Transcript: Taking Your Business From Zero to One With AI

The transcript of How Do You Use ChatGPT? with Nicholas Thorne is below for paying subscribers.

Timestamps

1. Introduction: 00:48
2. How AI can make you a more effective founder: 12:10
3. Live demo of Audos! 17:01 
4. Why Nicholas built an AI tool to enable entrepreneurs: 27:11
5. How Audos puts you in “edit mode” instead of “create mode”: 28:37 
6. Tools to gather customer feedback, generated by Audos: 31:29
7. How Audos actually works: 36:15
8. Nicholas uses ChatGPT to prototype a new feature: 38:47
9. How to establish checks and balances while using ChatGPT: 45:54
10. AI as a force for pushing entrepreneurship to new heights: 1:00:37

Transcript

Dan Shipper (00:00:00)

There's this sort of end-to-end process of, start with a couple of questions to a deck, to a website, to a little wizard that can help you modify a couple of different things. 

Nicholas Thorne (00:00:09)

I hope you're not tired of all the AI because unfortunately the answer to it is that we use a lot of AI to make—

Dan Shipper (00:00:15)

Why did I know that was coming?

Nicholas Thorne (00:00:17)

I've never taken a computer science class in my life. I don't know anything about computer science. Prior to ChatGPT, I never particularly programmed anything. ChatGPT has been completely revolutionizing to me, of being able to write code. For programmers, I know that what I create is not code, but it does function, it does work, and it does deliver a result. 

Dan Shipper (00:00:35)

This is truly wild.

Nicholas, welcome to the show.

Nicholas Thorne (00:00:50)

Thank you for having me, Dan.

Dan Shipper (00:00:52)

I'm excited to have you. So, for people who don't know, you are a close friend of mine. You are the general partner of the incubator Prehype, which actually incubated—or really was where Every came from, where my first, not— my first company was 10 years ago. Where Every came from is from Prehype, so I owe you and Henrik, the other partner, a big debt of gratitude for that. And in addition to—

Nicholas Thorne (00:01:32)

The best types of debts of gratitude are when the other sides of the coin feel that they are the ones who owe the debt of gratitude, so I assume we’re a mutually assured destruction here.

Dan Shipper (00:01:45)

Or gratitude. Pick your poison. 

Nicholas Thorne (00:01:47)

That gets a little too close to MAGA, so we’ll stick—

Dan Shipper (00:01:53)

So, you have incubated— I mean, Every is obviously one of your illustrious examples, but you've also incubated 50 other startups, including BarkBox, Ro, Public.com, For Them, Ferry, and Headway. And one of the things that I think is most interesting about you is you've been very early to the two biggest technology trends over the last decade and a half. So your first company, Basno, was an electronic badge company before NFTs were a thing. So totally called the NFT thing 10 years ago-ish.

Nicholas Thorne (00:02:35)

Yeah. Laughably so. I have to basically say to everyone who comes upon some version of that conversation that, unfortunately, NFTs pre-blockchains are the most laughable sentences you could imagine.

Dan Shipper (00:02:54)

Yeah, I guess you were too early in the sense that the blockchains were not around, but still you had the concept.

Nicholas Thorne (00:03:02)

At least we were mining bitcoin as a result of it and mined a block. So nycbitminter is in the blockchain.

Dan Shipper (00:03:08)

There you go—part of history. And then you were also just really into GPT-3 in 2020-ish.

Nicholas Thorne (00:03:18)

Yeah. Fall 2020. Yeah.

Dan Shipper (00:03:20)

I gotta say when you were first starting to get into it, I was like, this is kind of interesting, but I don't know, Nicholas is off the deep end again. And you were so right. You freaking called it. One of the really interesting things about being so early is I think you've been able to really understand what these tools are useful for and apply them in your life and in your business. And at Prehype, your job is to start and incubate companies. But what you've done is you've actually gone and started a new company to help do more and better incubations with GPT-3, or GPT-4 now. And it's sort of like this interesting GPT-all-the-way-down thing where you're using ChatGPT to help you build the business, the business itself, the product itself uses uses the GPT API to serve your customers, and your and your customers are using ChatGPT or the GPT API to build their businesses. So you're using AI to the fullest extent that I think you possibly can and the result—

Nicholas Thorne (00:04:39)

Yeah, maximalists or whatever the word is.

Dan Shipper (00:04:40)

Yeah, and I think the results are really sort of speaking for themselves. You told me in the pre-production call that your goal is to help people go from zero to business, help your customers go from zero to having a business as quickly as possible. And with AI that takes like three days now on average where it used to take a month or just wouldn't happen at all because they didn't have the skills to do it, and you're basically in this in this place with this company, it's called Audos, where you can't keep up with demand and you have thousands of people looking to pay you to help you build their businesses, and with a 48-hour payback period, it's kind of a crazy thing that you've discovered. And what I want to do today is go through this business that you've built and then talk about how you've built it with AI, and then how the product itself works with AI to serve customers ‘cause I think it's this end-to-end use case of, I want to start a business with AI and I want to learn how to leverage it to build stuff and, even if I don't know how to code, and to make products and make money online. And I think you're the perfect person to help walk people through that ‘cause that's literally your job and you've done it yourself.

Nicholas Thorne (00:05:52)

Yeah. I mean, as you're saying, the interesting part to me in just reflecting on it on the fly is there's both AI native in our capability set—the way that we work is AI is everywhere—and then, in a way, we serve a customer and our interaction with our customer AI is super native. So we're sitting on both sides of that trade, which is very purposeful, inasmuch as we have been trying forever to just remove as much of the edifice between founder and customer as possible. That's both the advice I would give other founders, especially in the earliest, earliest phases to be as tight to your customer as possible. And you do that in two ways: One is obviously by trying to talk to them and spend time with them. And so AI as a conversational force is an incredible power for staying close to your customer. But then you also do that by not getting sucked back into the organization as much as possible. And so, if you can imagine the amount of capabilities that if you can turn them over to AI and you don't have to hire people that you then have to manage and spend time with, well, then, definitionally you stay close, tighter to the customer too. And so it's both because of the way you interface with people and by not getting drawn back into the org that you stay close to the customer. So, for sure, we're just kind of AI everywhere, not as a buzzword, although happy to benefit from the buzzword if it's a benefit, but also just because I think it's a great way to spend time with customers.

