The best argument I’ve heard for why AI won't take your job with Box CEO Aaron Levie

This podcast is brought to you by Atlassian Rovo, the AI that takes your team from AI novice to AI native. Hey, welcome to Platformer. I'm Casey Newton. Recently at Platformer, we announced a new push into original journalism. We've always done interviews with tech leaders, but we wanted to experiment with extending those conversations into new formats like audio and video. So for the next several weeks, we'll be bringing you a series of conversations trying to make sense of what's happening on the ground in Silicon Valley today. I'm bringing in people I've known for a long time and people I've only met more recently to try to get a handle on a very confusing subject. What is the future of work in a world where the AI systems keep improving? I have to tell you, it's a really strange moment to be covering this industry. The companies building AI are convinced that they're on the verge of automating away most knowledge work. Meanwhile, the companies buying that AI are way less sure about that. And then on the worker side, you have a split between people who say that their jobs have already transformed quite a lot and other people who say nothing has really changed. And also, they kind of hate the way that their bosses are talking about it. Meanwhile, layoffs keep arriving with AI being cited as a partial cause, even as the executives ordering them insist that AI is somehow going to create more jobs than it destroys. The stock market thinks one thing on a Monday and something completely different on a Friday. Nobody seems to agree on what's actually happening, which is usually a sign that something is. So we're going to do what reporters do, which is talk to people. Over the next several weeks, I'm sitting down with founders, CEOs, researchers, and operators, the people whose day jobs require them to have a working theory about where all of this goes. And ask them, what's really happening? We're also going to do another thing good reporters do, which is bring you data. Each week, we're going to kick off the episode with fresh numbers from the most interesting research, surveys, benchmarks, and other sources that we can find to help tell the story of how work is changing. And I did not want to do that first part all by myself. So to kick us off, I'd like to bring in Platformer fellow and Gen Z AI correspondent, Ella Marcianos. Ella, how are you? I'm doing well. I've had a beautiful morning reading news stories, browsing Twitter, journalizing. I don't know what else you'd expect me to have been doing this morning. I mean, that's exactly what I expect you to do each and every morning because that helps you complete the task that I have given you, which is each week bringing us a story that helps to advance our understanding of AI and work. So I am very excited to learn, what are you bringing us this week? Yeah, so this week, we've seen like pretty two like pretty interesting large surveys about how people are using AI at work that, yeah, I don't know, it's like an amount of data on this that we haven't necessarily had before. And one thing that like really struck me in both of these surveys, one's from the Financial Times, one's from Gallup, is people who are higher earning in their jobs, people who are in leadership positions at their jobs, are actually using AI a lot more of the time than like more junior workers or people who are earning less money. So the big headline data from the Financial Times, for example, is they split up workers into like deciles of income. The like top 10% of income, 60% of people are using AI most days. The bottom 10% of income, 14% of people are using AI most days. And this is across like all sort of knowledge work jobs or it's every kind of job? Every kind of job. A lot of this sample is knowledge work, as far as I can tell. Got it. Okay, so what do we make of this? Does it seem like it's as simple as if you use AI more, someone gives you a raise? Or is it that if you are a higher earner, maybe you're in the sort of job where you just sort of understand, hey, I really need to be using this stuff? Yeah, so a few things might be going on. There's the simplest deflationary hypothesis, which is people who earn more money just have more computer jobs. They're writing more. They're writing more code. Because this like contains a pretty broad variety of professions, a lot of which are computer jobs, it might not just be that. Another thing that is brought up in the Financial Times article is just people who have are in general in a higher income percentile have more tech literacy. They will know more about these systems, what they're useful for. They'll be less likely to like be afraid of them. And so they'll use them more often. Okay, so let me just ask one question about this study before we move on to the other one, which is like, do you see anything actionable in here? Like, do you think that based on this, workers have an incentive to like use AI more in the hopes that it will mean that they get higher pay? Or are we just sort of observing that like well-paid people seem to be using a lot of AI? Right now, I think we're just observing that well-paid people use a lot of AI. Like the study authors have specifically said they want to do a follow-up that tracks like promotion patterns. And currently, like the age of these systems and the data is such that we can't show causation. Got it. Okay, tell