of those platforms, there's a lot of revenues coming from ads, right?
18:16
SPEAKER_01: So Amazon, Shopify, and so on, like they make a ton of money through ads.
18:20
SPEAKER_01: Now, what happens if there's no eyeballs to look at those ads, right?
18:24
SPEAKER_01: Right.
18:25
SPEAKER_01: That is a massive thing, right?
18:28
SPEAKER_01: So being a platform used to be being the marketplace used to mean that you owned not only access to the customers, access to the product, you were a two-sided marketplace, you were able to insert recommendations, insert ads.
18:43
SPEAKER_01: You had a lot of power as that platform.
18:45
SPEAKER_01: Now, all of a sudden, that power is being taken away from you by those bots because you're just becoming a source of information for someone else to make the decision.
18:54
SPEAKER_01: So you're losing that platform power, which is, you know, and and we we've seen that with some of the market responses with the with the market capitalization of eBay and and Amazon just dropping right after Muse came out.
19:10
SPEAKER_02: Absolutely.
19:11
SPEAKER_02: Absolutely insane.
19:12
SPEAKER_02: And hopefully we'll be able to show that chart here that you uh you referenced it.
19:17
SPEAKER_01: Yes.
19:17
SPEAKER_01: Um but so that's the the broader perspective and why there's gonna be, you know, a some platforms are gonna resist massively.
19:27
SPEAKER_01: So Amazon has banned Muse from doing any shopping on their platforms, right?
19:32
SPEAKER_01: Um others are embracing it.
19:33
SPEAKER_01: So Shopify, for example, which powers a lot of small and medium shops, has completely opened the doors to Muse.
19:40
SPEAKER_01: They're like, you know what, muse is great, let's just do it, right?
19:44
SPEAKER_01: Unclear exactly who's right or wrong.
19:46
SPEAKER_01: Um, but going back to your first question about, you know, how do we think about bots doing the shopping?
19:55
SPEAKER_01: I mean, it it becomes again, I would separate in into two different buckets.
19:59
SPEAKER_01: There's the I'm gonna shop from normal shops and then I'm gonna shop from you know individuals.
20:05
SPEAKER_01: Right.
20:06
SPEAKER_01: Like the story that you that that we talked about that you told about about the the guy who had a bot and organized a sale on marketplace and the the buyer showed up unannounced because that's crazy.
20:19
SPEAKER_01: That's insane.
20:21
SPEAKER_02: The bot gave his address out, yeah.
20:23
SPEAKER_02: Negotiated the price down ridiculously.
20:26
SPEAKER_01: That is wrong in so many, so many ways, right?
20:29
SPEAKER_02: Total violation, yeah.
20:31
SPEAKER_01: But I think, you know, I think for the audience, the the the the most interest interesting thing is, yeah, you are if what happens if you use a bot to shop on those on for you on marketplaces, right?
20:43
SPEAKER_01: On the shop.
20:44
SPEAKER_01: So you don't go to the Walmart.com, you don't go to Target.com, you don't really do the research anymore.
20:51
SPEAKER_01: You just say, hey, look, I need to buy a toothbrush for my kids.
20:55
SPEAKER_01: You know, what is the best toothbrush for my kids today and go find the best price?
21:00
SPEAKER_01: And I'm willing to wait six months, right?
21:04
SPEAKER_01: Very I mean, I'm saying that because I've nearly literally done that in the last 24 hours.
21:10
SPEAKER_01: Um and what what will the agent do?
21:15
SPEAKER_01: Well, the agent is gonna go read a bunch of websites and a bunch of aggregators, right?
21:19
SPEAKER_01: Um, from the wire cutters of the world and whatnot, and say, hey, look, these are the top three brands, the top three models for kids, and um, you know, these are the ones that provide you the best bang for the buck.
21:32
SPEAKER_01: Oh, and by the way, this one's on sale right now, or this one was on sale last week, and it looks like it's based on the history, it's probably gonna be on sale again in a couple of days or in a couple of weeks.
21:44
SPEAKER_01: So, you know, my suggestion is you buy this, but you wait, right?
21:47
SPEAKER_01: Um, so what happened there?
