Christina Wallace on AI and Entrepreneurship

The bar has been raised.

Yes.

For any type of creator— engineer, artist, writer— and good enough is no longer good enough. Christina Wallace is a professor of entrepreneurship at Harvard Business School. And so we're seeing teams of 2 and 3 founders, maybe 1 or 2 employees, able to do the work of a team that used to be 15. The rules of building a company have changed overnight. It's been really inspiring, but also, like, slightly terrifying, this pace of adoption. And not everyone is asking the right questions.

Is that helpful? Is that a tool?

I think it's dangerous.

That's dangerous.

I do.

Why is that dangerous?

The old advantages are gone, and the new ones may not be what you think. As you look for moats, what is your competitive advantage here? Proprietary data and judgment. Welcome to The Parlor Room Presents Hello AI.

Before we start, I want to take a quick moment to thank you for listening to this season of The Parlor Room. If you're watching on video, you can see that I'm in Klarman Hall right now in the Klarman Auditorium. Please follow on whatever platform you're using to watch or listen to the show. Thank you, and I hope you enjoy this episode. You've been working with founders for a very long time. When was it that you started to notice that the way these companies are getting built was changing in a fundamental way?

About a year ago.

Okay.

So kind of top of 2025, it became clear that entrepreneurs were using AI, and not just in a, like, getting a better version of Googling something, right? Like, when LLMs first came out, and we were all like, oh my gosh, ChatGPT is so powerful, the default usage, sort of like, well, let me just ask ChatGPT instead of doing my Google. And it was good for short-circuiting research. Of course, you have no idea if you can trust that research, but it allowed you to do things a lot faster than you used to— write an email, process a set of data, whatever. But that is still very much in the era of AI as a copilot, or as I think of it, as my intern. It helps me remember I need to review its output.

Yeah.

And about a year ago, I started seeing a lot more of this shift toward AI agents, where you're not using it to process information and give it to you so you're making decisions, but instead you're setting up key operating systems in your organisation so that the agent takes in the information from some sort of a sensor or input, does something to it that you have defined, and then makes a decision based on the output of that analysis, all without you having to intervene. And it really escalated. I mean, there's the speed of which this adoption occurred within startup world over the course of 2025 was pretty spectacular. If you think about what the adoption curves used to look like for new technologies, To the point where we are now seeing startups that are skipping their seed round of financing, right? They're going out, maybe they're getting a pre-seed. They need a little bit of money to work with, $50,000, $100,000. But that round where they used to go raise the million to hire some people and start building things out, they can skip. Because instead of needing to hire 10 people, they're building 10 agents. to do that work. And so we're seeing teams of 2 and 3 founders, maybe 1 or 2 employees, able to do the work of a team that used to be 15. And it's been really inspiring, but also slightly terrifying, this pace of adoption, because it requires an eye on what's developing in this technology at the same time as an eye on building your company. And that's a lot to—

Yeah.

to manage and take in as a founder.

Are there specific people that are better at this than others? Like, if you had a skill set for being that kind of founder right now, is it the Jack-Jill of all trades? Is it someone who's just super driven? What's the number one successful skill you're seeing, or personality type?

I mean, I think the number one personality is instead of throwing money at the problem, which is what especially venture-backed startups might do in a previous era. This is someone whose default reaction is, is there a way I can build this through a series of AI tools? Like, how do I check that first and tinker with that and try to build it myself before I go to the point where I bring in an expert or I go in and hire a full-time person for it? So it's just this fundamental shift of, Let me try to do it first, which has always been in the DNA of founders. So I don't think it's, you know, materially different, but it doesn't require you to be a technical founder to build any of these, right? A lot of these tools are really quite user-friendly. And, and, you know, I mean, I was building some agents for my Broadway production company. I have 2 partners, but we don't have any operating staff. We don't have any full-time employees. And, You know, we're, we're all 3 of us busy with full-time jobs. And I was like, how could I build out my investor onboarding flow through a Zapier agent that connects to the APIs of my bank account and my document signing processes? And the work is twofold. One, you have to really think through sort of step by step in an algorithmic way. What is the work to be done? What does the input look like? What do the outputs need to look like? How does that connect to the next step of the process? Because the outputs from the one step become the inputs to the next step, right? So it's a very logistical way of sort of breaking down a problem or a process into these really concrete steps. And then a tiny bit of not being afraid of the technical side of things when you're trying to connect APIs or in some ways get these different functions to talk to each other.

Mm-hmm.

You do have to get into the code, but you don't have to be a software engineer to do this, right? Like, Claude Code now allows very non-technical people to build entire prototypes of apps, of websites, without ever having to tinker with the code themselves. So I think there's just, if you have that mindset of, let me figure it out first, And I'm not afraid to have to tinker and kind of onboard to a bunch of different new ways of working. It's just, it materially changes your cost structure and the speed with which you can experiment.

So you brought it up, so I wanna follow up on it. I'm easily distracted by talking about the arts and things like that. So we're both from the arts world as well. And this time we're in right now is, maybe exciting, maybe upsetting with the way AI's involved. And it gets across all the worlds, the coding for sure. But you're talking about Broadway.

Mm-hmm.

I'm sure there are people out there right now writing Broadway songs with AI and Suno and all of those programs and different things like that. I'm trying to think, what do we do now? There are people that are just like, I know plenty of people who are just so incredibly talented. And they were born talented. And then not just that, they committed their lives to getting better and better at the thing they're doing. So they're amazing musicians, they're amazing authors, they're amazing painters, whatever it is. And we've obviously had that devalued financially. Like, they don't make as much money as they used to make. And I've got stories myself and stories of my friends where you're like, well, that song doesn't matter anymore. Like, I can make a fake one right now. Or not a fake one, but a different one. Fake one. I'm just going to say it.

Sure, a fake one.

What happens here? Like, how do we, or people who feel that they've developed that talent, then they have someone coming up hitting a couple buttons, and it feels like society is recognizing those things as equals, or maybe they can't differentiate the equals. What— this is not specifically business, but what do we say to the people who are feeling that right now, who are just like, well, what was all this for? What's my special ability that I was born with or developed for now?

Yeah. Yeah. It's a great question because I see— I have such a wide range of creative friends and their views on AI.

Hmm.

I have some on the far extreme saying, I'm not touching that. I'm not engaging with that entire world because it's fake, and what I do is real, and what the other creatives do is real. And the second I engage with that, it will use my work and build the algorithms off of that, and at some point it will replace me. And then I have on the other extreme, creatives that are saying, this is just another tool, right? GarageBand was a tool that made you able to make an entire song and produce things without having to have, you know, 4 people and a studio and all of these musicians. This is another tool that allows me to be creative.

Hmm.

Maybe to an extreme that I wouldn't have access to, or, you know, coming up with entirely different modes of expression that are gonna be native to this era, right? So I have friends across the whole spectrum. I'm a little bit in the middle of—

Mm-hmm.

