Making product on the S curve of AI. Plus, Anthropic Artifacts
42:28 Share

I don't think any founder ever goes into a major release for their company not expecting the world to catch on fire because for them it's the single most important thing like in their life. I think it is so brilliant not just because of it being a primitive that makes sense it's aligned with like their strengths right it is in service of the model the model is the product.

There's like one big dog. Right. And then there's a bunch of little people. They're all lemmings trying to feed the one big dog. That's obviously not true in the case of a core GPT-like interface, a core chat interface that is still very early in the S-curve. Yep. And it's also very true in lots of other verticals where you really believe that there's rapid innovation. Hey, welcome. Welcome to Hallway Chat. Yes.

I'm Fraser. I'm Nabil. We should get started. You only have a little bit of time before you're taking your kids to go get the best tacos in the world. Let's talk. You got to do it. I got to spring this on you. So a couple of times ago, you had this big rant that we're never going to even talk about models anymore. Models are no longer a thing. What was the first investment we made after that show? I don't know what you're talking about. Leapfrog.

Yeah. So this is a confusing job because sometimes you have conviction about one way of viewing the world and then you meet a group who tells you what they want to do. Yeah. And you just got to reevaluate.

Strong convictions loosely held. Strong convictions loosely held. I thought that was worth just dropping. Now, the unfortunate thing, of course, is like there's probably some founder who didn't even listen to the podcast, by the way. They saw the title, which I think was something like, should we just stop talking about models? They didn't even reach out. Like, we just lost a wonderful founder we're going to work with. No, I hope not. I hope not. I mean, we certainly are all trying to make sense out of the world right now. Yeah. But, man, you got to have humility. Like, what?

The whole point of any good founder is that they have the chutzpah to tell the world that they're wrong about how they're thinking about the world and then go do it my way, right? There you go. It's an interesting conversation, right, as to why we decided –

to have some loose beliefs. I still laugh about that regularly. There's nothing more exciting to me than having a conviction or having a thesis and then have a founder walk in and say, hey, have you ever thought about it this way? And just, you're just, oh my God, never. That's amazing. And now you have a new lens with which to process the world. That's it.

I've heard you and Santo and others here basically say some semblance of that. Yeah. Where the most exciting pitches have been where you have some ideas about technology trends and you've been interested in a few different things. And then some founder comes in and just lays out a vision for the future that you hadn't even contemplated. It's one of the reasons why we keep the team relatively small here, right? We talked about it, which is just like, this is the problem with large companies. As large companies, the things that are loosely held become...

written down as dogma and then they're carried on. Then the person, the product manager who came up with the idea isn't even there anymore, but everybody else carries on as dogma. So it's no longer malleable. I think there's a lot of that institutionalization and venture that has gone on, especially in the last five or seven years.

And that's super weird in a world where everything is changing right now, right? Yeah, yeah, yeah, yeah. And the more process you have, the more you work against the process and evaluate yourself against adhering to that. So on the topic of things being ambiguous and moving very quickly, Anthropic Portfolio Company launches something that I think is just too interesting not to discuss, and that's Claude Artifact.

Yeah, I mean, look, a little, we're going to talk a little of our own book, which I gave you crap about when you brought up talking about this today. But we're just, it's user behavior. The thing that I am doing right now, and you were right to call it out, like every day I'm

I have a little post-it note next to my laptop at home that says, should you build an artifact for this? And it's because I'm trying to train myself. Just like whatever it was a couple of years ago, you're trying to train yourself to start searching in perplexity versus searching in Google for something to get used to a new behavior. I have every couple of days now...

Oh, listen, man, I'm going to stop. I trust you and myself enough. You are not like a shill. We are like authentic to a fault as to like who we are and what we care about and our values. Like we're talking about this because it's actually like exciting and awesome. Exciting, yeah, yeah, yeah. Why don't you tell a little bit about Cloud Artifacts, set it up, what is it? So the typical version is it's what GPT Store should have been. No, come on. No. No, no, no.

