Patrick Collison on Stripe’s Early Choices, Smalltalk, and What Comes After Coding

Michael Truell, CEO of Cursor, sits down with Patrick Collison, CEO of Stripe and an investor in Anysphere, to talk about Collison's history with Smalltalk and Lisp, the MongoDB and Ruby decisions Stripe still lives with 15 years later, why he'd spend even more time on API design if he could do it over, and whether AI is actually showing up in economic productivity data. This episode originally aired on Cursor's podcast.

• 0:00Speaker 00

  It's interesting to me that we haven't experimented in some sense that much with the paradigm of programming over the past 20 years. Yes. You put those together, you now have the ability to, again, at the kind of level of the individual cell, to read, think, and to write. And this starts to really feel like a new kind of Turing loop and to have its own sort of completeness. I think that's a case where the right API design, the right abstraction design, ended up having just quite significant business ramifications. I think the basic idea of as development environment and not just text editor is really the right idea, and that's the thing I want to see a return to.

• 0:41Speaker 01

  Patrick Collison wrote his first startup in small talk. Its development environment let him fix errors mid-request, inspect stack frames, and resume execution. And he wanted that more than he wanted a mainstream language. He and his brother chose Ruby and MongoDB for Stripe instead. Those decisions still defined the company 15 years and 44 seconds of annual downtime later. Now Stripe is shipping V2 APIs, rewriting core abstractions first designed in 2010. It's taken years. Defining the new APIs is the easy part. Making them work alongside everything already built on the old ones is, as Collison put it, more like an instruction set migration than a product launch. This conversation previously aired on Cursor's podcast also gets into why AI hasn't moved

• 1:25Speaker 01

  productivity numbers, what today's dev environment could steal from Lisp machines, and Collison's work at Arc on foundational models for biology. Michael Truell, CEO of Cursor, sits down with Patrick Collison, CEO of Stripe.

• 1:38Speaker 00

  Well, it's great to have you. Thanks for

• 1:42Speaker 02

  Thank you for being here.

• 1:43Speaker 00

  Thanks for having me.

• 1:44Speaker 02

  Yes. I've heard that your first startup was written in small talk. Please explain.

• 1:48Speaker 00

  I don't know what there is to explain. It's the best programming language. Well, I had worked on Lisp and Lisp dialects before that, and actually I worked on Lisp web frameworks. And when we went to build our first startup, we first wrote it in, we first implemented it in Rails, and then I found, compared to Lisp, that development process kind of frustrating. And, I mean, we don't need to get into the full details, but I thought that continuation-based web frameworks were really the right way to implement web applications. There were no continuations in

• 2:25Speaker 00

  There's no continuation-based framework in Ruby. And I was kind of searching around. I found that there was a good one that had just been written in Smalltalk. And so I decided to play with it a little bit. And then I found that Smalltalk is actually this extremely interesting development environment that had a lot of the aspects of Lisp that I'd really appreciated there, like, you know, and a fully interactive environment with a proper debugger, so that you can edit the code while, you know, in the middle of some web request or, you know, deep in some stack trace or something. And, you know, you could, for example, encounter an error with some web request, edit the

• 3:06Speaker 00

  code to fix the error, and then resume higher up in the stack such that the entire web request would just complete. And so rather than this kind of annoying feedback loop of having to add some log statements and like do this binary search to like find the problem and eventually like deploy, you know, a fixed version, you know, a process that could take an hour, you could just like literally inspect the stack frame, see which variable has the wrong value, fix it, like, you know, jump back up, hit proceed and have the whole thing work. So anyway, the point is, in the hunt for this continuation-based web framework, I realized that Smalltalk in general had just a much more powerful development environment as compared to Ruby slash as compared to basically every other mainstream programming language.

• 3:46Speaker 00

  And so we decided to, yeah, use it for the company, which in hindsight was, I mean, I don't know if it was a terrible decision or not. The reason I think one would think it would be terrible is that it would be, you know, hard to hire people and hard to scale and, you know, whatever. It wasn't hard to hire people, or rather, nobody knew it, but it was easy to teach them.

• 4:05Speaker 02

  Did they know before they joined?

• 4:07Speaker 00

  No, no, but they learned really quickly. And then you have smart people learn languages really quickly. So I don't think that's really a reason not to use a non-mainstream language. The company didn't work, I think, for unrelated reasons. I think just the idea wasn't that strong. But we also chose Ruby for Stripe, so, I don't know, I think maybe the gains were not quite as large as I hoped.

• 4:25Speaker 02

  And was your Smalltalk enthusiasm shared by the acquirers of the startup? And what was the dynamic, you know, was there like this blissfully ignorant management that foisted this Smalltalk codebase on a bunch of unsuspecting developers that were then kind of like piling over, you know? Or yeah, what was the dynamic between the programmers' management sort of, what happened to that Smalltalk codebase?

• 4:45Speaker 00

  Yeah, yeah, yeah.

• 4:45Speaker 02

  Does it still live on somewhere?

• 4:47Speaker 00

  I wish, and I'm 99% sure the answer is no. It, the company that acquired us was mainly a talent acquisition, so the codebase itself was last rolled.

• 5:00Speaker 02

  Okay, and it was immediately just gone?

• 5:02Speaker 00

  Yeah.

• 5:02Speaker 02

  Okay, gotcha. I've also heard that one of your earliest programming projects was working on an AI bot written in Lisp.

• 5:10Speaker 00

  True.

• 5:10Speaker 02

  And it was something like, it was a client for MSN?