me about the Gallup survey. Yeah, basically, they, among other things, are asking people whether AI makes them more productive at their jobs, as well as how often they're using AI. And one big thing is when people were asked if AI makes them more productive at their job, 71% of people who said they were in a leadership position said that AI made them more productive, whereas like only 54% of people who said they were individual contributors said it made them more productive. And, you know, it has this graph where like as you go down the totem pole, people are a little less optimistic about whether AI makes them more productive. Yeah, like the closer you are to entry level, the likely you are to say AI is bullshit, right? And the closer you get to the C-suite, the more likely you are to say AI is everything. I wrote a piece earlier this year, the AI productivity paradox. It's about exactly this dynamic. And the finding was exactly the same. What hypotheses did the study authors have for why these higher-earning leaders are using AI more or like getting more out of it? Yeah, basically, the top two hypotheses are that they will know how to use the tech and that they will use more computer, they will have more computer-y jobs, which is basically the same as the Financial Times. As the first study, yeah. Yeah. I have a third wild theory on this, which is if you're a more junior employee, you have less executive decision over how you're, like, less power over how your job gets done. And if you try something new and it ends up, like, messing things up, you are just, like, in way more trouble because you don't necessarily have jurisdiction. So you don't necessarily want to go to your manager and be like, how much should I be using AI for this? And especially then be the first AI adopter when something goes wrong. Right. That makes a lot of sense to me, right? It's like scary to be in that entry-level role. And I think often people don't really feel empowered to say, like, hey, like, would you mind if I bring my own tech stack into the enterprise? Like, most people are just going to wait to be, like, you know, given their Notion login or whatever. All right. Well, it's an interesting set of findings. Any sort of, like, final conclusions about, like, what this means or, like, if you're a worker, what you think you should do with this information? Yeah. I actually, I want to bring up kind of a non sequitur, which is there's another finding in the Gallup survey that I thought was pretty interesting, which is how much people report productivity gains by sector. And some of it, so the greatest gains are in managerial and healthcare, which you would kind of expect. The lowest gains are in service. But then, like, the second lowest is office and admin support. And, you know, there's, there's kind of two things here. One is there is a clear reason why you might see lower adoption, which is one, knowledge, and two, like, some of your job in admin support is literally moving boxes around. You know, like ChatGPT is not going to do that for you. Can't do that, yeah. But, like, I've worked as an office assistant before, and, you know, like, some of the time I was moving around boxes, but some of the time I was doing stuff like, I open a spreadsheet, and I check if people are listed twice on the HR spreadsheet. I check if people who are gone are listed on the HR spreadsheet. Like, I port copy into WordPress. And, you know, that's really stuff where I'm not like, Claude can't do this. Claude is not useful for this. In fact, like, I can see plenty of roles within this sector where, like, AI maybe per unit time is more useful. And then another thing that adds a little nuance here is there has been a body of literature previously, like from 2023 to 2025, where people do trials within workplaces of how well, like, how much AI boosts productivity, where typically the finding is that lower skilled workers, people with less job experience, have their productivity boosted more by AI. And so I don't know, these two anecdotes, this anecdotal Software-as-a-service business model to me on a whiteboard, it was super useful to me. Well, little did I know, that wouldn't matter. So, yes. Well, this is kind of my first question, which was like, if you were explaining your business to a reporter like that today, How much of that whiteboard would look the same and how much would just be totally different? Well, I'd probably, if my recollection serves me, it was probably a lot of it was trying to compare the on-prem days to cloud and why cloud was such a big deal. And I feel like my predictive capabilities were pretty locked in, you know, sort of maybe short of AGI. But the whole idea was software is going to move from your data centers. It's going to move to the internet. And in the process, the real power of that is that it becomes available to way more companies, you know, businesses of all sizes, lines of businesses that never could have used software before, end users. And this was sort of this phase of consumerization of IT. So that kind of played out. And then obviously we're now in like the next frontier of what is software going to look like in the future. And I think a lot of the kind of core architectural components hold. Like if you are, you know, running a global, you know, supply chain at a Fortune 500 company, you want deterministic systems and software that power your ERP. If you are at a large B2B Salesforce, you want