Prominent Wharton AI professor: don't start software companies; the money is in AI

We can easily hand our skills over to AI, and we may never get them back.

You don't want AI to make those decisions, even if they're good ones. Yes, you want to be involved in that process. I'm deciding when to intervene with my own decisions in the system rather than simply sending agents out to do the work. You have to re-adapt from scratch every two weeks. That's what I'm feeling. So we have to add friction to our lives on purpose in certain key areas to maintain an advantage.

I wouldn't say never delegate...

That's always a tough question, right? Ambition seems to be what we call entrepreneurial efficacy in the entrepreneurship world, right? It's the same as initiative, and the truth is it's quite hard to achieve. We don't have a manual for that. But actually, I realize that people create barriers that you can help them overcome simply by asking them to be more ambitious and by being in an environment where their peers are more ambitious. So a lot of what I ask, and which is now super easy, is like, what else could you do besides that? And that used to be a hard question before AI, because you'd stop because you only had the capacity to do a certain amount of things. And now, as the frontier expands, pushing ambition is something much more important.

It's funny because the last version I taught was six months ago, and even then it was like the beginning of working with AI agents. My next semester starts in January. So actually, after ChatGPT came out — I had been following, you know, obviously the models for a while — but after Da Vinci, which came out just a couple of days before ChatGPT, and then ChatGPT came out, and then a couple of days later I had a class and basically canceled the whole class plan and said, 'I'm going to teach you to use ChatGPT.' And by the end of class, one of my students actually stopped paying attention and programmed his entire semester project during the same class, and showed me the results at the end. And so every year since then, the level of ambition has kept going up. The most interesting thing this time was all these people who were experts in their fields. I have surgeons, biomedical device people, finance experts, you know, people from NGOs and humanitarian aid, soldiers — all in this class — and suddenly everyone was building real things, right? And so very interesting ideas came up: how do we track our entire supply chain, because many companies don't actually know their whole supply chain? And how do we find where the financial weakness is in that chain that we'd have to watch out for? A very interesting set of ideas. There were a bunch of very interesting things about specific problems in medicine, like how to actually give patients information in a way they understand instead of just handing them forms. I mean, very interesting things, but very specific, right? Focused on something people know. And part of the interesting thing is that software can be many things. Once you're thinking about solving a problem you have a lot of experience in, suddenly there are a lot of possibilities there that are very exciting.

Where we teach entrepreneurship at some of the best business schools, you know, we used to swap notes about the topics we see people tackle over and over that we know are never going to work, but we always found ourselves in a dilemma: should we tell them it's not going to work, or let them try to figure it out themselves? So there are some that are kind of common problems that are actually a good example of where the technology reaches its limit. One of the most common is, we're going to replace real estate agents, right? It seems like a big deal, like you're paying around 2% to these people who seem to do less and less work for you. Obviously this is a disintermediation case, and the idea keeps failing because it turns out there's a deep social, political, and legal integration of brokers into the process that you simply can't displace. So sometimes the reality of the world is very different from what you believe. And the other is students who see their problems as students from too narrow a perspective. So everyone always wants, you know, better ways to get homemade food delivered, or better ways to, you know, store things between semesters, which probably aren't that universal once you leave college.

Well, I don't know, just try it, tell people, right? I mean, these are the ideas with the highest failure rate, but obviously because they want to do it, right? Because they keep coming up, they're not bad ideas. The point is you can't dismiss them at the outset. I mean, there should be a better way to buy and sell a house, right? And you know, people do want homemade food more easily than with a delivery service, which is roughly what they're asking for now. It's like, how do I get my mom to cook something for me as a son or daughter? That's basically what they're asking. So there should be a way to solve this. Hey, there's demand for that, but these are harder problems than people think, and in a way that makes them more ambitious, right? Because it's no longer the technical part, but the numbers look really good, and then they simply reach a point where they say, 'There's a challenge here, so whoever solves it will make a lot of money.' But it's also an idea we've seen people fail at over and over again. So maybe some of the people watching or listening to us will be the ones who manage to transform the experience of dealing with real estate agents. But it's not as easy as it seems, I guess.