I can see the fear, and it's valid, right? As you think about the last big strike in Hollywood, one of the big sticking points was whether studios would be able to use AI-generated extras. Because up to this point, that was one of the ways you got your leg in in that world as an actor, right? As you're trying to network your way around, you're trying to get an agent, you're working on your craft, getting that day rate of a couple hundred bucks to be a stand-in in the back of some big studio film was how you paid your rent. It might be how you end up at the craft table with you know, someone important, and you start up a conversation, and it gives you— it gets you in the room, right? So it is both a sustainable piece, and it gives you some sort of access. It's never a guarantee that anything is gonna come, but at least it creates the possibility for luck. And the second you recreate all of those extras with AI, because there's no great artistic intent required in having a background of a crowd, then you've cut out that entry point. In the same way that we're seeing some fear for even our MBA students who have historically been really valued by management consulting, by investment banking as that first layer of work, that work, a lot of that work can now be done by both copilots and agents. So how do they get in and get trained, get apprenticed for the second layer and the third layer, which is where the human judgment comes in? So I think part of this is not the fear of the technology itself, it's the fear of an already very fragile creative world is going to compress even further such that only people who have independent wealth, who already have connections, who can sustain that long period it takes to get established and build a career, only they will be able to pursue this life, right? So there's that piece, and I fully recognise it. There's the other piece, and this is particularly within generative AI, which says everything it makes is just a remix of everything that has come before it, right? By definition.

Mm-hmm.

It has been trained on all of this thing, this data, and now it's going to, pull together pieces in potentially a novel-ish way, but you're using existing materials. So what it creates, number one, can never have intent, which is at the core of any artistic expression. Number two, it can only use what has existed to this point.

Hmm.

And so any big shift in style, in culture, in In everything. Like, imagine modern art before Picasso's Blue Era. Like, AI could not come up with the Blue Era. It took a person who was going through experiences and tinkering with different modes of expression to suddenly say, I'm gonna have an entirely new technique and form of art, and it influences this whole shift of other artists. So AI can never go through those. big creative shifts because it can't make, it can only remake. So there's still obviously a huge need for artists and creatives for that originality. There is a huge need for intent, right? You can't make something that is meaningful if all you're doing is remixing raw components. Artists say something with their work. Without it, it's, you know, it's like buying a print of a painting from IKEA, right? Like, it might be good enough to decorate the wall of your guest room, but no one's ever gonna stop and spend 10 minutes staring at it and really try to, like, take in not just the beauty, but what it's saying, because it's not saying anything.

No.

So The replication of mediocre art or of things approximating art, the cost has dropped down to basically zero. And that's scary for mediocre artists. The access to building a life in it might be compressing, and that's scary for anyone who doesn't come from the resources to fund a an independent life without economic concerns. But there's still clearly a role for artists who have a point of view, who have something to say, who are making and remaking the vocabulary of how we interact, right? Setting culture, curation, taste. We still need humans for that.

Mm-hmm.

We still need the judgment and the truly, like, the human condition to find the story in anything that is being made. It just is going to, in my opinion, require a pretty fundamental shift in the business model of being a creative. And that's terrifying because the last business model of being a creative has sucked. It's been a while since it's been a nice sustainable business model. And so you've got a lot of creatives that are like, I'm exhausted. I'm in debt up to my eyeballs. I can hardly sustain in this world, and now you want me to start from scratch and reimagine how I make a life? Like—

Yeah.

So I don't discount how terrifying this is, but I fully believe we cannot lose these beautifully trained artists with something to say who have only one way of sharing that with the world, like, they will be needed until the day humanity goes poof.

I hope so. I really hope so. So I get scared because I think, you know, if we use Picasso as an example, if there were 7,000, or let's be honest, if there were 7 million other people making garbage around him, then his stuff probably wouldn't have stood out at that point. You know what I mean? So I just get so nervous that whatever's being created right now will basically replace culture and become this thing that we kind of lose track of. I mean, I heard an AI-created song that was good, you know what I mean? But it was based on someone else's broken heart.

Of course.

You know what I mean? So—

And it required that person to not only have the broken heart, but to express that broken heart first, right? Without that, all of culture ends circa 2023 when LLMs and generative AI came into the world. Are we there?

Is culture over, I guess, is my point.

I don't think so. I hope not. I think what's different is the bar for having something to say and saying it in a way that is not just beautiful, but technically just impressive, has gone up, right? There have been lots of people replicating Picasso since Picasso's time.

Definitely.

But the value at Sotheby's is not the Picasso ripoff. It's not the LLM-generated version of, you know, your portrait Picasso style. It's the original because there's only one of them.

Yeah.

And we know this was created by the master.

Yeah.

And there's prestige in having that original, right? There's still value in that versus the ripoffs.

Yeah.

So part of this requires us to still value the originals. And I do think among sort of the patron class that has always funded the arts, there is still that desire to be the one of one, right? That they have access and curation and ownership of creativity. Um, but I think the, the, the bar has just gone up. Right? If I can create or recreate a decent song, decent is no longer good enough. Now I need to have something that is truly original, à la Lin-Manuel Miranda in Hamilton, right? Where you're like, whoa, not only what he's saying, but how he's saying it, the expression, the vision, the execution, all of this we had never seen before. And it was so arresting. that it took the world by storm, and it still is, whatever, 12 years later. And so I think not that every piece of work has to be Hamilton, but the good enough sort of copying of the geniuses, right? You might used to be able to make a living making the equivalent of IKEA paintings for someone's wall, and that's not— like, that market has gone to zero now.

I hope that's the case. I'm really hoping that So I'm thinking about almost like a dinosaur situation. Like, if you were alive before AI took over, you have an original seed inside you somewhere. The question is people entering the world now, will they—

Will they ever have that seed or have a reason to cultivate that seed? I think that's what we, as educators, we as people who care about the arts, that's what we have to continue to support. and encourage, right? So everything from, like, my daughter who is 6 years old loves visual art. She's in art classes, and right now at 6, what they do is replicate what exists.

Mm-hmm.

But she's building the technique so that when she has something to say of her own, she'll be able to make that. And the longer that I can encourage this, and in the meantime, take her to museums and introduce her to artists and encourage that self-expression even when the output is not yet genius work. And the longer that I can push her off from being like, oh, you don't have to do that, there's a shortcut, right? And I think this is the same that we're seeing with AI in education. It's not that these tools are bad in the same way that, like, the calculator didn't obviate the need from understanding arithmetic.

Yeah.

Excel doesn't change the fact that you still need to know why you're doing the analysis. It just makes the analysis so much faster. But you can't hand the tools to a 6-year-old and say, now you can use the tools without ever understanding what they're doing or why they're doing it. And it's almost like you just keep trying different things until you like the output, and you never know why, right? There's so much of AI that is a black box and is only understood, if at all, by the people making it. And so I think there's some responsibility for educators and for people who want to protect the culture to say, like, maybe there has to be a— I don't know, an age limit or an ability to say, like, now that you have mastered the concept, the techniques, you've built a rich inner life, you have an understanding of why we might need to be able to make this from nothing. Now you've earned access to the tools such that you can use them responsibly and use them to enable the work, not to replace the work.

Yeah, no, I think that's a— that would be where we need to get. I don't know how we get there, but that's where we need to go. You know, you brought up calculators.

Yeah.

