Things Paul Says on Podcasts

yeah so i programmed the rules of preludes and fugues into this box and i rode my bike around i listened to it for a long time some of them are pretty good they sounded kind of pretty they didn't kind of go anywhere and i went home you like keith jarrett right you love him

i love keith jarrett

it's playing out yeah like we can see it uh people are drifting back to other ways to connect of other ways to connect this is very cool it's an i think it's it's a it's an artifact of your journey which i think is what's most fascinating i brought it

into our internal product management tool which is called cadence the company's gonna eyeball it and then we'll put it out by the time this podcast is i

think people will play with it and i think oh i'd love to see people build on it i think it would be very cool

I think what is, here's what I've been getting at throughout, and I wasn't quite sure where I was going with this. The fantasy is of resolution.

Yeah.

never going to work.

It's never going to work. It can do other things, but it will never resolve that for you. And so I think that that is, that's something that's got to get baked into culture: that this is not really a tool for resolving problems.

No, it looks like a killer move, and it's actually not.

You could use a push mower just for that. And I think

that's where we are today. A lot of things we ask of AI.

Yeah.

Okay. Shows up, and it's from this company, TypeSafe. And it is, they call it a system one model. But here's what's different: you don't talk to it, you give it a set of options. You say, and the options are kind of yes, no, multiple choice, probabilities, things like that, numbers. And you say, okay, Jev, and you prompt it just like you would a LLM. You give it, you're like, hey, Jev, I got this document for you.

Yes.

like i just like our our ability to be discontented and miserable and greedy far exceeds our willingness Willingness to be comfortable with this is what makes you hopeful. Yes. Yeah.

Yeah.

I agree with this. Yeah,

that's that's what I see here. What I expect is near term, everyone's going to roll out their own version of these and so on. But what it did is it opened up a category of like, hey, wait a minute, this is a lot cheaper and faster. It does a lot of stuff I want. I'm a little tired of like, Claude 9.x is going to do the burberry. And you're like, I mean,

it seems like it's every month at this point.

No, it's not. And that part's really hard. And the only non-violent revolution way to deal with that is the legislative process. And that's slow and it is not aligned with this level of change, even when it works.

That's right.

And hey, listen, you know, you and I have lived through mobile. We've lived through the advent of the web and some other things. And we're still here.

We are. I think it's just so, it's hard to imagine. You know, also, I just kind of thought that we were done. You know, I looked at Bitcoin. I was like, all right, it's a nice swing for the fences there, fellas. NFTs. Yeah, exactly. NFTs, you're going to burn a lot of cycles. You'll try to create scarcity where there isn't. And I really did think it was kind of a little bit over, right? There wasn't any new revolution in tech that was on the way. And blockchain stuff just felt like such a swing and a mist to me from a tech point of view, maybe not from people from an economic point of view, but from a tech point of view, I'm like, you're wasting beautiful CPU cycles. Don't do it. They're too valuable. You could be making pictures. And now we do have something that truly is eating all the CPU cycles it could, but it's doing all sorts of new things. So I don't know. I really, I didn't see it coming.

Well, all right. We got to wrap up, but I am, you've already given us your gift, so that's good. Even though it was earlier, I don't hold it against you too much. I'm a bad boy. Bad boy. I'm going to put you on the spot talking about language. Is there a word that kind of sums up how you feel about where we are at this point when it comes to AI? One word.

What if we spend another $2 million on marketing directly to doctors? And it'll kind of move the sliders around in its weird brain because it's just running a program. The outputs show up and it's like, here's what I think. And you can kind of get it to narrate that back to you. But you also get the stats. And so to me, these are ways of using this stuff where you never, I don't want to get far away from the process, the thinking process. It's my thinking process, not its. But boy, is it nice. I have it summarize the news every morning. And sometimes I have it just make up a weird product idea, run the business model and report back on how it did. It's never something I want to build. It's never good. But it's a way for me to keep thinking about the client and about their business, right? So there's ways to mess with this stuff where I think you get to move forward and you get to think your thoughts without letting it kind of take over. But

