Whose Jobs Does AI Really Replace?

(soft music) This session was recorded live at the 2026 ASU GSV Summit in San Diego. (audience applauding) Welcome everyone. Standing room only, this is a testament, I think, to you all and the topic of whose jobs does AI really replace anyway. We're gonna get some answers about the past from our steam panel today. Some answers about the present, and then we're gonna look into the crystal ball that no one really knows about. And see where we're going. We've got an all-star panel here, Matt Sigelman, of course, Burning Glass Institute all the way on my right. Then we have Shod Ahmed from Anthropic. We have Joe Fuller from Harvard and Allison Salisbury from Humanist Venture Studio to answer the question. I'm Michael Horn, I'm gonna try to fade in the background and listen to these folks talk. And I wanna start with the past, and Matt, I'm gonna start with you and this. My mentor and Joe's former colleague, Clay Christensen, used to say that the really inconvenient thing about data is that when God created it, He only created it about the distant past to really know what was going on. And so I figured we would ground ourselves there because AI has been around for several years now. We're starting to know some things. I'll let you talk tech revolutions in general, but also sort of what have we seen in the early innings that we feel like we actually have some confidence about what AI is doing in the labor market. - Well, terrific. In the spirit of the Ghost of Christmas Prass, I'll rattle my chains a little bit. And I actually go back and take advantage of my age here. I wanna rewind the tape to 1998 for those of us who can remember that far back when the internet was a new thing. And I think it's an instructive thing. We've been spending a lot of time actually going far back to your point, and I'll talk in a few minutes, either now or when we talk about the Ghost of Christmas Present about what we're already seeing in these early innings of LLMs and the AI revolution. But there's a lot to learn from past tech revolutions. If you were going back to 1998, which was the greatest mania since the Dutch tulip craze, what you would look at the predictions that we're going on, and you'd be right to be skeptical. And I think what you would find if you then fast forward to where we are today, you would say that in fact, the hype was understated. So I think we've done some analysis recently found that about 90% of workers work in a job that makes significant use of the internet. That's probably not what most of us would have expected in 1998. But what you'd also find is that those changes took a lot longer. Some jobs are actually still going away. Meeta Readers only just started to disappear in the last five to seven years as connected devices became more prevalent, parking attendance. There's an app for that. But what you would also see is that they are, and you would also see by the way that they are less predictable than you would expect. There's 65,000 job openings last year for social media managers. Social media didn't emerge for another 12 years, and would have been very hard to predict. Most of us in 1998, except maybe the thwarted CEO of WebVan wouldn't have really anticipated that the biggest category of internet job creation would be for delivery drivers and warehouse workers. But what I think is also, and I think this is where it connects to what we're seeing already in the early innings. The biggest internet and past tech revolutions did carve away some jobs. There are reasons why AI may be different and we'll sure we'll talk about those. But while with apologies for my former travel agent, what I will say is that most of what we've seen is that past tech revolutions have changed how we work rather than weather. And that's what we're seeing so far. We're seeing relatively few jobs that are showing signs of going away, in fact very few indeed. That doesn't mean that's not gonna change. But what we are seeing is very meaningful changes in the definitions of jobs. The skills that employers are looking for in those jobs. And in ways that very much align to the interplay of automation and augmentation and how they are rewriting work. So with that. - Well, let me ask one question and then you've shod with that because anthropic put out a very interesting report looking at historical tech revolutions and the current coverage, if you will, of AI versus its theoretical capability. So I'm just sort of curious where you agree with Matt on the past of what you guys saw and where you disagree as well and then Joe and Allison, I want you guys both to jump in there. - Yeah, thanks. So what we agree is this idea that jobs are a package of tasks and those tasks are reformulating as we speak. And so talking about job disappearance is challenging. What we're trying to do with our data sets is just monitor how those tasks are being refashioned as AI you starts to increase. And so we will see some jobs shift in change. We will probably see some new jobs emerge and some might disappear, but that's going to be a process that we all have to be watching very closely, especially sitting in the education sector where we're trying to monitor what's happening with the labor market. I think we might have a divergence and you didn't say this explicitly, but this is happening much faster. So based on our last data, what we're seeing about diffusion of AI is we will probably get to, and on the US, for example, national use of AI at the 80% levels in the next seven to eight years. For example, with the internet, it took almost twice that amount of time to