I feel anxiety whenever I see or people talk about calculators because I didn't do so well in math, and I never really understood The greater part, like if the bottom left of the calculator were the 9 numbers plus the 0 and the basics, plus, minus, division, and then there were all those fancy characters on the outskirts. I never really understood those. I never really figured out how to use those correctly. Anyway, that just gets me stressed out, but let's get back to some business and entrepreneurship talk.

To be fair, I still don't know what a couple of those do, and I was a math major who went to Harvard Business School, so.

And you didn't. Okay, good.

I mean, in part, There are lots of different types of calculators. We had financial calculators when I was a business school student here. And to some degree, I was like, if I push this, a thing happens, and I don't know why or what.

Yeah.

There are graphing calculators back from, you know, TI-89 still has some sort of monopoly. Like, the fact that TI-89 still costs the same thing today that they did in 2001 is amazing to me. Texas Instruments, great business model. But I think for the most part, like, don't let it intimidate you. Just use it for what it's useful for.

Yeah.

And maybe be curious, but at the very least, ignore it.

I barely— I'd have to try really hard to be curious, honestly, about math.

And that's fine.

Yeah.

That's fine.

But I'll get by. I got what I needed. But the minute they said you don't have to take any more math classes, I was out the door onto other things.

It just, as a math major, that breaks my heart a little bit.

I'm sorry.

Because what most American students learn as math through elementary school, middle school, high school, even, like, your first math class in college, is not, like, mathematics.

Mm-hmm.

It's arithmetic, it's calculations, it's computations, it's some of the basic building blocks of things that you do need to understand in order to do higher-level mathematics, but it's not mathematics.

Oh, you're making this mathematics thing sound exciting.

Math is creative. And math, just like the arts, require people to have an innovative point of view to say, like, what if I tried to solve this problem this way? And what if I deployed these tools? And what if I tried to, like, find a way in? It's so much more interesting. Than multiplication tables and calculating integrals.

You're making it sound good, but I missed the boat is the way I'm going to look at it.

It's fair.

It's fair. It almost gives me like a fever dream anxiety feeling. Opening the curtain, there's excitement there, but I'm also petrified. So let's go back to entrepreneurship. I'm sorry to take us off track, but I appreciate being able to talk to you about this kind of thing because I think you've got a similar mindset to this as I do. From an entrepreneur's perspective. What has AI genuinely changed, and what might we be overstating?

I love that you're asking that question.

Okay, good.

Because I am not just like a blanket cheerleader for these tools, even in entrepreneurship, not even putting creativity aside. So there are a couple of really big shifts that are materially shifting, especially early, early-stage entrepreneurship, when you're still in that point where you're trying to verify, to validate, is this a real customer problem? For which customers? Are they willing to change their behavior or endure some sort of friction, pay money, something to solve this problem? Right? Those are like 4 questions before we even get to the point to what is your solution? Because if it's a nice to have, not a need to have, it's like, yeah, it's annoying, but it doesn't doesn't rank in the top 10 things that I am gonna focus on in a day, it's a waste of time to build a business solving that problem. So there's tons of work up front before we get into product design or technology or pricing, go-to-market, right? All of those things that you think about as the work of an entrepreneur, there's all this work up front that really comes down to understanding the humans that you are trying to solve a problem for. And there is, in general, I think the default for especially a lot of first-time entrepreneurs is, well, I'll just do surveys. I'll send a survey out, I'll ask questions so that I can get that at scale, right? 'Cause I don't want 5 pieces of feedback, I want 5,000. And I'll send out a survey and I'll ask them questions, and then they will tell me like, yes, this is a problem. Yes, I'm willing to change my behavior. And we've always pushed students, away from surveys because, number one, they can only answer the questions that you knew to ask, so you might not be asking the right questions. Number two, there's a pretty big gulf between what people say and what they actually do, right? And certainly when you are being asked a question about it in a theoretical context where you're not having to pony up money for the solution, you're not having to change your behavior, you're just thinking about Basically, would you like to be the sort of person who has this problem solved? It costs you nothing to say yes, right? So the signal from surveys is terrible. So we say, don't use surveys. Go and do customer discovery. Go talk to actual people, and don't give them these leading questions. Elicit, pull out their experiences such that you can then deduce, right, what are the learnings from this? And don't just talk to them, right? There are these other tools of ethnography where you go and you exist in the space alongside them. You observe what they do. You don't even ask them to self-report. You watch as an outsider, and you can take notes or capture or document the behavior, because not everyone is super self-aware. And so being able to document is getting you one step closer to the truth versus self-reported. But it takes a long time to do that work, right? It takes you an hour to spend an hour with someone, plus the time prepping and finding that person in the first place, plus the time to pull out all the learnings after. So in sort of old-school entrepreneurship, as in 3 years ago—

Sure.

You might only have the capacity to do 10 or 12 of these, and then you're like, good enough is good enough. Let's move on. So one of the big shifts in a lot of these tools is it can really condense the amount of work that goes into getting more of this ethnography-based, interview-based customer discovery, where certainly the processing of data, you can put your raw interview notes into an LLM and ask it, like, pull out the themes that you're seeing here. You don't have to go through and code it by hand. It can do some of that processing for you. You can also use it to generate, like, give me, help me figure out a first draft of an interview guide. Here's what I'm trying to learn. I don't want leading questions.

Mm-hmm.

So help me craft this. Use the mindset. And this is where I often will, when I'm working with these LLMs or these Copilots, I will tell it what role I want it to play. You are, A PhD in anthropology who has spent 2 decades as a human researcher. Help me construct whatever this interview guide needs to be, because what I'm trying to learn are the following things. And I am not an expert in anthropology and human research. I've done it a handful of times, not 2 decades' worth. And so being able to have it take on that persona and go look through those sources— And come back with that perspective and help me craft it, I already have a better first draft of this interview. It can also help me think through how might I find more of these people? I want people not related to me, not attached to my existing network so that it's not biased in any way. So how do I find people I'm not connected to? It can help you brainstorm that. It can even, and I've seen this, and I I hesitate to say this because this is not a replacement for doing actual people-based research, but once you have a good body of interview notes, of ethnography observations that you can put into something like a NotebookLM on Google that constricts whatever you give it to be the body of knowledge, then you can ask it to take on these different personas, such that you can ask it questions. You can follow up, what is your day like here, or what are your challenges there? And it's based off of real people that you have actually spoken to, but it allows you to maybe have that regular follow-up cadence as you're working on something instead of going back and bugging them 3 more times, which you wouldn't do. And so instead, you would just not have that input. I think it's also useful to replicate perspectives that you might have a blind spot around. So if you have never been pregnant and you're trying to imagine, like, how might this be a very different problem for women who are pregnant? How might this elderly person experience this issue? How might this different culture— are there differences in culture that I need to be aware of before I make these choices? Again, it doesn't replace the human conversation, but it can help you approximate, and in some ways uncover the blind spots where you need to go and do more research.

That's outstanding, yeah.

So this is super helpful at the earliest stages. I think the next place where it's been really transformative is that instead of taking weeks or months to build a prototype, you can now vibe code your way into something in I've done it before.