there is a rule in there, right? The rule is you are not looking for it to do shortcuts for you. You're looking for it to be productive for you, but you're still doing the primary thinking. You're looking at it to create framework. I mean, a business model is just a framework like it is a program. Those are things that you've selected if you haven't come up with yourself. But I think most of us, I'll put my hand up here as guilty as charge is we are doing the shortcuts. We are okay with mediocre when, I mean, like, all right, I'll put it to you this way. There's, I do a lot of writing and I hate doing a lot of that writing. It's boring, it's tedious, and it's throwaway. And I use Gen AI to bring shitty shards of idea and information up to a mediocre average level. I wouldn't do that with important writing, but man, I'm so happy to do it with stuff that would take me an hour. Now takes me five minutes. Great. Is that the kind of shortcut that we should be avoiding because there's the slippery slope? Or what you think?

can't believe them.

Yeah, you can't, right? And so, like, and it just kind of is a sucky thing about life that you have to kind of always keep a kind of neutral affect and be like, well, maybe it is good, but we're going to have to verify that, right? You have a successful publishing company because you have a metric for what is good. And somebody might come up to you and be like, it's the best book on XYZ I ever read, but you might know that it didn't sell as well as something else. And you have to kind of factor that in as you're processing all that information. You got this thing telling you you're smart all day long. And smart people want to hear that all day long. It's very dangerous. That part to me, having it generate, having it fill in a blank page that you kind of don't want to deal with. And then you go and you're like, all right, this is okay. Let's get it over the line. You know, talking to you guys do you publish a lot of stuff, like talking to the companies that print, right? Like having it like just sort of draft an email for you. I don't give

me a quote. Give

This is a tricky one. First of all, I mean, AI as we mostly experience is made of language. So that's hard too, because it tends to dominate every conversation. You can't talk to anybody in technology now without talking about AI. Like it's just impossible. And so part of my last year has just been to kind of give up and just be like, all right, this is it. Like I fought for a little bit. I was like, I'm not going to let it write for me. And I don't. But like, I just had like a set of rules. And then Claude Code showed up. And I'm like, oh, boy, this is a little bit game over. You're not going to put, it's not that I lean into it and I go like, oh my God, here it is. We all have to get on board. Just, I just have a good sense. I think you do too. There's certain things that just won't go back in the box. And when Claude Code showed up, I was like, that won't be going back in the box. There's just no way that you can make coding radically easier in a world where coding is worth trillions of dollars and touches every aspect of society and people are going to be able to reject that. Like it's just going to go everywhere. And I think we're seeing that, right? You just, whether you like it or not or want it or not or whatever, this is an inescapable technology. So

what were the rules that you had until Claude Code?

better than you, but are you using it mostly for prototyping? Are you actually creating? No, I'm

creating large complicated systems that, and, and so, you know, there's, it's all the classic rules. Like you limit your, your exposure to, uh, uh, you know, where can data come in? Where can what do, you know, what are sort of the post endpoints? Where can people log in? Where is authentication necessary? I think one good threshold would be I would have no problem releasing something vibe-coded that is essentially read-only, right? Like it doesn't have personal information. It's reading from a database of facts or articles or statements. Go ahead and throw that up because the worst thing that will happen is they'll just hack you and use your computer, right? They're not going to be able to get anybody's information. The minute personal information comes in, then it gets really serious. At the same time, like, you know, I'm never going to implement, just because of the way the web worked for the last 10 years, I'm never going to implement auth again. I'm going to use an external law server. I'm never going to implement payments again. I'm going to use Stripe. So the surface is actually, you can really narrow your risk when you're working with this stuff. And then you kind of can let it go and see what it comes up with.

So what about agents in that whole risk scenario?

he hit a great line. And the line was, you can't have art without resistance in the materials. There it is. And you cannot have creativity without constraints. And it's not just that the constraints kind of give you a to-do list. The form and the structure of things are how people understand what you're doing. If I suddenly start singing a song right now, that would seem bananas. But if I said, Rich Ziade been composing funny little songs. I want you to hear a couple of them. You'd be like, all right, I'll evaluate based on what I think a song should sound like. Software was so expensive that we had zillions of processes to manage the costs because failure was a normal and natural constraint.

