Why this simple AI mistake could cost you your job w/ Sangeet Paul Choudhary

0:00 One of the things that I would encourage every knowledge worker to do is to really think about with AI becoming available with certain forms of capabilities becoming mass commodity available to everyone what is your unique asset [music] and how can you rethink how you deploy it to solve the customer's problem using these tools that are available or capabilities that are available. [music] Enjoy the conversation.

0:30 Welcome everybody to another episode of Merging Minds. Our guest today is Sangit Shidari. He's the best-selling co-author of Platform Revolution and the author of the new book reshuffle which received the 2025 thinkers 50 strategy award for the most impactful idea in the field of strategy. He has advised CEOs at more than 40 Fortune 500 companies as well as preipo technology firms. He is currently a senior fellow at the University of California, Berkeley and has presented at leading global forums including the G20 summit, the World 50 Summit and the World Economic Forum. Welcome Sange.

1:07 Really happy to have you. >> Thank you so much Gabriel. And Sanit if we can get started for those that aren't familiar with your work, can you give us a high level overview of what platform economics is and the overall macro framework of your thinking so that people can understand um can make a little bit more sense of your overall perspective? >> Yeah, absolutely. I think um you know the first body of work that I looked at was platform economics and then I've extended that into my work on AI and what underpins both of them is to really look at how our systems change whenever a new technological shift happens. So what do I mean by systems? I mean uh you know the way our industries are structured, the way our organizations are structured, uh the way work is done within organizations and the way jobs are structured. So if I give a simple example before the platform economy before the rise of uh algorithmic uh markets if you will um the taxi industry worked a certain way but with algorithmic markets and platform economy we saw Uber and uh you know ride hailing companies coming in and reorganizing it in a new way and we've seen that algorithmic reorganization of some markets but with AI coming in and this is the key thesis of reshuffle we are now going to see reorganization at the level of jobs, organizations, industries across the entire knowledge economy. And that is what's um most interesting about what's happening with AI today. It's not about just AI taking over jobs or AI replacing humans. It's really a question of what do the jobs of the future look like? What do the organizations of the future look like?

2:46 Uh will the industries of the future have the same boundaries as they do today? And those are the questions that AI is forcing on us today. So that's really been the theme of my work across these two topics. >> So it's very interesting how one flowed so well into the other, right? It seems very synchronistic because I I from what I read the idea and and correct me if I'm wrong, the the antithesis of the platform um business is the pipeline business. Am I did I understand that correctly from your book? Sure.

3:15 >> Yes. So, so we're going from and the pipeline would be the traditional um model where someone manufactures a product, sells that product for some kind of value ad and creates a profit. The platform is b more based on the ecosystem that's generated. They themselves may not have the product itself, but they're managing they're generating value through the plat. Did I understand that correctly? >> Yeah. Yeah. It's uh it's about uh uh coordinating that ecosystem to together either facilitating the end transactions or coordinating how they work together to make something. So it's about essentially shifting from a production logic of industries to a more of a coordination logic.

3:57 >> Got it. So we go from more like this industrial era u optimization to coordination logic. That's right. And in your in your opinion, if I understood correctly, AI is like a a step further into this coordination logic because AI enables these different systems that previously couldn't communicate and couldn't work together. Now they can coordinate and I there's analogy of the GPS, right, that you use where it's analogous to what GPS did to the the the let's say the the the transportation industry. Did did I get that correctly?

4:28 >> Yeah, absolutely. And I wanted to kind of bring that message out because a lot of uh today's intellectual frameworking of AI bothers from the the automot uh auto you know automation u history and what it fails to understand is that most u automation historically has played out in explicitly codified processes. So if you if you could codify a process as step one, step two, step three and you could bring in a machine for specific steps, you could automate that and that worked with explicit codified processes.

5:02 But today we're working with knowledge work and knowledge work is not explicitly codified. How you translate something cannot be codified. It's very tacit and yet AI comes after this tacit uh uh structuring of work. I won't even call it a workflow because it's happening in your head, right? uh the the the it's all very tacit and yet AI comes after that takes some parts of it away and displaces other parts and that changes the structure of the work and that's the point I wanted to make that it's not just automation it's really about recoordinating the work once it's been broken out of this uh you know tacet bundle that used to be there and so a lot of our um assumptions about how work is done. Yeah, how knowledge work is done is up for questioning. And that is really why I wanted to bring this coordination framework over there and shift um the lens away from how we focus on AI only as a way of automating things. Yeah, that makes a lot of sense.

6:05 And in the context of our industry of this translation and localization space, machine translation for example was used exactly as you described. It was just as one step further towards automation. So let's say translation agencies and buyers they used machine translation in a certain step of their production but they we're still thinking very much as like this is the sequence of production and machine translation steps into a specific sequence right >> now we're talking and one of the things I love about what I read in your perspective it's a much more systemic approach you're looking at like this greater transformation that's created if people are open to that and one of the the criticisms that I have um especially towards our our space and and that probably includes us as well as as tech as a software company but I think uh the thinking is has been it was very successful with machine translation. So machine translation came in they used it to basically make things faster, make things cheaper, make that translation process a little bit smoother and everybody won. More volume got translated, more more profits. It was it was a it was a happy day in the translation world even though a lot of translators felt upset about it and there's like some disgruntlement overall from an economic perspective very little actual reshuffleling just really you know more maybe a little bit of transformation essentially it's like making a car but faster just like you described in the automotive industry but you're still in that same production chain now with AI the the feeling I get is that people are still trying to say we've seen this before we just have to do it again and it's just let's put AI now into that same framework into that same workflow and everything's going to work out.