to have a clear set of business logic around how your CRM, you know, works and how your internal workflows around sales automation work. If you're managing, you know, documents for a government agency or a pharma company or a law firm or a large bank, you want to make sure you can secure that data, protect it, govern it, ensure that it's all in a safe place and available to the right people. So all that is staying the same. What is completely changed is probably the interaction patterns on those systems, where the interaction is coming from. And then obviously, like what you can now do with all that data and information. So the big idea here is that in the future, I would say, you know, probably if today, like 90% of activity on this software is humans interacting with the interfaces of these tools, probably in three years from now, it would be 90, 10 the other direction, which will be agents interacting with these systems, talking to the data, pulling up data from these tools, interacting with this technology. And 10% might be you going and browsing and looking through the software yourself. Now, the interesting thing, and this is going to be the open debate for the industry is, in that 90.10, did the human side go down by 90%? Or did we just have a 10x increase in now the agents leveraging these tools? And my argument is probably more in that latter category, which is agents are this explosion of new workers that are all using these systems, which make the technologies even more valuable and useful because you have all of these new workers on these digital platforms that need data, that need to be able to ensure that they're secure, that aren't leaking information in the wrong ways. So you need those guardrails still, but now you've got this massive multiplier of what people can do with their data because you have agents that can run in parallel. Right. That makes sense to me. There's this really interesting challenge. How is that podcast over? Yeah, that's all the time we have, but I really want to thank you for joining us because I think we all learned a lot. No, let's throw in a few more questions for the super fans out there because you actually just introduced what I think, you know, it seems like a pretty possibly profound change in the business model for what you all do. Right. In these SaaS companies, y'all have gotten used to selling by the seat, right? You have 10 employees, you want 10 of them to be able to use Box. You pay a monthly fee for that. And it seems like that business model is under a lot of pressure in a world where maybe I don't have 10 people in those jobs anymore. I don't actually need the 10 seats. What I need is a business outcome. And you just sort of laid out a case for maybe, well, you know, agents are going to kind of come along and do that. So how are you navigating that? And do you think this seat-based business model survives in SaaS? Yeah. So I think you posited a scenario that is probably the most open for debate, which is, did the people go away? And in my sort of map that I laid out, the people stayed the same number, but the agents sort of multiplied on top of the platforms. And I think the big question is, there will be some software categories where the literal seats are not as relevant because you don't maybe have as many people doing the work as you did previously. I would actually argue that for a large portion of software categories, that won't be the case. It'll actually be the case that you'll have the same number or more people, but then you'll actually also have 10 times the number of agents as people. And so then it's actually this multiplicative effect of more people or the same number of people or maybe a minor reduction and then vastly more agents. And the part that's not being kind of priced in by the market is sort of, is that scenario playing out? And if I just look at a lot of our software consumption internally at Box, there's not a large number of cases that I can make for many of our software products to reduce the number of people that that sort of exists as seats. But there's a lot of cases to be made that there's a lot more agentic use cases for that software. So if I take like an objective example, that's not Box, if I look at Salesforce as an example, we actually are going to have more sales reps at the end of this year than we had at the start of this year. So that's more seats within the Salesforce universe. And at the same time, I can already imagine, you know, you know, 10 to a hundred more agent use cases on the Salesforce platform than I could have two years ago. And those agents might not be again, roaming around the interface of Salesforce. They will show up inside of Claude CoWork or inside of codex or inside of ChatGPT. So the agent will, I will be interacting with the agent via a different interface, but the underlying seat that kind of says, Hey, Aaron is a user in this platform. They have this level of access to this type of data that actually doesn't necessarily go away in this world. And so we're already seeing this within our customers, which is you tend to want a seat for the person because you want some kind of stateful representation of like, what data does that person have? What are their entitlements? What information can they access? But then an agent might, might do an unbounded level of consumption on the software where I, as a person can only like click so many things per day, but an agent can go and do that