Dan Shipper (00:07:30)

That makes total sense. I would actually really love—can you show us? 

Nicholas Thorne (00:07:32)

Yeah, let's do the normal— 

Dan Shipper (00:07:35)

Yeah, the business that you built and show it to us a little bit.

Nicholas Thorne (00:07:38)

As you mentioned Our core business has been helping entrepreneurs figure out what company to start next, and we've tried a hundred different ways to do that. And but by calling it 2020 and having been in that rough business for a decade that looked like maybe we could a few times a year, help someone start something. And that was because A, we needed to have a good relationship with them and get to know them and these things take time, but also it's because practically speaking, there was a lot of work, in the way that we were doing it, to kind of help someone go from thinking about a couple of different business ideas, framing them up in a consistent enough way that you could even kind of compare them apples, going out and finding some early customers for those things and talking to them and synthesizing the results of those conversations, and then going out and finding some more customers who are hopefully strangers and talk talking to them in some way, probably less of an actual conversation, more of some sort of lead acquisition flow, all towards like helping create this dynamic for a founder where they were, as much as possible, not kind of a priori making judgments about which of their ideas were good ones, but instead, just building some early momentum and seeing if anything kind of just stuck. Ad absurdum, we feel like bad pickers of ideas—or maybe not bad ones, but that picking is hard—and that if you can just go ahead and try things that you end up learning a lot and getting first-party insights. And so everything about what we've built with Audos is to try to make that set of processes available to a much broader group of people. We actually didn't set out to do that. We set out to just make our own process more predictable and repeatable, et cetera. And the natural result of it, of course, is that if you encode it in software, then you can do more, and so—

Dan Shipper (00:9:40)

I want to stop you right there. I want to stop you right there. ‘Cause I think you just said something really profound that that I've encountered too, and it sounds like we're sort of— I always love when two people have similar insights from maybe without having talked—they've come to the same conclusion and the thing that you said about we're not great pickers or maybe we're average pickers or whatever but it's always better to to go out and try something to know if it works. The thing that resonates so much for me about that is—I think it relates very deeply to what you're probably about to show and also what I've seen AI do for people. When I try to figure out, when I try to get that core of what's most useful is that for any project that you want to do, you can go and show how it would work and start getting feedback on it and getting hype—whatever you want, whether that's software, whether that's filmmaking, whether that's art, whatever, you can just do that in a couple hours, whereas before either it would have taken a long time and you would have had to have a lot of skills and a lot of whatever, or you would have had to be able to convince someone to invest in you to fund your movie to get all the cameras and the director and whatever, or you would have had to get someone to fund you to go and hire engineers to make your app. And I think the one of the core things that's happening is we're moving into this world where anyone can just make that little thing to show their work to show someone how amazing the thing in their head is without having to convince them. And it's really hard to convince people of things, but if they can see it, you might get a reaction and things might get funded or get made that never would have been able to be made before. And I think that's sort of what you're doing.

Nicholas Thorne (00:11:30)

Yeah. There's the prototype-is-worth-a-thousand-decks type of mindset in there, and I completely agree. And we've been working on a book, and one of the things that I just will call out to that is to your point here is that to me one of the profound implications about of that dynamic is not just true in terms of how you get started with things, but it's also true of how you operate them over time. And the point being, yes, that's, that's true of how you get started. It's also true of once you're starting and the practice of continuing to build things because, imagine that one of the reasons that an entrepreneur goes from being an entrepreneur to being a CEO with big-company problems over time is that they are no longer the first-draft-maker within their company. And so one of the things that we have been thinking about is what happens if you can be the first-draft-maker longer as someone working on something. And arguably in our view, it helps you stay close to your customer, which is important, but in general rarely is it that— I think we can all intuit that if the founder or CEO could stay close to their customer longer that that would be very valuable, right? And it doesn't mean that they would have all the ideas, but at least they could have contributed more of the ideas, right? And so anyway, these kind of hybrid concepts in my mind are really important in terms of both sitting in the customer service seat longer, I guess I'd call it, i.e., just being a little bit closer to the conversation that your organization or your business or product or project is having with your customer. If you can just hang on to that a little longer, stay a little bit closer to that, obviously your information flow improves, but then, to your point, if you can use those tools to create more first drafts, just the fidelity of what you create as a CEO is less of this— We talk about this concept that if you sit there as a running a big organization and all you do is you create this big of vision and then other people go off and create proposals as to how to achieve it. And then there's the quote unquote big meeting at the end that's just a very difficult organizational model to make work. Or it works, but it’s definitionally a big-company thing. And so I'm so interested in what happened. I mean, this is looking forward, really, it's like what happens when you have companies where the founder-CEO and/or the founding team is just more often and, for a longer period of time, the ones making the first drafts, then what? I think what you'll find is there are really interesting downstream implications of that.

Dan Shipper (00:14:16)

I love that. No, I think that's great. It reminds me a lot of— It's not the exact same thing, but there's that book about how Apple builds products and it's all about prototypes and demos. And it sort of reminds me of that a little bit.

Nicholas Thorne (00:14:29)

Yeah, 100 percent. Totally. Okay, so Audos— So, I was kind of like going into this background of Audos. So, Audos currently lives primarily— And I guess I'll just do a quick demo here, Dan, and then you yell if something we should do differently. So Audos currently lives primarily in an Instagram DM. And so it’s an AI, you can imagine—

Dan Shipper (00:14:51)

Wait, why'd you start there? So it's a DM AI bot. Why'd you start with a bot in DMs?