7:52 >> But things aren't working out. like just from um uh a few interesting stats. Rodrigo and I went to um one of the largest translation events in the world, GALA, the globalization association of language uh organizations and uh it had 50% of expected attendees in 2026. Right? So there's a huge decrease in and that's very significant. Right? It's not necessarily um a deterministic indicator of how well the industry is performing, but it's it's it's a sign, right? And then from anecdotal conversations with people and any industry data that I've seen, there's there's already signs of decline um and etc. So, I I want to ask you what are your thoughts in terms of that? I know that you've been researching uh how this has been impacting translators, how and the industry. Would would love to hear your thoughts on that.

8:46 Yeah, I think you know I I came at it uh first from framing my overall reshuffle thesis uh which was agnostic of translation and then thinking about how it applies to translation and then um over the past um I mean more recently I've been building an index based on actual data that um validates or invalidates some of the thesis that I've been uh building for various jobs and translation is is very very much a part of that. uh so you know when you think about translation and especially when I look at it as somebody who's not been in the industry and coming at it from the outside in the the first fallacy that people usually have when they think of translation as well we're talking about large language models and translation is language and so if somebody has to be automated and removed it should be the translator because clearly you can't have a better technology to um to solve that problem and very often what we u misunderstand is that translation is not just about transfer of language. It's also about transfer of nuance and context uh and it's also about transfer of consequence in many cases. So there's a significant amount of risk bathing involved in many forms of translation.

9:59 And so what we will likely see uh you know structurally is that even if and I'm going to qualify the even if but even if the transfer of language can be substituted with a machine generated component and over time I believe it will increasingly the transfer of nuance and the transfer of consequences becomes much more difficult to automate. So that's one point. The reason I said even if is because even um though you know translation is almost native to LLMs.

10:31 The very fact that outside of English the corpus on which LLMs have been trained have lacked the variety of nuance that the English corpus has had shows that you know translation uh will be ether uh ridden even in the uh transfer of language if you will because it's very difficult to distinguish where it's just language and where's nuance. So that's why I said even if but then you know as the corpus increases over time u and as more translation hopefully gets done now that tools are getting better as well and hopefully work gets reshuffled around it then the corpus hopefully increases. So those are you know certain components to think about.

11:10 The second piece that um I I I would say is that if we look at the data itself um there is a very clear view that translation jobs are changing. Uh so I analyzed um you know around 150 different job postings and all of those job postings are some version of verify what the machine did. It's no longer apply nuance and translate. It's verify what the machine did. It's none of them give an indication of what the future translation workflow looks like. So all of them are uh stuck in what I call the you know what I would think of as the human in the loop fallacy. We often uh celebrate the human in the loop. But just because the human remains does not mean that it's a job that's uh interesting to do and that's actually uh uplifting. You are if if uh if the the value of your job came through understanding nuance and now you're focused on correcting machine outputs, it's not necessarily a good outcome. So there's a significant human in the loop fallacy that is writing all of these translation jobs uh job postings, if you will. Yet there are several players who are you know especially those who are um independent who are beginning to leverage AI enabled translation to rethink workflows but that hasn't really happened on the job site. So I think what I'm trying to say is that translation is sort of a canary in the coal mine in terms of what happens when an industry overestimates what AI can do to the core value that the industry provides and what happens when an industry even misunderstands the core value because the core value is not transfer of language it's transfer of consequence and uh nuance and depending on the context that may change so legal translation still whether there's because transfer of consequence is high but there are other scenarios where the translation of nuance is important consequence is low where we are seeing widespread AI adoption so that so that's uh you know uh one one key aspect over here and u um I I I would just say that um we uh this is an issue that is not unique to translation in general we overestimate what AI can do we overestimate what automation does to the job. So the other uh you know outcome that came out of this analysis was that I mapped translation uh or the translator's job into 10 different forms of value and um I'll just uh you know pull that up to kind of uh give an example. So there are many different forms of value. There's you know comprehension of the source, there's cationation of terminology, client scoping, domain expertise and so on. And AI you know absorbs some of these things like comprehension of the source and you know curation of terminology but there are many other things that AI does not absorb and those sources of value or those value drivers get redistributed across the workflow. So now increasingly other people start doing it. So things around you know client scoping things around anything that uh related to nuance now falls into somebody else's lap if you take away the translator's job. So this is again a pattern that we've seen before. Uh and I talk about it in in my book as well that when the word processor came in uh the typist job um you know went away even though the task of typing still remained because the task of typing got reorganized across the entire organization. And so one of the things that we miss when we think about the impact of AI is that it's not just automation, it's also some parts of your role could be deconfigured and thrown out to other people in the organization. So really if you are a translator or if you're any knowledge worker, you should be thinking about how do you rebundle or deconfigure your Z when it's being split apart. Uh some of it is being taken by the machine, some thrown to other parts of the organization. What's your new role going to look like? So I'll pause there. quite a few thoughts. But that's, you know, that's why I like translation. It's a very nice canary in the coal mine in terms of not what AI is going to do to jobs, but what organizations think AI is going to do to jobs, which is not necessarily on target today.