I like that part.

What did you go build?

Oh, a bunch of different things. Yeah.

Going to leave it at that.

Yeah, yeah.

Fair. So I just went and was part of the Startup Bootcamp that we run here at HBS yesterday. I was like a mentor to go see what some of the teams have been working on. And I ran Startup Bootcamp for 4 years, so I know kind of what the old version could accomplish in, it's 8 days, right? It's not an accelerator, it's not a year-long, you know, anything, it's 8 days. But these teams come in with an idea, they've done some research before this program starts, and then over the course of this, I mean, I liken it a little bit to cosplaying entrepreneurship, right? 'Cause we're sort of stepping through the steps really fast.

Yeah.

But back in the era when I ran it, The best we could get to in terms of output in 8 days was a pitch deck that sort of outlined the theory of the venture. And most of it was, like—

Yeah.

Your best guess, somewhat based on some data. Mostly it was indicating where you need to go collect more information. And so I go and sit down with these teams yesterday who are on day 7 of 8, And all 3 of the teams I sat with walked me through a working prototype of their product.

Wow.

Which is astonishing, not because they're going to go deploy this product after 7 days of thinking about the problem, but because when you're getting feedback on anything, whether it's from investors or customers or anyone in between, if you're asking them to imagine something in their head and then give you feedback off of what they imagine, The gulf between what you intend and what they imagine could be quite wide.

Sure.

And so their feedback is only as good as you hope they got the right picture. But when you can put an artifact in front of them and you can say, like, try to use this, you can watch where are they clicking, what are they frustrated by, where do their eyes light up, and where are they like, I mean, That's cool, I guess, right? Like, you can get those real reactions to something that is not production quality, but it is closer to the truth than whatever they're imagining in their head.

Do you think that that kind of process can get you to an actual MVP?

It can get you there so much faster. I don't— it depends on the product, right? There are versions of what people are building that are just wrappers around technology that already exists. And then there are versions of what people are building that are net new technology. This is not going to replace what used to be a very serious engineer and a whole team that has to understand edge cases and safety and, and, and whatever else, but it will get you closer to something such that when you start hiring that team, you have a much clearer way to explain what's in your head again, because you're looking at the same thing. They have an ability to get faster to what are the constraints or the edge cases that I have to be thoughtful around. And I think you just, you are so much more productive. And even great engineers are using many of these tools to make code faster, right? Just like creatives, this is not gonna be the death of coding.

Mm-hmm.

We are still going to need software engineers who understand the architecture and the the deep interaction of all of these different tools, who understand the guts of the program. But we're going to need a lot fewer mediocre coding boot camp graduates, right?

Yeah.

And so I think that it, it raises the stakes of what is needed. And I think it brings the non-technical founder much closer to understanding what they're hiring for and how to interact with their technical team. than they did before. In the same way that if you look at a team that used— every single team I talked to yesterday had a beautifully branded deck. They had a logo. They had a name, right? Our students are notoriously sort of not that artistic. These are people who come from commercial backgrounds for the most part. And when I was running it, they'd say, well, I'm not a designer. The best I can do is pick a great font in PowerPoint.

Yeah, that's later. It'll look good later.

It'll look good later. And the difference is now there are all of these tools for taking a first pass at a pretty generic brand. So none of this is going to replace the thought and the effort and the design and the kerning and the pixel pushing and the perfect Pantone shade of what a great brand will look like. But it gives you a first pass. that helps sort of fake a little bit of credibility that, like, you're a real company, even though you're not yet.

Yeah.

When you go out and try to get feedback so that the feedback you're getting isn't biased against, like, this looks like you made it in Microsoft Word.

Yeah.

Right?

Yeah, definitely.

And when you show up to that meeting with the branding agency where your brief before was, I don't know, like, we're a bank and we wanna sort of fit in with what banks look like, but we're a disruptor to banks, so we gotta be different. different enough. And, um, you know, some, some of our assets are gonna live, I don't know, in like all these different contexts. So also I like blue.

Yeah.

Right? Like that is—

It's a nightmare for—

It's a nightmare for a creative. Whereas if you can bring something in and you're like, this is not it, but this is in the neighborhood. Yep.

It's here for a reason, whatever we're at. Oh, it is.

And here's what I like and here's what I don't like and here's what I don't even know.

Yeah.

The number of passes through that creative team is going to have to do has dropped significantly.

Sure.

And at least anyone who understands design is going to be able to differentiate between something that has been spat out by a remix of previous designs and something that is truly original.

Yeah. So that's kind of where the overstated comes in a little bit, I think, is you're saying, we're always still going to need the coders to kind of put those pieces together. And there might be this understanding of, like, now I don't need them. And people will be like, well, I can make my own logo. But that's maybe overstated. Maybe you can if you open a little candy shop in your town, then it's nice. But actual finishing a whole brand style guide is probably not going to happen in a way that it can compete with other established brands.

Exactly. In the same way, I think about a lot of people, certainly my LinkedIn feed has become overflowing with LLM-written posts.

Yeah, the em dashes are everywhere.

The em dashes, the it's not this, it's that. The it's 17 paragraphs long, but if you read it closely, it says absolutely nothing.

Yeah.

Right? The approximation of deep thoughts.

That's a really disturbing thought, but it's true.

And And they all sound exactly the same. So as an individual output, yes, ChatGPT can write you a LinkedIn post, but to stand out, to do the job, which is thought leadership, content marketing, networking, whatever that objective is for you, you can't just do that in a world where everyone is doing it. It still requires really interesting thoughts. really well-expressed writing, um, the visuals that might help capture your attention as you're going through. Like, you can spot a Midjourney-generated image a mile away now, I'm sure.

Right now, but who knows?

True.

You know, coming up. Yeah.

But even then, like, you look at these beautifully created, um, even think about, like, uh, headshots, right? People, like, you can go and get a approximating a professional headshot using these apps. Look at them deeply.

Oh, impossible lighting scenarios also.

Impossible lighting scenarios. There's no depth behind the eyes. There's no soul.

Yeah.

You can still, I guarantee you, if you looked at a real headshot versus one of these, even if technically they had figured out everything else, you would be able to tell the difference. It's the uncanny valley.

Oh, without a doubt. Yeah. And the instant no blemishes, all that stuff that you're—

Exactly. And so I think the— it gets us to a prototype, it gets us to a first draft, it gets us so much faster to get started. And no longer do founders have, of any kind, no matter your background, the excuse of, well, I don't have that skill set, I can't even try. So now you have at your fingertips the bare-bones mediocre version of basically every skill that a founder would need to get started. But it doesn't mean you can take this all the way through to production and expect it to stand out and take over the world and scale without breaking.

Now, I wonder, we were talking about that in LinkedIn, just these piles of text coming out of LLMs. Will we— I wonder, to get to the unique place, wouldn't there be a scenario that a year from now, I am able to customize my own new voice by saying, here are 5 books by Mark Twain, and here are 40 letters that my mom wrote. Make a new person that is Mark Twain and my mom. and writes everything for me.

You can do that now.

Yeah, but that's what I'm saying. Like, it's already happened.

What do I do? It doesn't mean anything about that is original.

