Who Wins When AI Makes Knowledge Work Cheap? With Sangeet Paul Choudary

Speaker A: Not just cost of knowledge, but cost of knowledge work, because that's the advantage with agents, that agents can also execute the work on your behalf. And that makes it really valuable because information was accessible in the past as well. But usable information and actionable information, which can be actioned to get to a goal and accomplish a certain goal, was not. And that is what AI addresses.

Speaker B: My name's Mike Landa and you're listening to Higgle, the B2B Sales Club podcast, where we bring you actionable insights about sales, RFI negotiations and difficult procurement discussions from sales leaders, brand leaders, and procurement leaders. Please subscribe to get updates when new episodes are released. Sangeet, thanks ever so much for joining me on Higgle, the B2B Salesforce podcast. It's a real pleasure to have you.

Speaker A: Thank you, Mike. It's a pleasure to be here.

Speaker B: So, having read the book and followed you a lot on LinkedIn, I think this should be a very interesting conversation for our listeners about what AI is doing to the future of work. So, before, before we kick off, very simple. Who are you? What do you do, and, um, what's your favorite song? And, um, why?

Speaker A: I'm Sangeet Chaudhary. I've written a few books on how technology changes industries, how technology changes the economy. My first book was Platform Revolution, which talked about how the confluence of mobile, data and cloud was leading to the rise of platform business models. And my new book is Reshuffle, which looks at how AI, and particularly the latest round of generative and agentic AI, how all of that is restructuring the knowledge economy. So the question is that I'm really fascinated with. With are, ah, why do systems exist the way they do today? How does new technology restructure them? And then who wins and loses in the new games that emerge once the system gets restructured? So that's a little bit about my work on the question of, um, you know, songs. There was a time when, you know, early in my life I was big time into Bon Jovi's 80s albums. And so, yeah, yeah, I still am.

Speaker B: Yeah, yeah.

Speaker A: I mean, I still am. Yes. But since you said the favorite song, I think there were a bunch of songs that he wrote which were full of hope at the time, which had an impact on me. And so I would say Keep the Faith is probably right up there.

Speaker B: Yeah, it's a great song. Absolutely. Our, uh, son, who's 14, he's into all sorts of music, but of course he's picked up some of my collection. I collect vinyl and we've got all sorts of 70s, 80s rock in that collection. So I'm sure we could talk for a long time about songs, but we won't. We'll talk about how AI impacts industries. Do you want to first start off sangeet, as we discussed? It's an organic conversation, but let's frame it around something simple at the beginning. So an introduction to Reshuffle. So for listeners who haven't read Reshuffle, can you explain the core thesis and specifically, uh, around that kind of system constraints, control points, unbundling of roles and tasks and why, you think? Maybe just to add in, why is a lot of the current thinking missing the point?

Speaker A: Yeah, I mean, let me take a slightly adventurous take on that. Since we just talked about music.

Speaker B: Excellent. Definitely.

Speaker A: Since we just talked about music, I don't want to abandon music right away. So I want to talk about three lessons from what happened to music when we went from physical distribution to digital distribution. Right. And these three are not necessarily the most obvious implications or the m. Most obvious reads of what happened around that time. But that's what digital did to music and to content around that time. AI, ah, is doing something very similar to knowledge work right now. So I want to kind of link the two things a little bit together. The problem as you mentioned, you know, is that we don't think of AI as changing industry structure, changing the way our organizations should work. We just think of it as how did it help me with my work? How do. Does it get something, you know, automated or done quickly? Exactly right. How does it augment me things of that sort? But let's go back to music. So a few different things happened around that time. You know, the first thing was that physical distribution of music was structured around a constraint, which was that selling music through stores was expensive because you had to stamp it onto a CD or a cassette before that and move it into stores. And so it made sense to sell 10 songs even if the buyer just wanted to buy one. Right. So that was the logic of the album, because the buyer would not pay a premium for one song, but they would pay if they thought they were buying 10 songs, or at least that the industry could convince them to do that.