at a hundred X the scale. So the seat sort of gives me the ability to go in and use my information across these other agents. But then at some scale, you have so much data being used. You have so much access on the platform that then there's a consumption model on top. And this is why I think you're going to actually have this stacking element of the business model in software, which is, which is, you know, humans probably still will have seats, but then agents will be a consumption pattern on top of that. As a CEO, I'm imagining you're looking at all of the SaaS that you guys buy just to run your business. I imagine you might be happy if you didn't have to pay for all of those seats and you could just have agents do it. And you seem very bullish on agents. So like when you just look at your own spending on SaaS, your feeling is truly like, I'm happy to keep spending for all the seats that I'm spending on. Well, there's a difference between happy and practical. So, so, you know, I, I always would like our IT spend to be less, but I'm also extremely practical about how technology works. And, and, and, you know, the, the phenomenon, as an example, I mean, the, the challenge is that the bear case of software has actually like, like it's like a confusing amalgamation of multiple issues that people have. And it's kind of like a Rorschach test of just like, what, what do you hate about software? And then you'll see that in, in the bear case. And so some people are like, well, what, what we're going to do is we're going to vibe code CRM systems. And then other people will say, well, no, we're just not going to have employees. And so it'll just be agents. And then, and then other people will say, well, we actually just don't need all the features of these SaaS systems. So agents will go and do those. And, you know, some of them I'm sympathetic to some of them. I'm not the one I'm not, I'm extremely like on that continuum. The one I'm extremely not sympathetic to is, is we have no projects internally that I know of, at least that I've, I've kind of at least approved to kind of vibe code a replacement to an existing SaaS service. Because, because if I look at the stack of at least the ones that matter, if I look at the stack of like our ERP system, our HR system, our CRM system, our document management system, it would do us no good to spend our, our time and energy and IT resources trying to replicate functionality that, that is already kind of doing its purpose, especially Correspond to the level of trust you have to have in the platform. There's probably one more element, which is like, how much does the system benefit from a world of multiple agents needing the data as opposed to one agent needing the data? Because then that would point to whether the enterprise wants to sort of put all of your value into one of the labs or one of the products, or does it need to be a different layer that everything kind of talks to? This is why you still see companies like Databricks or Snowflake are growing quite well. And this is sort of what we're experiencing is, is actually you don't really want to move your data around constantly. You want your data to be be kind of, you know, abstracted from where the agent is because you want to be able to to structure your data, secure it, govern it, and then let all the agents talk to it. And that that sort of reinforces the need for that as a kind of control point. So I would just say there's a continuum and like if you, you could probably, you'd probably have like a quotient for like how durable is the is the platform based on those those factors. Right. The levy formula. Here's what I'm taking away from it. If you have a to-do list app for teams, get out of that business. That's not going to work based on what Aaron just said. I would say that business is actively pivoting. Yeah. Yeah. I actually think I know one that that might be. I know Box has talked about AI for a long time, but I'm just curious for you personally, you know, we do have these momentary enthusiasms in Silicon Valley. I think it's fair to say both of us had a crypto phase. I'm imagining that your AI phase started earlier, but maybe sort of took longer to kind of reach the fruition it's at now. But like, did you have a kind of moment of conversion where you like saw a paper, a product, something where you're like, okay, like I need to start taking this really seriously and spin it up teams. Yeah, well, there's been three moments. If I, if I, like, it never is this perfect, but like it almost, you could almost think about it as, as, as like this. Um, there was a moment 10 years ago, more or less, maybe, maybe, maybe nine, eight years ago where like vision models were getting good enough where we saw these scenarios where you, you know, you give, you give the vision model a, a document and it could OCR it, or you give a vision model, an image of a, of a retail product and it could kind of classify it properly. That was a pretty big deal because you could imagine that like, okay, like if we could just convert everything into into a visual, you know, kind of system that then these agents or these AI systems are pretty good at that. The problem on the tech side was you had to train individual models basically for every domain at the time. So this was sort of just before the transformer. And, and so if you wanted to do like, like document classification in legal, you had to have a different model than if you