So, Claude... So I'll tell you something that's very funny. You asked in a recent episode of us talking, do you think it's still going to be a chat interface? Yeah. I said, like, hell no. Like, the chance that Noah at OpenAI and Team has stumbled on a chat interface is like...

Noah, before we ended up on the chat GPT Enterprise, him and Val prototyped a version of Artifacts. It's a very natural way to co-create the Artifact. It's a great name. It's a great name. Right. It's a great name. Yeah. Calling out that prescience and then your comment there is too funny. Yeah.

Thanks for pulling that to the foreground. I did not know that. Awesome story. Artifacts, very simply, if you have not used it, you do need to go into Claude and have a paid account and you have to turn it on. You have to go into the bottom left corner, hit a button, go into settings and turn it on manually. But basically, if you then type into the chat interface, a thing that would be represented in software and in a UI, it will...

render that thing on the right-hand panel. So your interface changes from a single threaded chat that it would normally be to, I'm now going to build a little React app. I'm going to render on the right-hand side of the page, the thing that you're trying to talk about. And if you go now search artifacts and Claude on Twitter, you will see a whole bunch of wonderful things. You haven't already been exposed to this of people experimenting with this idea. I

I did some fun experiments with it that were completely unproductive. I've done some generative art, just like algorithmic art, for probably 20 years now, off and on as a side project. And so I had it make some algorithmic art for me where it was just rendering little line drawings and then had it render above those line drawings parameterizations so that I could type in, for instance, how many squares should it render per second and a little slider that I could move back and forth. For me, the first simple...

user-controllable instantiation of this thing that we would like academically call malleable software. Yeah, that's right. So let's see if we can even just...

Come at that from a slightly different angle. Sure. Yeah, you have your chat dialogue. Yep. And you say, I want to write a memo about X, Y, and Z. We don't write memos, so I don't know why I came up with that. But it then creates what looks like the equivalent of a Word document. Yes. And rather than in the chat dialogue having the document appear, it is an artifact. It is a piece of the UI that is pegged on the side of the window. Right.

And then you can iterate and riff on the output with the model through chat. And so you can say like, oh, actually, the first paragraph is a little bit more verbose than I'd like it to be. Can you tighten it up? And actually, can you make the entire theme to be more about X rather than Y? Right. And then rather than having to have it just like spit out the content into the chat window again, it just basically updates everything.

The artifact. It's a great name. It's such a great name. And that's what I mean by this concept of malleable software, which is you're not really coding. You're still speaking to the chat agent. That's it. But the chat agent is writing code and then executing code and then showing that code to you as a part of the overall process. That's right.

is trying to finish this deck-building game that we've been working on for a little while that we built almost entirely with AI, and not like AI art, AI balancing, AI content. Like, we've been trying to use every AI tool possible, and this is like the vessel for that. So this involves a lot of mid-journey prompting for the art itself.

across a lot of different categories of types of characters and types of art. And we've built almost like an internal nomenclature and language amongst us for how do you describe the foreground and the background? Basically a prompt language. So I had it build a little UI toolkit. Okay.

A mid-journey prompt engineering thing just for my game where it's basically a bunch of checkboxes. You pick, oh, is this a scientist or an engineer? Right. And then it like inserts the little six-word thing that we've encamped that makes it look the way it's supposed to look. Okay. And so I turned what is normally a, was a bunch of Google spreadsheets with a bunch of words in them that were copying and pasting over and over and over again and getting really bored at into a simple web view interface checkbox. Great.

that we just go through really quickly and speeds us up like crazy. Okay. So, it's

It's awesome. They've done a great job as well because you talked about the ability to share and remix it. Yeah. That launched, you know, a couple of weeks after they introduced the main feature and it was in response to what their users were kind of like pulling the product into that direction. And I think a similar one that we can probably tack onto this that's aside of this is the concept of projects, which they also launched recently, which goes very well with artifacts, right? So projects, very simply, if you haven't looked at it, is essentially I can...

have a repository of chats inside of one meta structure with a bunch of documents and things that are inside of it. So if I think I'm going to keep talking about a project over and over again, so in the context of

Maybe we're working on a deal or we're thinking about a new area of research. And so we have five white paper PDFs that we want to put in there. And then we started a bunch of texts against them. And we wanted to go back to the same knowledge over and over again. We start a project instead. And then all these things seem to layer very – none of them are like huge, huge. They're all not things that nobody would have ever thought of before. But they layer very well into each other.