But no one knows it. That's the part that scares me. People will just be like, what's so unique about this? Like, well, it's my mom and Mark Twain. That's pretty unique, you know? So—

It's certainly unique, although I would point out your mom could make the exact same one. So depending on—

She could, but we're not in competition with each other.

Depending on your source material, I think you're gonna end up with a lot of— Also, non-imaginative people are non-imaginative across the board.

Mm-hmm.

So if you are already trying to short-circuit the effort of having an original thought, I guarantee you you're not going to be thinking, like, what is the most— what are the components that make up my voice, and how have they been best approximated in history? Who is that writer, that speaker, that politician that I would mix No, they're all gonna go and say, take the Man in the Arena speech from Roosevelt and cross it with the eloquence of Barack Obama, and then throw in, I don't know, some of the real talk of insert bro podcaster here.

Yeah.

And everyone else is gonna do that same thing, right?

Yeah.

So also part of, as you think about, even if you're not the creative, Even if you are never going to go and write these things or paint the paintings, you still have to have a working knowledge of the creators, of culture, in order to direct AI in the precise way you want it. So I don't necessarily use your mother's letters and Mark Twain, but I will frequently, when I'm working with Gemini or ChatGPT or Perplexity, say, Take these combinations of whether it's people's resumes or their backgrounds or their way of talking and use that to inform how you now do the following thing. But it still requires me to have this working knowledge to be creative enough to put those together.

So we're saying that the majority of these paths or pools will be overcrowded with the most obvious scenarios. Yeah. Yes. And those will become dead ends or just oversaturated.

I think so. I think we— the pile of slop is real. And even as the tools improve, maybe the slop has a slightly higher resolution to it.

Yeah.

But it's still slop.

Yeah. So the more creative you are, obviously the more success or more differentiation you will have, probably.

I think so. I mean, again, the bar has been raised.

Yes.

For any type of creator, engineer, artist, writer, the bar has been raised, and good enough is no longer good enough. So now there— you either have to commit yourself to excellence and originality and put out ideas that might completely flop because they're non-obvious, or you need to find something else to do with your time.

I'm thinking— I'm thinking about Indiana Jones being chased by that giant ball that's chasing him. I haven't seen that since I was a kid, but that gave I'm talking a lot about what causes me anxiety. That caused me anxiety too. And I feel like that giant ball is all this AI stuff, and I'm just outrunning it for now.

I feel in some ways the same way, but I will admit, as, like, an elder millennial, that ball has been chasing me my entire life, adult life.

Yeah.

Right? Every step along the way of what we were told would be true about the future Yeah. Has been ripped out from under us. So I started college, 9/11 happened 3 weeks later. I graduated from college, I started business school, the financial crisis happens. I get out into the world, I start building things, I get married, I have a kid. I'm told you're gonna get a village that'll help you. You get a tribe of moms. And then the pandemic happens.

Yeah.

And I can't even hire help. I can't bring over my friends to hold my baby while I can take a shower, right? So I have none of those resources. Then I get out to the place where I have to pay for childcare, and the combined childcare of my 2 kids is greater than my mortgage, which I only managed to get out of sheer luck in the real estate, right? Like—

Yeah.

Every step along the way, the ball has been chasing me. So in some ways, this feels No new. It's just like, yes, it's terrifying. I knew there would be something terrifying. I just wasn't able to point to this in a specific way until now.

Wow. You know what's interesting though? As a young Gen Xer, my response to your response, the ball was not chasing me for a while.

Really?

And then it chased me, I think probably roughly around 9/11 as well, 'cause I was new to the workforce at that point. But yeah, so I have this perspective of like, what the hell are we doing? It was— everything seemed okay.

Yeah.

And now we're all going nuts.

Yeah.

Just part of the nuts. But anyway, yeah, all right, so if we would just— let's go a little bit deeper into the overstated. Where is it truly overstated?

So I think a little bit to what we were just talking about, it's overstated just how polished, how, you know, developed and robust the outputs can be. Um, the good enough is good enough, but the final production quality is not there. I think it's also overstated in— really, this is why I call it my intern and not my copilot. The quality of the work still requires a human to vet it, right? It's non-determinative, these outcomes. If you ask it the same question 3 times, you will get 3 different answers.

Yeah.

So as a result, you still need some amount of judgment when looking at the output. Before you just plop it in and build on it as truth, right? It's the opposite of mathematics, where once we have proven a theorem in mathematics, everyone can take that as a baseline for their next proof because it is proven without a shadow of a doubt. It's not, it's the opposite of that with, it's certainly with LLMs and with generative AI. So if you're going to use these tools to do market research, You can't just take the output and say, great, that's the size of the market. That's— those are the competitors. These are their price points. This is my pricing. You have to fact-check that. And you— there are ways around it. You can literally tell the LLMs, don't fabricate, don't hallucinate. If you don't know the answer, just tell me. You can ask it to give me the sources. And don't just tell me the source, link To the source so it can't fabricate the source, link to the source, and tell me what page it's on, what point in this source you got this information from. And that can decrease the likelihood of this being wrong, but it's never going to be— or at least to my understanding of how the technology works, it's never going to be 100% foolproof. So any of the work that you sort of outsource to AI, you still need a human who has enough understanding of that discipline to vet it. So this is where it can be dangerous if you're using it to teach you about an industry you don't know anything about, or role-play consumers that you haven't actually talked to yet, because it is very convincing in what it makes up. So certainly overstated there. And then I think even in the agent side of AI, where we're seeing the opportunity to, rather than hire a customer service person or an operations associate to build the agent to take in the thing and do the thing, and then the output is this, and then that triggers the thing, and then it all gets done, you still need human supervision. to oversee all of this, there will always be edge cases that break something. And where you, as a business owner, might want to instill judgment, right? Think about what broke when you were able to call a real human and explain your situation, and they're able to give you an exemption to policy, to the point where now you're on a chatbot or you're talking to an automated person on the phone, and there's no such thing as an exemption. They're like, we don't issue refunds. And you're like, my dad just died. Like, I literally can't do— I'm on my way. We don't issue refunds.

Yeah.

Right? So there's brand risk to businesses if you don't have that human involved. And then within the black box of how things work, you have to build in transcripts and records and logs of the decisions being made and how the data is being used so that you can go back and audit it. later and be sure that what happened happened correctly, especially if you're going to get people who, you know, this person is approved and this person is denied and they want to know why.

Yeah.

So there's a lot of human judgment still really necessary to this world. And there's a lot of sort of the clarity of intent. What do I want? I have to know that first and foremost before I can direct these tools to do it for me. If I ask it to design itself and then output itself, is it what I wanted? I don't know. Right. And so the judgment piece, the brand risk piece, is still very much real. And I don't want to overstate for founders that, like, you can just go and build an entire company yourself and let the machines, let the robots Let's do the work.

Can we try an exercise?

Sure.

All right, I've got an exercise for you. It's almost— people might say it's a game.

OK.

They might say it's a game. So I'm going to describe a handful of startup moments.

OK.

And I'd like you to then tell me on each one of those moments, is AI helpful, misleading, or dangerous?