Speaker B: And so, so sangeet in that system, where was the control point? Who had the control, the power in that system?

Speaker A: Well, so there are two pieces over there. What's the constraint? There are two constraints over there in that system. One is the constraint around production, which is, you know, how do you get to studio quality music? But then the second and major constraint was distribution and Access. And so labels and with, you know, in partnership with retailers, kind of own both of those constraints and hence own the control points associated with managing. Right, yeah. And, you know, something very interesting happened around that time when digital music came in. The constraint of physical distribution went away so you could get onto Napster and download a single song. So the cost of distributing music overnight fell to zero. So the logic of the previous bundle, which was the album, the album was a bundle which was structured around the constraint of distribution, that logic went away overn. Right? And so what happened was music got unbundled, and that's a very important phenomenon over there.

Speaker B: So just explain music got unbundled for the listeners. Just expand just so that everyone understands.

Speaker A: Yeah, absolutely. So in the past you wanted to access music, you had to buy the whole bundle, which was the album. Now you could just go onto Napster for those who lived through that time, and you could just download songs all night on Napster and then later on other peer to peer sharing sites. And essentially you were not forced to pay for 10 songs. Well, initially you did not have to pay even for one song because of pilots, but you could essentially download songs at will. Right? So that is the idea of unbundling, that the individual song got unbundled from the album and was free to travel. Right. Now, there are three interesting implications over there, which is what I want to talk about. The first is that unbundling does not help anyone make money. It helps the consumer in a big way because you are no longer tied to the old bundle. So everybody was happy on the consumer side, but then the labels were not happy and nobody else was making money. And that is where what then happened was that Steve Jobs came in and decided that the previous constraint had been removed. There were new problems with unbundled pirated music. You know, you could get a virus on your computer. You did not have an easy way to play that song. You could not easily move it onto an MP3 player. And, you know, there were the legal issues associated with it as well. And simultaneously, the music industry was keen to figure out, well, how do we make money over here? So what Steve Jobs did was he rebundled music around a single device, which was the ipod, and created an integrated experience of consuming music. But you did not need to buy the album. You could buy individual songs and hence you could now create your own personalized bundle, which was your playlist. Right? And until you get to the point where you rebundle, you don't make money in the new system. And that's What Steve Jobs did with the ipod, and the test for that is that the ipod is the only consumer music player, you know, in history to charge such a premium. Because music players in general were, you know, they were commodities. But it was really the fact that this was a player integrated into complimentary music store, the itunes store, and that was the only place you could get music and access the world's music. That is what enabled him to charge a premium for the ipod.

Speaker B: So when Apple owned the physical player that you used, so they got a one off revenue stream from buying the ipod. But the value was really in and access to the media platform that had all of the content. And that I think from memory was a subscription.

Speaker A: Yeah, correct. And you could purchase music at 99 cents a song. And so consumers were.

Speaker B: I remember, right, yeah, yeah, I remember that. Yes, yes, yes, yes, yes, yes.

Speaker A: They were paying 10 to 20 to $30 for albums and now suddenly you could get it for 99 cents. And so there are two or three things that are really important over here, right. One is constraints determine the shape of the system. So the previous bundle was the album, the new bundle was the playlist. Constraints shifted. Right. The second thing is that we often assume that when the system moves, the old gatekeepers will be completely removed. And um, there's a very interesting surprising story over here which I want to bring back to AI as well. But there were two gatekeepers in the previous model which we talked about, the labels and the stores, right. So the stores went out of business, you know, Radio Shack and everybody else. So they lost in the new model. But the labels, they actually became stronger because over the years there were a whole range of labels that had acquired artist licenses and all of them were going bankrupt and they were selling for pennies on the dollar. So three labels picked up all the artists licenses at pennies on the dollar without having invested all the money in, uh, producing the music. So they became really powerful, they owned the new control point. And so today what you see is, you know, if you look at Spotify, really the only winner with Spotify are the labels because they own the control point. They own that whole library of music.