Yes.

Yes.

And I think that right now, for all the power of these tools, they are not optimized for building that.

We've

So the pre-Hobbesian era means that... Okay, and we talk a lot about regulations and rules for AI. But this is a little different. This is like, wait a minute, everyone has been given this new power and there is no set of rules and guidance for it. And so they're kind of maximizing their own influence, their own ability to show off and impress others and take control and kind of gain a little territory using these tools. And look, most of us live in a society and we're doing it at work and we're saying, look, I made a PowerPoint deck in five minutes. Aren't I a good person? Boss is saying, hey, I want you to do more and more of that. And so we're in this state where we are just kind of manufacturing all this content that looks like the right stuff because that's going to make everybody think that we're really smart. And I do, I want to describe something that I see a lot of. And it's terrible because I decided to go in super, super deep on this stuff. And I vibe code all the time and I really wanted to understand it. I've written about it. I've talked about it. And there is this moment where you make something happen and you go, oh, I cracked it because it just used to take so long. And what it's meant is that whenever anybody demos me anything they built with AI, I can't be impressed.

You can't be.

that's right. And

then it's hard because they're actually, people are really hurt because you're essentially telling them their baby's ugly, even though it was a baby they gave birth to in about six hours instead of nine months. Yes. So anyway, we're in this pre-Hobbian state.

Yes, we are. There's a few things that have happened here. One, we're producing too much stuff. We had a prospect come to us, say, I have the requirements. I just need you to build it. She gave us, I kid you not, a dense 80-page requirements document. Clearly, she didn't write it. It was so complex and vast that no one read it. I told her, we're not going to read this. It's too much stuff.

And then while I'm working on that visualization, I realized that I could abstract it into a library so I could reuse it. And then you know what becomes the most important thing in my life?

Abstracting to the library.

No.

No one has ever said those words. No, let's

go back to the valley.

but enormous numbers of careers are downstream of it. Correct. What is it? Like two-thirds. If you're going to spend money on enterprise software, like you spend a dollar, you're going to spend $2 kind of supporting and maintaining and installing around that in the

tool. I gave issue one, which is building great software alone is really hard because there's a lot of different skills and discipline that are required and very few people have them all. I mean,

Bingo. But there's another problem, especially because the AI companies, because they're under enormous pressure to meet the X value expectations that they've put out in the world, they know they have to go to the next frontier, which is business software. They have to. Because there's just not enough money.

This is why you hear about... So here's what's happened. The company Palantir, which is very AI adjacent. Adorable company. Oh, boy. Let's not talk about... Let's just leave it.

Cuties. Let's leave

And you sent Grady a pizza, he finished it off for you, and you're live. That's incredible. But the differentiating value of that initiative, the truth is, the tools are incredible, but the tools are extremely derivative, extremely opinionated, and can't help but lean on the other candle stores. And if you are lazy at the outset, when the idea is being conceived, you're going to end up doing work later to somehow get money. You're going to end up buying ads. You're going to end up doing other stuff, which is the problem. So,

nobody wants to hear the trillion-dollar companies aren't going to get big by telling you, hey, it's still a lot of work.

it'll throw you some serendipity.

A little bit. And even the serendipity. Explain to the audience why it makes candle stores that look a lot like other candle stores.

It has metabolized all the candle stores in the world. Great explanation. It has eaten them all up and it's digested them. And now it is ready to poop out a candle store at any time.

A little bit. And even the serendipity. Explain to the audience why it makes candle stores that look a lot like other candle stores.

It has metabolized all the candle stores in the world. Great explanation. It has eaten them all up and it's digested them. And now it is ready to poop out a candle store at any time.

there's a way to work with this stuff where you can say, okay, we're going to start with the baseline of candle store. But I wanted to do, I want people to be able to come in and literally describe their moods by picking, you know, colors and or tell me how old they are. Yeah. And then I will have it do. Like, I could come up with a whole product and experience around your candle store.

Yes.

At that point, and I've been doing this, I've been doing stuff like this. Yep. It is absolutely indistinguishable from classic product work and development. It's not.