21:49
SPEAKER_01: Like, A, I've done zero of the research, I've not gone on any websites that could have shown any any ads for me, right?
21:55
SPEAKER_01: I've not gone on any of the Amazon or Target.
22:01
SPEAKER_01: The agents done that for me.
22:03
SPEAKER_01: I am completely not seeing anything else that you could be recommending for me.
22:07
SPEAKER_01: Like I'm not seeing the hey look, if you buy this toothbrush, you should buy the toothpaste, you should buy some that.
22:13
SPEAKER_01: Not seeing any of that stuff.
22:14
SPEAKER_01: No bundling.
22:15
SPEAKER_01: Absolutely not.
22:16
SPEAKER_01: That's right.
22:18
SPEAKER_01: And also the whole gaming of the TPRs with like, hey, look, the price is dot 99 or something, trying to play with human psychology is gone completely.
22:29
SPEAKER_01: Completely.
22:32
SPEAKER_01: And so now it also means that I think you will see a big differentiation between different categories of purchases.
22:42
SPEAKER_01: One of them is going to be the one the um you know there's going to be discretionary spending and non-discretionary spending.
22:50
SPEAKER_01: The non-discretionary spending meaning you need to buy it today.
22:52
SPEAKER_01: So you just have to you just have to go get it and that's it, right?
22:56
SPEAKER_01: But the for which is you know some, I don't know how much, but some percentage of your spend, of course.
23:02
SPEAKER_01: But then for everything else that you're like, you know what, I just should I need new winter boots because winter is coming.
23:08
SPEAKER_01: Okay, well, I don't need them today.
23:09
SPEAKER_01: I need them probably the next over the next four months.
23:12
SPEAKER_01: I'm willing to wait, right?
23:15
SPEAKER_01: So what does that, you know, it creates a whole other mechanics, right?
23:19
SPEAKER_01: Oh, and also the moment you do that, all of a sudden you also lose some of the impulse buying that you used to have.
23:27
SPEAKER_01: I'm looking for boots.
23:28
SPEAKER_01: Oh, you know what?
23:29
SPEAKER_01: Maybe I should also buy some new socks.
23:30
SPEAKER_01: I shouldn't buy this and that, right?
23:32
SPEAKER_01: And that may completely go away, uh, or at least be massively uh lowered.
23:38
SPEAKER_01: So I think whatever whatever I think about it, you know, if I take a big step back, whatever I think about it is that uh the direction of travel is clear.
23:50
SPEAKER_01: It's gonna happen, whether we like it or not.
23:52
SPEAKER_01: That is what people are gonna do.
23:53
SPEAKER_01: That is just too convenient to pass.
23:56
SPEAKER_01: And so I think the industry just has to adapt, you know.
24:00
SPEAKER_01: If you're not pro if you're not adapting, you're regressing, and I think you know, there needs to be a lot of adaptation at pace.
24:07
SPEAKER_02: Definitely.
24:08
SPEAKER_02: A hundred percent.
24:09
SPEAKER_02: And speaking of adaptation, something else that you hit on this week was really interesting because one of the biggest challenges we've always had in category management, I feel like this is probably very similar in any industry, especially in CPG, is there is a lot of institutional knowledge in people's heads.
24:27
SPEAKER_02: And each account that you work on, especially if you're with a larger CPG company, they'll have a Walmart team, a Kroger team who are dedicated to that account.
24:37
SPEAKER_02: So then you add all that institutional knowledge of how that account prefers to see data.
24:43
SPEAKER_02: Walmart, for example, always goes by Walmart week, not a standard Gregorian calendar week, right?
24:48
SPEAKER_02: So there's a lot of information living in people's heads that if they retire, move elsewhere, it goes with them.
24:56
SPEAKER_02: So the XAI team bots, right?
24:58
SPEAKER_02: So learning more about a team and how a team operates versus just one person, highly interesting from a category management perspective, because essentially you could save that institutional knowledge from walking out the door and really making that category management knowledge become a lot more usable moving forward.
25:17
SPEAKER_02: So what are you what are your thoughts about all of that with the team bot?
25:23
SPEAKER_01: Yeah, so look, so so the team bot, so that's grok's latest um offering, right?