Speaker B: Let's just go a layer beyond that as well. The artist. So what happened to the artist in that kind of 20 year period of transition from physical medium into digital? They produce the content. So you'd have thought they should be winners.

Speaker A: Yeah, absolutely. So again, you know, we haven't come back to AI yet, but this is, doesn't matter.

Speaker B: We'll come back to that in A minute.

Speaker A: All of this has really interesting implications for how you should think about your work, your industry and so on. What happened to the artists? Right. What happened to the artists? The first thing that happened, and I talk about this in Reshuffle as well, the first interesting thing that happened to the artist was that the artist realized that in the past their core product was the album and the complimentary product was the concert. So they would.

Speaker B: Exactly.

Speaker A: They would perform to sell albums, right?

Speaker B: Yes. Yeah. You went on tour because then people went, oh, I'll buy the album. And then they go the next day and buy the album. It was a marketing mechanism.

Speaker A: Absolutely. And if you look at the 10 years or the 15 years that followed since then, the core product became the concert, became the concert. What happened was the album or the music got commoditized. So when something gets commoditized, value shifts to the complementary capability. And that's really important today with AI as we'll get to shortly as well. But that's really what happened. And you know, it goes all the way to what you see with Taylor Swift's Eras tour. Now that's like taken that all the way to the end, that value has shifted all the way to this really huge rebundled concert experiences, if you will.

Speaker B: So now what happens as an artist? And I guess the problem is it's only for the premiere artists, the ones at the top of the tree is songs are free, uh, effectively on Spotify, yes, we subscribe, but you can have any song you like 24 hours a day. So the artist records a song that becomes someone's popular. The artist then puts on a concert tour because the audience likes the song, so they go to the concert. But now at the concert, the concert tickets are a premium. So you make money actually on concerts where you never used to make money, I don't think on concerts. So now you make money from the concert and the merchandising that goes around that and the branding associated with the concert post the concert, that all becomes the new value bundle.

Speaker A: Yeah, absolutely. And there are, you know, complementary shifts that have happened that enable that. Earlier the label used to bundle both the creation of the music and the distribution of the music and management of the talent and essentially all opportunity that came your way. Now, artists can have their own social media following and so that helps them build a fan base which they can redirect to the concert and redirect to whatever they launch on the side in terms of merchandise and so on. So all of that plays a role as well. Having said, all of this music is a unique category which is very winner take all in the sense that music, you like a song, you keep listening to it. So you don't necessarily want to keep discovering new songs every day versus general content. On TikTok, you rarely watch the same thing twice. You want to keep moving to the next thing and the next thing. Right. So the creator economics are very different from that perspective, which means that it's more of a winner take all. That's a separate topic, but that's, I just wanted to call that out about music.

Speaker B: So on that topic we will get to AI eventually. But if you look at the influencer, uh, economy and that now is a very vibrant, valuable economy, the influencer economy within marketing, where do you think that's going in terms of constraints and control points? Because that instant consumption and that disposable consumption I think is part of the challenge.

Speaker A: Yeah, well, when you think about the influencer economy, are you talking about the creator's perspective?

Speaker B: Are you talking about the creator's perspective? Yeah. So I think if a brand would obviously before use advertising and they'd use SEO and all sorts of different channels. But of course now the influencer channel has become very valuable. Consumers like to tune into influencers that have authenticity, that can carry the brand and the brand image. So but that influencer economy is vast. So who makes money in that industry in that system? Is it the brand that makes the money? Is it the influencer that makes the money? Who's got the control? I think it's because it's new. It's quite an interesting. By the way listeners, we didn't script any of this as you can tell. So I don't know what Sangeet's going to say, but I have a lot of audience who is marketing agencies. So they'll be interested in what's happening with the influencer, uh, economy.