do that. So we're seeing that rapid uptake as we go. I think the second thing based on this data drop, so we have the spider chart that went viral even by anthropic standards. And it was like a test of if you're an investor, you saw investment opportunity, if you're a student, you got nervous about what your major was. If you're in higher ed, you would see what some of the areas of focus could be. And but in that work, we realize that we have breath right now. We have about 50% of jobs, about half of jobs are using about 25% AI on 25% of tasks. So that number is starting to increase. And we see depth only in very few areas. So 4% of jobs like software engineering, we're seeing AI use across 75% of tasks. So I say that to say we're very early in that, in this spectrum. The last thing I'll leave is productivity gains are starting to accrue with those who use AI the most. So our last report showed that if you've been using AI for about six months continuously, you're starting to see outsized ROI at the individual level. And we're starting to see that at the firm level as well. And so again, we're very early. And as these models have just after the cloud Christmas, and these models have rapidly improved, you're now starting to see those productivity gains start to hit firms and individuals. And so again, we're early. We'll see how those gains play out and what that means for you. If I could just bill on that for a second, I think it's very dangerous to try to extrapolate on the current data for a couple of additional reasons. This is not only but also point, the vast majority of companies are not doing anything that remotely resembles organized training as to how to use this. And data we put out quarterly at Harvard working with Vanderbilt is you're almost twice as likely to be using AI at home as you are in a job for the average person. I'm not talking about people in this room. So if you're using a home, what version are you using? The free one. And you guys blew my socks off in February. And the advanced subscription tools are for worries compared to the tenural preuses of the free versions, no offense meant to preus. I drive an electric car myself. So we don't know yet what people unafettered using sophisticated tools with an adequate training. Bizarrely, 80% of white collar employees expect their supervisor to teach them what to do and tell them what they want done. And 90% of supervisors expect their white collar subordinates to come to them now with AI-driven recommendations and ideas. So there's a football just bouncing along the ground. No one's falling on it. Some companies, though, are really working this hard. And I know some of them, and I'm seeing the productivity impacts there. And it isn't so much jobs going away. They're getting reformulated, but much more efficient. So total positions in that emerging role, I would anticipate will show. New rules, new ideas may come up to offset some of that. But when I looked at your spider diagram, I said, I'm seeing that in action in better companies. I'm seeing they're pushing out to that efficient horizon much faster and that's going to have a profound impact on their economics. - One last idea I'd introduced about the past because I think it'll be relevant to our conversation about the future is that when you look at, not just AI, how technology has changed work over the last 10 or 15 years, it's not just at the task and job level, it's at how organizations work and function. So you think about even just our alliance on email, things like Slack, shared knowledge bases that we pull from, the ability to do conversations live with any in the world, any time through Zoom or other platforms. It's changed the way that we set up in design organizations. You think about the modern product and engineering team, for example, looks nothing like the equivalent would have looked a decade ago. And so I actually think one of the biggest waves of change that we're gonna see around the impact of jobs is how organizations choose to redesign themselves. So not just at the task level, not just at the job level, but at the organizational level. And you can see a lot of this going on right now in early stage companies. How do startups choose to organize themselves, increasing looks fundamentally different than how startups would have chosen to organize themselves even a year ago. And also you see this at the other end of the spectrum. You see folks like Jack Dorsey and all the headlines he's gotten in the last few weeks around Dorsey mode and redesigning block, really putting AI at the center of middle management and redesigning all of his jobs and his entire organizational structure to all be doers and builders. And so I actually think this organizational transformation will end up having one of the most lasting and transformational effects on the future of jobs. But for some reason is not something we choose to opt in reference or discuss. - Well, and it echoes right the analogy of electricity in many ways and right initially you didn't see those productivity gains when you were subbing it in for steam and then the real redesign of both the factory and the organizational model rounded was the quantum leap forward. And I guess that brings us to the present because my wife and I hosted a dinner party the other day. There was like six people from very different professions. Every single person had a very different lived experience at the moment with AI. And it goes a little bit joe to what you were just referencing in terms of some of these productivity gains, these big lean in that certain companies are doing. And my sense