• 5:15Speaker 00

  I don't know where you found that, but that is true.

• 5:17Speaker 02

  And I heard that you got kind of nerded tonight by the idea of trying to get it to pass the Turing test.

• 5:21Speaker 00

  Yes.

• 5:22Speaker 02

  And I'm curious, what did you miss? Why didn't you make ChatGPT? And, well, maybe a little bit more seriously, how did it work? And what was the state of neural networks at the time? And did you consider using any antecedents to the technology we use today?

• 5:36Speaker 00

  Yeah, so that was the project. It was a little critter that used MSN Messenger, which was all the rage at the time. I guess that's like maybe a specific kind of sedimentary layer in the chronology of different instant messaging solutions and probably dates me quite precisely.

• 5:52Speaker 00

  And it was a really simple Bayesian next word predictor, like there was nothing really that sophisticated there. To the extent there was anything sophisticated was maybe that it used, like the training data was the conversations itself had an MSN Messenger rather than kind of general text corpus. And it worked reasonably well and better versions looked a couple of words ahead and what have you. And, I mean, it never really passed the Turing test where people have actual suspicion. They're trying to exercise this discernment, but it certainly passed some weaker versions of the Turing test where they were unsuspecting and people ended up having quite lengthy conversations with it. And that was part of how I discovered Lisp. And I remember Paradigms of AI Programming by Peter Norvig being a really formative

• 6:36Speaker 00

  book and had all sorts of interesting approaches there. It didn't have anything on neural networks, I'm almost sure. And I never, I mean, I'd read some Marvin Minsky stuff, Society of the Mind or whatever, on neural nets, but I never really seriously looked at them. And I actually experimented a lot with genetic algorithms. They were, I guess, more practical on your own computer. Like, it takes a lot of computer training in neural nets. So I experimented a lot with genetic algorithms. And actually, I used Vorjak at the keyboard layout because it's more comfortable to type on than QWERTY. But, as does John, my...

• 7:11Speaker 00

  My brother, so no one can ever use our computers. But I wrote a genetic optimizer to figure out what the optimal keyboard layout was. And it turns out, it is in fact, basically Vorzak using a genetic approach. So I went deep down that rabbit hole, but I never really played with neural networks. And I guess that's why, you know, that, but probably 70 other reasons is why, you know, I did not create ChatGPT.

• 7:38Speaker 02

  There is an old video of you being interviewed, I think after selling Automattic,

• 7:45Speaker 02

  where you're asked about Smalltalk. That's where I found kind of that weird fact. I think at the time, people asked you why. And one of the things you said was, I mean, you liked some features about Smalltalk, list-style languages. And you predicted, and I think that this was circa maybe 2008 or something like that, that the mainline C-style programming languages would increasingly borrow ideas from these older programming languages. And that kind of has been the case in the JavaScript, Python ecosystems.

• 8:08Speaker 00

  Yeah.

• 8:08Speaker 02

  Do you think that there are any underrated ideas buried away in kind of older, more esoteric programming languages that should be borrowed by the mainline?

• 8:16Speaker 00

  Yeah, it's been interesting how a lot of the ideas have been kind of, have been borrowed by the JavaScript ecosystem and in a strange way, like through the web inspector.

• 8:25Speaker 02

  Yeah.

• 8:26Speaker 00

  Where you have this, I mean, that's one of the richest runtimes in some sense that people have, you know, general exposure to. I don't think JavaScript has first-class stack frames. Maybe there's some weird extension or something where you can get that. But ECMAScript doesn't have that, I'm pretty sure. First-class stack frames actually let you do a lot of other things for kind of obvious reasons. So maybe that's very, that's kind of too specific. I mean, I think the idea of, and maybe this would cursor becomes. I think the basic idea of as development environment and not just text editor is really the right idea. And that's the thing I want to see a return to. That's the thing that the list machines had and genera. That's the thing that to some extent Mathematica has.

• 9:11Speaker 00

  That's the thing that Smalltalk has. And I think it's just such a mistake that we have ended up with development environments where there is such a separation between the runtime and the text editing and the,

• 9:24Speaker 00

  and the environment in which the code, I mean, well.

• 9:29Speaker 00

  The runtime and the place where the code runs can be the same or different, but there are kind of three maybe slightly deconceptionally different things. And in those three environments, they can all coexist in the same place. And I find, like, I mean, still to this day, I use Mathematica a lot, not because I'm doing some particularly arcane symbolic mathematics, but because it's just a more efficient development environment. Now, that's maybe a bit less true with LLMs because the Mathematica, you know, Mathematica does not support cursor-style prompted development. But that, I think, is the core idea that I wish others would borrow. And VS Code has been a step to some extent, slightly in that direction, but I think we could take it way further. And it'd be really, I mean, what I'd love to see, for example, is when I hover over

• 10:13Speaker 00

  a line of code, I would like to see, you know, profiling information about, you know, just the runtime characteristics of that code or that function or whatever. I would like to see logging and error information overlaid. When I hover over a variable, I would like to see how, like, the most common values that it takes on in production. These kinds of, like, just rich, deep integrations. Are you a fan of inventing on principle and those talks? Yes, yes. I think Brett leans too much, I mean, I'm a huge fan of Brett. Just, he's such an incredible... Have you been to Dynamicland? Yes. Okay. Yep. And I've supported it, so huge fan of Brett.