Speaker A: Yeah, for sure. So I think, you know, the thing is that when a new economic system emerges, we try to put our uh, previous lenses to it, we try to put our previous logic to it and then we soon realize that there's a lot more to be done than the previous logic. So if you look at the influencer economy itself, right, the first port of call was to just say, well, it's attention, so let's advertise. And so, you know, you had advertising based models and influencers realized that this is interesting, but not as interesting because if you can really amass a big base then you can have brand partnerships and sponsorships. So you know, again, in the next version of it was the brand partnership and sponsorship. But that's also a previous era logic that is being transferred into this new era. Right. What has been happening for the last five to seven years is that there have been two convergences or two trends converging that are changing how the influencer bundle will be monetized if you will or will make sense. Right? So one is that the more the influencer economy proliferates, the more customer journeys get fragmented. What I mean by that is customers start their customer journey with a brand at an influencer right now, right? So for instance, if you are a hotel, you know, in the past you would work with a TripAdvisor but then you would only get to serving active demand. You would never get to serving latent demand which is I'm m dreaming of traveling.

Speaker B: I'm thinking maybe I'd love to go to India. I'm not sure when exactly. I wonder what it would be like, what could I go and see and do?

Speaker A: Correct. And so the influencer uh, catches somebody at the top of the funnel over there, right. And is uniquely positioned to convert latent to active demand because of the nature of their content persuasion and so on. Alongside that, generative AI has dramatically collapsed the cost of creating content and hence creating, you know, marketing material if you will, at scale, at scale in an automated way. And hence highly personalized specific marketing collateral can be created per influencer, per, you know, content type or topic at scale. And so the real opportunity for brands today is to bring the two things together which is your customer journey that's starting elsewhere. Marketing collateral can be created at scale. And so the constraint is how do you bring these two things together which is where a whole range of new platforms are coming up which are kind of trying to bring these two things together. We'll help to generate the marketing collateral at scale, we'll link it back to the commerce and we will also link it back to the influencers network. So take an example, you know, global distribution systems, right? Which in the travel industry that they distribute the hotel inventory to a ah, booking.com today. Tomorrow they can partner with the influencers by creating a platform like this and distribute the same inventory to an influencer because they own the marketing collateral generation and the commerce engine behind it. And so that's where this is really heading, to really do personalized commerce at scale.

Speaker B: Interesting. And who do you think again, how does the influencer uh, win in that game?

Speaker A: Well the influencer wins in that game because they are closest to the transaction than they've ever been. Right. Think of advertising, brand partnerships and now commerce advertising. They were farthest away from the transaction. It was transaction agnostic, it was just attention brand partnerships, they came closer to the transaction, but they were still at the point of brand creation, etc. Now they are in the transaction, so they get a cut.

Speaker B: If it's a commerce platform, they could do revenue share models.

Speaker A: That's the idea of a platform like this. You know that if you can create a platform that links, and there are companies that are doing this, that links the commerce engine at the back end with the marketing collateral generation with the influencer at the front end, then you're essentially getting a percentage. You know, essentially it's a distributed travel agent model. Every Instagram influencer is a potential travel agent, Right?

Speaker B: Yes, exactly. And now the influencer genuinely gets value for the audience that they've built because they're part of the commerce transaction.

Speaker A: Yes, yes.

Speaker B: So let's move on to the book, Sangeet, because that's a nice intro into the book, because the book is effectively a, uh, I'm guessing, but it's an assimilation of all of that thinking into a number of theses, concepts, principles backed up by really interesting illustrations. So I thought the container illustration was brilliant. So basically, anyone that's not read the book, what Sangeet cleverly does is lays out how the system works and the constraints, control points, but then illustrates it with real stories about what happened in real industries. So just talk us through the kind of the core thesis of the book.