is that there's different sectors, different companies are approaching this very differently. So I'd love to in the present moment get a little bit of that nuance a little bit. Let's name some names, let's get into some sectors and start to help us unpack why is data different from coding, different from marketing and so forth. - Let me just start with a couple of observations. I'm sure my fellow panelists have a lot to offer on this. Even today over 60% of executives describe deploying AI as a technological task. In my estimation, that's just a complete mystery to the situation. AI is about managing work and it's about organizing processes. It's about, and someone who bolts this on to their existing process like the next SAS software package they could buy, they may get some incremental efficiency gains, but they're missing the point. The general purpose technology. You have to create your process of just if Dorsey is. And I can't say I've ever really invoked him as a model of management previously. So there's the first for everything else. We won't tell your employer. - But you have to create the process around what this technology does. The vast majority of companies are prepared for that. The large companies I see making good progress here have isolated several, usually not more than three, what I'm going to call main sequence processes. So very fundamental to their competitive position. They have effectively followed a model not unlike Apple in their revolutionary pride development. They put the process they're going to use, deploy the AI and they insulated from the rest. This is what's driving some of you be familiar with the J curve concept. Companies are running parallel processes right now until they make sure they've got their arms around the new process. But you will see companies and I will just mention a few. JP Morgan, Jason Wells Fargo would be good examples. Proctor and Coca-Cola would be good examples. Where they are pushing it hard and fast in those main sequence things like risk analysis for the banks and like marketing and promotion in the consumer package goods companies. And the economics of those which I am familiar enough to say would you with confidence are very, very compelling. So now there's another issue that crops up here which is if you're the biggest bank in the United States or the biggest beverage company in the world, JP Morgan, Jason Coca-Cola, if you are standing now on the accelerator and learning faster than your competition, you will get to a point where it's mathematically impossible for the other person to close that gap. So the opportunity for early leaders to gain significant year you have to be very careful because I don't want any of my friends quoting me in court in the future. But let's say greater levels of competitiveness in some of these markets is very, very real and got through all sorts of interesting questions out in the future about industrial policy and antitrust law and whatnot. But I just for the future conversation. - So many good things you could go through. Just to build up Joe's comment, one of the things we are seeing among our customers which are among the most early adopters and many of them are at the frontier with us is there's a choice. And I think to Allison's point, there's a lot of choices and that's why we're early and we can shape some of those choices. Market forces, quarterly shareholder pressures will enact a certain set of choices. There's also alternatives that we're trying to explore. So for example, when you take the productivity gains from AI, we are seeing certain companies take those productivity gains and invest them back into growth, innovation and their people. And Walmart's been very public about trying to do just that. And so, and you have others who go straight to the bottom line. And those are discrete choices in the same industry in some cases. In some industries, you don't have as much growth. You're competing on margin, you're competing on cost and you're gonna see a lot more efficiency gains and jobs taken out of those industries. But in other places, you can see this idea of doing more with the people you have. And while we're wondering, we're trying to build the evidence and understand what are those companies doing? Because we're seeing anecdotally, they're getting outsized return on investment. They're creating safer spaces for other people to experiment. They're doing some of these more top down strategic projects. And then, Rob, we talk about, this technology can help your employees work smarter. It can make your processes faster. It can make your products better. And usually the companies that are, and organizations that are outgaining the others are picking two of those. Usually they're getting the technology in the hands of their employees. And they're being very intentional about specific processes or products that they're trying to put this AI on. And then what they do with that, insane productivity gains. And we see it in the topic every day. We've baked them in, is you can do more with the people you have and it creates even more opportunity. And our organization, even our AI driven organization, is not yet structured to deal with that massive productivity influx. But there's a universe of more opportunities and back logs were seeing with engineering teams that they're finally getting to, that they could never have gone to with the resources that they've had. And so there's some interesting effects that will start to see play out, especially in growth industries. I've been really oriented recently on trying to do my best to share where I'm changing