15:29 >> I I I agree that makes a lot of sense. I I I I I so if I understand correctly, what you're saying is uh even though let's say um large language models and this whole text stack enables functionally something to be translated, it doesn't mean that it carries all of the consequences, all of the nuances, all of the um uh tangential let's say values that are in in that right and one of the things I always think about is when we project our software, we always are thinking about this um um idea.

15:59 Okay, even at language parody, why are we here? Right? Why why why does this why is the software necessary, etc. >> But one of the things I I want to ask about this reshuffleling is um the word is great. I love it. Uh I noticed that for translators, as an example, the ones that I know, the ones that I've talked to, it's extremely painful often to change the way they've configured their brains over the past 10, 20, 30 years becoming very relevant in their profession. It's also very hard from a a self-worth perspective because they went from being let's say the knowledge holders without them the process was um simply didn't exist and now the process can be done without them but they do add like you said a lot of value but it's I noticed that it seems to me that a typical human response is that it's easier to do something entirely different it's easier for me to say you know what I'm no longer going to translate I'm going to bake cakes than it is to recon configure the way I've wired my brain, the way I've educated myself. And because a lot of translators were so used to that heavy lifting of let's say researching if it's the right term or not typing these activities that are very easily outsourced to a large language model for instance it's hard for them to shift gears into this more uh critical role because as you described right if you're thinking more about nuance and consequence it's a lot more more critical a lot less about mechanical. So I wanted to know your perspective on like this more behavioral process and cognitive process of transformation.

17:34 >> Yeah, that's um you're absolutely spot on. That's not uh very straightforward. It's um it's easy to say reimagine your job. But sometimes as you said it's easier to just stop translating and go bake cakes because you know that's still the same. you don't have to reimagine anything over there and you still feel that that's uh more defensible. Um and in certain cases that that might be the right or unfortunately the only option left. Um, having said that, one of the things, and I'm I'm not going to pretend that this applies across the board, but one of the things that I would encourage every knowledge worker to do is to really think about with AI becoming available with certain forms of capabilities becoming mass commodity available to everyone, what is your unique asset and how can you rethink how you deploy it to solve the customer's problem using these tools that are available or capabilities that are available. What that means is and you know I'm I'm focusing very heavily on solving the customer's problem because that is why our jobs exist today. Our jobs don't exist for any other reason. Our firms don't exist for any other reason. It's all to deliver a solution. So it's a even though this is not easy what we need to go back to is first principles with the new capabilities available today in terms of technology with what I specifically bring to the table. Are there fundamentally new problems that I can solve? Is there a fundamentally new solution I can provide to the same problem that I used to solve? So think you know change the starting point is to step out of your workflow because we feel the threat in the workflow. The workflow is what we have protected. We have not protected the solution because we felt the solution was stable and we were competing on how well we could manage our workflow. Now when AI comes in, we keep on trying to protect our workflow.

19:27 But I would encourage step out of the workflow and go back to why did your job exist? Why do you have value in the economy? What problem were you solving? Could you solve the same problem with a better solution? Could you solve a different problem for the same customer with a better solution? Could you solve a different problem for a different customer or the same problem for a different customer? Now that these capabilities are available. So the only way to reimagine your job is to move away from your job, your workflow and go back to where's the value, what's the solution and start reimagining from there. These are not easy to do. But if we remain stuck in thinking about how do I stop being the human in the loop and how do I improve my workflow? Um that's sort of you know you're working from within the frame instead of looking at what are the new realities and hence what new problems can I solve. That's the in a way I'm saying that you know all of us have to be more entrepreneurial because the capabilities that are being made available are dramatically changing on a weekly monthly basis and that calls for entrepreneurialism and problem solving rather than just workflow deconfiguration.

20:32 >> For sure. Yeah. I I love what you said. It requires much more abstract thinking too. And I think along with abstract thinking, this is the part that I personally find hard and I think that for a lot of people is hard to f to stop protecting the workflow and start protecting the solution >> requires a certain sense of selflessness or because again if you're very attached to the workflow, it's hard to give that up because you believe that your value comes from the workflow. Whether you're a translation agency, whether you're an a translator, whether whatever role you have, I think people derive sense of self-worth from the workflow. And like you said, because it felt very protected, it was very comfortable. But I I do think that psychologically it's a hard thing for people to get over. And and I do agree that the pressure is going to be inevitable and people are either going to figure a way out or they won't. But there is I think a a pretty big um psychological uh component to these transformations.