• 10:55Speaker 00

  The place that I've maybe differed, or at least that just resonates with me, you know, somewhat less, is Brett is really into this idea of, obviously, of graphical and visual representations for phenomena. And I think that works very well in certain domains, like the kinds of dynamical systems that, you know, he has demonstrated some of the ideas with. I think it's often very hard to find such useful spatial continuous representations for arbitrary systems. Like for various parts of Stripe, I'm not quite sure what that would be. And I'm not sure, even if we could find it, you know, exactly how useful it would be.

• 11:35Speaker 00

  Maybe it's just me, I reason much more kind of symbolically and sort of lexically than I do visually and graphically. It might just be personal preference, but I don't know. The kind of paradigm breaking that he's been engaged in, I think, is hugely admirable. Are you going to make a true integrated development environment?

• 11:53Speaker 02

  So we are playing with ideas around letting the AI increasingly take time into the background to run its code and react to the output. And we think that this should all work well together. Like, you know, we focused a ton on inflow, speed and control. And we think that that's really, really important for AI is, you know, to give programmers the control over everything, have them understand everything the AI is producing. Also to give them really, really fast iteration loops. Programmers hate waiting for things. But in some cases, we think it's now becoming possible to go tell the AI to think for a bit and then come back to you and have the API be a little bit more like the API of another human being. And we think you want that all to work well together. So, you know, the AI can come back to you with 70% of something and then you can bring it into the foreground really quickly, work with it,

• 12:37Speaker 02

  and then spin it back off to the background. And, you know, as part of having the AI spend a bunch of time thinking in the background to make that thinking useful, you kind of needed to run the code and then react to it. Or else it's just kind of staring at the thing that it wrote and thinking more.

• 12:51Speaker 00

  Maybe I'm supposed to be the one answering the questions rather than asking them. But do you think in five years, the main thing that I'm looking at in Cursor will be code or something else?

• 13:05Speaker 02

  I think it might be something else. I think that there are Big, big, big simplification, but when you're defining what a piece of software is, there's the logic component, which is what engineers spend a lot of time on, designing exactly how the software works. There's also for end-user applications and things that have GUIs, there's this visual component. And I think that there is, maybe it's going to be us, maybe it's going to be someone else, there is a future version of the world where the way you interact with AI is a little bit less like it's a human helper that you're delegating work to or looking over your shoulder, predicting the next set of things you're going to do. And instead, it's a little bit more of an advanced compiler or interpreter technology. And it could lead you to a world where programming languages actually change. And they can start to get a little bit less formal, they can start to get a little bit

• 13:49Speaker 02

  higher level, they can start to be a little bit more about what you want and a little bit less about how you do it. And I think that it won't look like a Google Doc necessarily. I think that there are things you want to keep around from programming, like the naming of not logic somewhere and then using that in a bunch of other places. I think that there's also this other element too of the visuals of what a piece of software looks like. And I think, maybe us, maybe some other tool, but I think there's a world where direct manipulation of the UI starts to play a little bit more into it. But these are far-flung experimental ideas.

• 14:22Speaker 00

  In general, I will say, and it's not terrible, but I feel like there, it's interesting to me that we haven't experimented in some sense that much with the paradigm of programming over the past 20 years. And the main things we're discussing here are from the 80s or the 70s. And there are way more developers, obviously, now than there ever have in the past. But in some sense, the aperture of experimentation there feels like it's really not that wide. And again, the JavaScript ecosystem and a couple of others have done some cool things. And there's a lot of experimentation at the language level with Rust and Go and everything else. But at the development environment level, I don't know why it is, but maybe it's just

• 15:07Speaker 00

  too hard and complicated now, but there's been less than I would have expected.

• 15:09Speaker 02

  Yeah, I agree.

• 15:13Speaker 02

  I think maybe something we're working on.

• 15:16Speaker 00

  Maybe this explains Cursor's success to some extent for you know you guys are the first people to really take it seriously in quite a while.

• 15:23Speaker 02

  Well, I mean, yeah, I think we also benefit a lot from the why now of like there's now this, you know, great new color to paint with. Or set of colors to paint with. I think also there's just a ton of lock in with programming languages around both the neurons in your head of like. Programming languages are kind of complex UI for programmers to define exactly how the computer should function. And so, you know, people learn languages and those, you know, people don't like to learn that many things. And then there's also the lock in of you have a lot of logic sitting around in one language and you need to maintain that. And I actually think that that's a pretty interesting, or one of our hopes is that as AI programming gets better and better and better. One of the downsides of working on professional applications with hundreds of people dealing with many millions of lines of logic is the weight of the code base really starts to weigh on you.

• 16:06Speaker 00

  Yeah.

• 16:07Speaker 02

  And so the the feeling of being in a net new code base where it's just everything feels effortless goes away, everything's, you know, a chore, you have to. You know, change one thing here, breaks, you know, something else here and it becomes kind of this big ball of mud. And making that effortless, reducing the kind of weight of an existing set of logic, I think is, you know, one of the areas in which AI can, you know, make programming better.

• 16:28Speaker 00

  Someone said on Twitter today.

• 16:31Speaker 00

  Maybe it was Andre Karpathy, but maybe I'm misattributing that and, you know, made too many things. To do with vibe coding, get attributed to Andre, like, you know, to Churchill or Einstein or something, but I think that's. But this person, whoever it was, was making observation that, you know, it's one thing to be prompting the creation of code. But another place where AI could conceivably do a lot to help is in the beautification and the refactoring of code bases. And you can imagine that, you know, you're producing all this, you know, a little bit ungainly, not quite correctly factored, you know, detritus at the front. And you have this, you know, and then nocturnally, this thing comes up behind you and makes it all, you know, beautifully factored.