25:30
SPEAKER_01: Just to to live to level set, um, which I think is interesting, um, especially you know the the latest grok models have been extremely powerful, like grok 4.6 that came out probably pretty recently has been uh very, very strong.
25:45
SPEAKER_01: Um I st I don't know that grok is again the best placed to win that war.
25:53
SPEAKER_01: Let me explain why.
25:54
SPEAKER_01: Um I think a lot of companies already store knowledge in other systems.
26:02
SPEAKER_01: Right.
26:03
SPEAKER_01: So emails, Slack, Teams, Confluence, OneNote, Notion, Obsidian, whatever like note-taking approaches.
26:13
SPEAKER_01: Exactly, right?
26:13
SPEAKER_01: CRM and so Exactly, right?
26:16
SPEAKER_01: Um any place where you store communication documentation that you've had with your customers essentially is an accumulation of knowledge.
26:25
SPEAKER_01: And if you have all the decks you've ever presented to Walmart, an agent could look at those and say, oh yeah, there's not a single deck that goes by the Gregorian week, they go by the Walmart weeks, I think you call them.
26:37
SPEAKER_01: So I think that is pretty straightforward to extract.
26:41
SPEAKER_01: And so my my supposition here is that the key here is not going to be about the who's gonna provide the best agent.
26:51
SPEAKER_01: I think the key is gonna be who has that data that can be exploited and synthesized, right?
26:57
SPEAKER_01: Because look, if if Bob, who's on the Walmart team, who's been on the Walmart team for 15 years, wins the lottery and retires tomorrow, right?
27:07
SPEAKER_01: Well, guess what?
27:07
SPEAKER_01: You still have 15 years of emails from Bob, right?
27:11
SPEAKER_01: And you may not have Bob anymore, but you have all of that information.
27:13
SPEAKER_01: You have 15 years of Teams and Slack messages and and all of that.
27:17
SPEAKER_01: And you can probably recreate a reasonably accurate digital version of Bob with email trace.
27:27
SPEAKER_01: Exactly.
27:28
SPEAKER_01: Yeah, yeah, just based on his trace.
27:29
SPEAKER_01: And look, and we're seeing that already, right?
27:31
SPEAKER_01: You already have um a number of people who create their avatars where it you you show to an agent this is everything I ever wrote, my blog posts, my emails, everything.
27:45
SPEAKER_01: So you know my ton of voice, you know what I know, blah, blah, blah.
27:48
SPEAKER_01: So now, if someone comes to me and asks me a question, try to see if you, the agent, can already answer just based on the knowledge you've acquired about me.
27:57
SPEAKER_01: Right.
27:58
SPEAKER_01: And so I think to that extent, um if I if I go back to the point about the the team norms and that sort of knowledge, the winners are going to be again those platforms that own that data.
28:13
SPEAKER_01: So yeah, the notions, the one notes, and so on.
28:16
SPEAKER_01: And because it it's a real problem.
28:18
SPEAKER_01: It's a very real problem.
28:20
SPEAKER_01: Um it you always have the like onboard a new employee on a new team.
28:26
SPEAKER_01: It's never, it's never simple.
28:27
SPEAKER_01: Even the best companies struggle with that.
28:29
SPEAKER_01: And you know, it's one of those things that can be so much easier and so much better and get you more productive very quickly.
28:36
SPEAKER_02: Absolutely.
28:36
SPEAKER_02: And I think it brings up, you know, I agree, Grok isn't necessarily the platform to do it, but does it make sense to bring some kind of knowledge base or store a knowledge base with AI because of turn and burn, right?
28:54
SPEAKER_02: So instead of saying, oh, you know, I haven't worked on this account, let's go ask Susan, do you instead ask an AI agent and do training that way and use that as your knowledge repository instead of human brain, right?
29:10
SPEAKER_01: So well, look, I think if we believe that, and again, I think that's the direction of travel, if we believe that each one of us will have an always-on agent dedicated to helping us do our job or in our lives, right?
29:27
SPEAKER_01: Um if we believe that that's true, then it becomes very reasonable to say, before I bother the real human being, let me talk to that person's assistant to see if I can get a quick answer, right?