21:29 >> Yeah. Yeah. I I mean you you mentioned uh quite a few important points over here and um the self worth point is absolutely u very central to it but a related point I would also call out uh that every knowledge worker should think about is that while we try to tied our selfworth to the workflow we have not necessarily been very self-aware about what we typically bring. So we have assumed always the workflow as being a constant and our our ability to do it much better than our peers as the way we compete and what I think everybody needs in the age of AI is to be more self-aware about what is unique that you bring to the table and again it might sound you know a little abstract unless you start applying it in your context but just the simple frame move away from how can I get better at translating ing to thinking what what was I really bringing to the table? What have I what skills have I honed over time? Which of those can I still leverage on? Which of those can I not? And so that calls for greater self-awareness and and it's not very natural for us to do that because we have been trained to live in an economy which had very stable jobs and we just had to figure out how to get better at them. And now we have to move to an economy where we have to think about what the new jobs are going to be and how can we uh you know proactively build that for ourselves. So it's not very easy for most people to think in those terms.

22:58 >> I agree and one thing I wanted to bring to the conversation because it also impacts our space from a but from a coming from another field which is I've been reading about companies like Cloudflare and ClickUp and Block. Cloud and ClickUp recently fired, if I'm not mistaken, 20 22% respectively of their workforce with very strong mandates of you're either basically a builder or you're a seller. And essentially you're you're either and and and what reading about these um transformations, one of the things that I found particularly interesting is that um the the the the idea behind it is you can't really um what was I going to say the the the idea is you you even the tech stack that's behind these these organizations is being or entirely thrown away. So maybe a content management system is now being repurposed as an AI pipeline and >> right >> uh and and middle management has been particularly impacted by this. So the idea is we don't need that many managers anymore managing these things. We need people who can either build or or who can sell. I would add another category that's not really emphasized but you kind of hinted on it which is the expert. So the person who can be let's say not just not necessarily building but advising and analyzing and um improving on the process. But I think when something like that happens that creates a lot of disruption because there's this idea that um instead and I think it goes back to the idea of going baking cake. I think for some of these organizations they know that the transformation is there but it's it seems like it's very hard for them to get their current people to get on board this new way of thinking. So for them it's simpler to just wipe a clean slate and then kind of like start over. And I wanted to ask your opinion on these kinds of transformations on whether you think this is the only way forward, whether you think this is something that is going to be the new status quo or whether do you think this is kind of like a little bit of a of a hype curve.

25:01 Um how how do you see all of this? >> Yeah, so I think you you you hit on all the points over there as well. Um it's uh we we are at um I I I I don't know if I want to call it a transition period but it's it's somewhat singular. It's not the shape of the future. U but we are at this point where we are doing two things together. One there is very clear structural um logic and guidance that the jobs of the future the organizations of the future will look different and we don't need the same people inensive structures that we had. At the same time, there also is pressure on all these organizations to fit an AI hype narrative which has pumped a lot of money into infrastructure because there were a few players who were very clear winners over there but is yet to realize where that money gets realized further up the stack and all these companies whether it's block whether it's Cloudflare these are all companies further up the stack who are trying to prove that the value of [snorts] improvements in AI is going to show up in their stock. Now, how do you prove that? Um, Silicon Valley is or or you know, Wall Street measures companies um on revenue per employee. So, if your revenues aren't going up, and some of them are, but if in general revenues aren't going up, what do you do? You bring down the employees and your revenue per employee goes up. So revenue per employee is a preAI metric but it's a classic metric to measure you know that that that feels like you're measuring the impact of AI because you know when you use the term of agents are coming in and we are now using agents instead of employees and when you start using those terms you're automatically saying well if all of that is true revenue per employee should shoot up and so I think a lot of the layoffs today are not just actual structural layoffs. They are pandering to the hype layoffs with some structural truth to it. Which is why I say that this is a unique moment in time. We are going to figure out what the new model is going to look like, but right now it's overweighted towards the revenue per employee issue.

27:07 >> Yeah, that that's interesting. I I think another question I have regarding that is um and and and and again that that makes a lot of sense. But my qu my question is whether or not for instance how much value is being destroyed in the process because just as an example when you toss out a content management system just as an example >> and you rebuild that with agents and AI pipelines and you use engineering power to do that.

27:35 >> Yeah. Maybe that's interesting from a conceptual perspective, but I'm I'm pretty sure that if you squeeze the dollars and cents that are being used to manage that and all of the transformational cost, it's probably not going to be something that is a net positive from a P&L perspective. It's something that looks good, >> but I don't know how sustainable it is. I don't know how scalable it is. I wanted to to hear your thoughts on this process of just throwing out, you know, tech. So I I'm going to say something a little controversial over here and um that is basically this that if you are uncertain about where future value is going to come from then whether you do it the right way or the wrong way you don't really know what one what's going to work. So you don't just don't know what is the the way to go and so the only way that you feel comfortable betting on is the fast way because speed creates learning. things go wrong, you have a lot of capital, you can change your mind, you can learn and I think all the large companies which are capital rich today um are actually veering towards that. I I I believe that's what's happening in my um discussions with executives who are making some of these decisions. That's this is typically the dilemma that if I have four ways of doing it and I don't know which of those four is any better than the other, what should I do? The answer is whatever is fastest and will help me learn fastest, let me do that. And I think that's what's happening. Um so while we will see a lot of casualties because um you know throwing away your text tag and trying to re-imagine it and then throwing away your employees and trying to replace that asset knowledge all of that is going to have implications but that doesn't mean that not doing it is the better way to do it either and so I think unfortunately um or or rather um um I I I I'll say structurally we are moving in the direction of a world where If you do not know what is the right system to build in the future, you just build things faster, learn faster, and figure out your way to get to it.