Speaker A: Yeah, absolutely. So we talked about music, I'm happy to talk about the container, but, uh, working across all of this is this simple idea that the way our industries, our firms, our workflows, our jobs are structured today is on a certain logic which made sense in, you know, in the previous regime of technology and assumptions. Right. And when those assumptions change because technology changes and technology dismantles a previously existing constraint, then the whole system restructures. So giving, you know, two quick examples, you know, with music, we talked about physical distribution. With a constraint, once that gets removed, you now have the ability to charge a premium on the ipod in a way that was not possible in the past in any other, you know, scenario. And if you look at the container example that I gave, essentially what I say is that prior to the shipping container, global logistics was unstructured. And so you could ship in bags and barrels, which made logistics very unreliable because you could not move things fast at the port. And alongside that, you could not move things seamlessly between trucks. Trains and ships. The moment the container came, the ports got automated, but then the trucks, trains and ships agreed on a common format. So logistics became the liable, which then removed logistics as a constraint and hence created global supply chains and opened up this era of globalization. Right. And with AI, we're in a very similar point because I believe that most of the knowledge work industry, and I would argue that most value in most industries today is associated with knowledge work. Even if you take an industry like mining and metals, the value is not in the excavation, the value is in the exploration, which is a knowledge work component.

Speaker B: It's knowing where to dig. It's not digging.

Speaker A: Exactly. And predictive AI helps you know where to dig because it can work through that and validate that. That. So the key constraint or the assumption was that access to knowledge work is expensive. Right. Whether you know, it's the mining example, but you know, a simpler example back home is just that since we're all in the knowledge work industry, it's access to highly trained talent, highly skilled talent was expensive. You had to hire somebody who was expensive and then you had to keep training them. And so essentially accessing some forms of knowledge work was expensive. And increasingly, as AI improves, many forms of knowledge work can now be more cheaply accessed. And so the constraint falls.

Speaker B: That's the key point in a very simple way. In my simple thinking before access to knowledge, expertise was expensive and it was human based, it was tacit knowledge, often in our heads. And you hired an expert to find out where should I go digging. Now the cost of knowledge is collapsing because as the models are trained on the world's knowledge assets, then I can ask any of the LLMs very similar questions. It may not get the exact answer, uh, but it gives you a pretty good starting point. So the cost is going to zero.

Speaker A: Yeah, absolutely. I would just embellish that a little bit and say it's not just cost of knowledge, but cost of knowledge work. Because that's the advantage with agents, that agents can also execute the work on your behalf. And that makes it really valuable because, uh, you know, information was accessible in the past as well. But usable information and actionable information, which can be action to get to a goal and accomplish certain goal, was not. And that is what AI addresses. So yes, that becomes very cheap to access.

Speaker B: So what happens next, Sangeet? So I'm from the consulting world. I've been involved in consulting for 30 or so years now. And obviously in its simplistic form you would say, well, what's the need for consultants anymore? Why would someone pay for an expensive consultant? Where do you think that economy is going? The whole knowledge worker, uh, consultancy, professional services economy going?

Speaker A: Yeah. So there are three or four things that happen. The first thing that people assume happens is that AI either automates work and takes jobs away, or it augments people and makes them better at their job. But that's a very another way of looking at it. I'll just take an example over here. Why it's not as simple as automation and augmentation. Think of translation, right? Translation is knowledge work, right? And you would expect that. Even if AI is really bad at everything else, at the end of the day, it's a large language model. So the one thing it should do really well is translation. It understands language, right? So it should do translation really well. What we don't understand is that the act of translation, translation is actually a bundle of different activities of which language mapping is only one activity. That is the only thing that AI takes. And so you cannot just automate the rest of the activities. The rest of the activities are understanding the context, understanding sensitivities, which ones are risk bearing versus not, and then translating with all of that in mind. Now the challenge then is that what happens to a job is not that AI takes it away or makes it better. What AI does is it unbundles the job. Translation is a classic example where one component, which is the most visible component, goes into the machine, so it can do language mapping. But if you are doing translation in a, uh, domain where risk is important, context is important, somebody has to hold that and manage that. In the past, the translator was doing all of that in their head at the same time. That's the idea of tacit knowledge. Today it's getting very complicated because you see seemingly good translation coming up, but somebody else is on the hook for managing the context and for checking it. And I mean, I spoke at a translators conference recently and they were not very happy about the fact that they, you know, for them it's easier to translate a document than to check poor AI output, because that's how they think, right? You cannot retrofit context onto the AI translation. So for them it's double the work. But they're being paid half the money because, you know, the clients assume that it's AI.