• 17:13Speaker 00

  And the only CS class I ever took was this from this class from Jerry Sussman on on. He was basically focused on, I mean, he called it large-scale symbolic systems. But really, what he was trying to focus on was the idea of creating code bases and environments and abstractions that were easy to modify. And like, there were no assignments in the class where you had to write something from scratch. Every assignment was about modifying an existing system and thinking about how could you design things in such a way that those modifications become, and there might be quite deep modifications, become straightforward. And I think that's a lovely idea. Obviously, in practice, it's often very difficult to do that, given all the exigencies and pressures of the things you want to ship, you know, today and next week and so forth.

• 17:56Speaker 00

  But if you could have an AI, like often when you're writing this stuff, you realize, well, I really should be doing it the beautiful way, but I'm not.

• 18:01Speaker 02

  Yes.

• 18:02Speaker 00

  Maybe we could have an AI coming up behind us to clean it up.

• 18:04Speaker 02

  Yes, yes, maybe soon. One thing that happens to a lot of developers that care, or a lot of people come to development because they care about building things. They want to make things happen on the computer screen. And so then that leads them to coding. And then something that happens to, you know, a big group of developers is they eventually realize the software they want to create is too big that they can't write, you know, all of the code themselves, and they have to go to humans to help them write the code. And so maybe they then become an engineering manager, a director, or whatever it is. Maybe they start a company, right? And then most of the work becomes not typing code. It becomes, you know, coordinating amongst people. Do you think that there are any ideas from programming that are helpful for the act of kind of programming amongst the organization to get a group of people to build software together?

• 18:46Speaker 00

  Interesting. I think taking APIs and data models really seriously. If I was to do everything that Stripe began, I mean, there's a million small things that you would do differently, and even some kind of big things. But the thing that I think we could maybe foreseeably and beneficially done differently would be to have spent even more time than we did on APIs and data models. And

• 19:17Speaker 00

  Part of the reason is the, I guess, Conway's Law effect of how both of those things end up shaping the organization. So I guess if you don't deeply internalize that, then maybe you've less control over the organizational dynamics than you might otherwise like to have. But also, I think it ends up shaping not only... I mean, the weak version of Conway's Law is that it shapes your organization. I think the strong version is that it substantially shapes your strategy and just your business outcomes. And this isn't exactly maybe a version of that, but I often reflect on how the iOS

• 19:56Speaker 00

  software ecosystem for a very long time, and plausibly still today, was so much more vibrant and kind of vital and successful than the Android app ecosystem. And there's a lot of things that are different across those two ecosystems. There are now way more Android devices in use, I believe, than iOS devices. But I think much of the fact that app developers tended to prefer building their apps on iOS and releasing their apps first on iOS and maybe the iOS version being better than the Android version or whatever is because the frameworks and the abstractions for iOS were just originally better than

• 20:41Speaker 00

  the Android ones. But I think that's a case where the right API design, the right abstraction design, ended up having just quite significant business ramifications. And I think there's kind of a sense that maybe it's not worth dwelling on these things because everything in technology changes so rapidly and whatever assumptions you make, they'll be obsolete in two years or something. I think in practice, that's not true. And that's like the right API design and the right abstraction, the right data models can really endure. And for the first versions of iOS, many of the classes that one used were prefixed with NS, NS, of course, standing for next step, right? And so that's a case where the API design survived for two decades or more. And in the case of Stripe, Stripe is now

• 21:23Speaker 00

  15 years old, and there were lots of things that we designed 15 years ago that are still in use today, which is kind of good and bad in the sense that they endured, but also we are still...

• 21:35Speaker 00

  We are still under the

• 21:37Speaker 02

  Living with their faults.

• 21:38Speaker 00

  Exactly. And so, anyway, that's maybe the thing that I would that's the first thing that comes to mind.

• 21:44Speaker 02

  In fact, on that final note, I was talking with an engineering leader at kind of a preeminent, successful Silicon Valley private company. And they were talking about how their code base is largely in Scala. And they said that they like to think of kind of the beginnings of the startup as this big bang moment where these tired, overworked, maybe over-caffeinated founding team members are willy-nilly making these initial technical decisions that then dictate the lives of hundreds of professional engineers in the future. And that Scala choice was one of them. And they sort of live with the faults of that now. But what are those kind of what were the consequential, it could be good or bad, initial conditions of the Stripe Big Bang

• 22:29Speaker 02

  that you guys still live with right now?

• 22:31Speaker 00

  I mean, I think that metaphor is well, it sounds true to me, is the first thing I'd say. I mean, maybe there's a little bit of survivorship bias, where like the actual statement is the early decisions that we made that we never changed are decisions that we lived with. But there's a kind of tautology there or something. And there are certainly design decisions we made pretty early on that are not true today. So early versions of the Stripe dashboard or something were built extraordinarily differently to the dashboard of today. And the converse is also true. So initially, we decided to use MongoDB at Stripe and decided to use Ruby at Stripe.

• 23:15Speaker 00

  And those are still quite foundational technologies at Stripe. And we had to build a lot of infrastructure in order to make MongoDB as fault tolerant and as distributed and as durable and as reliable and everything as We needed it to be and as it now is. Like we had Stripe's critical API availability last year was 99.99986%, which is 44 seconds of unavailability through the whole year. Which is, we thin-- and others don't publish statistics that are kind of granular, but we believe that is the best in the industry. So, you know, everything that our storage team has built and many other teams,

• 24:00Speaker 00

  you know, it ended up really working there. But that was a quite important critical decision, initial decision. And Ruby, similarly. I guess companies sometimes change languages along the way. But I feel like the initial language chosen tends to have a--

• 24:16Speaker 02

  I heard there were debates in Stripe about, or one of, actually, one of our co-founders, internet Stripe, early on. Or not early on in Stripe's history, early on in kind of our collective personal history. And he remembers there being documents upon documents about a potential job of migration.