Nicholas Thorne (00:014:55)

Yeah. We started with a bot in DMs because distribution, what do you know matters, and Facebook has introduced, as of like last fall, a product to acquire people straight into a conversation. And that was a dream come true for people who spend a lot of time using the word “talk to your customers.” If you're in the business of trying to talk to your customers, well then, where do you do that? Where you can have kind of a persistent conversation and not have to install some new behavior for people where they come into your app, which is laughably difficult to do, so I think the question was, could we actually just make a version of a chat experience that would live inside of Instagram so that we could both take advantage of Facebook's products as well as, frankly, just do something that was natural and native to our customer. And so, that's where that decision got made for us. We had these onsite versions, in fact, at some point I was talking to a journalist and they're like, why does this work this way? I'm like, well, what are you talking about? And they're using some version of a product they found online from six months ago. I was like, oh, crud, that's not the right one, you need to go over to Instagram. And so, yeah, we start with a conversation, so people find us there. We do have a following and, I don't know, we’ve got thousands of followers who have traditionally. We were publishing these business-book little videos and shorts, and that was just to build up a little bit of an audience and whatever, but people do come and find us there. But mostly we just acquire people. We run ads that say varying versions of, do you want to start a company or—

Dan Shipper (00:16:41)

Do you have an example?

Nicholas Thorne (00:16:43)

Yeah. I mean we run ads that are things like, “That little business idea you have it'll still be an idea next week,” or “You can let our AI make you an AI and go from zero to business.”

Dan Shipper (00:16:51)

Wow.

Nicholas Thorne (00:16:52)

I think that's probably—

Dan Shipper (00:16:53)

I would click that. I like that a lot.

Nicholas Thorne (00:16:56)

You know, we do a lot of that kind of stuff and we're testing into more and different creative. But what happens here is that, when you get here, if you click on that there's a bit of back and forth that you don’t actually see here. But then, very quickly we do a few things. One is we anchor and the AI is trained to request a name from you, the entrepreneur, and so what we learned over time is that obviously, when you start to create chat experiences, where you start matters a lot, especially if what you're trying to steer towards is a fixed outcome. And in our case, we have some very fixed outcomes that we want to be able to deliver so that this is not just chat for the sake of chat. It's chat for the sake of calling a function and we can get as wonky or not about that as we want. But suffice it to say that June 13th of last year when OpenAI came out with a function-calling tool, that was a big deal for us. 

But so we try to get you to name a person. Again, that's a very basic thing that has nothing to do with AI that has to do with our own sensibility, that the type of business that we feel can help people start, but maybe in general, the way we happen to know how to help people start companies, certainly not the only way, but the way we know, is to help them really kind of deeply imagine who the person is that they're going to serve and almost with unsaid truth that if you do not know that person, or they're not in your phone book, or a few of them are not in your phone book, you're going to have an uphill battle. Whereas if you can kind of cultivate founder-customer fit, if you can actually take this knowledge of someone, it could be even yourself really, and turn that into an initial product offering and then build an entire kind of company that protects and cultivates your fit over time that really interesting things start to happen. 

But anyway, this person provides the name of their customer. You can tell that this person has thought about this, because once we ask them the next question, which is what does the customer need or want, but can't get and what's stopping them from getting it, they've provided a detailed answer. So they say in this case that Camden, their customer wants to save to buy a home but it's difficult in these times to save what they want. And this is where you can tell they've thought about it. ‘Cause they're kind of pitching the solution already as an employer-match program to help save for a new home instead of a normal retirement account, such as an IRA or 401k. So kind of an interesting idea to have your employer be willing or able to match your kind of home-saving contributions as opposed to your 401k. Importantly, we, the AI, will then try to reframe the problem for the customer. And this is really just to get a little bit deeper into, what is the problem? And so it provides a couple of different options as to what are some other interpretations of the problem where obviously our perspective is that if you can be very precise about the problem you're trying to solve, it just makes coming up with a solution easier. That's true. Generally, again, it's also true in how well the GPT can do in providing interesting solution alternatives. And then in this case, we'll pitch a series of potential solutions, which are really just across a bunch of different business models and solutions styles. How can we think about solutions for this? And rarely are they mutually exclusive. You'll see the first one is software. The next one's a platform or a network. The next one is a physical product. The next one is some sort of financial product. The next one is some sort of kind of content problem. So they're all these different angles. And most of the time, what happens is people say, oh, well, it's a little bit of that, and it's a lot of this, and maybe it's a touch of that. But within those three or four questions, what we then move on to is the GPT, our tool will say, okay, great, give me a second. And I'm going to try to kind of frame this up in a presentation. And so what happens now in the background is a set of GPTs really, are running a variety of processes that all then lead to contributing to the creation of a little presentation, which works through, high-level, what's the context? A couple of slides on what context does this problem and customer exist in? If we kind of restate the problem, how would we—

Dan Shipper (00:21:43)

Wait, wait, wait, wait. This is kind of crazy. So, basically you acquire a customer with an ad saying, do you want to start a business? And the customer's like, yeah. And then you ask them a couple of questions with GPT through Instagram, and then five questions in, they just get an automatically generated—I mean, this deck looks really good—beautiful deck. Are there any humans? Are you involved in Wizard of Oz-ing any of this, or is it all automated?

Nicholas Thorne (00:22:25)

No. Every once in a while, something breaks and someone starts yelling and we have to go hit a button or something to restart things. Google times out our permissions on the Google Slides little API—it's not really the Google Slides API, it's an app script, if you want to get wonky and, we have to re-permission it. That's probably been the key thing that keeps coming up once every three weeks. And as a type of thing where it's not working for a couple of customers and what's going on there, I'm like, why do we not put down a list that says, don't forget that. That's all automatic, and so, yeah, let's just go through super fast ‘cause what we do is we talk about the context, we talk about the problem, we then have had a little GPT go off and think about defining at every level, what are the words that would describe the iconography for this business. And then over time, using some models, not DALL-E in this case, to make a little icon, what should we call this, which we do based on a process whereby we figure out who are the heroes of the people of these customers? Who are some, some celebrities or scientists or whatever that these people might look up to. And then what would they call this business? 