Speaker B: AI is doing the work, so it's half of the price. That's an important thread to pull on, I think also sangeet in the marketing world, the problem we now have is, and I think it's getting bigger, and the evidence I've Got is it's definitely getting bigger. When an agency would produce a creative campaign for a client and um, they do the strategy, they do the creative, they do all of the asset creation. They'd load it onto the platforms, they'd monitor it. That value chain. A lot of clients are saying, great, but now with AI, surely it's cheaper for you because it must be cheaper, you're using AI, so why are you still charging me the same price? That becomes very hard to defend for agencies.

Speaker A: Yeah, yeah, no, for sure. I think, you know, there are three or four effects over there that become really important. One is that a lot of things that agencies used to do get commoditized because now the tool can do it. So the profits associated with that will in the long run move to the tool. It won't move to the agency if the work does not change. Right. I mean that's the operative term, um, over there.

Speaker B: So if you just kept the system the same, the profits move from the agency into the tools. So the agency profits would decline if they just operated in the same system.

Speaker A: Correct. Or the profits or the value would move back to the customer. Because the customer can now access the tool and get some of the work done themselves.

Speaker B: Because why would they need the agency to do something that they can do themselves?

Speaker A: Exactly.

Speaker B: Self serve models.

Speaker A: And it doesn't mean complete displacement, but I think complete displacement is very easy to argue against. What is more complex is that the bundles of services you used to sell, not all of it you could sell any further. Right. Because some of it can be done by the customer. Simple example, if you're a consultant, you could have sold, um, a market analysis and you could have sold a strategy downstream of it. Now the market analysis is a commodity, so you certainly can't sell that. By the time the consult, the client has done the market analysis themselves using the tool, they know what the strategy should cover and which parts they need a consultant for which parts they can do themselves. So if the system doesn't change, what goes to the consultant gets progressively narrower. So I'm not saying it's a death of consulting, but it becomes narrow.

Speaker B: Yeah, exactly. So therefore you have to rethink where your place is in the system and where value shifts to. So let's move on to. Because I'm conscious of time as well, let's move on to the marketing ecosystem. Let's think about the agencies existing in the marketing ecosystem for a reason. It was typically capacity and it was capability is that uh, brands didn't have the number of people or the expertise to create and run marketing campaigns across all channels. So there was a valuable role for agencies to play with AI. First of all, what are the constraints in the existing system? How does AI change those constraints? And thirdly, what does that mean for agencies in terms of reinventing themselves in the new world?

Speaker A: Yeah, yeah, I think there were multiple issues with the sort of the marketing services value chain, if you will. I mean think of anything in terms of where is the work getting done. So the work is getting done either in tools and software or the collateral you are creating or in the services you're providing. So there's a whole range of different locations of that work. Right. And there are challenges everywhere. So on the software side, the software stack was very fragmented. So one of the things that agencies would do was that they would pull together a custom stack that would solve your workflow and then on top of that they would create a custom collateral engine and you know, and a reporting engine. Right.

Speaker B: So they were coordinators, correct?