my mind on things, because I think that's where the most interesting nuggets are. And you sharing that made me think about something that I have started to change my mind about just in the last month. And actually watching some of what's unfolding a block was some of the inspiration where I had this mental model that the companies that will win are the companies that use AI for productivity and then take the extra people and reinvest those people and their talents in next horizons of growth to do things that they fundamentally couldn't do or couldn't imagine doing before. And those the companies will win. And so the container of headcounts should kind of stay within an equal liberium as we go through kind of the next, let's say one or two chapters of this. I have started to change my mind on that. With the block layoff, it was pretty significant. It was like a 40% layoff. In some parts of their organization, it was 70%. And Ethan Malik was quick to tweet, that's actually a losing strategy. You should take that extra headcount. You should invest in doing something different in driving growth and driving innovation and driving creativity, which is sort of where I was at up until that point. But now working with my teams with early stage companies, actually you move slower when you have more people. and the sort of joke that I say is like humans are lossy. There's a lot of shared contacts you have to have with a team of humans to go do something innovative and do something quickly. It requires meetings, it requires communication, it requires alignment. And if you actually have fewer people with a tremendous amount of lever to my eye, in some cases you can accomplish far more than with the team twice that size. And so I've really started to, because of the way the technology has developed, started to question this notion that the container, even for companies that are taking this as an opportunity, not just to drive productivity, but to drive growth and innovation, it is still possible that the container will continue to shrink rather rapidly. So just sharing this is kind of something just the last few weeks of sort of updated my mental model around. You're right. I'm right, he says. I think there's a lot of attractiveness to on the extreme side of that the sort of mythology around the single founder company and we've all been hearing about that. I think it's a little bit early to write off the advantages of scale, not just in terms of assets, not just in terms of data, but in terms of people. I think Joe pointed specifically to the importance of using this as a moment to rethink, not just work, but how we organize work. And I think that's what's going to allow this to be a moment of greater productivity. I think it's important to, as since we're at the ghost of Christmas present here, to see, just bring some evidence to notwithstanding the counter of the block layoff, what we see actually more broadly bearing out across the economy, which I think is more supportive of Ethan's hypothesis that this is a moment of greater productivity. Here's what we see. We were actually pretty surprised by this ourselves, but we do a lot of work trying to just measure what are the parts of jobs that are going away, what are the parts of jobs that are going augmented, that whole sort of weighing in on the whole tastes great, less filling of automation and augmentation. And what we've been seeing is rather than some jobs being characterized by a lot of things being automated away, other seeing a lot of activities being augmented, we instead see that they're happening in the same jobs. As I've my fellow data nerds here, it's a 0.87 correlation between those two. This is quite significant. It means that nature pours a vacuum. When you start to take tasks away, people start doing other more valuable things. But that's also why it's so important to Joe's point before that we redefine the unit of analysis. As far I think we've been thinking about this in terms of jobs, maybe we get down to tasks. But the real unit of analysis here needs to be use cases. How specifically sector by sector, by role by role, how are people using AI? Because it's only when we can define the use cases that AI is being applied to. And we're doing a lot of work to actually track that and say, OK, how are those use cases spreading to Joe's point right now? It's a relatively small fraction of jobs that actually can define a use case where the job isn't just broadly saying, hey, we're an AI first company. But when you actually get to that precision, it is spreading. It is growing very rapidly. But when we get to that level of being able to say, how are people using AI, then we can actually start to understand the broader set of questions. I know we're all here to discover which jobs go away. How do we change what we do inside a job? OK, well, here's-- and when I say use cases, what I'm really talking about is the commercial imperatives that people are applying AI toward. Here's the things that we can do in our sector that AI is going to help us solve. If I'm Brian Nichols at Starbucks, I say I want every customary served in four minutes. How is AI going to help us deliver the four-minute latte? When you do that, then we can say, OK, here's how we have each person's job changes. Here's who's getting more efficient, here who's doing more. And then we can figure out how this all plays out from there. I want to stay on this because as we start to move into the where's it going, and the ghosts of Christmas future, I guess, Matt, tease your analogy. The question I'm curious about is, I hear a bit of-- it depends along