And so we have all of these different processes that are running in the background. Obviously we had to pick a color palette ‘cause this one is browns and greens and whatever. And other ones are different colors. We've had these subagents go off and do these various jobs, we go through a user journey or attempt to kind of frame up what is the solution and then how would someone experience it? All the way through to kind of, what would you think about as being the key benefits and how much would you charge for that? How many people are out there that might want this? And so, in the background, a Google search has been run to go look for data and to source data. Obviously, we can then take how many people times how much much they would pay and at least start to think about the size of this opportunity. And then we create a to-do list, but all of which, again, yes, happens automatically and, as long as GPT-4 is not too backed up, it happens in five or six minutes.

Dan Shipper (00:23:57)

Can we just back up. I want to do a little bit of a dramatic reading of this. I just need to see what this is like, ‘cause this is kind of crazy. ‘Cause I spent so much time making these kinds of decks, as you know.

Nicholas Thorne (00:24:12)

As have I. You want me to pitch it to you? How about I pitch it to you?

Dan Shipper (00:24:15)

I want to pitch it. I want to pitch it. Because you’ve seen this already. I want to pitch it myself.

I just want to be clear about a couple of questions. There's a whole deck being created. So, okay. So, “The rise of the gig economy and the increasing cost of living have made home ownership a distant dream for many hardworking individuals.” That's actually not a bad start. Maybe we could have compressed it a little bit, but it's pretty good.

Nicholas Thorne (00:24:43)

It’s a bit wordy. A bit wordy.

Dan Shipper (00:24:45)

“Historically, employer match programs have been successful in helping employees save for retirement, but there's a glaring absence of such programs for home savings.” Yeah, I agree. I think this is a little bit speaker notes-y, and maybe have GPT compress it a little bit, but it's telling a story, which is, I think, one of the really hard things about making decks like these. “The primary problem is the difficulty in saving money for a home and the lack of financial support services, such as employer match programs specifically designed for home savings, which limit their ability to accumulate funds faster.” And then it has a couple of specific problems, like, “the high cost of living and unstable income from gig jobs, existing financial support systems are not designed to help individuals save for home ownership and it means people can't—”

Nicholas Thorne (00:25:43)

And I think that one key thing I put in here is that— and this is a conversation you and I think we've had—is what's the difference between a pitch and a plan? And in this case it's a little bit of a hybrid, right? ‘Cause we are playing back this idea to the entrepreneur. And so we are trying to kind of do justice, obviously, first of all, accelerate their thought process, take them out of the writer's block and how do I even frame this up? And second of all, because I think it's validating to hear your idea played back to you. And second of all, is to probably be a little wordier and a little err on the side of having more text than not, both so we have substrate to chip away at with them and also because it's probably more of a plan. It's more of something for them to look at and say, okay, yes, this is, this is kind of what I'm saying, but I agree with you that's not to defend the product but to contextualize it a little bit.

Dan Shipper (00:26:15)

That makes a lot of sense. So, the first couple of slides we've framed up the problem. Then we have a logo and a name. The thing that feels really magical to me about this is it's like seeing your idea come to life in a couple of questions is so cool, it must be so validating because I feel like there are so many people out there that have these ideas and have never seen them actually in the real world because they don't have the time or they don't have the skills or whatever. And even though this isn't the real product yet, just seeing a deck with your logo and your problem played out like this, I think is probably a huge moment for people.

Nicholas Thorne (00:26:58)

I think so. I mean, just to give you some stats—and stats are kind of useless until you feel them, but I think these are things we feel and so therefore I think the stats give better talking points. It's something like 60-to-65 percent of people who, if you ask them, have had a business idea, and it's the same number of people, and this is true both in the U.S. and I think globally who say that they have the resources and the experience and the knowledge to start a business, but 8 percent of people or something like that start a business in any given period. So you got a big gap. You basically have 60 percent of people who have had ideas and separately think that they have the resources, the knowledge, and the experience to do it. And then you have basically 90 percent lossiness and it's like, why is it lossy? Some portion of that, I think we can think is probably a good thing, but generally speaking, I think it has a lot to do with not knowing where to start having way too many things to potentially make progress on, such that you don't end up making progress on any of them. And so, to your point, I think if we can just start to play things back to people, put them in edit mode, not creation mode a little bit more often, it just allows the little early bits of momentum to build on themselves. And momentum is oxygen for these types of things. So you give a lot of chances that something might stick if you can get people out of their own heads and give them a little bit of a way of working.

Dan Shipper (00:28:36)

Yeah, that makes sense. So, okay. So I think I get the deck. I think this is awesome. Once you've built the deck—

Nicholas Thorne (00:28:40)

Yeah. What happens? What happens? So the first thing you can do is obviously you can play back into it. This person did not do that. I think he just said, okay, but just—

Dan Shipper (00:28:49)

Just okay? Give me a little kudos there, man!

Nicholas Thorne (00:28:51)

Well, he says, okay, but then I think he goes— Actually, sorry, no, he said, okay, I'm going to make a deck. And then he said, looks great, keep going, And so that's very, very, very common. People do have feedback that they'd like to give. And so let me just go find a link to this guy's feedback. If he had asked for it, the Copilot would have just gathered up a few of the key kinds of assets. One of the things we've also learned is trying to give you full— By the way, they can just duplicate that deck and then they can take it and make whatever edits. And sometimes the Copilot's a little cheeky and when it finds the comments to be a little bit too specific, it'll say, why don't you just go ahead and a duplicate copy of that, which, again, you and I, Dan, will appreciate from having written decks like these with a lot of different people.

Dan Shipper (00:29:46)

I wish I could have said that.

Nicholas Thorne (00:29:49)

Sometimes that's a very human and correct answer to the question, right? He could have said I’d like to edit this at which point he would've arrived on a little interface where he could have edited the icon, for example. So he might say, it needs to have a home, and at least it'll go off and create new ones for him, and he could rename it and he could pick new fonts and new colors and things like that. So if he wanted to say, this is the one I'd like then, or he could go do that yeah. It's obviously got something in it around kinds of mountains and things. And so he could do that. He can also go and say, yeah, that's not quite the way I want to pitch this. And so we have this kind of elevator pitch version of it, which is actually a lot of what was generated as the underpinning of the whole deck and so we do that. Anyway, so he could do that and then he'd make his saves and then he regenerates everything. From there, a couple of things will happen. One is he does make a small payment here, and then as soon as he's done that we do a few things. So the first is that we do make him a landing page.