Speaker A: Exactly right. And the coordination was a combination of what was tasket versus what was in the tool. But it was still coordination then. You know, there was this whole piece around planning the campaign, running the campaign, checking results and so on and improving it on that basis. Now so three things happen over here. First, the tools are not as siloed anymore because you can construct agentic workflows and achieve that coordination much better. So you don't necessarily need a, ah, highly human intensive workforce to manage that coordination. Collateral development, again, a lot of the costs associated with collateral development with campaign planning starts going down. It's really the judgment on top of it, which is what kind of campaigns work versus not, uh, you know, the ability to think through the creatives, innovate on that. There are very few constraints that remain,

Speaker B: very few constraints that remain. I think that's part of the challenge is that the constraints that existed. The capacity problem, AI takes away a lot of the capacity problem, the capability problem, AI takes away, as we've just said with given its knowledge work, quite a lot of the capability constraint. And now clients can go, well I'll just use a platform that has all of this capability embedded and it's got almost infinite capacity. The quality may not be as good, but as I wrote the other week, dollar Shave Club have done this. It frees the brand to experiment on their own without having the constraint of an agency.

Speaker A: Yeah, yeah, yeah, exactly, yeah. And you know, on the quality problem, I will just say that we have never had a technology which evolves at the rate at which this does. And so the quality problem is, to me, it's, it's a matter of time. So you can only play an arbitrage game if you're saying your quality is better. But what's truly a defensible game is not about being human. In the age of AI, the truly defensible game is, I'll, uh, use a term here that is a bit cliched, but I'll explain it. The truly defensible game is that you stop thinking in terms of outputs and think in terms of outcomes. Everybody's saying that. What does that mean? The first half is simple. Don't think in terms of output, don't think in terms of, here's what I give to the client, because the client will say, claude, design can also do that for me. So don't think in terms of outputs. Don't get mad at that. Figure out how can you create a proprietary learning loop to constantly improve the outcome for the client. That means that, that that learning loop cannot simply stay in your head. You need to restructure your work so that every choice you make at the marketing tech stack, at the collateral design, at the campaign design, even if those choices are being made by AI, there's a way for you to track those choices, track the outcomes, and then create a learning loop between the two. Unless you do that, unless you figure out how do I change the nature of work over here so that it helps me get to better outcomes, you're in this losing game of, you know, you're in this trap of constantly, uh, going for more and more output when AI keeps getting better alongside you.

Speaker B: I think that's very well put. Yeah, very well pulled together as a thought. You're basically acting as what people are saying is an orchestrator. Your job is to be the orchestrator of the work and build the learning loops so that it gets better and better as time goes on. The question for you then, Sangeet, who should own the workflow? The danger here is, as an ex buyer, if I was working with a third party and they built the orchestration layer and they had all of the, they owned, um, the learning loops, I can't off board that provider because if I offboard them, all of that learning knowledge, that learning loop disappears. So I go, well, I have to own all that. So now you have to build it in my environment and you operate it for me. And that's quite a different model. What are your thoughts?

Speaker A: Yeah, I'll make a distinction between two things which people tend to mix today. One is the Intent and the other is the actual work. You know, people say that in the age of AI, data is going to be king and all of that. It's one of those things, you know what. Which I collectively call true, but utterly useless. There might be some truth to it.

Speaker B: I couldn't agree more, to be quite frank. Yes.

Speaker A: Right. So there might be some truth to it, but you know, we can't do much with that. What's really important over here is that we need to think about how do we restructure, how do we restructure how the work flows, what decisions are made

Speaker B: at every step in order to improve the outcome.

Speaker A: Exactly. The learning loop does not necessarily have to go through some complex data lake and all of those things. The learning loop has to be something that helps you figure out what are your heuristics on how to improve the way you run the work so that the outcome improves. That's the key thing over here. I'll give an example over here just to illustrate this. I talk about this in the book, which is this example of Shein, the fast fashion company, Chinese, brilliant example.

Speaker B: Definitely.

Speaker A: So traditional fashion used to work on this non learning model, which was a merchandiser, would go to Paris, Tokyo, Milan, figure out what was happening, come back, give a design brief and then the fashion house would launch a whole seasonal collection. And then as soon as that launched there would be another four month or six month cycle. And what Shein does is it has this learning loop which is very interesting. It constantly senses and uh, captures new trends on social media. It converts it into a design brief, gives it to a designer, you know, some on demand designer they have and then they have an on demand factory network which creates a micro batch of, you know, it could be like a

Speaker B: hundred shoes, 100 dresses, 100 shirts.