In general, I mean, I think if you're trying to build a business from scratch, you know, aim high because maybe you won't get there anyway, right? And ambition also helps motivate people, it helps you bring in the best people to work with you. It creates a real self-fulfilling prophecy, you know? But I also think a business has stages, right? So that initial stage can be hard no matter what. Scaling presents its own challenges compared to running a small business, right? So they're not exactly equivalent, right? Like trying to get to the next level. If you're going to raise a Series B and a Series C, that's very different from running, you know, like a local business of some kind, right? So I think they're different skills. Yeah, yeah. I mean, if you're starting from scratch, right? And you want to conquer the world, be totally ambitious. But also, by the way, I think, you know, we keep talking about ideas. I think people aren't being ambitious enough with the kinds of ideas they pursue, partly because they have a kind of Silicon Valley mindset about software, right? And I think a lot about — I've studied the history of technology, among other things, and I've worked in it. And if you look at who actually got rich during the Industrial Revolution, right? It wasn't the people selling steam engines, even though that's still the analogy in Silicon Valley, like I want to sell picks and shovels, right? Do you want to sell picks and shovels? Or do you want to dig a gold mine with new technology for gold mines? And I feel like we're overlooking that point. If you can make a business more efficient, just do it yourself right now, right? Break through those barriers directly.

I think, you know, it's good news for entrepreneurship in general. The early data looks really good, and you can understand why it's a very good idea. I think everyone is a little too obsessed with SaaS right now, right? I mean, what do you do to make money? Software as a service. And I think that's threatened, right? Like, it's a different environment than before. Now I've spent a lot of time with big companies as well as with entrepreneurs, and there are people who are getting it. There's a lot of conflict about what the future of software will look like. On the other hand, I think software is becoming more and more a basic requirement for carrying out many other interesting activities. So I'd love to see more entrepreneurs diversify instead of software being the only thing they sell, to start manufacturing or making something real, because there has never been a better time for that.

It's an interesting thing, right? A lot of things are happening at once. On the one hand, now everyone is a programmer, I mean, and that's not just theoretical, right? There's an excellent series of studies from Anthropic suggesting that, in fact, managers now write better code than programmers, at least in the first versions, right? Because it's about giving clear instructions rather than being a coding genius. Things are transforming very fast, right? You have people 100 times more productive. As a result, companies are also a complete mess, but they all have their sights set on, you know, SaaS providers, right? There isn't a single big company out there that doesn't want to start thinking about how do we become more independent, how do we build our own tools. It will take a while before they figure all that out, and many of them are still very happy. I mean, as far as I know, you know, both OpenAI and Anthropic still use Slack, and you know, probably still use Salesforce and Workday tools, so they'll make more decisions about what kind of software provider to choose. But I think the most interesting thing is that companies are realizing that many of these startups selling them things are only selling them what they could already build themselves, right? They don't have their own LLMs, so they're selling them LLM capabilities they already have, and that's creating a very interesting dynamic.

I think software is going to be democratized, right? I mean, the whole thing — I teach at Wharton, I teach MBA students, you know, I teach undergraduates, and many of the MBA students don't have a technical background. So the key to success over the last 10 years has been finding a technical person who can program for you. I think it's incredibly liberating to see this as other fields where you can approach it in different ways, right? There will be very software-focused companies where you really need the best engineers working with AI to succeed, and then software will be like a toolbox we use to solve things, and I think democratizing that is incredibly useful, right? I mean, you don't want to be a software company, you want to do your newsletter, you want to do your videos, and you know, there were things in your way. And software, by the way, was one of them, right? It was also that you needed to be able to handle your own finances. I mean, when I created my first startup, I didn't realize — it was the late '90s, the first internet bubble — but I didn't realize you could pay someone a few cents per pay period to handle your employees' taxes. So I was calculating taxes in Excel every week during the first few weeks without really realizing that you could just do that. So even just the advice, all those resources within your reach, make this much easier now than it ever was.