wanted to do like financial equity research analysis, which meant that you never had the takeoff moment of AI. So that, that was, but, but we, we, we, we got the first sense of like, oh, this is a big deal. Obviously big deal number two, ChatGPT, because, because while we, we kind of, you know, we're following GPT-2 and 3, and we had a hackathon where somebody did like GPT-2 inside of a document and, but it was like, it was like producing like garbled text. So it like really wasn't, you know, maybe it could like type ahead, you know, five extra words, but like, was not going to game change your productivity. So it didn't have like a complete real religious moment. ChatGPT was sort of the first time in this era where it was like, okay, this is like a very big deal. We're going to be able to now wire up these LLMs and connect to your data and then do meaningful things with them. And then I think probably the most recent one in the past, let's say year plus has been kind of this kind of, you could probably, you'd probably have like a quotient for like how durable is the, is the platform based on those, those factors. Right. The levy formula. Here's what I'm taking away from it. If you have a to-do list app for teams, get out of that business. That's not going to work based on what Aaron just said. I would say that business is actively pivoting. Yeah. Yeah. I actually think I know one that, that might be. Um, I know Box has talked about AI for a long time, but I'm just curious for you personally, you know, we do have these momentary enthusiasms in Silicon Valley. I think it's fair to say both of us had a crypto phase. I'm imagining that your AI phase started earlier, but maybe sort of took longer to kind of reach the fruition it's at now. But like, did you have a kind of moment of conversion where you like saw a paper, a product, something where you're like, okay, like I need to start taking this really seriously and spin it up teams. Yeah. Well, there's been three moments. If I, if I, like, it never is this perfect, but like it almost, you could almost think about it as, as, as like this. Um, there was a moment 10 years ago, more or less, maybe, maybe, maybe nine, eight years ago where like vision models were getting good enough where we saw these scenarios where you, you know, you give, you give the vision model a, a document and it could OCR it, or you give a vision model, uh, an image of a, of a retail product and it could kind of classify it properly. That was a pretty big deal because you could imagine that like, okay, like if we could just convert everything into, into a visual, you know, kind of system that then these agents or these AI systems are pretty good at that. The problem on the tech side was you had to train individual models basically for every domain at the time. So this was sort of just before the transformer. And, and so if you wanted to do like, like document classification in legal, you had to have a different model than if you wanted to do like financial equity research analysis, which meant that you never had the takeoff moment of AI. So that, that was, but, but we, we, we, we got the first sense of like, oh, this is a big deal. Obviously big deal number two, ChatGPT, because, because while we, we kind of, you know, we're following GPT-2 and 3, and we had a hackathon where somebody did like GPT-2 inside of a document. And, but it was like, it was like producing like garbled text. So it like really wasn't, you know, maybe it could like type ahead, you know, five extra words, but like was not going to game change your productivity. So it didn't have like a complete real religious moment. ChatGPT was sort of the first time in this era where it was like, okay, this is like a very big deal. We're going to be able to now wire up these LLMs and connect to your data and then do meaningful things with them. And then I think probably the most recent one in the past, let's say year plus has been kind of this kind of, you could probably, you'd probably have like a quotient for like how durable is the, is the platform based on those, those factors. Right. The levy formula. Here's what I'm taking away from it. If you have um a, a, a to-do list app for teams, get out of that business. That's not going to work based on what Aaron just said. I would say that business is actively pivoting. Yeah. Yeah. I actually think I know one, um, that, that might be. Um, I know Box has talked about AI for a long time, but I'm just curious for you personally, you know, we do have these momentary enthusiasms in Silicon Valley. I think it's fair to say both of us had a crypto phase. Um, I'm imagining that your AI phase started earlier, but maybe sort of took longer to kind of reach the fruition it's at now. But like, did you have a kind of moment of conversion where you like saw a paper, a product, something where you're like, okay, like I need to start taking this really seriously and spin it up teams. Yeah. I mean, it's like, it's like, uh, so pedestrian that I'm, I'm, I'm embarrassed to share, but like, I was, uh, I was going into a city and uh like a week later, and I needed to map out a bunch of customers that I should be visiting. Um, and because I, I had like, you know, half my schedule filled and I, I had room for some more. So, um, and this, and this specific example that I think I was referencing, but this happens like three times a week. And this is the example I was using perplexity computer, which, which