Today, we're discussing the six big questions that are shaping AI. Now, today we are discussing six big questions shaping AI. This is sort of a quintessential weekend long read {slash} big think episode. And a good way to sum up the high level before I land back in the office chair next week after a week of being away. The six questions that I'm going to discuss are one, how much job displacement will there actually be? Two, to what extent AI becomes a political issue and in what ways? Three, who gets to decide the limits of how AI gets used? Four, how deep the market's pockets are for the infrastructure buildout and how much external factors will impact that? Five, how fast will differentiated enterprise adoption compound? And six, just how much agency do agents really give people? Are we on the verge of a wave of the greatest flourishing of small business entrepreneurship that we've ever seen? Let's start with what certainly has been one of the dominant public discussions. How much job displacement will there actually be? Now, one of the things that

makes this conversation potent right now is that we have a tricky combination of one, some very real announcements, but two, those announcements being nascent enough that we don't know for sure how much we can extrapolate them out. Meaning effectively that our imaginations about the possibilities of job displacement are running wild with just enough nascent evidence to really feed into those fears. And of course, it's not just the block and other layoff announcements. A working paper from the National Bureau of Economic Research found that out of a survey of 750 chief financial officers from US firms, about 44% said that they plan on some AI-related job cuts. Although as Fortune points out, while the number of estimated job cuts from that survey will be nine times higher than the AI-related job cuts last year, the total number is still expected to be a tiny fraction, just 0.4% of all roles, of some of the doomsday predictions out there. And the doomsday predictions are flourishing right now. Senator Mark Warner recently suggested that new college graduate employment will spike to 30% plus the next couple of years. Dario Amodei continues to sound off

about the idea that AI will eliminate 50% of entry-level white collar jobs within the next 3 years. Basically, you can't really throw a stick without hitting some prognostication about how we're all going to lose our jobs. Now, obviously I did a whole long show about my optimism and why I don't think A AI is going to take our jobs and frankly, the will AI take our jobs conversation is even the right one to be having. And what's encouraging to me is that we're finally starting to see a bit of a counter discussion. Chicago Booth's Alex Immerman and Harvard fellow Sumitria Shukla recently dropped a blog post called how will AI-driven automation actually affect jobs? Now, this is not some full-throated argument that AI isn't going to cause disruptions, but a reminder that simple exposure to AI is not really the critical thing. In a summary post on Twitter, Alex writes, "AI exposure measures are not meant to predict displacement or job automation. Exposure can lead to job loss or it can lead to more hiring and higher wages. It all depends on how one automated tasks interact with non-automated tasks, i.e. to what extent they're complements. Two,

how consumer demand in that sector responds to prices, i.e. elasticity of consumer demand. And three, the dimensionality of the job, i.e. the number of tasks a job has." Even more optimistic is this recent report from Lenny Rachitsky of Lenny's podcast and Lenny's newsletter called State of the Product Job Market in early 2026. Lenny writes, "In spite of the headlines about layoffs and AI taking jobs, we're actually seeing a lot of promising signs in tech hiring and some interesting new trends. One, product manager openings are at the highest level we've seen in over 3 years. Two, AI hasn't slowed the demand for software engineers, at least not yet. Three, AI roles in general are absolutely exploding. And then seven, yes, we're skipping a couple, despite ongoing layoffs, the overall number of tech jobs continues to grow. And I anticipate that over time, there will start to be more focus on where new jobs will actually come from. For example, a recent Goldman Sachs report analyzed how AI would shift the job market. It found that AI could automate tasks that make up about 25% of work hours in the US and that around that, roughly 6 to 7% of workers might face

displacement. However, the report also points out that the technology will create entirely new categories of work. For example, just the physical infrastructure for AI is going to require massive labor. They point out that the US alone needs 500,000 new workers by December 30th to handle electric power demands. Since October of 2022, construction jobs related to data centers have already grown by 216,000. The AI companies themselves, despite being some of the leaders in how to use AI, are still planning on growing. OpenAI apparently plans to double their workforce to 8,000 by the end of this year. And even the ECB has found that the companies that are most AI-native right now are actually hiring more than they're firing. It makes sense to me that alongside this major jump in capabilities, there are major renewed conversations and fears around job displacement. But I am hopeful and encouraged that in the months to come, I think the conversation about those effects will get a little bit less black and white and a little bit more nuanced and varied. Now, of course, quite related to the job conversation is to what extent AI becomes a political issue and in what ways. There are a few different ways in which