29:39
SPEAKER_01: Hey, I know that Bob is on vacation, right?
29:43
SPEAKER_01: But how would Bob present this sort of information to Walmart?
29:48
SPEAKER_01: Or how has Bob presented that information to Walmart historically?
29:52
SPEAKER_02: Right.
29:52
SPEAKER_01: Right.
29:53
SPEAKER_01: And then the agents should be able to go back to all the decks that Bob created and blah, blah, blah, blah, right.
29:58
SPEAKER_01: Um, and so I I think that is a very real thing that's I I think it's happening right now.
30:07
SPEAKER_01: Like it's not a and and this this is not a oh me, it may happen in the next five years.
30:11
SPEAKER_01: I think I mean we are working on some systems ourselves right now to build those those personal agents.
30:19
SPEAKER_01: Having being in a world where you can ask someone's agent a first stab at an answer is a something that's going to happen in the next six to twelve months, not five years.
30:28
SPEAKER_02: Right.
30:30
SPEAKER_02: Absolutely.
30:31
SPEAKER_02: Absolutely.
30:32
SPEAKER_02: I think another thing that was really interesting about this week so you and I self-admittedly geek out anytime there's a new model released by AI platforms.
30:41
SPEAKER_02: I wasn't necessarily super excited about Sonnet 5.5 because of it being a powerful dynamic model.
30:49
SPEAKER_02: What was more exciting is the economics of it for me.
30:52
SPEAKER_02: So 30% faster, 30% cheaper.
30:55
SPEAKER_02: Definitely a lot more palatable from category management perspective when you're doing those deep analytical pieces and you need to burn a lot of tokens.
31:04
SPEAKER_02: Talk a little bit about what you think Sonnet 5.5 kind of challenges the AI market to do.
31:10
SPEAKER_01: Yeah, so I think, you know, if if I'm gonna I'm gonna be a good politician and I'm not going to answer your question directly.
31:18
SPEAKER_01: I think what the American people really want to know.
31:20
SPEAKER_01: No, but I think I I think Sonnet 5.5 is is just one example of this compression curve that we're seeing.
31:32
SPEAKER_01: So the the top end models today, so like the Astras, the the Fables and whatnot are super expensive.
31:40
SPEAKER_01: Like you can you can you can burn through your allowance of tokens in a few minutes if you're not careful.
31:45
SPEAKER_01: Right.
31:46
SPEAKER_01: Yes.
31:48
SPEAKER_01: And what is also very true is that you will probably have access to that level of intelligence for about half the price, or maybe a tenth of the price, in the next six months.
32:00
SPEAKER_01: Right.
32:01
SPEAKER_01: And so I mean look, we're we're already seeing that Sunnet 5.5 is nearly as good as um Opus 5.5, right?
32:11
SPEAKER_01: Um so what does that mean?
32:14
SPEAKER_01: Well, I think it means that we're going to see a lot of tasks that were still difficult or too expensive to run through an LLM become all of a sudden more economical to do so, to run through an LLM.
32:29
SPEAKER_01: And the other thing that it's that it's telling you or should tell you is that given the speed at which the industry is moving, being locked in one model is a really, really bad idea because look, your your Sunnet 5, yeah, it was pretty good.
32:43
SPEAKER_01: Sonnet 5.5 came out much better, much, much, much better.
32:46
SPEAKER_01: But also, guess what?
32:47
SPEAKER_01: Your Sunnet 5 is also likely to become um turned off in the next 12, 18 months, right?
32:56
SPEAKER_01: So, you know, there's a real risk if you are not if your processes are so tailored to a specific model that they will break if you change.
33:08
SPEAKER_01: Because it you will have to change, it's guaranteed.
33:10
SPEAKER_01: And I think that the pace of change is accelerating.
33:13
SPEAKER_02: Yep.
33:14
SPEAKER_02: Speaking of accelerating, so our last one that I found super interesting, which is ironic for me, because usually I hear banking and I get this sudden existential dread look on my face and and zoom out.
33:26
SPEAKER_02: But Barclays and Claude working together.
33:30
SPEAKER_02: And really what was surprising was scale for me.