Ooh, I like this game.

All right, but it's not over. For the people who are watching, I have paddles that I just created before we started. And you can play along at home with your list if you're listening or you're watching. This is helpful.

Okay.

This is maybe misleading.

Okay.

And this is dangerous. So when I give you these scenarios, you get to hold one of those up. We were just talking about, you know, extreme talent and creativity. And look at that. You know, those are beautiful.

I feel like I'm a judge on The Voice or something.

Yes, this is our version of The Voice. Super exciting. All right. So scenario 1.

Yeah.

You're going to tell me if it's helpful, misleading, or dangerous.

OK.

OK, a founder uses AI to generate customer personas before talking to any real customers. Is that helpful? Is that misleading?

I think it's dangerous.

That's dangerous? Why is that dangerous?

Because the crux of your defensibility is building something that people want because you understand, you have an insight, a deep insight into the problem being solved. And if you have a good enough insight off of what may or may not be slop, that entire— that's the kernel of your business is understanding the customer need. You can create personas after you talk to real customers and use that as the training data. But I would fundamentally discourage any founder from using it first.

Dangerous.

Dangerous.

All right, ready for scenario 2?

Yeah.

A solo founder uses AI to design branding. We talked about this a little bit. But let's go into it again. A solo founder uses AI to design branding, pitch decks, and a landing page before hiring anyone.

I think that's helpful.

All right. We talked about a little bit already.

Better to get started.

Yeah. Simple.

Yeah.

Let's go on to number 3. Okay. A startup uses AI-written copy to test value propositions across dozens of ads. Not so clear, maybe. I don't know.

Less clear. I'm going to do a joint helpful, but maybe misleading.

AI wouldn't do that. I know. I know. AI wouldn't put both of those up. All right.

So, okay, so here's the benefit. Rather than, we talked about this earlier, rather than talking to a handful of customers, you can talk to more. Rather than running a handful of value prop tests, now you can 10x that number off of the same time of human input. So that's helpful, right? You can have more confidence because you have more data points. Where it could be misleading is when you're testing value props.

Okay.

Right? We get back to what is the customer problem and how well do you understand it? Really excellent writing can make the difference as to whether you are communicating that value prop in a way that not only spans the translation divide between you and the person receiving it, but also Gestures to them. I see you, I know you, right? There are all these little nuances that can either say, like, I'm deeply in your world and I get it, or I am an outsider trying to solve your problem and I'm never going to be able to. And it's tiny little things. Word choice, length of sentence, the tone, the vernacular, even where you run those tests. And so a thoughtful founder who knows that ahead of time and uses those guardrails to create really nuanced tests at scale, probably pretty helpful.

Hmm.

A founder who is not aware of that and just says, give me 40 versions of these 4 value props, and then goes and puts money behind them on Facebook ads—

That's misleading, absolutely.

I think it's misleading. You're going to get mediocre data because you had mediocre inputs.

Great. All right, let's do another scenario. That was a good one. A founder relies on AI to analyze customer feedback instead of reading interviews themselves. Ooh.

Okay, this is a could be helpful, could be dangerous.

Oh, this is a 2 extremes.

2 extremes.

Okay.

Okay, so it is absolutely helpful to be able to just drop all of your raw data in, qualitative data, and ask it to help you synthesize these insights. You're gonna get to probably a lot faster of a first-level understanding, and maybe even have it pull out things you wouldn't have caught. Okay. Where I caution you to not stop here, but also go back and read. And it doesn't have to be every single word of every single page, but go and spot read. Pick spots where you're going deep into the verbatims, because sometimes there are insights off of these specific things that customers have said that require, again, human judgment or human ingenuity to say, oh, Wait, that could connect with this, and that reminds me of this thing that I experienced in my professional life 12 years ago that AI doesn't know about. And I'm pulling all those 3 things together to make this leap toward an insight that I might be able to build off of, right? So if your intern who's great at data synthesis does the work for you, you're going to get hopefully great data synthesis, but you might not get that same level of world-changing, product-changing insights from something that only you bring to the table. And this is why when we talk to founders, the 2 top questions anytime someone's building something is why now and why you. And this is a perfect example of the why you has to matter. for this to be defensible and for you to be able to build something that matters. And if you're taking the why you out of the equation in the analysis, in the insights, then maybe there is no why you.

You need the why you.

I think so.

Great. All right, our last scenario.

Yeah.

I'm going to bring it closer to my eyes. Oh, man, getting old. A startup pitches investors using AI-generated market sizing.

I put that in the could be misleading.

Okay.

I want them to use those tools as part of their work. But if they take it out of the AI and plop it into their slide deck and have no— A, have not vetted the facts, but B, have no understanding of the drivers of that market, they're gonna fall apart on the first question from the investor.

Mm-hmm.

Because any market is not static. So having understanding of these drivers, is it growing? How fast is it growing? Is it shrinking to a level that we are concerned about? Is there latency in the market, untapped demand? So the current size of the market is not actually indicative of what could be, right? We saw this with Uber. When they first launched, the size of the taxi market in San Francisco, you're thinking, like, well, if you take a portion of that, 10% of that, how can you build a business around this? Years later, their business alone in San Francisco usurped the size of the original taxi market.

Wow.

Because the solution kind of sucked. And so a bunch of people who wanted to use a taxi didn't. So there's the opportunity to open up demand that doesn't exist. And be able to make the argument around why and how much and, right, the drivers there. There's also the layers of complexity around, like, how are behaviors shifting? Are there signals that you are seeing that might dramatically change what this market looks like? Are there submarkets that you might want to focus on for specific reasons and make that your beachhead? All of this requires, again, back to the why you, human judgment, and not just plop it in. So great as an input should not be the output.

Perfect. I'll take those back from you now, 'cause they're distracting me. So well drawn. Okay, so we talked about how AI can potentially get you to good enough for certain things. And we talked a little bit about taste.

Yeah.

Maybe a lot about taste. It's hard to tell. I was trying to talk about it as well. How do we evaluate taste? Because there are all different kinds of tastes out there, and some people are louder and have bad taste, and some people are quiet and have good taste. So we're kind of in a who's louder world wins right now is what it feels like in some cases. So how does taste play into that?

So this one's— this one's tricky because we're very much in an attention economy.

Mm-hmm.

And the thing that the platforms reward, the algorithms are engineered to do, is to give attention to extreme emotion. Which means you can't just be loud, but you have to be angry about it.

Oh, right.

You can't just be loud, but you have to be like, this is the only way, and there's no room for anyone else. And so that is who the voices are that we're hearing over and over, and therefore what they believe becomes the taste. So I hear you, right? It's a very different world than when we had the gatekeepers and the curators and the editors, for good or for bad.

Yeah, it was different.

That at least because we knew there was a process for, like, they're rising up in their career because their taste is Good, that what they choose for us has some level of curation or editing involved that we can trust it. So the space is very different. And I fully hear that, like, what if you have great taste and you're quiet? There's clearly— there will always be a room for tastemakers. Not everyone has good taste, and most people know when they don't have good taste. Not everyone. And so they're looking for people who can guide them in that, right? As a postpartum mother, when I was first trying to figure out what my style would be after my second kid, because after my first kid was the pandemic, I just wore sweatpants for 2 years.