Because obviously if you're making an artifact, an artifact would sit inside of a project so it had access to all of these PDFs and other things, knowledge. And then what you want is sharing, obviously, of a project with an artifact. And that's the thing that launches later. They kind of like overlap really naturally. It's great. And so I'm just smiling because you said that like it's not stuff that people wouldn't have thought about earlier. I'm just laughing. I'm like, we should just go talk to Noah to figure out what we all should be launching in like two to three years.

I think for this conversation, what is really interesting to talk about is product strategy and how to think about that depending on where you are on the S-curve. And so a couple of times ago, we were like, hey, how is this all going to play out? And we commented on Diane at Anthropics comment that like,

depending on where you are in the S-curve. And then we were framing it through that point of view. I think that artifacts and then the world's response to it is the first strong evidence that we are very early in the S-curve. Why do you say that? ChatGPT comes out as just a chat dialogue, sets the world on fire. Sure. We are even discussing, do we think chat is the final UI? Of course, we don't think it is. This is the first time where we have seen

a product of this sort introduce a meaningful new element to the UI. A new primitive. There's lots of people building entirely new interfaces on top of Jaxx and BT, blah, blah, blah, new verticals. We see lots of that. But this is a new primitive. A new primitive, but also the product is that broad consumer AI product. What we said is, what will us and our siblings and our parents be using when they think of an AI product in three years' time? Yep. And

This is the first time where we've seen a new UI really resonate in this type of product experience. Why do you think it does? We can get back to the S-curve thing in just a second. Yeah, yeah, yeah. But I just want us to reflect for a second on...

what it is about what they did here that we're impressed with and that we teach, that would teach a founder any lesson or like, what lessons do we take away from this that we might apply to the other founders we work with and the other products we're working with? I'm not privy to anything that they, like the thought that they put into it, but I've sat through and worked through it with Noah when we did like basically a prototype of this. And it is a very natural way. Like, what are you using today

Chat GPT-4 or CLOD-4. It is to create software. It is to create an essay. It is to create, like in your case, the prompt generator for mid-journey, like a little bit of software, an app.

And these things don't get it right in first shot, zero shot. Like any good project, you have to volley it back and forth and you have to co-create it with the AI. And this is the first time where the co-creation happens in a way where you are playing more of the editor role

And it is then going to do all of the heavy work behind the scenes and showing you the product, the byproduct of the work. I agree with everything you just said, except I pushed back slightly as you were just talking about it. I thought the other instantiation of this is code interpreter. Yeah. Yeah, yeah, sure. I think, you know, the other thing that feels very similar to this that ChatGPT pulled off is when I'm making charts and graphs and things like that, and data analysis inside of ChatGPT, like it has the same... Yeah, yeah.

Yeah, yeah. Well, listen, I thought that the ChatGPT launch of the embedded tables or charts or whatever of it was directionally aligned with this and beautifully done. Yeah, I agree. Do you know who did that? Who did that? Noah. Noah.

Yeah, of course. You are now editing the output and the AI agent or assistant is doing all of the work for you and presenting it in a way that's easier for you to critique it. So back to the S-curve. I think it is so brilliant, not just because of it being a primitive that makes sense. It's aligned with their strengths. It is in service of the model. The model is the product.

All of those things. It's brilliant in how they introduced it. And we've had this discussion before. It's like when you're in a mystery versus a puzzle moment, what is the right approach to product strategy? And I'll ask that to you and then frame it in the context of what they've done with artifacts. The question is just...

how certain are you at how right you are. The thing that happens for a late stage Red Ocean market is that you have talked to all the customers, you know what all of your competitors are doing, you have a sense of what's going to happen for the next year because everything's slow and things don't get released that often. It's a known market with known customers. When you get something right, you know it and it's very durable.