does some pretty good workhorse stuff. And I just gave it the task of like rank order all of the, you know, top 50 companies in this region, uh, get me the LinkedIn of every single uh CIO of those companies. So I could just make sure I'm like, okay, who, who have I connected with, whatever. Um, and, um, uh, and then, and then I didn't even know what I would do next. Um This is absolutely true, by the way. Why are some people so quick to think that AI can automate away a whole job? I think we, first of all, this is amazing technology. And like, and like it is it is the coolest technology, you know, at least that I've ever played with in my life. And so to some extent, it's like, it's like deceptively cool because, because it's like, like, oh my God, I think I just did my, my, my taxes or like, oh my God, I just built this amazing marketing website in like five minutes. And, and then we, we sort of like look at the output and we're like, gosh, that must like totally replace the job of like XYZ profession. And there's a, there's a few kind of like core flaws with that, which is like, well, what is that profession, you know, doing for, for all the hours in their day? And like how much of it is, is, you know, just doing the final calculation of your taxes versus no, it's like getting all of your data in order. It's reviewing all of the work in the process. It's asking you questions back and forth on like, what are you optimizing for? And it's knowing the right questions to ask. It's, it's like, you know, dealing with the, the three missing things that, that you didn't even remember that you're supposed to add. But, but if like an AI system had done it, it would have totally glossed over. Like that's what the profession does and the automation of one or two or five of the steps are just these individual tasks that it's able to automate. In the, in the case of development, like we're seeing this day in and day out. It's like you or I can go and tell Claude code, generate me the XYZ product. And we could be like, wow, that must automate the engineer out of existence. Well, there's a couple issues. Like one, like, like the code quality is, is probably horrendous. Like the ability to like now ask it to do 40 other things over a 12 month period is just going to stack in complexity. The moment that there is, you want to actually get that, that software hosted and run and make sure there's no downtime and ensure that you have a good distributed system is already a hundred times more complex than you just prompting the code to get written. The moment that there's a security event, all of a sudden, somebody's got to like wake up and respond to that. You know, I can name 30 other things that a developer has to do. Like you have to like understand like, what's the roadmap? Where's the company going? All of those things. And so all of a sudden you're like, oh, the job of the engineer is absolutely writing code. And there's a lot of people that say like, you know, the job of the engineer was never to write code. It was to do X, but it's like, no, no. They're like, they're writing code most of the time in the, in the prior, you know, world of work. The problem is, is that you were highly constrained by how much code they could write in a day. And then, and then they were just automatically bottlenecked by that to do the other things that their job could be. And so what is the future engineer? It's like, yes, it's to understand the, the, what are you trying to build? It's to make sure that it gets built properly. It's to ensure that there's no security issues. It's to ensure that it gets released. It's to ensure that it's high quality, all of those things. And so if you or I go and VibeCode something, we think we've replaced the engineer. We think we've replaced the accountant. We think we've replaced the lawyer that we get advice from. But then you actually go and look at like, okay, that was like, that was like, you know, that that was the first 80% of the job. But the extra 20%, it turns out is like all of the value creation of that profession goes into that last 20%. And all of the expertise and domain knowledge is in that last 20%. Not the, not the text that got generated. And, and so, and so that's like the misunderstanding. And then, and then the converse is like, we look at AI and we're like, we're like, we want to use it for XYZ thing. I'll use it for like analysis of like a market that I'm thinking about. And if I just took the output of that analysis and I ran with it, I know it would not work because I know that it's missing context that, that, that it doesn't know because either I didn't give it or I know something else about a different trend. But somebody else might see that and say, wow, Aaron's job is like incredibly easy. And like, like AI just gave me the answer of what he's going to go do. And, and then I'm like, I'm like, no, actually my job is way harder. I promise. And I think that's kind of what, what is sort of happening is like the moment you have a five or 10% flaw in what the answer was, the moment you have a security event, the moment you have to actually like see the whole thing through, that's really what you're paying that worker to go and do. And, and I think you're just, you're, we're going to have this duality of like, we're going to see these amazing things and then you're going to have to go and, and, and see, and then that last mile of real work that has to happen, it doesn't