AI could become a political issue. There are issues of X risk and runaway takeoff AI that threatens human life. There are the more here and now concerns around jobs and data centers. There are also questions around children, mental health and a lot more. Which of these issues gets the most traction will, I think, shape pretty dramatically the way that AI becomes a political issue. Now, it could be all of them, of course, but that is a question to watch. A second question is the extent to which it is partisan or not. Right now, the discourse isn't all that clearly partisan, although I anticipate that getting a little bit more challenging as the midterms heat up. For example, AOC recently tweeted, "Politicians, especially Dems, should pledge not to take AI money. They are buying up influence ahead of the midterms and Dems who take AI money will lose authority and trust as the public bears the cost. Their money will end up being toxic anyway. People are catching on." Still, when you look across the issues, it would be absolutely 100% inaccurate to say that there is a Republican position on AI or a Democrat position on AI. In the wake of Bernie Sanders and AOC

introducing their data center moratorium bill, you had Senator Mark Warner, who we just mentioned, say that it was a dumb idea. John Fetterman slamming it as China first policy. And then on the Republican side, there's no consensus either. In fact, AI regulation and the White House's relationship with AI companies is kind of a major schism right now. Steve Bannon's whole crew are getting increasingly loud. And if you put Donald Trump, Josh Hawley, Steve Bannon and Ronda Santis in a room, you're going to have very different Republican views on what we should be doing and thinking with AI. Now, here are some of my predictions. I think that while X risk is going to try to make a resurgence, I just don't think it becomes the resident issue when it comes to AI. I think it's only getting a second breath because Bernie Sanders has decided to put a focus on it and because anytime there's a big new jump in capability, it's kind of a natural time for people to ask those questions again. I think that data centers and jobs are much bigger, more politically potent issues. However, in some ways, I think that how bad the data center issue gets is going to be largely driven by the job situation. Yes, there are real community concerns with data centers, but there's

also a lot of room with data center construction to shift the balance. We've already seen the White House with its ratepayer protection pledge get all the AI companies to commit to making sure that people's electricity bills don't get up because they need new capacity for their data centers. And I think you're going to see a lot more agreements like that. Where it gets really challenging is if data centers become the visual embodiment of 10 or 15% unemployment. That's where things really start to get hairy. Obviously related to politics is the question which smashed its way into our consciousness this past month. Who gets to decide the limits of how AI gets used? This was an inevitable conversation. It just happened a little bit faster than we might have thought. Now, I've talked about this ad nauseam, so we don't have to get too deep into it. But suffice it to say that the very public rhetorical real and now legal battle between Anthropic and the Pentagon has big implications for AI going forward. Hold aside all the details and specific personalities involved and at core, this question is a question of ultimate power. One of the uncomfortable realities is that the likely

significance of AI across so many different sectors of the economy and human social life will make people increasingly uncomfortable with it being controlled by singular private companies. I haven't seen any calls for nationalization yet, but I would be shocked if we don't see them before this is all said and done. At the very least, you're going to see more conversations like the one sparked by Stanford Professor Andy Hall, who recently proposed new constitutional conventions to determine how the governance layer of AI should work. Our fourth question actually evolved a little bit from when I first started thinking about this episode a few weeks ago to where it is now. One of the big questions facing AI coming into this year was how deep the market's appetite and pockets were for the infrastructure buildout. Over the course of 2025, we went from a buildout that was largely financed by hyperscaler balance sheets to one that was increasingly financed by investors in private credit markets. To the extent that those investors continue to have high demand for that debt, the AI boom could build unabated. Of course, the risk is the more you move off balance sheet and into the credit markets, the more risk of those markets

claiming up and that causing ripple effects, which because of the extent to which AI has propped up public markets for so long, would have implications far beyond just AI itself. However, over the last couple of weeks, obviously this no longer is just a question of market's appetites in general, but also how broader geopolitical and economic challenges are going to impact the private market's appetite for AI debt. I'm recording this episode about a week in advance from when you're hearing it and so a lot could have changed between now and then. But at the time that I am writing, one of the big conversations across all sorts of different outlets is how the war in Iran and its impact on energy costs could have, among its other downstream effects, fairly big implications on the AI boom. The World Trade Organization's chief economist warned about this, saying if the price of energy continues to be elevated for the whole year, that could put a crimp on the AI boom. On the oil price blog, why the Iran war may have just killed the AI boom. The war's effects, writes Michael Kern, including the collapse of shipping insurance in the Strait of Hormuz, attacks on data centers, and a spike in oil prices are structural problems that