29:37 >> And that makes a lot of sense since there's so little clarity. The quicker you iterate, the faster you arrive at it's like a scientific experiment. The more you hypothesize, test, and iterate, the faster you are to some source of truth. Um and and yeah, personally I I understand that it could sound controversial to me. It sounds actually very logical. Uh I think the the again I want to go back to this idea that for I based on what I've heard and based on what I've read and based on what I've seen, it seems that a lot of people in these orgs have tremendous challenge in stepping out of their workflows and stepping into their solution hat. So a lot of them have opportunities given from execs to reimagine what it is they're doing or to rewire what it is they're doing. But it's hard. It's hard for the the typical tendency is to try to protect the workflow. It's trying to explain okay no this makes sense because of A B C D. This is why it says I've seen this in localization before. No, if you don't do localization this way, you're going to be embarrassed. The whole market is going to crumble.

30:42 everybody's going to laugh at the company and they'll they'll produce all these slides that explain the value. This is why we have to spend a million dollars on our team on this and blah blah blah. >> And I think a certain thing that I'm seeing is people no longer buy these narratives. It's like the protect workflow narrative. It just it just feels like noise to the ears of a lot of these execs driving these decisions. and um and and I'm wondering you know your thoughts on on that and what people can do to to uh step out of this mindset.

31:15 >> Well, you know what I I mean this is some of it is going to be anecdotal but what I have seen happening more and more as I talk to people who are you know going through these changes in the midst of layoffs etc. There there are people who want to reimagine the solution and who want to reimagine things and not just get stuck in their workflow. If they try to do that within the organization, there are always people who have a different opinion about how it should be reimagined because when you don't know, you know, my opinion is as as good as yours and so on, >> right? And so the incentive for somebody who truly wants to reimagine things is to actually leave and do that on their own and actually benefit from the spoils of the reimagined workflow and reimagined solution and so on rather than sit in the organization where the benefits are not coming to you and the bureaucracy is coming to you. And so what I am beginning to see and that's why I said it's anecdotal but uh I'm seeing this a lot that the people who are increasingly staying back or uh you know voluntarily staying back are uh you know people who already have the uh political power to do all this reimagining or people who do not want to reimagine. So anybody in between starts feeling the trade is much better outside.

32:31 >> Yeah that makes sense. That makes sense. Um and when we're talking about this reimagining for instance going back to the the translators now right um with m with machine translation there was this very um clear dynamic that was established between translators and machine translation. Translators did not like it. They thought it was like a necessary evil. Uh they they thought it was uh painful. They thought it it was um in many regards um took away from their prestige, took away from their skill and it was easy for them to um to diminish it, you know. So their their defense was always pointing fingers at it, pointing poking at its flaws and things like that. And again, people would kind of buy it, you know, if you executives didn't want to have um their company become the source of an embarrassment. So, they would buy that narrative. And now that narrative is like, yeah, whatever, you know, I don't care if there's something wrong on my website in German, I'll go around and it'll fix it. And you were talking about, you were alluding to this earlier, but I think this iterative mindset seems to be more and more prevalent as opposed to this final deliverable mindset. And I wanted to hear your thoughts on that and how that changes how you see your own work when you when you know that it can always be iterated on as opposed to a final crushing deliverable.

34:00 >> Yeah. I mean um [snorts] so I think there are you know there are two two different points that I want to make over there. Um the first is and and please uh let me know if this is not directly answering the question but the first thing that comes to mind is that we have been taught to overvalue the output and we have not been sufficiently taught to understand how that output will be used what is the final use of that output going to be what's the outcome it's going to lead to etc. And we are now in a world we were we are coming from a world where it was difficult to generate seemingly good outputs. You know even seemingly correct outputs were difficult to generate.

34:41 Today you can generate a seemingly correct output very cheaply. And so when when we've been trained to overvalue outputs, we get into this bias of generate the output, fix it, generate the output, fix it, teach it a little more, fix it a little more. And so we are still working always within how do we get to the output as soon as possible and then try to figure it out because earlier that was the constraint. Getting to the output was the hard part. Now that is the easy part. And so when that becomes the easy part, then by overindexing on that and staying stuck in the output, we degrade the quality of getting to the outcome. But we also make the process of getting to the outcome much more difficult because if we overvalue outputs and earlier all the friction and the heavy lifting was done to get to a good output and now you can get to a seemingly good output but with many others very fast you you end up catching yourself in a u you know getting stuck in a workflow that does not really help.