Speaker A: Right. Not a whole collection. They test that and that helps them do two things. With that you learn either to abandon or do more of that if the test works or fails. But at the same time it creates the learning loop of which design worked, which factory worked. And with that it's absorbing the designer logic. Over time it's getting better than the individual designers and it's creating, it's improving its ability to figure out how different factories do for different kinds of designs. And so that's a uh, you know, learning loop in the work. The choices connecting to the outcomes.

Speaker B: Brilliant. So very conscious of your time. Just in three minutes we'll close.

Speaker A: Sure.

Speaker B: One question for you. The Jebbings paradox. So you wrote it? Yeah. You wrote at some length about this about why it's all an incorrect assumption. Can you explain what it is and then explain why you think it's incorrect thinking?

Speaker A: Yeah. So essentially the Johannes paradox says that when a new technology comes in and when efficiency gains result, you know, you're no longer in a zero sum game where the technology is taking away jobs from workers, demand expands as well, and hence the surplus that gets generated flows, uh, back to the workers as well. And so, you know, it's a very astute observation for when it was made. The problem is that today a lot of people use it to just take one of the two extremes and say, you know, one extreme says that AI is taking away jobs, the other says no, well, Javon's paradox, that it's going to create too many new jobs, which we can't even imagine, and you know, again, it's true, but utterly useless, you know, with jobs how. But I think the most important thing is that what's different from Javon's time to now is that A, the technology is constantly improving, so there is a constant race against the technology. But B, and this is the more important thing, there is a distinction between who is using the technology and hence the technology is working for them versus who is working for the technology. Because if you are the data labeler, you are working for the model. If you are the translator checking outputs on AI translated text, you are improving the model. You are working for the model. You're not augmented in any exciting career improving way. You are augmenting yourself away to irrelevance on that particular role. We've seen this classic situation with what happened with Uber or Ride hailing, where it created two classes of workers. There are people who are above the algorithm and creating Uber, that's the Uber data scientist. And there are people below the algorithm who are the drivers. You can say they are augmented, but they are really on a race to the bottom and they are doing commoditized work. So the point is, did ride hailing increase the demand, increase the pie? Absolutely it did. But did the spoils come back to the drivers? It did not. It all went to Uber stock. And so the point is that you can quote Jaiwon's Paradox, but what matters is not whether demand expands. What matters is you are above the model or algorithm or below the model or algorithm. So that was the key point that I wanted to make, which is different from the time when it was originally framed, which was, you know, back in the coal mining era days, I think

Speaker B: as a kind of a summary to the show, this episode. I think that's a, uh, perfect summary. If you're listening to this episode, think about really carefully. Are you sitting above the algorithm and the infrastructure or are you sitting below it? If you're sitting below it it, then it will be a race to the bottom. And the sitting below it to me is if you as an agency, keep doing the things you've always done, but you do it a bit faster, that will only go in one direction.

Speaker A: Yeah, absolutely.

Speaker B: Um, you must sit above the algorithm. You must sit alongside the CMO and be the advisor and the growth advisor to the CMO about how to redesign the entire ecosystem. Then you can create value. Sagit, it's been amazing. Thank you very much indeed for being a guest on the show. Where can people find out more about you?

Speaker A: No, thank you so much. You can obviously have a look at the book Reshuffle. You can look at the companion site, which is reshufflebook.com it has a map of related concepts that you can dig into and understand, play around with before you get the book if you want to. And I write a newsletter regularly at platforms substack.com Brilliant.

Speaker B: Sangeet, it's been a real pleasure. Thanks ever so much for joining me.

Speaker A: M likewise. Thank you so much.

Speaker B: Thanks for listening to hello Higgle, the B2B Sales Club podcast series with your host, Mike Lander. Please subscribe so that you'll catch all the next episodes.