will increase component costs and slow the AI buildout. Compounding issues, including higher costs for fuel and fertilizer, coupled with elevated electricity bills from data center demand, will shorten the political window for AI transition and fuel consumer backlash. Time magazine also wrote about this. In this case, reiterating that like it or not, what's bad for AI is bad for the economy writ large. Writes Time, the AI industry, and specifically its data center investments, are essentially holding up the US economy, accounting for 39% of US GDP growth in the first three quarters of last year, according to the Federal Reserve Bank of St. Louis. Now, one very specific issue, even if the worst prognostications don't come to pass, is that at the very least, the war is likely to have some impact on the UAE and Saudi Arabia, who have been some of the biggest investors in AI. Miles Krupka from The Information writes, "The war in Iran is complicating plans by Gulf nations to spend more than $300 billion on data centers, chips, and other AI investments. These effects are not theoretical. When you've got drone strikes on Amazon data centers in the region, it makes the calculus on building out in that region look very,

very different." The Information writes, "Gulf nations won't rush to divert resources away from AI investments because of their economic and strategic importance, but they might have little choice if the conflict stretches on for a long time." Said analyst Steven Minton, "If that turns into months or even longer, there could certainly be a disruptive pause to some of that investment." Now, our last two questions that will shape AI are a little bit more back in the realm of operations and AI in practice. And the first is, how fast will differentiated enterprise adoption compound? And so, the key terms are differentiated adoption and compounding. You've probably already heard me talk a lot about efficiency versus opportunity AI. Efficiency AI, in short, is doing the same with less. Opportunity AI is recognizing that the real power of this technology is not just to be 30% more productive, it's to do things you never could before. Now, right now we are living in the shift from efficiency to opportunity AI. The changes that are happening right now are not little. They are insanely huge.

We've gone in the last 3 months from people viewing agents as these things which might be interesting in some vertical areas or functional areas, to people building massive agentic teams with open claw that are changing literally every single thing about how they work. In that process, the split between the fast-moving startups who are reinventing how they work and the big companies is getting insane. And what's very clear is that there is absolutely no doubt at all that company building is going to look just absolutely, totally different. The org chart is going to get completely upended. The speed of execution will be unlike anything we've ever seen. We will see tiny companies with one or five or 10 employees doing millions and then tens of millions and then hundreds of millions of dollars in business, and there will be implications for things like venture capital, which has to deal with this very different reality. Now, if that is pretty much guaranteed in the realm of startups and small companies, how does this look for enterprises? Certainly, there is a world where things continue to diffuse very slowly. Michael Chen from Applied Compute recently wrote

what to expect when you're deploying AI in the enterprise, and effectively it was a big reminder that things move very, very slowly. That the capability overhang is not just a concern, but an existential state. Data ready, for example, he says is just a state of mind, with the gap between we have data and we have data in a format that AI systems can learn from being enormous. He calls timelines optimistic at best, with the challenges not being just that enterprises are slow, but that they don't even realize that there are all these things that they have to do like data provisioning and compute access that make them even slower than they think they're going to be. Third, he points out an absolute truism at this point. The challenge of AI adoption in the enterprise is not a technology challenge. It is an organizational and management challenge. Period, full stop. I don't even really need to get into this. Everyone knows this at this point. The way that Michael frames this is the real deployment environment is the org chart. He writes, "With one of our recent projects, one of our biggest onboarding challenges was simply learning the org chart. Not the one on paper, but the real one. Who actually controls data access? Who can approve a deployment? Who's working

on adjacent projects that might overlap or conflict with yours? There's never one single point of contact, and getting work underway often means figuring out the answers together." Increasingly, there is even chatter, even as AI companies invest so much more in their forward-deployed engineering model, that that alone is not going to cut it, and that there really needs to be mass-scale changes to the way that organizations adapt that even having a bunch of embedded engineers aren't going to change. So again, there is a world where AI, despite all of its capability acceleration, continues to diffuse extremely slowly. But what matters is not so much just the average speed of enterprise AI diffusion, it's the difference between fast organizations and slow organizations. If all big companies adopt AI and get transformed by AI at the same pace, even if they're behind, theoretically that's fine, because their competitors are behind, too. My guess, however, is that we see some very significant breakouts that massively upend the playing field. I would guess that the way it actually happens is that the majority of the