And so you want to do, you know, if you're a person releasing a new first-person shooter for the PlayStation, we know the history of all the first-person shooters. We know all the other first-person shooters. You can do research on what casual versus hardcore users think about in terms of time to kill and reload while waits. And you can just go so deep that you do the research and you do a big research

huge launch. You have utter ideal certainty about what you're trying to do if you've done the work. And much like a puzzle, you can just do more work deeper to get to the answer. I paint that picture because I think people can understand what that feels like, right? And then

AI, if you pass it the opposite direction, is different in almost every market, even static markets, even older markets right now, as we're trying to figure out how it is going to transform those markets. There are some markets where it's not going to do enough. The incumbents are going to sprinkle in a few little things, and it's not going to be disruptive enough for a new startup to do a thing. That's obviously not true in the case of

a core GPT-like interface, a core chat interface that is still very early in the S-curve. And it's also very true in lots of other verticals where you really believe that they're disrupting innovation. In a world where it's disruptive innovation, you have to match your marketing and your product velocity to your certainty that you're right. And your certainty should be low. It

It has to be confident when you're trying to pitch a VC or hire a new VP or whatever that you've got a read on the future. But the cone of uncertainty on that read has to be very wide. You're really certain about next month. You're really uncertain about next year. We're moving through the curve. And if that's happening, if you're early in the S-curve,

then the answer is that everything should map to that uncertainty. So it should be a tweet, not a launch party. It should be baked for a couple of weeks, not a...

six months. It should feel in every way you should communicate to the customer that they are also early with you on the journey. You know, the late S-curve adoption company, the last mover advantage company of all time is Apple. And you can just look at everything as a rubric of how Apple launches or does a thing. And if you believe you are actually early in an S-curve, then you should be doing the opposite of everything Apple does with a launch. Right.

And that means you should try to feel more amateur. You should try to feel more authentic. You should try to feel more familial. You should try to tell them this is a beta. This is a labs. This is going to change next week. And you're going to try and ship regularly because you're not trying to tell people that the story is done. You're trying to get them to turn the page.

I love it. You're trying to hook them and then give them a new page next week and a new page a month later so that they're going to be on the journey as this market develops. That's right. Not to try and present yourself as the next Salesforce or the next blah, blah, blah. Yeah. You're just saying it's early, it's crazy, come on the journey. It intuitively makes sense. And by the way, it's also self-evident in AI because in a way that it's not normally with startups because if it was...

If it was all baked right now on first launch, then that would almost necessarily be a sustaining innovation. Yeah, yeah, yeah. Which almost necessarily means it's not a startup opportunity. So you shouldn't be doing it. Yeah, yeah, for sure. So if it's a startup opportunity, it's probably crazy and insane and it's probably still malleable and still soft clay. And so you should just communicate that to users. And so you should –

The old story I tell a lot about this that I really have respect for is, which I probably even said on this podcast before, because you get to the point where you start selling stories over and over again. But like, Oculus, after the acquisition by Facebook, now Meta, I thought they did an excellent job at the third version of the Rift, I think it was, which was the first version released after the acquisition. And I

I think they read the market incredibly well, which was that everybody after, you know, Unicorn Plus acquisition is like, oh my God, VR is here. And internally, that was not the conversation. Internally, as we are very early in the S-curve, it is not ready for prime time. Everybody is getting way too overhyped. And so that first version, which could have been as pretty and wonderful and amazing as an Apple product, obviously they had the budget, the resources. The first version that came out afterwards was a simple cardboard box. Love it.

And like dialed as down as possible. You want to tell you expectations, right? Whatever that expression is, is like life is really about how, whether you meet, exceed or fall short of expectations. Yeah. And that's really what this is about. You, there's the other expression that like a startup is an idea maze. And like the one path where you're saying, if you are confident that you're toward the top of the S curve and you know what users want and you can like,

map that roadmap out for 12 months, and then just tick it off and have a big launch event that's anchored with the product release. That's like saying, go three steps forward, take a left, five steps forward, take a right through the maze. If you already know the map to the maze, you can do that, but nobody knows that in this world. And so you need to kind of meander and see where the path takes you.