• 24:34Speaker 00

  Yeah, so we, that partly happened. As in, we have rewritten a bunch of key services on Java. So some services for which throughput, throughput in particular is really important. And if you torture Ruby enough and maybe rewrite parts of it, parts of some hot paths in C or something, you can get it to be pretty fast. But you're often fighting against the allocator and, you know, various parts of even just like Ruby strings are not that efficient and stuff. So we've rewritten certain services in Java and now we use both.

• 25:13Speaker 02

  Why did you, or did you consider anything other than Mongo? And why did you pick Mongo early on? And what was the RFC process, RFP process, decision-making process for that?

• 25:25Speaker 00

  Well, it was just me and John. So, you know, we were sitting on the couch, it's like, choose Mongo, yeah, fine.

• 25:30Speaker 02

  Did they, did they get through to you with like a blog? Or was it just the reputation of Mongo at the time and open source communities, something else?

• 25:38Speaker 00

  I think it was-- So I wrote a data store for our prior company, an object-based data store. And I didn't really like SQL. I thought it was, there was too much of a translational mismatch between the domain of the application and that which SQL natively makes expressible. And so with SQL, obviously, you have to collapse down into a relatively restricted set of primitive forms. Whereas in your application, you might have a concept of, I don't know, let's say in the case of Stripe, of money that doesn't exactly comport with how the particular SQL database you're using happens to represent money, or whatever the case might be.

• 26:24Speaker 00

  And so I just had this principled objection to SQL. And I'm not endorsing this or saying it was good, but as this interview shows, I suppose, I had all sorts of strange notions about technology. And with Stripe, we wanted to be more mainstream and a little bit more, a little bit less heterodox in our technology choices than our prior company. And so instead of using Smalltalk, okay, we weren't going to go to Java, but we went to Ruby, which at least on a relative basis seemed to be mainstream. And similarly, rather than write our own object database, we went relatively more mainstream and used Mongo, which still give a lot of flexibility by virtue of being a kind of object data store. So that was fine.

• 27:10Speaker 00

  Everything I've said might disqualify me from ever making technology choices for another company, but... Would you do anything differently about Stripe 2? We haven't talked that much about it publicly yet. And the answer might be a bit like, you know, there's the Ju and Li quote about the, was it Deng Xiaoping, about the French Revolution, you know, it's too soon to judge. And so back in 2022, I believe,

• 27:43Speaker 00

  We, I mean, to this discussion about data models and abstractions, we realized that a couple of the core abstractions in Stripe were just not the right long-term abstractions, and we had to fix that. And so we designed a bunch of V2 APIs. Fortunately, we had contemplated the possibility of this earlier at Stripe. So, you know, most of the REST URLs that people are familiar with in Stripe are prefixed with slash V1. They've been prefixed with slash V1 from, you know, 2010. And so then in 2022, we decided, okay, we might use the, you know, we might increment the namespace. So we designed those new APIs. They started to ship this year.

• 28:28Speaker 00

  Congratulations. Thank you. And we're extremely excited about the functionality that it's going to enable. And, you know, without going into the arcana of it, you know, they will enable things like, historically, we have drawn distinctions and represented separately, things like and customers, things like subaccounts, things like recipients for different kinds of payments. And we're unifying all of those into being, you know, into the same kind of entity representation, which is, on some level, clearly the right answer and, you know, makes a lot of will

• 29:07Speaker 00

  and is already changing the businesses of some of our customers because they can enable their users to do various things without having to re-enter details or maybe to bring the same account across different countries or whatever the case might be. So, anyway, it's been a long journey. And the reason it was a long journey is, I guess, because...

• 29:34Speaker 00

  It's not that useful to just define these APIs in isolation. If we just wanted to define them in isolation, that's a pretty easy thing to do. The thing that's difficult is to make them interoperable with all the existing things at Stripe and to build translation layers and so forth.

• 29:50Speaker 00

  And then to figure out with our customers what a sensible upgrade path might look like, because we control our code base, but we don't control theirs. And so it's going to be, I don't want to exaggerate it, but in certain respects at least, it feels a bit more like an instruction set migration for a chip architecture or something where the instruction set by itself is easy, but it's all the kind of coexistence questions that become hard. But it's hard to ship this year and we're excited about it. I mean, I guess your question was maybe what lessons we've learned from it and...

• 30:23Speaker 02

  And maybe, do you think there's anything bigger to draw out of that? On either projects that are rewrites or thinking about these kind of decades-long abstractions?

• 30:32Speaker 00

  Yeah.

• 30:32Speaker 02

  And how to do that well?

• 30:34Speaker 00

  My trite answer to that is to unify everything you can plausibly unify and...

• 30:43Speaker 02

  How do you test design ideas for V2?

• 30:46Speaker 00

  Well, the people designing it, well, I'll give you one other lesson and then I'll answer that question. So just the other lesson, just maybe a bit tongue-in-cheek.

• 30:52Speaker 02

  And also, is there some chief API designer who's the mastermind? And it's one person? It's not some sort of working group?