enterprise pack remains slow to diffuse, call it 80%, and pretty much all the actions happen in the other 20%. But those other 20% don't just add 50% efficiency gains while the laggards get 25 or 30% efficiency gains, those 20% wildly outperform. We are talking shifts that totally challenge the comparative rankings and positions of companies. We're talking mid-markets jumping up tiers. We're talking about companies moving into adjacent product areas. We're talking about companies dominating press coverage. And the key difference will be not just how fast enterprises move, but how they reinvest their AI gains, because that's where compounding differentiation comes in. The companies that win this next phase are going to reinvest their AI gains in more AI innovation, more AI enablement for their people, more product development, more R&D, more sales efforts, more of all the things that allow them to become a bigger, more successful company. Let me put it a different way. Stock buybacks, a common way for companies to

reinvest profits, are literally never going to be more expensive than they are when you could be putting that money into reinvestment in AI. Simply put, I think not only are we going to see a huge and increased gap between leaders and laggards, I think that space is going to compound over time, and the laggards will never be able to catch up. Last major question is almost the sort of positive inverse of the first question. The first question we asked was about job displacement. The final question we're asking, and the one that I think is dramatically important in shaping how AI plays out, is how much agency these agents that we're all trying now actually give people. There's a strange duality in our discourse about agents. On the one hand, the premise of all of this job displacement discourse is that companies are going to try to replace people with agents, and the thing that makes that resonant is that companies can clearly do all of the work that they are currently doing with far fewer human inputs than they could before when they are using agents well. The mistake, of course, is in thinking

that there is a fixed amount of work to be done, and that companies or the market will ultimately view doing the same amount of work that you do now with less human input because you're using agents as a success. In practice, when you're looking at the people who are getting the most out of agents right now, they are not shifting the end of their day from 5:00 p.m. to 1:00 p.m. because of agents, they are massively, radically expanding their outputs. They are working more than ever because the leverage they have to do more and do more faster is unlike anything they've ever experienced. And while the adoption pattern of organizations won't be exactly the same as individuals, it should be fairly telling to us that the actual practical, lived effect of highly successful agent usage right now is 100% not people getting fired, it's the people using those agents having more work than ever because they have more leverage than ever. So again, one path is companies keep a fixed amount of output and pay less for it, the other is they reinvest that back in, and a lot of what that looks like is super-powering everybody with agents. But let's say that that doesn't happen.

Let's say we've got all these people no longer working their traditional corporate jobs. Let's say that in a transitional period, the overall number of white-collar jobs does go down. So, those people displaced can't naturally flow into some other industry. Again, I don't think this is exactly how that plays out, but just for the sake of argument, let's say it is. Well, then the question will be, how much agency do those newly unemployed folks have to chart a new career path that looks different than just getting a different job of the genre that just let them go? How many of them can actually start businesses? How many of them can become successful consultants? The opportunities of agents are not just a question that determines the beginning of that unemployment story, they're the key thing in determining the end of that story as well. If we just assume that there is a fixed number of people who can be entrepreneurs and small business leaders, then maybe we're just up a creek without a paddle. But if, on the other hand, knowledge workers and all of the recent college grads that aren't getting traditional

corporate jobs can pair up in pods of four and build interesting, meaningful things, not only will they be fine, they will thrive. I am increasingly of the belief that we are massively underselling people's adaptability. Sometimes the jobs discourse feels like we assume that this entire generation of people coming out of college now are going to sit around moping until someone, anyone, gives them a job. Sure, that might be the story of some, but I think that's a pretty depressing view to take of people's agency. My strong guess is that what actually happens is that after a bunch of frustration, hundreds of applications that they probably sent out with AI written cover letters and no callbacks, they say screw it. If the corporate world doesn't want me, I don't want it. And they go try to do something different. Now, the best of times, that is not an easy path. And I think part of our policy engagement around AI disruption should be around making it a more viable or at least somewhat less risky path. But I think we have barely begun to scratch the surface of what type of superpowers AI is going to give the

people who are willing to go out there and do the work. And I think, based on the people that I've seen sign up for Claw Camp and AI Dev New Year and all of these sorts of programs, that we are going to be shocked by just how many people actually fit into that category. Call me naive, call me an optimist. I think people are going to impress us. Anyways, guys, for now, that is going to do it. Six Questions Shaping AI. Appreciate you listening or watching, as always. And until next time, peace.