33:34
SPEAKER_02: The ability, what was it, 120,000 emails a day, 16,000 users?
33:39
SPEAKER_02: I mean, just astronomical when it comes to workflow optimization and harmonization.
33:46
SPEAKER_02: So interestingly enough, right, what I was interested about to kind of get your perspective of is banking is such a regulated industry.
33:55
SPEAKER_02: Very surprised that they would kind of be one of the pioneers to kind of use Claude as an operating model from an agent agentic perspective.
34:06
SPEAKER_02: So talk a little bit about that, what your thoughts are.
34:09
SPEAKER_02: You think we're gonna see a lot more forward integration as far as using, you know, a certain model?
34:14
SPEAKER_02: And do you think that some of these the other challenge that I was thinking about is because Barclays didn't marry Claude, does that set them up in the future to always have to work with Claude?
34:27
SPEAKER_02: Or do you think it's a situation where the model can be agnostic in some way?
34:32
SPEAKER_01: Well, there's a lot of questions to answer here.
34:35
SPEAKER_02: I just pepper it you with a lot.
34:36
SPEAKER_01: Um So, number one, yes, the the banking industry historically has been beh a couple of generations behind, right?
34:44
SPEAKER_01: 100%.
34:45
SPEAKER_01: Even a year ago, 12, maybe 18 months ago, you know, a lot of the bankers that I know told me, hey, look, we're just we're stuck using Copilot because we're not allowed to use anything else, right?
34:57
SPEAKER_01: I think I think the the, you know, and that was due to the heavy regulation that they that they are under, which makes sense.
35:05
SPEAKER_01: I think now the banks, all the banks are realizing that they have to move on, they have to evolve, right?
35:10
SPEAKER_01: And I think all the banks are using AI more and more, right?
35:15
SPEAKER_01: So whether it's through a direct provider or through an aggregated provider, um, I know for a fact that some banks are also creating their own layers on top of the model.
35:29
SPEAKER_01: So basically um they're creating an interface where you can consume the API of the models, um, but where you leave no trace on the model side.
35:40
SPEAKER_01: So it's much more secure, you're fully in control, you can swap things in and out, but you need to maintain the system, so it's not necessarily easy.
35:47
SPEAKER_01: Um I think the cat's out of the bag now, right?
35:52
SPEAKER_01: All the banks are using AI, they're gonna use it more and more.
35:54
SPEAKER_01: I think that's absolutely true.
35:56
SPEAKER_01: Um I think the they've clearly the the legal councils have and the the CTOs of the bank have absolutely found a way around the regulation and the compliance around it, right?
36:09
SPEAKER_01: Um I also think that no, if you're like a claude or a Chat GPT, being able to say that you have a a very large client that's signed in for like a one or two-year deal is massively attractive.
36:26
SPEAKER_01: Yeah.
36:27
SPEAKER_01: Because what did what I know about and what we've learned from uh Anthropic's IPO prospectus is that um a very large fraction of Anthropic's revenue is tied to a very small number of companies.
36:42
SPEAKER_01: Right.
36:43
SPEAKER_02: Absolutely.
36:44
SPEAKER_01: And to that extent, if those companies go away, they're in real trouble.
36:48
SPEAKER_01: And so there's a real incentive to keep those companies on the platform, giving them good deals, massive discounts in exchange of a long-term commitment.
36:58
SPEAKER_01: Um and so for the banks, I think it's a and look, and and the bankers are smart people, right?
37:06
SPEAKER_01: So I think none of them is gonna be, or at least I hope, none of them is gonna be uh cornered into being stuck with one single model and being held hostage by a provider.
37:18
SPEAKER_01: I don't think they would ever let that happen.
37:20
SPEAKER_01: I think their strategy is very much to say, look, we we will make a partnership with a model to try to get as much of a good deal as we can, and we also need to have a a side route, an exit route if things go badly.
37:34
SPEAKER_01: Because otherwise it becomes a major, I mean, this is a life or death operational risk for the business at that stage at that stage.
37:44
SPEAKER_02: Absolutely.