• 30:59Speaker 00

  There is a working group, there are working groups, but there is also a singular person who understands and is more than anyone else responsible for the whole. And I think that's necessary. And my other kind of trite exhortation would be to make anything that plausibly could be an N by M relationship to support that, because if you only support one to N or N to one or whatever, and even if it's not obvious how it could possibly be N to M, just inevitably you'll end up needing that and you'll think, well, you could never have a company that's owned by two different companies or something.

• 31:40Speaker 00

  But it turns out that sort of every permutation in the space is, in fact, eventually explored.

• 31:47Speaker 00

  As to, you know... How to do that well? I really feel like it's like these new APIs. We think they're the, well, you asked the question, how do we know they're the right APIs? Partly from showing early versions of them to customers. Partly because the people who designed them had spent many, many years in the, witnessing and living with the shortcomings of the prior version. So we were kind of coming with strong opinions. But even the strong opinions, I mean, one can sometimes predict wrongly or extrapolate wrongly or over engineer something or whatever. So I think the cycles of customer validation, customer feedback are extremely important.

• 32:29Speaker 00

  And it's also very important, we did a lot of this, to literally write the integrations that would exist in the new world because, I mean, you really, I mean, I think Java is maybe an example of, yes, it fixes a bunch of problems with memory management or whatever that existed with C or C++ and antecedents, but at the cost of a lot of prolixity and overhead. And in order to kind of safeguard ourselves against inadvertently over engineering things, we forced ourselves to write a lot of API code specifically describing how we would implement various business models and flows and so forth, just to make sure that when you look at it, it feels right.

• 33:13Speaker 00

  But I don't want to endorse our approaches too strongly just yet. I mean, I'm feeling very optimistic, but you know, it's, we're, I don't know what fraction, but 60, 70% done or something, but not like 100%. And so I don't want to prematurely declare any victory.

• 33:30Speaker 02

  How do you, Patrick Carlson, use AI?

• 33:33Speaker 00

  Well, I, the main ways are the predictable ones where I use LLM chat tools a lot. And

• 33:45Speaker 02

  What are you using them for?

• 33:47Speaker 00

  Mainly for answering factual or empirical questions I'm curious about. So for for deep research style questions, I don't always use deep research. And now that the LNs are getting better at tool use and just navigating the web themselves, you don't need deep research as much, but for answering empirical or factual questions. I wish they were useful for for writing, but I usually end up dissatisfied with the writing that they produce. So I don't reuse them very much for that. And even for editing or grading my own writing, I mean.

• 34:25Speaker 02

  Have you seen any improvements as the models have progressed in the writing? I agree also. It's surprisingly generic.

• 34:30Speaker 00

  Yes.

• 34:30Speaker 02

  And.

• 34:32Speaker 00

  I'm trying to prompt it to not be generic.

• 34:34Speaker 02

  Yes, yes, inserting names of people.

• 34:36Speaker 00

  Yeah.

• 34:37Speaker 02

  And it just doesn't work. And so I have been disappointed the times when I've given it.

• 34:40Speaker 00

  People tell me that the base models are better at this. And it's the sort of normification of RLHF that puts it in in some kind of a tractor basin. Yeah, I have not succeeded in using them effectively there. People say that Claude is better and 0.3 is better than earlier OpenAI models. And on a relative basis, that might be true. But I don't want to sound self laudatory here and suggesting that I'm some particularly talented writer. I don't think I am. It's just like my personal style differs from the personal style, so to speak, of the models.

• 35:21Speaker 00

  And in some self centered way, when I write, I want to use my personal style. So I use them for the factual stuff a lot, and I find them terrific for that. And even when I'm reading a book, I'll sometimes activate, I've been recently using Grok's voice mode, and I'll just passively ask questions while I'm reading. And Grok is just listening in the background, and the answers are very helpful.

• 35:49Speaker 00

  And then I obviously use LMS for writing code and typically mediated through Cursor.

• 35:55Speaker 02

  So we are interviewing you, Patrick Collison, as kind of the most, if you had to pick the architect archetype of a software industrialist, I feel like you would be kind of straight out of Central Casting for a number of reasons. One is that you are running a large software company, a successful large software company. Two is you started as a programmer and then you know, moved to running the company. And then three is the company also builds things for developers. And so it's kind of the intersection of, you know, many circles in the Venn diagram. And so it's helpful to hear about, you know, discussing kind of experiences with Stripe. We are also interviewing Patrick Collison, the moonlighting economist

• 36:40Speaker 02

  and student of the world. And so are progress studies doomed now that AI is here? Is there any need for them?

• 36:47Speaker 00

  Well, I was going to say, I think the need for progress studies has increased. But again, I don't mean to suggest that proper noun progress studies sees increased need, but I think the kinds of questions that progress studies tries to answer are now more pressing and urgent because I think the degrees of freedom are increasing. And I think there is some Pan-Glasian view that AI will just magically solve all the problems. And predictions of the future are hard, but one, I don't think that's true. And two, in as much as we have, you know, evidence to date, I don't think that's been the track record. So I think that, you know, how we use these things, what kinds of decisions

• 37:29Speaker 00

  we make, what kind of, you know, considerations and, you know, margins of human welfare we seek to further, you know, I think all those judgments are going to really matter. And maybe a critique you could have leveled at progress studies or progress studies style of thinking five years ago is, these are all nice questions, but the world is on a kind of foreordained escalator path to, you know, some kind of teleological outcome. And I don't think the world feels that way today, or certainly feels much less that way today than it did. And so

• 38:05Speaker 02

  Because of global affairs or something else?