And that's what they did. I think they've actually done a very good job of coordinating this as well. They've released most of these features with

Simple tweets, small web pages, very little fanfare, just like, here's what to do.

but have a go with it. And I thought also in the past at certain times, similarly, like I think OpenAI has for many of their best features, it has been this kind of like,

you know, Greg tweets a thing, like it's a simple thing. You put it out there. For sure. There are counterexamples. You know, the GPT store, which was a counterexample where they really... Yeah, the gravity was quite high. Yeah. This is a thing. This is a huge... And ironically, maybe one of their... I personally would say maybe one of their worst releases. And it mapped to very high fanfare. You don't have to say anything about anything.

I'm just here to put you in awkward situations. I will say at the launch of ChatGPT was two weeks after Galactica or whatever that meta thing was, which was a chat, but yeah, of course, you're looking at me like, what the hell are you talking about? Nobody knows what it was. It was a meta release from their AI team, which was a chat model that you could interact with. And it was pushed out through the science team or something like that. And their team was...

saying this is going to revolutionize the way that science gets done. Three days later, they had to pull it. Oh, I remember that. The comms team, Hannah and Steve, were like, no, no, no. We're going to call this thing, rightfully so, a research preview. Right. And we're going to tamper down the expectations. Do you think without Galactica messing up publicly, bravado, chest thumping,

Do you think that Chachapiti may have been more? No, no, no. It was destined to be on a low-key release. That was the – it was a low-key release. That has been the MO from, like, Stephen Hanna on many things of that nature. Yeah, yeah, yeah. For sure, for sure. They're great at what they do. You know, one of the things that we just talked about is just the feeling that feels amazing when you get this right is one of stacking wins. Yeah, yeah.

on top of each other. And I'm curious, you've been inside of high-performance organizations. It feels so great as a consumer when you get that feeling. But...

It feels like so few companies do that. I'll come at it from a different perspective first. Sure. And we were talking earlier this week internally about how a founder had teed up a very large release. And then it kind of felt like, you know, the air escaping a balloon, you know, at the end. Yeah, that's right. Yeah, that's right.

And the comment was like, oh, I think their expectations were a little bit out of whack. And I don't think any founder ever goes into a major release for their company not expecting the world to catch on fire because for them it's the single most important thing in their life. And they wouldn't be doing it. I think actually the likelihood that you deliver something that actually resonates is rare. Yes. Right? Yes. And what's the response? If you deliver something that resonates –

catches the world in a small way on fire that's rare what's the what's the instinct well I think the instinct is exhale honestly like

Like, it's just so hard to ever get something that resonates that I think the instinct is to, quote unquote, take advantage of it. Uh-huh. By now we get to go. We were all taking a bunch of incredibly risky bets to try and get this thing out in the world. Yeah. High, super high bid. I don't know if it's going to work, but if it kind of works, then every instinct is to have the pendulum swing the other way. Now we get to go, okay, it's out. Somebody cares. Yeah. Yeah.

Now we can talk to engineering who is completely stressed out. That's right. And we just did an offsite where we all like, I don't know, got a home in Tahoe and lived together for six weeks to get this thing out the door. Everybody slept four hours a night for six weeks. It's finally out there.

Now we're going to bug fix. Yep. That's it. Now we're going to stabilize, right? Is this where you're going? Yeah. Right. It's like, we're going to stabilize it. We're going to do bug fixing. We can now have a six month roadmap. We can build up out the robustness of that one product. And this is another good example where if you're late in the S curve and you think it's an enduring innovation that was going to last for a real fine. Yeah. But most of the time that's the wrong call. A hundred percent. Yeah. Most of the time I think it's a hundred percent the wrong call. Yeah. Like, yeah, I,

I don't know. There's counters. I said, well, listen, like, you know, eBay had one trick, right? eBay released a, they had one product innovation. And after that, I'm not saying there were all the very good strategic choices that need to be made and a lot of scaling and a lot of tactical shit. But like, it is basically the same website has been since 1997. They're lucky they didn't disappear because of the network effects of a marketplace. Like otherwise they'd be gone. Yeah.