Yeah.

There was a moment when I was really missing the fashion magazines that I was able to turn to in the 2000s. To be like, what are we— you're telling me skinny jeans are out. What's in? And how do I translate that for a 40-year-old body who's had 2 children? And I went through, you know, pretty deep in the muck trying to find tastemakers to guide me. I ended up in TikTok world with a mix of people whose taste I appreciated.

Mm-hmm.

And I basically You know, followed them, pinned them to say, I agree with you on this, so what are you gonna tell me next? Oh, that's new. I'm gonna take that in and use that as an input. So part of this— there will always be a need for tastemakers. How we find them, how we amplify them has pretty dramatically changed. And how truly— and how they build a business about being tastemakers has dramatically changed. Right now, TikTok creators make their money off of views, comments, engagement, buying things through their links. If we go to a world where we are in generative engine optimisation and not search engine optimisation, where you're buying things directly through these LLMs and not through affiliate links, a huge portion of their business model has now gone to zero. I think in general, we are just in a moment of flux. And I don't know what the answer will be.

Yeah.

Except that there will be a need for tastemakers. There always will be.

I hope so. So I feel like, you know, I grew up in music and I think about who those tastemakers were and are. And I feel like I've got this understanding of what I like and why I like it, whether other people don't and all that stuff. I don't know. And then I remember watching The Devil Wears Prada and Meryl Streep kind of playing the Anna Wintour.

Back when it was in theaters for the first time?

Yes, back then.

Like era 2006?

Yeah, I was— that was definitely— yeah, I was probably 50 at that point. But the part I found interesting, which kind of had this like equated to my music thing, is she did that whole long thing about that cerulean blue.

Cerulean, yes.

And I was like, what? What? It's just blue. Like, I didn't understand it. But then I started to— in this, you know, kind of cheesy dialogue. I was like, oh yeah.

It's not cheesy, it's iconic.

Well, yeah, sure. And but that felt like a parallel in the fashion world to what I've kind of known, at least in the music world. And I thought, oh yeah, yeah, that's why I didn't realize it. But a lot of what I'm choosing, I'm not choosing, right? I'm just seeing the things that people are telling me is good. And in most cases, they know what they're talking about, and that's why it's good. In most cases, But right now, I see—

Yeah. There's some self-fulfilling prophecy though, right?

Oh, yeah, yeah, yeah, yeah.

Like, they know what they're talking about because by them saying it, it becomes true.

I guess so, yeah, yeah, because they're—

Like Pantone decides what the color of the year is, and then everyone's like, that's the color of the year. So I—

I never agree with that. Yeah. There was a green one that was not a good green one in 2017 that upset me. I'm not joking.

Tangerine Tango, 2012. I still remember it.

Really?

And I am not an orange person, and yet I went out and tried to incorporate—

The pressure we feel.

Right.

The pressure we feel.

It becomes self-fulfilling because of the tastemakers in some cases.

But sometimes those tastemakers are good.

Yeah, sure.

And we get good things out of it. So I think the ultimate scenario would be the tastemaker plus other opinions.

Okay.

But I feel like maybe right now we're not getting the tastemakers anymore.

We're just getting the other opinions.

Yeah.

It's possible. I mean, think about this from An editorial perspective of the news.

Yeah. Oh, yeah.

Well— Right?

Yeah.

We used to have the Walter Cronkites, the whoever the equivalent was on the written side at the New York Times, right? We had these editors, these figureheads at the top of the newsrooms that would decide, would use their taste about what is news, what should go on the front page, what should be the leading story at 6 o'clock. and what isn't news. And that editorial judgment shaped everyone's understanding of what was going on in the world.

Yeah.

Now that editorial judgment is being provided by algorithms that we have no insight into how they are written or who is writing them.

Yep.

And what biases or agendas they might have. We're not getting disclosure statements from the engineers behind TikTok's algorithm. We're certainly seeing this in the sort of polarization of what used to be sort of historical news outlets, where you say, oh, you go here if you wanna hear this opinion, and you go here if you wanna hear this opinion.

A little clearer.

Right? And so it's already demonstrating that there are tastemakers who are still deciding how cerulean shows up in your closet.

You just flipped that on me so much. I don't want the tastemakers, if you put it that way. I don't know what to do. Yeah, you're right.

So the democratization in some cases is freeing because there are stories I'm seeing on TikTok that I'm not seeing on the front page of the New York Times. And I can have an opinion over like, that should have been news and it's not getting attention. because it didn't make the tastemaker's agenda, but if enough people see it, that amplification will eventually be the story and they'll have to cover it or some, right? There's some ability to like use the democratization to say, that matters to me, even if it doesn't matter to you.

Yeah.

And so I think when you're thinking of tastemakers in the positive sense, you're thinking about, these are the areas of my life that I don't feel like I'm an expert in, Making the best choice, whether it's cerulean blue or it is— I don't, I don't know what the right font is for my logo. I'm not a kerning expert of what wide-spaced letters might subtly psychologically indicate to a customer versus very narrowly spaced letters. I don't claim expertise there. And so I'm looking to experts to guide me.

Mm-hmm.

And I hope that the ones I'm looking to have good taste.

Let's hope.

And so there's a piece of this around, like, how do I find the ones with good taste? How do we make sure they're visible in a world that might at best be drowning them out with noise, and at worst might be actively suppressing them? And then on the other side, they're like, these are the spaces where I do have some discernment in, and I want to be able to see the full raw data. I want see the pool of information so that I can have and use my taste to pull together the components that I think matter most to me. And so I think that's part of this tension of we want tastemakers when we like what they say.

Yeah.

And in some ways, we want to be able to counteract the noise so that signal can be found amidst the outrage and the loudness and the whatever else, while still having some amount of access to the raw inputs to make sure that the taste being handed to me is not the only one I have access to if that's not the one I want.

In the classroom, Are you teaching differently now because of AI?

Oh, yeah.

In what way?

So at HBS, we have 2 forms of education in the MBA. There's case method, which has been our bread and butter for over 100 years. And that is still— we're still teaching case method, right? It's Socratic investigation of ideas that requires engagement from the students to co-create this learning opportunity, right? We're not showing up to lecture them, here are the facts. We are there to have a conversation and prod into the judgment and the edge cases and the understanding of what is going on here and how might that apply somewhere else. What has materially changed about teaching case method is the preparation that took me 2 to 3 hours a night per case when I was an MBA student, You can now feed the case and the assignment questions and the exhibits into an LLM and ask it to do your homework for you.

Cliff Notes.

Cliff Notes. So that is forcing, in a good way, professors to up their game, where we can't just go in and have the cold call be verbatim one of the assignment questions. Because they've all printed out what ChatGPT has told them, and they can—

They're ready.