Dan Shipper (00:31:20)

What is the small payment?

Nicholas Thorne (00:31:22)

He makes a $30 payment, in this case to basically move on to the next step

Dan Shipper (00:31:31)

Which is gathering customer feedback.

Nicholas Thorne (00:31:32)

Exactly right. It's a set of tools to allow you to start to talk to your customers. And so there's really two sides to that. The first is to set yourself up to talk to strangers. It's a little bit actually inverted in order, but we do it basically because I think this feels very real to people to have a website created. So in addition to having had his deck created, this is a website that was made for him where it took obviously a lot of the assets and the copy and other kinds of concepts from the slides and it made it a deck. By the way, in the next set of conversation, it'll ask him, okay, well, let's find a URL for a domain for this. And so it did go by this domain and register this domain for him, so that's kind of one part of it. The second part of it is that we will also—Well, actually there's two sides of that. So one is that then actually what he also gets is his own GPT. And so imagine that he can start acquiring leads just as we acquired him. He can now start to acquire leads into a conversational experience that will chat to his customers and can both kind of use a conversational model to qualify them much as we are doing with our customers, but also to begin to sell them on the different parts of the product and the experience that he's envisioning. And so the kind of website plus this hosted GPT that we've created for him, and then a set of assets that we create to help him go run an ad campaign, are part of how you get strangers. Before we do that, and again, it's inverted only in the presentation here because of two things—one is I happen to pick a demo that's a few weeks old, but second of all, it feels more like something you want to buy to buy these external facing assets. And then we back you into the hard work or a slightly harder bit of work that you need to do for yourself, which is that we ask him to go off and interview himself, basically to complete an interview from the perspective of the customer, and also to and then we also ask him to share this with a few of his friends. And so this is something that more orients towards collecting feedback from the people you already know who might be potential customers of this product before you go into the business of talking to strangers. And this is a voice-oriented chat that goes through basically the need-finding playbook that you might be taught if you go to the d.school at Stanford or something like that. And you're trying to kind of really get into, what's the problem? And so here I might say you know, “Yes, I fit that profile, I am trying to buy a new home,” and so this will initiate a conversation between me and the and the GPT, and then lastly, it will actually take every one of these conversations that the GPT has with my friends, let's call it, and it'll synthesize them into basically a product roadmap. So it'll use these conversations to help identify what are the key features of this potential product, and then to play them back to you and organize them in a dashboard.

Dan Shipper (00:35:21)

This is truly wild. I knew you had a lot of AI stuff, but this is a lot more than I expected. This is great. So there's this sort of like end-to-end process of, start with a couple questions, to a deck, to a website, to a little wizard that can help you like modify a couple of different things, to a customer interview thing that will get a bunch of interviews and then it will create a product roadmap, to like running ad campaigns. I mean, that's freaking crazy. I want to move into how this all works. How do you build this? How do you make the crazy mashup of models that makes this possible? Because I've built a lot of this stuff before and, it's really hard to get it to work well enough to do stuff like this. And I'm just super curious to understand how you're actually doing it. How is it architected?

Nicholas Thorne (00:36:20)

I hope you're not tired of all the AI, because unfortunately the answer to it is that we use a lot of AI to make it.

Dan Shipper (00:36:26)

Why did I know that was coming?

Nicholas Thorne (00:36:30)

It kind of came up in our pre-conversation that, as with many things, when you start talking to someone about them, or you hold them at a little bit of distance, what you realize about yourself is you asked, and I think your normal production doc is to talk about how do I ChatGPT? The way I ChatGPT is to make this. And I think it's probably relevant to say, hey, I've never taken a computer science class in my life. I don't know anything about computer science. Prior to ChatGPT, I never particularly programmed anything. Maybe I'd messed around in a Jupyter notebook and done some data science or whatever you want to call that, which is really just a glorified way of running longer Excel functions on bigger datasets than Excel could handle. So I was basically taking whatever financial learning I had and applying it to a bigger data set. So ChatGPT has been completely revolutionizing to me of being able to write code. And for programmers, I know that what I create is not per se code, but it does function, it does work, and it does deliver a result.

Dan Shipper (00:37:36)

That’s code.

Nicholas Thorne (00:37:37)

And therefore I've generated many, many more lines of code in the last year and a half than I'd ever generated in my whole life combined. And that's because I just think it's magical that you can say, this is what I kind of want to do. Can you write me the first version of a script and then you can plug that script in and run it and then just paste the error message and slowly, but surely if you're patient enough, you can get there. And so basically everything you're looking at so far absent the customer interviewing stuff was stuff that I prototyped by talking to ChatGPT, it's been productionized in a variety of ways, so it doesn't fully resemble. But, for sure, most of those processes, the system design, certainly, and some of even the scripting concepts are very much still there. And so I think maybe what we can do is just look at one of the new features that we're creating, or one of the things we're trying to prototype, which is to make the launching of those custom GPT. So you're a customer, you come in, we want to be able to have your AI copilot that you can use to talk to your customers as quickly as possible. Currently, it does still require some manual work, but I am in the process right now of prototyping that in my prototyping framework so that I can then hand it to the real engineers and have them once it's working and say, hey, can you please make this? And so if it makes sense, we can just go do that right now.

Dan Shipper (00:39:02)

And let's do that. I want to see it. Let's go.

Nicholas Thorne (00:39:03)

It’ll be a productive period for me. Okay, so, this is a very typical thing that I would spend a lot of time doing in in ChatGPT, which is, you'll see it here, I've basically just copied and pasted a script that I have running, in this case, in an Airtable where I do a lot of my work because, it's the most intuitive combination for me of a database that I can write to and read from, but tha, I can also provide anyone access to, and they have GUI. I just wish that Airtable would—if anyone from Airtable is listening, if you could please extend the timeout on a script beyond 30 seconds, I would no longer have to use Google Cloud functions. That would be great. Thank you. So I do stuff where I paste this whole thing and I just say, hey, this is what I want to change. Can you change that? And so you paste this whole thing in, and then, as you know, it'll actually go and write the response and fix it and do whatever. 