I spent some months at Airbnb wondering like- It's a search box? A single- Airbnb, it's a search box? That's right. A single page. I spent some time wondering like, how are there organizations like these that exist versus an Apple and Amazon and Facebook that can punch, punch, punch, punch, punch? And I think it does come back to this. I think the rare ones are the ones that you cited where they do one thing and they strike gold, so to speak, and then they can just ride that. And those require network effects-

of some form or another where you just can't mess it up. Right? Versus the truly iconic companies, they don't deliver something that catches fire and then, like, let's do a week of bug fixing, let's do a week of, like, stabilization efforts, and then let's get to work on, like, the roadmap of long-tail feature requests. It is really hard to be a CEO and make that ask, though, right? But you're right. You're right. You know, Facebook...

comes out and they get traction. And then you see the Facebook feed. You see... Oh, yeah. The Facebook Connect where they were trying to authenticate logins across the rest of the internet. You see Facebook apps. You see Facebook coins. Like, it was... It was...

high-risk product release after high-risk product release for years. I used to think of it as you want to find the red-hot core of your product, and everything else on the periphery is kind of irrelevant. You and I have talked about it before. It's like, what is it that you're polishing, and polish matters. I think about it as you need to get the red-hot core, and then the next thing is not to go build out the adjacent features. It's like, how do you get that red-hot core hotter, bigger,

And that is like the Facebook newsfeed was a high risk thing because it put the previous Red Hot Corp at risk.

And it took real founder conviction to look at it and realize it was the right thing to do and push it out. Yeah. For people who don't have the history of it, Facebook at launch looks a lot like MySpace. It's a profile page. That's right. It's not even a feed yet. And it was working. And so you had to kind of throw out the paradigm of what worked after it was already working in order to find the next paradigm. Right. And I think there's examples of that.

Some examples of that happening in companies today. But you're right. The natural inclination is when something starts to work, your instinct is to build the scaffolding around it immediately. Your instinct is to bug fix, now we'll go build the SDK to that, the API to that. We'll go broad. Or the other thing, which I don't think we're saying at all, is...

You don't do things that are orthogonal to the Red Hot Core. You don't suddenly say, great, now we make shoes. It is how do we double down on the Red Hot Core, but how do we take another big, huge risk on top of the thing that just is working, which is smart. You started this entire thing by you take a beat or you take a breath. And I actually think you should take a beat and take a breath in the sense that I don't think you should do anything. And then you should start...

experimenting as quickly as you can to figure out what is the next thing that pushes the RedHawk core forward and bigger and burns brighter. Who do you think is doing that right now in AI? I think 11 Labs.

Right. Like I think you can argue that they launch long form audio text to speech. Then they do dubbing. Yeah. And then they do a consumer app and then they do licensed voices within a consumer app like it is pushing and pushing and pushing and pushing and pushing. Yep. You might be able to make an argument that like there's a little bit of like, OK, let's go and make shoes at this point.

Yeah, how much are they going after different versus doubling down on the center? The Facebook example was so great. It pushed in the same aligned direction around the mission and the vision with the one product. I think Descript has been doing this for a long time as well. They had transcription, knew it was going to be a commodity some number of years later, and they have continued to ship innovation after innovation after innovation. Whereas the counter example for me is like,

AI meeting bots, generally. Uh-huh, yeah. Right? Like, there's a whole cavalcade of AI meeting bot companies, and most of them came out with a single interesting good idea. And how much have they changed since then, other than just building out the cruft on the edges, right? Well, you want one that basically came back to the world and said...

We can't make this bigger, better, bolder in that direction. And so we're going to go plumb some other area. That's actually very true. Right? Yeah. Copilot was one you mentioned the other day. Yeah. That is also a great example. GitHub Copilot. They launched something that's profound. Yes. Right? Creates an entirely new market. I think they're still killing it on revenue. Sure. I think they did like half of what you should do. Right. They stood up.