35:43 So focusing you know moving away from the output to the outcome is to to in my mind the first thing to to really think about. Um, a lot of people talk about outcomes in the age of AI. We don't necessarily talk about how much even though we use the word outcome, we're always still stuck in output. So, really making that distinction is is uh very very important. And the the second point that I um I'll just make is that this overemphasis on output is what keeps you trapped in your workflow. This overemphasis on output is what keeps you focused away from the customer's need because yesterday the output was a proxy for the customer's need because the output was being generated only when things had reached a certain state.

36:32 Today we are at a place where if outputs become much easier to generate the entire downstream workflow is going to be completely different but going forward even the upstream workflow before you actually think about generating an output is going to change dramatically. So I'll take the the typest example again because I believe that's very important for where translators are today. So when the word processor came people thought that the typist would be you know augmented because you still had to type right. So the the word processor did not take away the the task of typing but it took away the job of the typist because the job of the typist was structured around the main friction that they were solving in the organization which was that editing was expensive and because editing was expensive all the hard work was done before you got to the output. So you would not create a document until it was fully structured all the iterations had been done and then you would go to a typist poolool and create the document.

37:27 So the hard work had been done before. Now that editing was cheap, anybody could type and so everybody was working on the output. What that you know what that really leads us to is that the second order effect of that was what was interesting because it because editing became cheap because now it was much easier to get to a first draft output much faster in that instance. The downstream implications of that completely changed. So the first thing was that iteration became very very u um you know um uh very much a source of value because in the past you had to um get to the right um text to be typed before you got to uh typing it. Now you could just whatever uh you know early drafts could also be typed out and when early drafts could be typed out and passed along to other people you could iterate on knowledge documents and that's really the source of the knowledge economy that we have today.

38:21 the fact that you could document clear thoughts, clear processes, send it to others, get them to uh react to them, work on that. And so if you really think of it, the ability to document an organization's processes started with the word processor. The ability to um use that documentation to run multinational organizations in different geographies by standardizing processes and and uh operating mechanisms and sending it across happened because of that. Most importantly, the ability to do strategic planning effectively happened because of that. Because strategic planning in the past was much more expensive and you could do it through closed door meetings, but you could not really do it through the kind of document driven iteration that the word processor provided. So the point that I'm trying to get to is that whether the word processor impacted the typist or not became a footnote. What ended up happening was that the entire logic of the knowledge economy changed around document-based knowledge organization. And those are the kinds of transformations that are about to happen because of AI which we don't yet see both before we get to the tool and after we get the output that whole range of uh you know how value is performed at both ends is going to change dramatically and we don't see it yet.

39:39 >> Yeah, that's super interesting. you're you're you're pointing much more to the the the impact the ripple that that little change produces and the change in itself and I agree that we probably we don't we don't see it yet but in your opinion just curious here um the word processor comparatively sure it displaced typists and it was revolutionary in its time but it was it seems relatively small compared to the impact that we're seeing with AI um would you agree or do you think that there's there's some kind of similarity in levels of um magnitude.

40:14 >> I think um you're absolutely right that the level of magnitude is different but the structure and the nature of change is quite similar. The fact that you could take a constrained uh point capability and make that available across the entire economy that restructured how the economy works. That's what the word processor did. That's what AI is doing today. And so from that perspective um I think the structural mechanisms are important to learn from. However because the word processor only impacted what could be codified. AI impacts any form of knowledge work. The the impact the magnitude of the impact is much bigger.

40:52 But it's quite instructive to look at what happened then to just you know open our minds to what's possible now and where we are at the very early stages of that. >> That makes a lot of sense. I mean I'm I'm I'm feeling more hopeful talking to you Sange. I mean maybe maybe I'm I'm I'm I'm just interpreting it in a brighter day but it seems to me that just recapping on some of the concepts that you've mentioned. So you've talked about uh people overestimating what AI does because of this output thinking as as opposed to the outcome thinking both in terms of the people producing the work as as well as the people consuming the work. this overlooking of consequences, this overlooking of nuances, this overlooking and in my opinion, you didn't necessarily say this, but in my opinion, I think there's also the point that language is something inherently human and it needs some kind of human bearer to have value.

41:44 Otherwise, it's just data flow, bits and bites of data floating around. The minute I say something, because I wanted to say something, the intentionality, the ownership, I think that plays a role in language. But again, maybe that's just my romantic perspective, but uh that that seems to I mean, what I'm hearing from you is that there's still a significant role for um language to play and um translations to play. >> Yeah. [snorts] >> But um but at the same time, it doesn't it doesn't seem like it's the end of the space. It's just a very significant transformation that's going to be way deeper than people can imagine or anticipate based on the analogy that you gave with the the the word processor. Am I uh mis misunderstanding something or do you think it's more or less along those lines?