• 38:07Speaker 00

  No, I mean, maybe somewhat global affairs, but the trifecta of global affairs writ large. Second, I think that aspirations and ideals are becoming contested more actively. And there's an ambiguity these days in the US as to what the left and the right even stand for. And I guess we currently have one party endorsing tariffs and another party opposing them, but with the veilanes kind of shift flipped from what one might have expected historically. And then third, yeah, obviously, technology and first and foremost, AI.

• 38:48Speaker 00

  But in our industry, stable coins, the rise of China as the preeminent manufacturing power in many technologies of the future, like drones and robots and batteries and solar, et cetera. So, yeah, in many different ways, I feel like the future is... Peter Schwartz has this concept of the Schwartz window as the window of contemplatable futures in whatever number of years hence. And I feel like that Schwartz window, as of, say, 2005, as we contemplate the world

• 39:32Speaker 00

  of 2015, was fairly narrow and was correctly fairly narrow. I think the world of 2015 did, in fact, unfold largely the way we would have expected in 2005. And I feel like today, in 2025, that window for 2035, like, it feels extremely broad. So, yeah, I think the progress studies cell questions are more pressing.

• 39:54Speaker 02

  So you were on the record. In saying that people should focus more on the question of why we don't see improvements in productivity numbers as information technology increases and also as more people have started working on science and technology and more money has gotten into it. And what do the numbers look like now? Do we see AI in the numbers?

• 40:15Speaker 00

  There was a new paper published on this very recently, like the past couple of days.

• 40:21Speaker 02

  Exciting.

• 40:21Speaker 00

  That I've not had a chance to. I just queued it today to read. So I have, at this moment, only read the abstract. Its claim is that one does not, in fact, observe productivity improvements stemming from from use of language models. Now, I certainly can't.

• 40:40Speaker 02

  Do you know what they're looking at?

• 40:41Speaker 00

  They appear to be undertaking some kind of natural experiment looking at the individual level based on intensity of LLM usage. But I certainly cannot endorse their methodological rigor. And upon understanding it better, you know, I might be either really impressed and find it very credible or horrified. I don't know. But that was just a finding I happened to stumble upon today. I mean, look, overall, GDP growth in the US looks, well, over the last two years, it's been somewhat better than we expected. Obviously, we're speaking right now at a kind of volatile time. We certainly don't see any evidence for exponential takeoff. And if we, inasmuch as we thought

• 41:25Speaker 00

  that the encouraging GDP figures we have seen in the US over the last two years are attributable to some of these new technologies, I think you would also expect to see them in other countries, right? Because these technologies are quasi-public good. Anybody can use these LLMs. GDP growth outside of the US has not been that encouraging. We're not living in some massively accelerated period of economic growth for the world writ large. And so, you know, obviously, it's early days. But I think, you know, I think we're seeing that the diffusion of these technologies through the economy really takes time and involves substantial complexity. And maybe just last point of that is, I believe Jack Clark said in an interview with Tyler Cohen, one of the co-founders of Anthropic.

• 42:08Speaker 00

  And you know, Anthropic, to some extent, has, well. And Atabik has always taken the concept of AGI and even ASI, I feel extremely seriously. And you know, Dario speaks about this publicly. He's written about it, et cetera. And again, Jack Clark, one of the co-founders. And he said that he expects AI to increase GDP growth by half percent a year. And I thought that, I mean, I interpret Jack as really an optimist. And you know, half a point a year is in fact, a lot of incremental GDP when compounded. So I'm not saying that that's small, but I think it's interesting that that was his figure.

• 42:45Speaker 02

  Yes. Do you think that with the form factor that AI is taking in the economy right now? If we just kind of stretch the line forward, do you think we're going to need new measures in economic productivity than we have right now? So like, assume real productivity goes up, assume that AI keeps getting better, it gets kind of deployed in the ways you would expect. Do you think we'll need new measures? Or it should show up in the numbers?

• 43:06Speaker 00

  No, no, I don't think so. I'm not saying that GDP is perfect. I think GDP can be improved, but in any world where what we generally think of as the economy is massively enhanced, it'll show up in GDP.

• 43:23Speaker 02

  Yep. When will we be able to program human biology?

• 43:27Speaker 00

  I'm very excited about this. At ARC, which is this biomedical research organization, which I was involved in founding, we're working on training foundation models for biology using DNA and things like that. We're working on a virtual cell. And generally, we're trying, I mean, a thing that I think is, that I didn't appreciate until really spending more time in biology is, we've never, like, we humanity have never cured a complex disease. So, you know, one ontology or schema or something of diseases would be, you know, you have infectious diseases, the flu, the cold, COVID, whatever, and tuberculosis and, you know, very diseases with high mortality rates.

• 44:10Speaker 00

  Then you have monogenic diseases where, you know, there's just sort of one genetic mutation that is responsible for the disease, like Huntington's. And then you have complex diseases. And the complex diseases are kind of the residual that are now left after we've cured most of the problematic infectious diseases, at least in the Western world. Most cardiovascular disease, most cancers, most autoimmune disease, most neurodegenerative disease, et cetera. For certain of these conditions, we have maybe treatments that help, like statins with cardiovascular disease, but for none of them can we really say that we've cured it, that we like understand the causal pathways in meaningful detail and that it's just we can we can vaccinate against it or something.

• 44:50Speaker 00

  And I think this is our hypothesis, it could be wrong, is that this is in part because we don't have experimental and kind of maybe epistemic is too grandiose a word, but kind of epistemic technology that's up to the task, like the the pleiotropy of the genes in terms of all the different parts of the body and the systems and the mechanisms inside the cell that they affect is so, you know, there's so much combinatoric complexity there. And then, you know, the environment is such a vast and, you know, difficult to quantify thing that just it's really hard to understand for any of these conditions, you know, the etiology and the dynamics and so forth.