Yes, it was an interesting study. It was GPT-4, right? A system that became obsolete long ago. Now there are 180 models better than GPT-4, of which 19 can run on your phone. So there's a big difference. And it just gave advice. That's why I think the figure was an 18% increase in profits for the best entrepreneurs, which is crazy. We do studies all the time and try to make a real difference in entrepreneurship. That's an outrageous figure, but it only helped the entrepreneurs who were already succeeding because it allowed them to expand in the way you mention. The entrepreneurs who were already failing couldn't implement the advice; they were already in a nosedive and there was no way to turn back. So it was — we know this can be effective, and that was before programming, right? That was literally what should I do now. It turns out, by the way, on the topic of entrepreneurship, about 50% of Americans have had a good idea or think they've had a good business idea in the last five years, and almost nobody tries because of all these barriers. So even just the advice to overcome those barriers represents a real golden age for entrepreneurship compared to what we had before.

I talk about this in the new book, but I think there are like four things you need to use AI successfully, and that's deep knowledge, right? Certain experience in something. And that, by the way, is like — entrepreneurship is about acquiring specialized knowledge methodically, right? You go into an industry sector, you interview people, you know, you learn about the industry, you test ideas — I mean, that's developing experience, and I think that's as important as it always was, right? Otherwise you're just sharing mediocre ideas, and mediocre ideas all look very similar to each other, right? The second is broad knowledge. Something increasingly important for entrepreneurship is knowing a bit about the fields you're not in, right? So if you're in software, you should know a bit of design and a bit of accounting because that helps you apply AI more broadly. And then the initiative we've been discussing a bit here, and finally, certain judgment — like selecting good ideas from the pile of junk — is also very useful. So I think what we're seeing with the more agentic systems is that they can work for longer periods, right? While GPT-4 could do four minutes of work independently with a 50% success rate, now we're well above 17 hours of work, but you still need to be able to give these systems some guidance, and even when they become really good, it turns out the human touch really matters because variation is what makes entrepreneurs succeed.

Yeah, I mean, it's funny because you realize that people are more important than we thought, or than some people thought, right? So everyone who was among the first to use Claude as a good writer, you know, when, you know, Opus 4.5 or whatever, wrote very well, and for a while nobody knew what Claude's writing sounded like. And now if you go — it's instantly nauseating, right? It's like, what has value? Wait, what is it? It's just — it's always the same, and it's not just the little phrases, but how it approaches things, the kind of language it uses, and you think, well, they should — if you didn't hire a marketing specialist, you're probably making a mistake. Now, the marketing specialist could be much more work than it used to be, but what each person brings really does matter a lot. And I think we're seeing that in many other cases, which is that AI pushes people toward a kind of consensus in strategy or vision. [cough] We have some experiments where we found we can make ideas almost as diverse as a group of people's, but if you map the ideas AI comes up with and map the ideas a group of humans comes up with, AI still doesn't cover the full range of human ideas, even if you're pushing it to the max. So this aspect of variation will become increasingly important. For example, what are you bringing to the system to guide or direct it, so that you don't end up with the same opinion as everyone else, above all? And then we'll be fine. I mean, it's fine if you use Claude for finance and that's not your competitive advantage, but in the area you're specifically aiming at, you have to stand out, you need to differentiate, or you need a competitive advantage that in some way has to be human, even when the models become incredibly capable.

We know a little bit about how to teach people to have good taste, right? It's like you need to see good and bad examples of things, so you want people to explain to you how to take something apart. So one of — I invite an art professor to my classes now who talks about how to think about the way a designer sees the world, right? So you get this kind of perspective, like, what makes a font good or bad? What makes an image interesting or not? And I think we need a vocabulary to talk about our tastes and judgment, right? If you talk to someone who really knows film, they have a taste in movies that is individual, but they can probably talk about it at length. In the same way as in venture capital, investors rely totally on taste, right? They develop a judgment for evaluating certain types of problems, though not always. So I think seeing a lot of variation is very important. Learning to be disciplined about it is another example of why I think education has some value here — being exposed to many options, I mean, that helps you develop the good taste you need.