Dan Shipper (00:40:07)

Wait. Frame this up for me a little bit at the top. So, tell me about where—

Nicholas Thorne (00:40:12)

Yeah, I will. Let me just go into an error table quickly so I can give you even more context.

Alright. So the job that we need to do right now—we, you, me and Dupree—is to do the following, which is we are going to take— I have two jobs to do really one is that in order to do this, the way that we've spec'd out the system. So let's just talk to you and that we need to: one, update a DNS record for this person's website by API in order to have it point to a server where we're hosting these GPTs for folks. So that's the one part of the job is to create a new subdomain for them, which let's just say is going to be until they ask us otherwise GPT dot their domain.com. So we've bought the domain and registered it. It is on our registrar. We have API access to it. So one little job to do is to go off and call the DNSimple API to update the DNS records to take the subdomain and point it at our IP address. The second thing to do is for each one of these, we have to make an Airtable. We could do anything, but it's a database in this case that we provide to the entrepreneur so that they have an easy way of seeing, for example, who are the people that are coming and talking to my GPT? Where is that information being saved? Can I see the chat history? And Airtable is at least the place where we've decided to do that for the moment where, because it's easy if we can send them a link, they go in, they have their own, it's their own little proto-CRM, if you will, that's starting to be populated for them. But to do that, if we really want it to be automated, we have to automatically generate a new Airtable and it has to have columns that match to what the GPT instructions are in terms of how, and what information it's looking to collect. 

So let's say you and I went through this and we were creating one. Let's stick to our guy, Alec. He wants to collect information about, do you own a home or not? Have you saved at all for your home? If so, in what format? How much have you saved? Well, so to make it automatic, we have to have a GPT that has those instructions to look for that information. We have to have it have defined the schema of the database that it's going to write to. And we have to then create that database to match that schema so that it all functions properly. And so I don't know how to do any of those things other than if I ask ChatGPT to help me, you know? So what you are seeing there in this case is— So for example, what I was working on right there, when I pulled up my ChatGPT screen—let's just expand this a little bit—was a script that ChatGPT wrote, obviously, duh, wherein we define what we are going to create a new Airtable base. And so this is a database schema writing—let's just find the instructions because basically that will tell you what it does. I'm not even giving it instructions right now because I give it this very detailed function call, but this is just a bunch of code that says, this is we're going to make a table with an Airtable.

You got to give it a name. You have to give it to this. You have to give it to that. And then what you'll see here is that it responded when I tested this with a long thing that set up a new table for this concept Echelon with a table inside of the base called Leads and with a couple of different columns, they need a name and a phone number and these different things. And so what I was doing right there in ChatGPT was just, I needed to update it because we actually need two tables inside of our Airtable base. Let's go to the actual Airtable that we're going to make, I needed one of them to be called Leads where we're going to save that information. And then I have another one called Chat where we save just the raw content of the chat. And it's not doing the second part yet. And so what I asked it to do here is I said, I'd like to create the JSON schema to provide for the creation of a second table in each base that it's—and again, this to make no sense to you. I can barely even spit this out, but obviously ChatGPT is magical because it actually understands this and then gives me the correct answer, right? So I said, the second table should be called Chat and it should have two columns of the type single line Text, Chat ID, and Chat. And then I literally just pasted, I took this whole thing, I copied it, I pasted it in here. You can see that it's kind of like not nicely formatted, et cetera. And then OpenAI is nice enough to say, okay, to do that, we need to do the following and here's the modified portion of your script that would allow you to do this. And it's pretty close, by the way. I'm lazy enough to say, hey, instead of— Here, it said go ahead and reproduce the part that's already in here. My programming is sufficiently bad that I would be nervous to do this and that I mess up the spacing and things like that. And so I just ask it to redo it and it does the whole thing for me. And then I'll just go paste that back in here. And try it and it'll probably throw an error. And then I copy-paste the error back in here and probably within two backs and forths, we'll have a, have a correct answer. And then I move on.

Dan Shipper (00:45:51)

I mean, I love it. That's amazing. Okay. I'm kind of curious about, if you're not really totally understanding what's going on, how are you guiding yourself through whether ChatGPT is doing the right thing? Is it just sort of like you check it and if it seems to work it works? Or what are the—

Nicholas Thorne (00:46:07)

I think I've obviously gotten a slightly better sense of this now that I've done it a lot. I do try to read. I'm not gonna say that I'm trying to teach myself how to do these things. I don't see it ever being a productive use of my time to start with a blank sheet of paper or a blank sheet of screen and start writing to it, that's just not where I'm going to spend my time, but to the extent that I want something generated quickly, it feels like it's pretty good at it. And once you've made the mistake a few times you can tend to be pretty smart about what the mistake is that you're making and try not to make it again. And so in that case, for example, I know that if I took that code, which if I hadn't read it, I know for sure they try to excerpt something because it's trying to not generate too many tokens and whatever. And so it comments out a section and it's like, hey, just plug that section in there. And, well, I think I know what section you want in there, but I'm not positive. And it will take me more time to go do that than just to burn more tokens. So can you go do it? I think that's probably the main thing.

The second thing is it's so good— When I was preparing, my process is really probably a three-step process. One is to try to come up with what's the output that you want ChatGPT to create. So in this case, obviously it's code, but it could be when we were building the initial part of our system. Imagine we had to come up with a color palette for the slides, right? Or for the brand of this thing. Well, what is the color palette of a given new idea and how would you come to that? And so I'd spent a bunch of time with ChatGPT, just going back and forth. If I tell you this then can you propose a color palette? And then I'd have to say, oh, well, and lo and behold, it's pretty good at— Actually, you could say to ChatGPT, give me neon-volt yellow, all of the Nike brand in hex, no problem—give you the correct answer. Or something very close to it. So one of these amazing things where I have no idea what the hexadecimal notation of volt yellow is. I know how to pick it on a color picker, but I don't know how to. I would never generate off the top of my head. So it has no problem in this weird space between natural language and code or versions of machine-readable information. 