a couple vectors. So one is, is it growing? Is it not? And then the other is, does it create more friction to have more humans? Or is it like radiology? We'll just actually order more exams. And so there's more work for them to do. I'm sort of-- Matt, I'd love to start with you here. Start to break it down a little bit. Where-- if you look at that task level at the actual jobs, where are you seeing future growth? Where are you completely uncertain? Because this may be different from past technologies in terms of growth curves. And the fact that the capability of it is increasing at the exponential rate it is and so forth. And where are you seeing evidence that, yeah, no, this is going away. Get out. So I wish this were just going to be sort of like a-- everyone turns out to be tiny tom or tiny tom. Tiny tom? Tiny tom, OK. You know, get up on my dick and see her. But I think exactly to your point, a lot of this is going to hinge on this question of which jobs and which workers have agency to innovate. Ultimately, AI is where AI is going to do more than create efficiency, more than just sort of free up assets or the like and create access capacity in the workforce. Where that capacity actually gets redeployed is in the jobs where people have the agency to use AI to create to say, OK, I'm going to innovate inside my job. That isn't necessarily just a set of white collar jobs. There's a lot of jobs out there where people have the agency. And again, this goes back to the question of structure to really do ingenious things. Just look at even just something as simple as a real estate survey right now. We're going to-- we use drones to do that, to do a whole set of things that had to be manually done or just weren't done before. So where do we have that agency? And in those kinds of places, we can productivity as a ratio. It's a ratio of output to input or of what the value of what we produce to the cost of what we produce. Those are going to be the jobs where the ratio is going to tilt toward the numerator toward the value, toward instead of productivity happening by using less labor, using lower cost labor. It will be yielded by where does people's endeavor come to be worth more. I'd jump into for a second. What I think-- I'm going to give a different calf to what Matt was just saying-- people who can do that have a durable asset in the labor market, which I'm going to call contextual intelligence. So they may have a deep understanding of the purpose of the process they work in, where they may have a deep understanding of a market or a marketplace or a competitive reality. And when they move to that more expansive frontier match describing and exercise more agency over what they do, in part, because AI is creating more of a buffer to some of the constant-- kind of for those of you familiar with Cal Newport, the Georgetown's work, constant interruptions in the workplace and productivity inhibitors known of things like email and Slack and your smartphone and end. They will have the ability to work with the AI to get ever-improving outcomes. And that is-- that's going to unlock a lot of productivity in companies. In jobs that really don't require a lot of contextual intelligence, but which involve lots of routine cognitive work. Those jobs are going to be very much imperiled. They tend to be lower value added. And one of the things that's kind of misreported is everyone thinks, you know, AI is coming for your recent college graduate kid with a 39, in a serial discipline from a selective school. It's going to hit lots of clerical and lower value added work often occupied by people with some college degree, maybe a community college, or just good work experience. And it'll do that because those jobs just don't require the cultivation of the level of contextual intelligence that makes that human being a good partner with the AI. We should talk-- it's going to be on a panel. We should talk about what I think we're all starting to see on early career. I think we're-- that's-- That's the place where the data is unclear, but it's moving in the direction where there are less jobs empirically available that used to be available for early career and I think these entry-level white collar rolls. And so, I mean, it is probably a no-higher, no-fire situation right now, and we can, among many of our customers and partners, I think they're being very cautious. And so, I think there's implications of that. What is that in software engineering, for example, where we've done a study on our own work and how it's changing, most of our early engineering talent are managing multiple agents, right? So, they're playing an engineering manager role. And in our AI-fuelency framework, discernment is really critical. And discernment is a skill you learn through your first few years on the job. And so, the question we're asking ourselves is how can you accelerate to get to that level of discernment and that pattern recognition that you often learn in the first two to three years in your job? And maybe there's a role for higher education and other town developer programs to do that. The other concern is what Joe's pointing out is the recent work from Opportun Work and Brookings Institution looking at the gateway jobs for stars, for those who don't have degrees, and how those are under threat. Many of those are those clerical jobs and other jobs that get you into corporate America and then that you then prosper over time as the data has shown. If those disappear, what those ladders of opportunities start to look like becomes very unclear. And so, I think these are the questions that we're