42:33 >> That's very much the central idea of reshuffle that um they are going to be winners and they are going to be losers but they will not be playing the game that they are playing today. Because when we think about any shift and especially with AI, the room splits into it's all going to be good or it's all going to be bad. So we have already chosen a win-lose situation. We're not really asked whether the game is going to change. Uh what I'm saying is that the only thing we can be certain about is that the game is going to change and that there will be winners and losers in the new game as well. So I think that's very important. The other point that you mentioned which I just want to very briefly touch on is this point about language being inherently human. Uh you're absolutely right. Um what we have to uh factor in also alongside that is to what extent and in what use cases is the human element of language important versus just the functional element of language because in the past both of them were bundled with the human. Today they get unbundled from the human but they have to then be reallocated appropriately. the functional element maybe it can be moved to the machine but the human element maybe needs to go back and so we often uh uh you know overestimate also what the human advantages because we think about the human advantage from a previous world where multiple forms of activities were bundled with the human not because they were all human in value but because the human was the only one who could perform it. So relational work is high value but operational fixing of something going wrong is not necessarily high value human. If a machine could do it you might want to get that done but relational nuance uh you know all of that is is u uh high value work. So you have to um and when I say you I mean anybody who's listening you know really think about whenever we use the term um human how are we redefining it for the future because there is definitely a very valuable role for the human everything has to be structured around humans that's where you know ingenuity lives that's where innovation comes from but that's not going to be the same as how we defined it in the past and so that the definition is very important I love that. That that makes a lot of sense. And one of the things I want to also push a little bit further. You were just talking about and correct me if I'm if I understood correctly. So it seems the analogy is it's like let's say we're playing soccer right now. That's the sport we're playing. And now AI comes and people think great now I can score more goals. Now I can score goals faster. Now I can defend better. But it's still soccer. What you're saying is it's no longer soccer. Maybe now it's baseball or basketball or some other kind of sports and we have a it's challenging for us to imagine what this new sport is. Did I am I like overstating that or >> Yeah, absolutely. I mean just uh two or three textures to that. One, it could be baseball which means it's not just a sport but even the contours of the playing field that are changing which means your industry, what you do, everything might change. two, it is a new sport, but you know what? You have to start playing it before the rules are settled. So, you have to get into the field, you have to start playing it, and you have to figure out the rules as you go. So, that's the the only two text textural points I would add to that. Um, that's where we are today.

45:50 >> Yeah, I I love it. I mean, it's um I think it it reflects well how I feel in general. I think I I I keep saying this. I've never felt as much opportunity as I do now, but I also I've never felt as much um risk as well because I think it's like you said, it's it's like we're trying to play the game as the game is being created as there there. It's like you're playing something that looks like baseball maybe, but there's not a clear rule. And maybe there's a catcher, maybe there's not a catcher, maybe there's a pitcher, maybe there's not a pitcher. So yeah, very u unsettling but also very interesting from just a historical moment for us. Um and and and the other thing I I I I meant to ask you is from what I've read you have a very interesting take on scarcity. I think most people at least me for instance scarcity is typically seen as something pretty negative but I know that for myself I typically produce my best work in scarcity because it's when I get more creative and things like that. But I wanted to understand your perspective on scarcity and all of this conversation that we're having around AI.

46:51 >> Yeah, I think um you know the the key uh reason I focus on scarcity is that we always talk about the fact that we're going to a world of abundance and there's going to be abundance on every aspect. Uh access to knowledge work is abundant etc etc. But the basic principle of economics is that the economy runs on scarcity and companies the logic of uh you know competition and the logic of uh uh how industries work is based on scarcity. What I mean by that is that in order to differentiate yourself and succeed as a business and as a person in the workforce, you don't rely on what is abundant because that becomes a commodity. It becomes a utility. You rely on what is scarce, what you can uniquely own that your competitor cannot. And so when AI comes in, when any new techn technological shift happens, but even more so with AI, many previous forms of scarcity suddenly become abundant. So knowledge work, I mean translation, you know, uh seemingly good output becomes very abundant. So it's no longer scarce. What that means is that when something becomes abundant, it starts being used in many new places where it was not possible to use it in the past. Maybe it's a different kind of company that could use something that was more expensive to acquire, now can use it more cheaply. What that then means is that as these capabilities become more abundant, companies will have to stop differentiating, companies and workers will have to stop differentiating on these capabilities that are becoming abundant and find what's the next scarcity that they can own. Because all of us have implicitly been owning the scarcity of what I do requires a lot of training, requires years of experience. You cannot just have somebody else do it that easily.

48:37 And suddenly you see a tool coming in and improving every month and doing more and more of it. And so knowing where the new scarcity lies is very is is very important. And the related point, you know, uh just to uh riff off what you were saying is that um scarcity or more specifically uh friction um or constraints uh are of two types. There are positive constraints and negative constraints. So when you said you do your best work, when you you know put in some boundaries to it, that's an example of positive constraints, right? But you're stuck in traffic and you're having to do work and you're not enjoying it. That's a negative constraint. So look for positive constraints because that helps you also identify how you find new ways to get to the outcome. And um the reason I personally feel positive constraints are so important today is because if you start using AI, you suddenly feel that you can do hundreds of different things with it.