GitHub Next, where they have all sorts of like crazy science experiments happening. And it's clear that these are science experiments. Right. And I don't. That sounds good. Yeah, it sounds good. Whether or not it's because they haven't been running the right experiments or they just haven't been able to find something that resonates, like they haven't pushed beyond the first copilot. Yep. I mean, like in any meaningful way.

Yep. Right? I don't know. We meet a lot of founders who roll their eyes and scoff at them. That's probably not the place that you want to have in the market. You want to have fear. Although, let's be clear, they're also because they're the founders trying to take... No. Get a copilot down. They're there. I

I think there's – no. There was a moment where, like, people were really fearful of Google. Yeah, yeah. And they were, like, fearful of them for good reason. That's right. But that was, like, 10 years ago. No, no, no, no. But I'm saying, like, there are moments. Why were they fearful of Google at that time? It's the same thing that we were just talking about. It was the – it's a few – you know, you launch Google, it's working, and then suddenly you get to this, like, thoroughbred horse stage where you're like –

Gmail launches, Google Maps launches, like Android phones, like they're just stacking win after win. That's right. You don't get a win for seven years and things are different. That's right. Nobody is concerned of their like

Kill zone for a really unfortunate term, right? Yep. Right. It's like this is clearly should be in GitHub's roadmap. And we don't care because they're morons. Yeah. Whether they're right or wrong. They're not striking fear into everybody's heads. That's right. That's right. And, you know, we don't have a good rebuttal to that. Right. We're like, yeah. Yeah. All right. Like, yeah. Like they have not done much. And this is quite interesting.

Yep. I have a question from venture perspective that might be useful for founders. Yeah, hit me. What do you do if you come into the partnership or the firm through one partner and then you realize during the process that it's not the right partner? Yep. It depends on the firm.

Okay. Like, you know this obviously from Spark's standpoint, there's like, you know, there's seven early stage partners. Flat, there's not a lot of credit. We all just play together. It's easy. The simple version is like, if you go look at the website and you think about the number of investors and it's relatively small, then there's probably not a whole lot of problem switching unless...

You can kind of quickly get the read that there's like sometimes the smaller partnerships you get, there's really it's like a one person firm. There's like one big dog. Right. And then there's a bunch of little people. Right. In which case, ironically, it's actually pretty easy to switch because they're all lemmings trying to feed the one big dog. Okay. And so that's actually fine. And then there are times where it is a big, large firm. You're just screwed.

That's just the fact of it is like it's an attribution culture. If you are trying to scale a very large firm, then credit needs to be attributed appropriately to figure out who I'm supposed to promote. And so there's some CRM somewhere that as soon as you made a phone call or had a coffee or showed up at a dinner, it's their job to tag as many humans as possible so that if one of those humans becomes an IPO, they might get promoted.

And so other partners know that and you're screwed. I'd say the easy test for this though is at least as a founder, I want to feel like the whole team is helping me anyway. Yeah. And so I would just be honest with the partner

about switching to another partner and saying, I think this other person might be a better fit. And then by the way, if they're not okay with that or you get the heebie-jeebies or whatever, that is telling you an awful lot about this relationship. That is informative to you in picking what firm you want to work with. So I think it's great either way. Yep.

If it screws it, I guess as a founder, if it screws it for me, I'd be like, awesome. I just learned something that I wouldn't have learned otherwise about their relationship anyway. That's a great way of framing it. And what I have been saying to founders, whenever there's like a hard question that comes up, is like, okay, you know, this is the start of what should be a very long-term great relationship that will be defined by working through a series of very hard problems. And so how we're going to work through...

X, Y, or Z is going to set things up. So let's get to work. Even better framing. Yeah. You just presented yourself with a conflict. Awesome. All the relationship is with the board over a period of time is trying to learn how to work through good conflicts in good ways. That's right. And listen, I've had to work through that in the inverse.

Right. And we've done we've done versions of that where it's like, OK, listen, why? Why? Why do we think it's the different partner might be helpful? Yep. And in that case, like you understand, there's like there's FUD. There's like real valid concerns. There's all sorts of like different things that help as you work through. Just like any of these things, you can like cut through to the core of the issue.