37:46
SPEAKER_02: And I think it's really interesting just from the past two podcasts that we've done, where we're seeing this AI modernization, integration, forward integration, whatever you want to call it, from using AI as a tool to a teammate, to an infrastructure kind of you know, knowledge base, and then to a shopper and an intermediary, which is really the piece that we all have to put our brains together and figure out.
38:12
SPEAKER_02: And to, you know, everyone's credit, everyone's trying to figure it out in real time.
38:19
SPEAKER_02: No one really has it locked down yet.
38:21
SPEAKER_02: Hopefully, we can help provide some guidance to that.
38:24
SPEAKER_02: But that being said, it is coming on the horizon.
38:27
SPEAKER_02: We're gonna have to continue to watch out for these developments.
38:30
SPEAKER_02: I mean, this is just two weeks since you and I spoke that all of these new things are coming to the head.
38:35
SPEAKER_02: And it's just crazy how fast tech is moving right now.
38:39
SPEAKER_02: So I know that a lot of people are excited that we're doing this podcast because they don't have time to stay on top of what's going on in the AI world.
38:46
SPEAKER_02: So we're super happy to help there.
38:49
SPEAKER_02: And what I'd like to end on is we actually got an email from one of our tens of fans of asking a question about AI.
39:00
SPEAKER_02: And I think that it's really applicable from what we're talking about.
39:04
SPEAKER_02: So they write, hello, Eve and Chelsea.
39:08
SPEAKER_02: Hi there.
39:09
SPEAKER_02: Really love your podcast so far.
39:12
SPEAKER_02: Love your your chemistry.
39:13
SPEAKER_02: Thank you for that.
39:14
SPEAKER_02: We appreciate that.
39:16
SPEAKER_02: My industry is obviously category management.
39:20
SPEAKER_02: Love that once again.
39:21
SPEAKER_02: I'm feeling pressure from my company to use AI.
39:25
SPEAKER_02: They really want us to integrate AI in everything that we're doing.
39:29
SPEAKER_02: I have two questions.
39:31
SPEAKER_02: Number one, it feels like a double-edged sword.
39:34
SPEAKER_02: At some point, when will my company say we're using AI for too much?
39:38
SPEAKER_02: We potentially don't need you.
39:40
SPEAKER_02: Secondly, do you think that my company is monitoring my AI chats?
39:46
SPEAKER_01: Well, okay.
39:49
SPEAKER_02: That's a multi-layered question.
39:51
SPEAKER_02: So thank you for your what for watching.
39:54
SPEAKER_02: Um, Eve, you take that first and then give me some time to think on that one.
40:00
SPEAKER_01: I'll I'll start with the second question.
40:03
SPEAKER_01: Is your company monitoring what you're doing?
40:05
SPEAKER_01: The answer is hell yes.
40:06
SPEAKER_02: Absolutely.
40:08
SPEAKER_01: Absolutely.
40:09
SPEAKER_01: At least from a volume perspective, right?
40:11
SPEAKER_01: At least from a span perspective.
40:14
SPEAKER_01: Um and no query that you make in one of those tools, of course, is logged and can be audited, right?
40:22
SPEAKER_01: That's just part just very much like your emails.
40:25
SPEAKER_01: Right.
40:26
SPEAKER_01: Like your work email does not belong to you, it belongs to the company, right?
40:28
SPEAKER_01: So same thing for the chats and the queries.
40:31
SPEAKER_01: So 100% you're being monitored.
40:34
SPEAKER_01: Uh how actively you're being monitored, I think is going to depend on each company, right?
40:39
SPEAKER_00: Definitely.
40:40
SPEAKER_01: Um, but it's also very true that there was a trend of token maxing in some tech companies where you know there was a clear incentive or a clear cool factor to be able to say, hey, look at all the tokens I've spent.
40:53
SPEAKER_01: And there were teams that said, hey, if you if you don't spend enough tokens, it's because you're not using enough AI and you're in trouble.
41:01
SPEAKER_01: Right.
41:02
SPEAKER_01: So absolutely the monitoring is in place.
41:05
SPEAKER_01: Now to the other question, you know, is there a risk that I will lose my job?
41:10
SPEAKER_01: Well, I think the real question is, is the job going to evolve?