49:34 But which of those things are going to be valuable? That's where you had to set in clear positive constraints. You know what's the problem you're trying to solve? What will deliver the right outcome? Now that I have these capabilities, how do I do it in a better way? Because today everybody's measuring adoption. Who's using the tool versus not? But that does not tell us anything because you are just using something that's abundant. What will determine value is how you use it to defend something that is scarce and unique to you. So that's that's how I see scarcity overall. Yeah, that's very interesting because what you described is th this this relationship between scarcity and abundance changes the entire market as well because like as you mentioned as translations seemingly good output becomes very abundant. It starts to show up in pe places that people didn't expect. Maybe companies are using now more languages or things that didn't get translated. I don't have any stats about this, but uh I would I would be uh inclined to believe that in this pre-abundant language era, most companies had way more content that they would like to translate, for instance, than they could. And now that's kind of shifting the the the scales there. And then the other thing that I find very interesting of that is if my focus is actually looking at scarcity as source of value that's that's a very interesting um north star or direction because instead of focusing on the because I think this as you described like in going back to the psychological thing when something is scarce but it becomes abundant there's a loss in value right the abundance has no no value and I think that's hard to overcome but if you overcome that from a psychological perspective and you can begin to direct your focus on this on the value of scarcity and where that scarcity may be. That's what's going to generate value in the future. I think part of the the hard the challenge of this moment is things are becoming abundant at a speed that they didn't become before, which requires us to be way more zeroed in to what is scarce and and drive value through that.

51:34 I don't know if that's correct. >> Well said. Well said. Yes, absolutely. This is very interesting and um I think we're we're reaching close to to our time sit but I I I learned so much. I'm very keen on um publishing uh thoughts on on your study because it's super detailed the one around translators and all your work around reshuffle. As closing thoughts uh what would you say to to our listeners? What would you say to people both that are in in in all parts of the spectrum? People that are feel feeling very impacted and and and hopeless by this. people that are kind of somewhere in the middle and people that are also uh very um enthusiastic about all this transformation.

52:16 >> Yeah. Um I I would say that you know it depends a lot on where you are in the spectrum today. If you have not yet been impacted but are seeing things coming your way, be very thankful that you have this on ramp to move to what you should next do. But don't be complacent because it's coming your way and you have to move in that direction. If you've already been impacted, it can be very uh disconcerting be depending on what kinds of uh what your situation is as well. I mean we have a lot of data from the gig economy that shows that welltrained people with you know post-graduate degrees once they got thrown into the gig economy they could not work their way back. Uh it's easy to say you know work your way back but if you're driving Uber 10 hours a day you don't have any time to work your way back. So um I I I can um I I I mean it's it's difficult um to have very simple fixes for somebody who's already impacted. But if you have the ability, if you have the bandwidth to start, you know, to to to play a slightly longer game and not look for tomorrow's job today um or or simply tomorrow's job today, but have the ability to look for next month or next year's job today. uh try to look at that because otherwise you're constantly chasing a shifting target versus building towards your next career move with greater deliberation and so I think uh having that bandwidth that buffer is also going to be a source of unfair advantage if you will those who have that will and have the clarity um and the self-awareness to move in this new direction will figure things out uh those who do not even if they did have the clarity, they don't have the bandwidth, they'll get stuck in that loop. So those are unfortunately some of the structural issues that we are going through today >> and and overall would you say I mean there like we're like we're going through a very deep transformation as as you described we have still to see the deeper repercussions of this transformation for now we're in that stage where oh yeah that's cool we can accelerate certain things with AI or we can make certain things that were scarce abundant but we're yet to see the systemic repercussions of all of In your opinion, 5 to 10 years down the road, do you think as a society we emerge better for this or do you think we're going to have let's say brutal inequality issues due to this? Like how do you see this playing out in a 5 to 10 year uh time time lapse?

54:53 >> I think the issue that um leads to inequality and any of those outcomes is not necessarily AI. It's the capital structures and the risk structures that we have created. We have pushed risk to all sorts of people who do not have the ability to access the returns associated with it and we have concentrated returns with a few people and that happened first with financialization that then happened with the platform economy and algorithmic coordination that I talked about and it's happening even more with AI. So whether it's AI, whether it's something else, the the issue to be solved is over there. And just using, you know, AI is um such a convenient uh excuse for anybody who's a capitalist to essentially blame the technology rather than their own choice on concentrating things further. And that is the unfortunate loop we need to get out of if we don't want, you know, another French revolution someday. So that's really the issue that's causing um um greater forms of inequality. I I do think inequality will continue. I would hope that the forms of inequality become slightly more equal rather than more unequal because today everything points to inequality itself becoming more unequal. Um so that is the real issue is over there. It's, you know, it's easy to blame technology for layoffs and that's what everybody's doing. But ultimately, it's all incentives for driving greater that turns to capital.

56:20 >> Agree. That's all all very interesting. I'm very grateful for all of your uh thoughts and comments today. I've learned so much and we're going to put this in the comments as well. But for anybody who's interested, reshufflebook.com has a lot of the information that we've been talking about. Thank you very much for the work you produce, Angit. I think this is very important in a time and age like today. I think giving people some kind of clarity and direction. I think that's huge. Uh but not just from like um economic perspective but just also from a human perspective. I think people are in need for some kind of grounding and very thankful for the work that that you do and thanks so much for being a part of our merging minds today.

56:59 >> Thank you. I I appreciate that so much and this has been such a pleasure. Thank you for having me on here. >> Thank you all. Till next time. Bye-bye.