I don't like the idea of judgment itself, because AI is obviously judging. You can't do a 17-hour task without AI making judgment calls, just as we don't have creativity on the list because AI is obviously making creative decisions. The question is whether they match the kind of judgment and creative decisions you would make, right? So we can classify that as taste or knowledge, but it's a different kind of question, right? So judgment is something very hard. We really don't know how to teach it, and we don't have a good example of it, but for some people, you know, the question becomes, in what context does that judgment occur? So it's something difficult to frame, right? It's like, what is judgment? And AI is exercising some kind of judgment. Obviously there's a human element of judgment that's different, but we haven't had to build the language to differentiate these things. Think a bit about what's happening in mathematics right now, right? Which is this kind of explosion of AI-generated proofs. And one of the things mathematicians will tell you when you talk to them is a bunch of revelations about their field, or maybe clarifications about their field that outsiders didn't know about, right? There's an aesthetic element in mathematics that really matters, you know, for a mathematician — that if the proof doesn't help you learn something beautiful or interesting, it has no value for the field of mathematics. Even if we solved this problem, who cares? How? So we're starting to differentiate things we didn't have to differentiate before, because the only kind of judgment and intelligence we had was human. So we didn't have to invent a different category.

A couple of things are happening at the same time, in a way, when you think about markets and education, right? First of all, we're temporarily in a crisis at universities. I mean, we're always in crisis at universities, so this isn't new. Anyone who's been in education knows this, but we're in a temporary crisis because of AI, right? And we already know roughly the outlines of that solution. AI is an incredibly good tutor when used as one, and it's terrible if you're just cheating and getting answers from it. Even if you think you're learning, you're not, right? So over time we'll have AI tutors outside of class. Inside the classroom we'll apply more hands-on learning. That is, if you've been to business school or law school, you've already done this kind of case studies, debates, and activities. And you know, we'll do in-class exams, we'll solve that. The real problem is where people actually develop judgment and critical thinking — it's in the real world, right? And we had this great arrangement for the last few thousand years where I would send you, you know, a newly trained employee who knew nothing, and we had apprenticeships, right? That's how the last 4,000 years of professional training have worked. So you'd get someone very junior who was desperate to prove they were good. They worked for little money because they wanted to climb the hierarchy, the rungs, and they did the work that a middle manager didn't want to do. And in return they got a bit of pay, they could prove their worth, and they learned how everything worked. And the middle manager handed off their work to someone who could at least be part of the job for them and make their life easier, right? And that's how they climbed the chain of command, and all of that broke, right? Because every — you know, your incentives as an intern or a first-year associate or whatever you were, your incentives are 100% to deliver the best possible product. And who's better than you? Claude, ChatGPT, or Gemini. So you're going to deliver Claude, ChatGPT, or Gemini answers. And this isn't like being a digital native, because you don't actually know what you're delivering, you're just passing things along. And all the middle managers have realized — why give feedback to a human who takes forever to do the work? Better to just have Claude do it and be done. So we've just broken the entire system that forms professional judgment and critical thinking, and we still haven't found a definitive replacement.

I think there are a couple of things. One is at the individual level. Now we have to start being very disciplined about when we fall into cognitive surrender, when we're handing things over to AI, and that's very hard to do. We're programmed — look, first of all, entrepreneurship, I mean, that's what I teach, right? And what we've been talking about here, it's about trying to make your life as easy as possible so you can focus on the hard stuff, right? And you know, but then you don't learn anything, right? As a result of that. But it also happens in any other aspect of life. We can easily hand skills over to AI. And I think we all feel tempted to do this every day, and maybe we'll never get those skills back. So we have to add friction on purpose to our lives, in certain key areas, to maintain an advantage, but we don't have incentives to do it. So at the individual level it will be about making decisions. At the organizational level, companies are going to do one of two things. They're going to follow either the easy path or the hard path. And I worry about the easy path, which is hiring only senior professionals. And we're already seeing some signs of that. I know you've had some economists talking about this. There are some early signs that entry-level positions could be at risk. We're not 100% sure yet, as you know. Or the other option, the harder one, is that we actually make training a serious part of the job, right? So really, you have to let people be inefficient and learn on the job. And we, instead of learning being accidental, will have to teach them — maybe we'll have to add an extra year to universities where we teach you to be, you know, a Google or Microsoft employee or whatever, and open a year of personalized training to develop you, because we know how to teach and evaluate people in classrooms, something we no longer know how to do in the real world. This is a very hard problem.