41:14
SPEAKER_01: And the answer is yes, 100% the job will evolve, right?
41:17
SPEAKER_01: Um and so I think the the reality is that um can you help?
41:25
SPEAKER_01: My advice to that person would be can you help shape the next generation of the role that he or she has so that that role becomes um AI enabled, like do not fight against AI, it's gonna be there, it's gonna stay, it's gonna grow.
41:39
SPEAKER_01: But what can you as a human being add that the AI cannot do?
41:42
SPEAKER_01: And can you make sure that you become the best at using AI because then you know you become very well positioned very much so to stick around.
41:53
SPEAKER_02: Yeah.
41:53
SPEAKER_02: Your job is to not make your job obsolete.
41:57
SPEAKER_02: So you still need to be in, and this is something that you and I talk about a lot.
42:03
SPEAKER_02: Use AI to push off almost as a lackey, right?
42:07
SPEAKER_02: To push off all the work that you don't want to be doing, but also as a strategic partner.
42:11
SPEAKER_02: But at the end of the day, you are still the brain, right?
42:14
SPEAKER_02: So you should be driving the strategy, you should use AI as a sounding board, but not necessarily to develop your strategy, just kind of more as testing hypotheses using AI is really how I would kind of look at your role.
42:29
SPEAKER_02: Will it unfortunately supersede analytical roles?
42:34
SPEAKER_02: Potentially, absolutely.
42:36
SPEAKER_02: You know, we see AI getting better at math, we see AI getting better at reasoning, it's learning as it goes along, and you are helping your AI model at your company learn.
42:46
SPEAKER_02: So just kind of think of it that way.
42:48
SPEAKER_02: I think that to Eve's point, though, it's an innovate or die situation where you have to adopt it.
42:57
SPEAKER_02: And if it's being pushed down by your company, that's not something that you want to retaliate against.
43:02
SPEAKER_02: You want to use it.
43:03
SPEAKER_02: But I do understand the thought process of walking a fine line.
43:07
SPEAKER_02: If I'm using AI to do everything, what's the point in having me at the organization?
43:12
SPEAKER_02: So I think just like Eve said, the point is to be really strategic about where you use AI in your career and where you don't.
43:19
SPEAKER_02: But you do have to use it.
43:21
SPEAKER_02: And it can, if used correctly, right?
43:25
SPEAKER_02: It's that once again, that kind of double-edged sword.
43:28
SPEAKER_02: If used correctly, it will help you elevate yourself and make yourself relevant from a perspective of, hey, we need this person.
43:35
SPEAKER_02: They're really good at AI.
43:37
SPEAKER_02: They can help shift and you know, develop the strategy for us of how to implement AI in different aspects of the business of category management.
43:47
SPEAKER_02: And that way your job's secure.
43:50
SPEAKER_02: So it is, I understand the trepidation almost is the word I want to use in implementing AI into your daily life from a business perspective.
44:01
SPEAKER_02: I do.
44:01
SPEAKER_02: However, that's kind of the crossroads we're at right now, right?
44:05
SPEAKER_02: So thank you for watching our podcast, like we said.
44:09
SPEAKER_02: Um, we will be back, you know, assuming that once again we escape nuclear winter, that AI doesn't break out and somehow figure out the cheat codes, even though A was uh Eve was telling us an interesting story about what was it, ChatGP, was it open agent?
44:24
SPEAKER_01: Yeah, like some some open AI agents hacking government websites and finding ways to escape the sandboxes they were locked into.
44:32
SPEAKER_02: Right.
44:32
SPEAKER_02: Luckily didn't get any personal information, so we're told.
44:36
SPEAKER_02: But that's we know of.
44:38
SPEAKER_02: That we know of, exactly.
44:40
SPEAKER_02: But you know, just like we said, AI is evolving, and we will be back in two weeks with more intel around how you stay on top of the news and continue to implement it in your everyday life.
44:52
SPEAKER_02: Eve, thank you so much.
44:54
SPEAKER_02: You're great to talk to as always, and we will see you next time.
44:58
SPEAKER_01: Thanks, Chelsea.
45:00
SPEAKER_02: Thank you, Eve.