go away. And, and, and it just, the last mile keeps moving and moving. And so, you know, you look at, you know, Dario has this interesting thing about, about, about kind of like, you know, you first automate the 90% and then you automate the next thing. And then, and then, and then you kind of increasingly sort of like are automating out the software development life cycle. But I would almost, I would, I would say that there's a different axis that, that, that maybe, you know, people need to think about, which is if you took today's static work, that actually maybe would work. It's like, you'll get the first 90% and then we're going to automate the next 9% and then we're going to automate the next 0.9% and so on. But actually what happens is, is there's a dynamic part of, of the equation that's not represented by that, which is the market is starting to ask more from the provider because they now know what is possible. And so just as you automated that first 90%, all of a sudden the market shifted on you and that 90% is now like the new 50% because actually the demands of what you ask an engineer to do just go up tenfold because you're like, I think you can do that thing way faster now. So I'm going to give you a much bigger project. Or I think that, that, that my, you know, tax accountant or my lawyer should do way bigger analysis of the, of the topic at hand as opposed to just give me like a rudimentary answer. So you have this other dynamic system that's happening, which is our needs and demands are just growing as a result of what we can go and automate. I mean, I, I love the idea that AI will just sort of let us re-envision what our jobs could be in a sort of a more expansive, creative vision, right? My fear is that the last mile of super, my fear is that the last mile of human supervision will just turn out to be like kind of boring, right? It's like, I feel like we're already starting to see this in some jobs where it's like, there is a piece of this that is automated and my job is now just to review AI output and that is just like pure drudgery. So is that a factor here? And does that like complicate the picture that you've just painted? Well, it doesn't complicate the picture. It adds a, uh, because, because the big question is, are there jobs in the future? And the answer is yes. Now, now the question is, do we want those jobs? Like, like I'm reviewing the output of AI agents. Um, uh, so maybe we all just like opt out of the economy because we're like, God, I don't want that job. But, but so interesting philosophical question of like, what is the new way to get fulfillment and creativity out of these jobs? And, and you can see it. You see burnout of, of engineers on Twitter or X that, that are basically like all my job is, is to review, you know, slop from the AI and like that's, there's a limit to how fun that is. Um, and so I, I, you know, I, I, I think super interesting question. I think there will be a, a continuum of jobs, though, as an example, because, you know, engineers are facing this first and they're also facing the existential kind of dread first. But engineering is, is a very unique job compared to the rest of the economy. Um, and we can kind of get in some of the differences and actually why I think diffusion will take longer than most people think when they just look at the engineering work. But engineering work is like, you know, most of your day is like, like, I'm not, I'm not trying to be reductive, but like your, your job is to, you know, like, obviously think about a problem, think about a system, write code and that code is text. So you're just write a lot of text and then somebody else reviews the text and you ship it. And so if the agent just did They have to like verify that the advice is good, or they have to go and review the sort of contract that got written. So, what we do is we're lowering the barrier for everybody to participate in these things in kind of like a touristy way. Like I can like be a tourist in software development, I can be a tourist in legal, I can be a tourist in healthcare, but that eventually still needs to get verified or the work actually has to get done in that last mile. And that eventually still moves into then you need some kind of like semi-expert to go and do this, which is why I actually don't know that you can yet claim what degree should you go into in college. I don't think we yet know that one of the degrees are off the table. Like, I think you should totally go into CS if you're really excited about software development. You just shouldn't expect to go build a little app that you press a button on. You should expect that you're going to use CS skills to go and do clinical trial automation at a pharma company. Well, it's I think a great place to land because it's the, it leaves me with a feeling that I have so rarely when thinking about the tech-enabled future lately, which is optimism. So thank you, Aaron, for giving us a jolt of that and would be great to maybe check in with you again in a year and see if the picture still is rosy. Awesome. Aaron, thanks so much for joining us. All right. Thanks, man. Platformer is produced by Lindsey Chu and edited by Fitz Harris at Story & Sound. You can watch this whole episode on YouTube at YouTube.com slash Casey Newton. My email is Casey at platformer.news, and we'll see you next week. Take your team from AI novice to AI native with Atlassian Rovo. Go to rovo.com to learn more today.