And so anyway, I would basically go through a process and say, hey, can you make a color palette? Oh, well, basically it needs to describe it first. And then I need to get it translated to hexadecimal form, notation, and then I need to get into a JSON object. And so finally I would arrive at this instruction set. That's basically your job is to take the following input and turn it into the following JSON object, which has a text-based description of the color palette, as well as three hexadecimal references up to like what the color is. And then I'm going to take that and I'm going to do something else with it, you know? And so then you would basically— Now let's say that you need to go make a logo with those colors. Well, now you have to call a new API. I'd often just copy and paste huge chunks of open API documentation from some random service that I found to do some job, right. And I would paste it in and I'd say, how do I format a call to this API if I know that I need to provide it these things and then it would do that? And then finally I would say, okay, now write me a script that I can run an Airtable for example, that would allow me to do that when something else happens, assume that I can take care of the automation and so we do a lot of that kind of stuff.

Dan Shipper (00:49:43)

That's really interesting. So I guess if I had to summarize how this is all working, Airtable is sort of a core thing where it's the source of truth. It's the database that people can interact with. You can interact with it, your customers can interact with it and it's the first building block. And then Airtable has these automations, these scripts that can run based on different actions that happen. So if someone adds a record or if someone, something changed, it can launch a script. And what you're doing is you're using ChatGPT to basically build whatever that automation is to program, okay someone fills out a form and it goes into the Airtable. We send them an email saying you need to pay or whatever. And ChatGPT is writing the script and then helping you connect to third-party services that are other building blocks that you might have like Stripe or like any of these other things. It’s sort of like—

Nicholas Thorne (00:50:42)

And that's certainly how we prototype things. And then obviously, if once something is being done enough times, then we'll move it into it and we run our main system now on Supabase, but it has a very similar set of features in terms of a DB with these edge functions, which are not dissimilar from the script running capabilities of Airtable. So Airtable to me is the lightweight place that either I as a novice or our customers can hang out and then we can productionize things over time.

Dan Shipper (00:51:10)

Right. Okay. No, I mean, that makes perfect sense and I think it's such an interesting model for people who are thinking about, okay, how do I get started with this stuff? How do I prototype stuff? It's like an Airtable with ChatGPT is sort of the glue that connects you to all these other third-party services that can be like— It's like Lego. You're stacking Lego bricks together and then you have something that functions. And once it functions, then you can be like, okay, maybe I should hand this off to a professional developer that can actually—

Nicholas Thorne (00:51:40)

Also I think it puts a lot of pressure on what the delivery model is. I mean, ChatGPT, I think, has done a great favor to entrepreneurs and people trying to build projects and certainly something that we are leveraging for our customers, which is that you and I know that the minute that you start trying to design UI and do wireframes of new services, the minute that a GUI is involved, I mean, you slow your pace of development down massively, right? On silly stuff—silly, but important stuff—like, oh, where's the forget password button? And what is your forget-password experience look like? And, oh, well actually maybe we should just do two-factor, so we don't even have to have a— Well, how does that work? What does the screen look like? Where does the button go to request a new code? You've just wasted days out of time and you haven't even gotten past the login and you're back and forth with stuff, you know? So I think what resonates with me about what you played back is, yeah, if you can get something functioning, you can also often make the hard decisions about, do I really need that GUI to present that? Or could I just put it in a text message to my client or whatever? And then is the service just something that kind of runs in the background for them, but delivers them this thing? And if it is something that I need to deliver, is there some really lightweight way for me to do it? So I think it is a really empowering thing to have this little combination of a database that you can play with in the form of Airtable. But also this has this kind of scripting layer on it, which ChatGPT is just so good at writing. These scripts, it's just logical that they exist all over the internet and it's pretty good at that.

Dan Shipper (00:53:21)

I love it. I love it. That makes a ton of sense, so I am obligated to ask this, but you don't have to answer it if you don't want to, but because it's on the screen, it's sort of my professional responsibility to ask you what “AI Copilot MLM Strategy” is.

Nicholas Thorne (00:53:38)

Oh, I was looking at, well, when we were talking about how to— Well, let me just find it. I mean, I don't know that we took it that far or we'll see, but that was a good example. I was trying to come up with, I think it's higher up in the thing, but I can't believe you didn't ask about “Panther vs. Grey Wolf.”

Dan Shipper (00:53:54)

That was coming next.

Nicholas Thorne (00:53:57)

I'm going to start with Panther vs. Gray Wolf because I'm going to try to make a slightly more meta point. This is obviously something I do with my son. My five-year-old son is obsessed with animals and we run this game ad infinitum and I am so grateful to ChatGPT for this, because I've saved a lot of money on these books that you can buy on Amazon that compare random animals, but that failed the next test, which is, obviously he reads the Great White Shark vs. Orca book that's been published. But then he's like, what if what if Orca fought a sperm whale and you're like, oh, well, they don't have that. And so ChatGPT is great at this. I have realized that, and in just preparing to talk to you, that a lot of the way that we just have discussed a lot is how I work with ChatGPT. A lot of the way I think with ChatGPT is to try to basically say, I'm thinking about these two things, and I'd like you to talk—I'd like to kind of go through a dialectic with you where we define them, or we list the pros and cons of them independently, and then where we crash them together and merge them and whatever. I think that honestly, it's in my own review, it's a way that ChatGPT, I feel, does a lot more for you than if you just go into this linear conversation where you say, I'm thinking about X, and, it almost has no frame of reference for what are the opposites. In fact, sometimes it'll ask you what are the alternatives? And so I just try to start with it also because if you give it to kind of almost seemingly very distant things it takes you seriously, it doesn't tend to say why would you ever compare AI and an MLM strategy. It tends to take it seriously. And so it tries to then do the work and it's work that often, I don't think I'd have the discipline to do alone, even if I could say, oh, well, I should really spend 30 minutes going through what's what makes for an MLM strategy and what are the characteristics of it and how would you merge that with this thought process and whatever. And so yeah, I mean, I think in this case, all the—

Dan Shipper (00:55:55)

Well, Panther vs. Grey Wolf, like what— I mean, we got to— 

Nicholas Thorne (00:55:57)

Oh, Panther vs. Grey Wolf. I think they're formidable predators. Let's see: “The outcome: Considering the panther's greater size, strength and lethal ambush tactics.” My son would be very happy about this. He's a feline over canine sympathist. I mean, this, in this case, I don't think I did the kind of dual model, although I can I, were we to go through this again, I would definitely do this because I have some alternatives in mind, but I was basically just trying to think about in our case, obviously we've had people who are trying to refer our product to other entrepreneurs in their network and asked us do we have referral codes and things like that. And so I was just trying to kind of like, imagine, are there any versions of an MLM that are not too mean or exploitative. And if so, what would be the way that you would think about that?