concerned about. We're trying to do some action learning with partners like CodePath and others. How do you move people up that spectrum very quickly within an education program so that they could be ready for these jobs? Given knowing that these jobs are changing every six months and the AI that's underneath them is changing rapidly as well. And those, if we were to train people on anthropic certification six months ago, would have been completely outdated. And so, how do we embed that, and view that dynamism among these partners so that we're all moving at this transition because if we don't, that by-modal distribution of those really using the technology and seeing it, and then those who aren't are left out, it will get extreme. And that's something we talk a lot about is that access question because if you send Silicon Valley people say, "Oh, this technology is great. Everyone's going to get it eventually." Yes, that's true, but we know that it takes time and specific communities to get access to that technology, and they're oftentimes behind and given the outsized productivity gains you see, it becomes more of a concern of how you participate in the labor market even in the next five years. So I'm curious, and Allison, I'll go to you, and then everyone can jump in more on this future question. You've actually, as you start to think about the role higher ed and other parts of the education system play in entry-level work and getting people that early experience, and so forth. You're seeing students actually rapidly, of course, correct in terms of what they're enrolling in. We have the software engineers who are freaking out headlines and papers and dramatic reduction after years of increases in those sorts of majors and things of that. Nature. And Allison, you've observed that some of this debate depends on the perspective you take around which part of the stack you will be in, if you will be, with AI. Can you expand that a little bit and tell us how you see, like, why are we having such sharply different takes on where work is going? Yeah, so on this specific question, I am a very visual person, and I came up with this visual of the bot sandwich, where you have humans above and below and the bots in the middle. And I came to this because I interview experts on the future of AI learning and work all the time for our publication, and I would get these, like, evangelists on how it was going to change work for the better, make it more purposeful, only agency and creativity, shorten the time cycle between idea and actually bringing that idea to life, and I'd get people with the exact opposite. AI will be used and already is being used to surveil, to optimize, to dehumanize work. And I was like, how is it possible that these incredible experts, I so deeply admire and are so data centered in their work, are coming to such radically different conclusions on the implications for purposeful work in the future. And it struck me that what the difference was what roles they were looking at. Are they looking at roles that sit above the bots that are directing them and getting leverage from them, expanding their span of control as a result, and being able to do things they could never have imagined before, or they below the bots? Are they being managed by them, told what to do, surveilled every day, every ticket, everything? And increasingly, it's getting harder and harder to maintain the skills and the competencies and the capacities to stay on top. Because it requires judgment and discernment and intuition and the ability to leverage these tools, not just in your own work, but in a cross-functional way with your peers. And so one of my concerns is that as the AI gets more powerful, that top piece of bread gets smaller and smaller and smaller, and more and more people start falling sort of below the line. And I think those are the jobs that by definition start getting more wrote, more rules-based, and as a result eventually also get automated away. And so that's sort of like one of the ways that I interpret when people are very excited about the future of work versus very dystopic about it. Usually it's what people and what jobs are they looking at. To shift a little bit into the future conversation, there is sort of a framework that I like, another nerdy framework. Raise your hand if you know the three-body problem, the book or the show. Okay, great. A room full of nerds as well. The physics, the three-body problem is when you have three gravitational objects in space moving around one another and it becomes very chaotic to predict the path they will take because they're all interacting with one another as a result of, as in contrast, it kind of two bodies where it's actually quite a linear trajectory, you can really model and predict the future. And so I call it AI's three-body problem. As a way of like understanding why is it so darn hard to predict the future of work? Like why is it that we're doing all this like, you know, chaotic sense-making? And it's because reasonable experts can agree across all three bodies. The first is how capable will the models get how fast? Period. People disagree on that question. The second is how quickly and how thoroughly will they diffuse into work? Which is impart a technology question, but as we've established on this panel, is just as much a change management question and org design question. And I think very quickly will become a regulatory question. And you know, one of the things I think a lot about is we still have people who pump your gas, a New Jersey and Oregon. We've chosen to preserve those jobs. And I imagine there are