They're ready. So it requires us to be really thoughtful about what are those questions that require the application of the analysis or the inference from, like, take it one step beyond the data in the case. the fact patterns, even the first-level, most obvious judgment, and say, but what happens if this? Or how might you apply this there? So it already is making us up our game in our cold calls, but it also requires us to stick with students longer in the interrogation, in the dialogue, and not allow these chip shots of comments where they—

Mm-hmm.

hop in and say the one thing, and then they're like, great, I made my comment. I can sit back and I'm good, to say, well, why do you say that? Where did you— what data are you using to come to that conclusion? Under what circumstances might that decision not hold? You have to really stick with them and push that conversation much deeper. So that's on the case method side. The other methodology that we use It's learning by doing. We have a lot, especially in the entrepreneurship unit, we have a lot of courses, especially in the second year, where students are actively building their companies as the content of the class. So I taught a course on startup operations last semester, and my students were all founders building their companies. And the entire premise of this class is, in addition to building the product, You also have to build a company, and there are all these things that you have to A, know exist, and B, be able to approximate a good enough solution for until the moment you're big enough to hire the expert. Everything from legal choices about the structure of your entity, where you incorporate, how you're thinking about equity splits between you and a co-founder or advisors. How you're building out your financial models, cash flow, and the implications of different levers you might be pulling in the early days of your business model, how that then implicates who you take money from and how you structure that financing, things around culture and hiring. And you don't have a human resources director, so I hope you're really good at writing job descriptions. and doing interviews? And if you're not, who's surrounding you to help you build out that skill set? And so there's all of this work of company building that has to happen.

Mm-hmm.

So all of my students are, are doing that while also working on their products in the course of my class. And I think one of the biggest changes over the last 6 months is literally how I would teach operations back in July, is not how I would teach operations now. If I'm telling you, put together your headcount, your org chart for your next 5 hires, that's assuming you're going to have people as the next 5 resources.

Oh, yeah.

Where there might be a world now where you're like, yeah, I'm going to hire a head of marketing or a head of insert function here, and then they are going to have The opportunity, the requirement to go and build the agents that support them. And so it's not about headcount anymore, it's about what are the skills and the resources and how do I make sure? And that might change who that person is. Maybe instead of hiring the kind of middle manager marketer who could do it all, but also understands the strategy, maybe you go closer to— the CMO sort of judgment marketer. Because execution is no longer required, they can use these tools for the execution, and the judgment is actually what you're hiring for. So fundamentally changes how you set that up. Also how you instrument your startup. You're thinking about what are the sources of data I need as inputs to these agents? And how do I make sure I'm capturing them at the best sort of resolution level in the closest to real time as I can, so that I'm not bucketing these data and then getting them a month later only to realise it has a meaningful output on my company? Right. How can I get this in real time, feed it into analysis in real time, and let that real-time information come through a dashboard and indicate what changes I should be making. But to do that, you have to capture the data first. So how are you instrumenting your company? All of this is different.

Yeah, so you're kind of standing on 2 sets of shifting sands.

Oh yeah.

Where you're having to think how—

That ball is rolling fast.

Oh man, I'm in the desert with Indiana Jones and that ball and everything, but yeah, It's very much like you're figuring out how you're going to teach this, but also being aware of what the students can do with this new technology and making sure they overlap in a way that's working. That's a little crazy.

I mean, it's no wonder that most of us in our 30s and 40s just have low-level anxiety at all times.

Yes. It's there. It's going to be there forever. Well, at least till maybe 50. I don't know. We'll see. If you're looking— if you took away all the tools that exist right now, Not using the tools, that is.

Took away the tools as in they no longer exist?

Well, you're not using them to solve a problem.

Okay.

They exist, but you're not gonna use them right this moment. But you're an entrepreneur in this world.

Yeah.

And we're just talking about your mindset. You can't use a tool to fix it. It's just your mindset. What's the mindset you need to have in this AI-saturated world?

Yeah.

Meaning you can't just be like, well, ChatGPT, what should my mindset be?

I think the 2 things that continue to come up as I'm meeting teams and I'm listening to my colleagues who are actively investing in companies in this space, the doing is no longer the moat because everyone can do, right? The headcount, the scale, the operations, all of that, it can be done. So as you look for moats, what is your competitive advantage here?

Mm-hmm.

I think it's shifted to proprietary data. What do you know about the world that other people don't? And judgment. What can you infer and connect and Extrapolate into an insight that a computer can't. And so, you know, we are seeing a year ago, we had all of these companies coming out of the MBA that were like, oh, we're AI for blank, we're AI for law, we're AI for accounting, we're AI for consultant decks. And all of them were using this tool that exists, to do the thing that humans used to do. But, but like, so could the company, right? Like, McKinsey doesn't need to hire, buy your software if they too can create an AI for decks, right? Like, if we have access to it, everyone has access to it.

Yeah.

So the access is no longer the moat. The output, the, you know, headcount, the, the, The scale is not necessarily the moat. So what is, is the proprietary data and the insights. If you have something that others don't, now you have an ability to solve something in a different way, or go to market in a different way, or have a not just incrementally better, but exponentially better solution because of those insights. And that is fundamentally what matters for this next era of entrepreneurship.

What do you think we're still not asking about entrepreneurship and AI?

I think the societal implications are larger than we are recognising right now. So as you think about what do we need to train entrepreneurs in, what sort of guardrails Or tests do we need to— the lenses we look at what they're building through, I think it is even more incumbent to include the ethical lens and the impact lens. Right. So from an impact point of view, we know that vast amounts of AI require vast amounts of energy, and that energy also, in many cases, requires vast amounts of water. And these 2 resources have to come from somewhere, and the use of these resources can materially impact what is available for everyone else, or the cost of that availability, or the depletion of those resources. And when it's small, it doesn't matter.

Mm-hmm.

But when it's everywhere, it's huge. And if each person is only making their decisions on an individual we'll never have that collective, how are we designing the system or mitigating these impacts at an industry world level? So there's the impact piece here that I'm not really hearing much of any conversation around other than, like, the environmentalists who have been—

Yeah.

—beating this drum for a while. But then there's the ethical and the personal impacts of this because so much of this is a black box. So much of this is, again, it's algorithms, many that you are controlling as the creator of this, but some of them you're not, right? If you're taking in someone else's product or someone else's tools as part of your process, you are ingesting their algorithms and their choices, their judgment, and you don't have insight into what that judgment was.

Mm-hmm.

So as you build something, you have to be even more aware of what are the biases that this data might be based on? What are the biases that might be introduced through the judgment? And all data is biased.

Mm-hmm.

So it's not that we can get perfect data, but you need awareness of which ways this is biased so that you can mitigate it. So that you can build in what is the opportunity for, I want to escalate this to a human. What is the visibility into how decisions are being made that if something were to jump the tracks, a human can step in and redirect it? Because once you set these things in motion, if there's no visibility into the system, it's just going to run and run and run, and it could shift. over time, especially if you're building on anyone else's tools, because as their models change or as their judgment changes and it gets ingested into yours, right, we end up going to Sydney instead of Oakland. Like—

Yeah.

It's a very different outcome, and you don't know it's happening. And yet the people you are serving are very much impacted by the fact that it is happening. So I really hope that we have even more than we have in the last 10 years bringing in the ethical conversations into the education and into the analysis and the evaluation of these startup ideas, because you have to be even more intentional about this now.