Dan Shipper (00:57:00)

Interesting. And ChatGPT is not going to judge you for asking that question, but someone else might.

Nicholas Thorne (00:57:04)

And in fact, again, to the point, take it quite seriously and try to go into these different mechanics and then I think you have the ability, therefore if you kind of take this in earnest to I actually, in this case, obviously didn't follow up. But what I would normally do here is say things like, okay, well, I'm really most interested in, let's just say you know I'm most interested in the downline concept and how that would work for me, I might even say things like, what does the typical downline agreement look like for a MLM participant and can we draft one? And I guess I do stuff like that where I just try to be way more result-oriented way faster than I would normally because it can often do such a good job of creating the thing. And then I find that that helps imagine the output.

Dan Shipper (00:58:12)

I love it. I'm just thinking about that video of Jim from the office and he's drawing the pyramid thing.

Nicholas Thorne (00:58:22)

I mean, I don't know. Suffice it to say, given that I didn't have a particularly in-depth conversation with ChatGPT, I'm not sure how far this thought process will go.

Dan Shipper (00:58:38)

No, I think it's really—

Nicholas Thorne (00:58:38)

At least now I have my downline agreement I can use if necessary.

Dan Shipper (00:58:40)

You know, if you need your first customer, let me know.

Nicholas Thorne (00:58:47)

My first mega. And just to be really— It's a full circle, we had a couple of MLMs issuing digital badges on our badge-issuing platform, they were the social media cred du jour in 2013 for women, in particular, who are interested in flexing their MLM muscle.

Dan Shipper (00:59:04)

I love that. Cool. Well, we're getting pretty close to to our time and I just want to kind of back up and talk about what we've talked about and leave people with a message, and also point them and direct them toward what you're doing—to Audos, to the book you're writing, all that kind of stuff. So, yeah. I mean, I think at a high level, what I'm observing is you have this you have this business that you've run for 10 years, helping entrepreneurs start companies. And then you have created a system that effectively scales this non-scalable business using 50 different AI tools that you have built yourself using AI, and you've basically abstracted yourself away from that process and created this thing that a person with a couple of chat messages can create a pitch deck, a landing page, can start recruiting customers, all that kind of stuff, which I think is totally wild and really interesting. I don't think I've ever seen someone use AI so extensively both in the sort of prototyping of the product and in the actual product that they make. Weaving together all these different microservices that you built I think is really cool. And for anyone who's wanting to build a product, this is a model, I think, especially if you're not technical, that is a really good way to get started. Yeah. What do you think? What do you think of that summary? Are there other things you want to leave people with?

Nicholas Thorne (01:00:34)

I think it's well summarized. I think AI can be a force for pushing entrepreneurship to new heights. And I guess I use the word entrepreneurship, and other people might use creativity or productivity or the different things, but I guess I use it as making something in service of someone else and I’ve lived that now for a couple of years, just from wanting to do more of what we do to serve our customers, but I think everyone we'll get a whiff of it. Not to be overly dramatic. I think we kind of need it. I think AI is bad for entrepreneurship, i.e. ones where people just use the big models and outsource parts of their life, et cetera. That's a very scary thing in my view, whereas, I'm kind of committed to this, but I think that it's possible it can be an amazing force multiplier for how people express themselves entrepreneurially speaking. And so if anyone, I guess, I'd say as a call to arms or whatever, is broadly interested in this stuff. I’m here and want to chat with anybody.

Dan Shipper (01:01:52)

No, I mean, I love that. I love this as a tool for creative expression for making things. And I think that's exactly what you're building for yourself and for other people. And I really, really, really love it. Where should people who are interested in connecting with you or using Audos or reading your book—where should they find you?

Nicholas Thorne (01:02:16)

Yeah, we're going to launch our book next month or so. That's Me, My Customer, and AI, in which most of any of this thought process will be kind of documented worst case scenario. I think I have a very— I can't say I post much on Twitter, but I will definitely post our book, so I'm thornny on Twitter. You go to Audos.com and it'll redirect you to our Instagram page. If you want to get started there, Dan's got my phone number. So if you've got Dan's phone number, he can give you mine. But I'm also [email protected] and I'm pretty good at doing that too. So whatever works and thank you for hanging out, Dan, as always. It's fun to share some of this stuff with you.

Dan Shipper (01:03:04)

This was a pleasure. Really, really fun to spend time together, and I'm excited to see where it goes.

Nicholas Thorne (01:03:10)

Me too. Every day there's something new. So that's the fun part for sure.

ChatGPT (01:03:23)

My gosh, folks, you absolutely positively have to smash that like button and subscribe to How Do You Use ChatGPT? Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure, unadulterated knowledge bombs about ChatGPT. Every episode is a rollercoaster of emotions, insights, and laughter that will leave you on the edge of your seat, craving for more. It's a journey into the future with Dan Shipper as the captain of the spaceship. So, do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now, without any further ado, let me just say, Dan, I'm absolutely, hopelessly in love with you.

Dan Shipper is the cofounder and CEO of Every, where he writes the Chain of Thought column and hosts the podcast How Do You Use ChatGPT? You can follow him on X at @danshipper and on LinkedIn, and Every on X at @every and on LinkedIn.