So it's still very early to say anything, right? I don't know, but does it ever happen to you, I mean, when I think about where we're headed. [laughter] I mean, but also, how long will being good enough be enough? Right? The problem is, I mean, that also keeps progressing. So, you know, this concept we developed in our first paper where we analyzed this, we call it the jagged frontier — that AI is good at things you wouldn't expect and bad at things you also wouldn't expect. And it's hard to know in advance what will be. As long as that frontier remains quite jagged, there will always be bottlenecks that humans will have to cover. And those can be very strange. For example, if you think about AI replacing lawyers, well, it turns out a lawyer does many things in the real world that involve obstacles AI can't necessarily do, even if you had a robot doing the work, right? They interact with human systems, they also have to meet legal requirements. Some people want to talk to a human, so this creates these bottlenecks that really reward talents different from before. And you see this also in programming, right? What makes a good programmer now is completely different from three years ago because you're no longer writing code. So I think one thing I'd say is that 'good enough' is a set of skills, and it could well be the people who are mediocre programmers but have a great vision for developing software, or communication. Exactly. Yes. Or communication. And those will also change over time, right? I'm not going to say AI will never do that task, it probably will. So I think we'll see which skills are valued change quickly, without an easy answer.

One of my favorite phenomena is seeing people who know nothing about a field get AI to do things that seem impossible. And you're like, 'Wait, how did you do that?' And they say, 'I just asked for it.' And you say, 'You can't ask AI for that.' And they did it, and it works. And you're like, 'Great, I mean, what are we doing here?' So, well, there's this strange phenomenon. I mean, there are two things we know make you better with AI. Before, AI used to be more of a leveler, right? If your performance was poor, it gave you a boost. It still is. But what we're seeing more and more — and there are some good results from Anthropic, plus other studies — suggesting that people with more experience are the ones getting the biggest benefits from AI at this stage. And that wasn't the case until agents appeared, more or less, in the last year. But the other thing we know is that there's a certain special skill for communicating with AI, which is like knowing how AI thinks, which could be a kind of theory of mind — like, I know people really well and I come to know the mind of Claude or ChatGPT or Gemini or whatever I'm using, I become more effective working with it. And so those are like two different skills. And then there's like a mysterious third category, which is like some people seem to be good at this in ways we don't yet understand. There's a great study by Alex Imes and some others. He's now at Chicago — no, he's now at DeepMind — but they measured people creating agents for negotiations and a couple of interesting things. One thing they discovered is that your skill multiplier becomes much higher because people participate in many negotiations. So a small difference in the skill of creating good agents has a big impact on outcomes. The second thing they discovered was that people who are good at creating agents aren't necessarily the ones who would be the best negotiators in real life. I mean, there was even a gender gap. One of the things we find repeatedly in studies, for a wide variety of reasons, is that women tend to outperform — sorry, underperform in negotiations, and a lot has been researched about this. It's been a big topic of concern, but women created negotiation agents much better than men, and you know, again, we actually don't know — I mean, I don't even know how to necessarily teach you to do that.

I feel like that's a bit made up, I mean, maybe it's true. For me, the analogy that makes the most sense for someone who has run organizations and managed people is to think of AI as a team I'm delegating work to. And if I can do a good job delegating to a team and realizing where a team goes wrong — which is very different from knowing my own process, right? It's not going to replicate my process, and part of what I try to do is preserve things for myself in that way too, right? Like I write everything, you know — I mean, it might seem crazy, since I use Substack and X and everything — I mean, I write all my own material, right? I write the first drafts of everything myself. I mean, yes, I type it all out, right? Because if I don't, AI is very seductive and it's definitely going to shape what I'm thinking in any direction, and then you're just going to read like Claude's results filtered through me. Do you like that? Do you use voice for anything, or...?

Sometimes I use voice for this. I think that simply typing on a computer is easier for many of these things, but yes, voice — and as you already know, you keep discovering how you generate ideas. For me, curiously, my best ideas come to me while doing SoulCycle, where I think I'm the only man and the oldest person in the place. Someone, you know, but somehow being in a noisy room where I pedal and people shout at me to reach my goals, I disconnect from everything and good ideas come to me. I don't know why, but everyone learns their own way of thinking. And I've talked to another scientist who told me, 'I need to be extremely exhausted one day a week and then my best ideas come to me.' So you discover things in that respect, but you also have to keep parts of what you do that are very human. I don't think it's necessarily about replicating my human process. It's about preserving a human process and then delegating the work the way you would with a team. For example, I have a very good idea. This is how I want you to execute that idea.