• 45:32Speaker 00

  Okay, then over the last 10-ish years, I mean, a bit longer, but a lot of the development has happened the last 10 years, we've gotten, we've gotten three new classes of technology in biology. For reading, we've gotten much better sequencing technology, single cell sequencing, the ability to sequence, you know, single cell sequencing of RNA and those improvements. At the kind of think level, we, you know, we've gotten neural networks and deep learning and transformers and everything, everything there, I mean, they've existed for a long time, but we've gotten the recent improvements in them. And the transformer in particular.

• 46:14Speaker 00

  And then on the right side, we've seen, obviously, huge improvements in functional genomics and CRISPR and, you know, bridge editing, which is a technology that kind of hark, but the ability to kind of make very specific directed perturbations in cells. But if you put those together, you now have the ability to, again, at the kind of level of the individual cell, to read, think, and to write. And this starts to really feel like a new kind of Turing loop and to have its own sort of completeness. And, you know, we will see how much this can do against these complex diseases and whether sort of this systematic approach is up to the task of shedding new light on

• 46:57Speaker 00

  their dynamics, but we are hopeful and excited.

• 47:01Speaker 02

  If we, here at Cursor, and also others in the industry, are successful in automating lots of programming as we know it today and, you know, replacing it with a form of software building that's much higher level and more productive and is much more just focused on defining what you would like the software to look like. If we, you know, succeed in that, who are you long? You know, people talk about the designers and how, you know, this will be like a renaissance for them, but, you know, are you long the grad students? I mean, there are lots of really, really amazing grad students who are awesome and then maybe are less skilled at making things happen on computers. But who do you think, you know, is the most unexpected beneficiary of a world where both many more people can make things on computers?

• 47:45Speaker 02

  And then also, you know, especially if it's an evolution away from programming, you know, the people who are already making things on computers are much, much, much more productive?

• 47:52Speaker 00

  I don't have a high confidence answer to that. You know, there's all sorts of trite stock answers like real assets, especially constrained real assets, you know, maybe we should be long SF real estate or something because... You know, it is one of the most beautiful cities in the world and will be enduringly so. Maybe we should be long the inputs and the ingredients to these systems because demand for them will go parabolic and so maybe we should be long copper. Maybe we should be long positional goods and celebrities and Taylor Swift's music catalog. There's a lot of, I think, compelling theories here, but part of what I think is interesting

• 48:33Speaker 00

  at this economic moment is the unpredictability and the contingency and kind of sensitivity to the precise assumptions in the technology trajectory itself. And the shape that it takes in five or 10 years or whatever, I think, is going to do a lot to determine the answer to that. And as I look backward the last couple of years, I'm struck by how many predictions have held up reasonably poorly, even for people who are, on the face of it, extremely well-informed. And so I've asked a lot of people this question and I have not heard any answers that are so compelling that I feel like I have conviction.

• 49:11Speaker 02

  So we are very happy to be serving Stripe and your guys' mission. What would you like us to build? How can we make Cursor better for you, either you, Patrick Holson, or you Stripe?

• 49:21Speaker 00

  Well, you guys are already making Stripe better, so keep doing what you're doing would not be a bad outcome from our vantage point. Cursor has today hundreds and soon thousands of extremely enthusiastic Stripe employees who are daily users of Cursor and they report that it's a very significant productivity enhancement.

• 49:48Speaker 02

  We'll wait for the economic numbers.

• 49:51Speaker 00

  Well, the economy is pretty big and these diffusions take time. So it seems kind of greedy to want more if you're already making... Stripe spends more on R&D and software creation than we spend on any single undertaking. And so if we're making that process more efficient and more productive, then maybe it seems greedy to want anything more. If I'm being selfish... Okay, three things.

• 50:17Speaker 02

  Perfect.

• 50:18Speaker 00

  The runtime characteristics and integration stuff that we just discussed, I think would be really valuable. I think the the refactoring and the beautification stuff that, again, we also talked about, I think would be extremely helpful and and I think really change our degrees of freedom. As in, if you could lower the cost of future changes to Stripe and improve the quality of the architecture. And then, third, we really care about what we call a Stripe craft and beauty. And we want our we want our software to be well-designed and pleasant to use and pleasant to use, not only in the superficial, you know, pixel sense, but also in the deep,

• 50:59Speaker 00

  it works very well sense and is something you can set up and largely forget about and just trust or forget about it in as much as you want to. You know, there's obviously a concern with AI that it leads to the creation of more slop and more kind of crappy things, but not more of the best things. I don't know what it would be that Cursor would do to ensure that the world is creating more of the best software and not just more software. But I think that's an interesting and important dimension. So those would be my, I mean, besides all the obvious things to do, those would be three suggestions.

• 51:39Speaker 02

  Amazing. Thank you, Patrick.

• 51:41Speaker 00

  All right. Thank you for having me.

• 51:42Speaker 02

  Yes.

• 51:45Speaker 01

  Thanks for listening to this episode of the A16Z Podcast. If you liked this episode, be sure to like, comment, subscribe, leave us a rating or a review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X at A16Z and subscribe to our Substack at A16Z.substack.com. Thanks again for listening, and I'll see you in the next episode.

• 52:10Speaker 01

  This information is for educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product. This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z. Such advertisements, companies, and individuals are not endorsed by AH Capital Management, LLC, A16Z, or any of its affiliates. Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.