a lot of choices we are going to make as a country when it comes to diffusion. And then the third is new jobs. What new jobs will be created as a result of this technology? And even there you have reasonable experts who disagree. You have people who say technology has always created more jobs and it's destroyed. Why would this be any different? And then you have people on the other side who say, well, past waves of technology have automated the body. What is the next frontier for humans to do economically productive work? And that becomes very difficult to imagine. And then you have these three vectors, capability, diffusion and new jobs that then interact with one another. And so for me, this framework was really clarifying because now when I engage in data or in headlines about navigating the future of work, I can say, well, which vector are we currently debating? Is a way of sort of hanging that additional insight and perspective kind of in the right place in my framework? All right. You all get basically lightning round, I think, on this future as we wrap up here takes things that we haven't explored yet. Or if you want to give a hot take and prediction, we'll take that as well. Should we run it down the list, Matt? Sure. So I'm not going to end with a prediction, but actually quite the opposite. And it's to say this. There's no shortage of prognostication when it comes to AI. We don't need more of it, notwithstanding the intelligent defenses of whatever positions we put forward. We do need is adopler radar that says, I'm spreading my picnic blanket out on a nice July afternoon. And it'll be helpful to know if there's a thunderstorm coming in the next 45 minutes. When we can do that, that's going to allow us to know where the changes are happening at the leading edge of any occupation will allow us to, instead of re-skilling, which has never historically worked to pre-skill, it will allow us to be able to actually anticipate where new jobs are emerging and how we get people there. Yeah, I'll just share that, again, this, I did that we're early and we have to act in tensionality. But for me, it's just learning and seeing what people are doing with this technology. I was in Cleveland a few weeks ago doing a workshop with small and medium businesses and the things people created, not knowing a thing of code in one hour for their business, where absolutely astounding. And these are things they wouldn't have done otherwise. They wouldn't have bought that extra software. They wouldn't have done, they couldn't afford it. They wouldn't have done it, made the time to do it. But it helps their inventory management or pay their taxes on time or just the different solutions. When I just think about the silver lining of this explosion of many, the costs to do things will come down and many more people can build hyper-local solutions or things that matter for them. It's just truly exciting. Another anecdote is take a higher education research, life sciences, where before to do advanced bioinformatics and genomic work, it's the purview of only a handful of institutions in America. Now with this technology, you're seeing state universities, many of them now can get in the game because they can use AI, they can create more more roles for their postdocs, it's not just centered around an elite group anymore. It actually many more groups can get in on it. So anything about what that does for scientific discovery and with these tools. I just share these two anecdotes like this unlock and this what diffusion could mean. By the way, SMBs employ about 47% of workers in America. There are these other things that could start to happen. Three is in Thropic or excited to look at the data with our partners to really understand what is emerging as well as what is disappearing. Joe, I also need 30 seconds, yeah. Two quick observations. One, the logic underpinning most of the rules of human resource management in large companies are not fit for purpose and they're all going to have to be overall. And the second is we haven't used the word agent once. And a gentage AI is going to be the next, I think, much more substantial wave because the inference and token cost to use LLMs for lots of tasks that can be done off a rigorfully trained smaller, I'll just arbitrarily call a smaller language model. It would be much more efficient to process that way. Now the brilliance of anthropic strategy is they're the tool provider for whomever LLM, SLM doesn't matter, they're it. So, when agents are being, a gentage AI is being funded at a colossal rate. And that's where you see those jobs that are preserved, that's a market signal for people to come up with a business plan to solve that economic problem. So we're not even at the end of the beginning yet, but when we really ring up the current and on agetic deployment, we may be a fourth body, I don't know. My probably most confident prediction over the next decade around the future of jobs is that the future of good jobs will be non-routine, non-rules-based work because AI is good at the inverse, it's good at routine and rules-based work. But that, and I think that starts to Shaw's point all the way at the entry level. And I think that has huge implications for our systems of education and learning because most of our systems of education and learning have been designed to prepare people for rules-based routine work period. And we absolutely have to change that. And I think that has to be the number one provocation for the future of high school, for the future of college. And with that, join me in thanking Allison, Joe, Shad, and Matt. Thank you.