And then just address it. We've absolutely had these conversations with the founder where it's like, oh, I think this other partner, well, we've certainly done the, it could be either of us. You pick. Yeah. It doesn't matter. Sure. And then there are other situations where you try to talk to the founder and they're like, well, I think it should be X person. And you're like, why? And you listen to the answers and you're like, I don't know if your head's on straight on this one. Like, I think you're getting bad advice. For these reasons, let's talk it through. Yeah. Yeah. That makes sense. All right.

Two other things. Sure, man. We have non-AI products. Oh, yes, we do. In our lives. In our lives. I'll go first. We can't help but apply our own sense of product nerdery to all the things. That's right. I'll go first. My wife and I bought a Roborock, some sort of horrible name. Roborock Max V8 Supra or something like that. It is a robot vacuum mop.

Okay. That then has a kind of eye robot thing. Yeah, yeah, yeah. Sure. But better. Okay. Okay. Why is it better?

Listen, it's the same as iRobot. It's got like a docking station where it returns to and it empties the vacuum. It empties the dirty mop water and it refills the mop water. It has LiDAR, so it navigates around your house. It identifies with AI objects in the way and navigate around them and stuff like that. I'm bringing this up because it is a life changer. We're pulling out of the house with the kids. Yeah.

And you just be like, oh, yeah, let's clean the floor. And you hit the button and it goes out and does it. But I'm bringing it up very quickly because a lot of discussion over the past year has been around the role that robots are going to have in our lives. Oh, yes. And like a big branch of effort.

Both from founders and investors has gone into humanoid robots. And there's like the, I don't know, what's the trite, almost insufferable comment that like the human world is built for humans. And so therefore you want to have a robot that's built. Yep. It does the job miraculously well. Yeah. Like amazingly well. Yeah. It costs not the same amount of a humanoid.

Right. Right. And yeah, we have stairs in our house and you know what we do? Yeah. We carry it up or down the stairs and we hit the button. Yeah. And the cost is so low.

That like eventually, undoubtedly, is so great of a product that we'll just buy another one. Well, there's two things that I think worth touching on. Yeah. One is we spend a lot of time thinking about how to make things magical for consumers so they have no effort. And instead, I think the better way to build product is if you have created a truly magical, wonderful new thing. Yeah.

Users will meet you halfway. Yeah. They will do things to get there. This is especially true in AI. Think about the crap you have to go through to get from mid-journey in Discord to use that product. Yeah. And many other people would have said, no, the only way to launch this product is an iPhone app because I have to meet users where they are. Yeah. Like...

Makes sense out of those two dichotomies. And the truth is that like, if it's really magical, users will, again, early in the S-curve, crazy new thing, they'll jump through hoops. It's okay. Late in the S-curve, it's marginally different than the next player. You better meet them exactly where they are. Sure. We got like a new outlet installed. Like we moved heaven and hell. That actually is a great way to bring it back is like, I've always thought that. Like I remember jumping through so many hoops to get free music.

Yep.

And they decide, they're like, no, the funnel's what we need to fix because people are not converting. So the flip other side of this late in the S-curve product is a coffee machine, of which there's millions of them. And they all serve different uses. I use Comotier sometimes. Frozen coffee, it's awesome. Highly recommend it. It's still so down-to-earth. A spin coffee for making espressos. Like, pour over coffee. There is no better product in the world than the X-Bloom, which I know I've been talking to you like crazy. Oh, my God.

They released a second version. I have not used a consumer product that has made impeccable choices. And everybody who just wants AI stuff and software stuff could just turn off the podcast right now because now we're talking about coffee machines. But it's a great example of something that for an early stage startup...

that is early in the S-curve, I would have pushed back pretty hard on basically every product choice they made. But for where they are, it was like every single product choice was great, which is to say it already made great coffee. It still makes the same coffee that it made in last year's version, but they have made like four or five

small, iterative choices. They changed the architecture of the framing of the product. So just three things. They changed the framing of the product so that it was a little less fussy and a little more clean. Great. That sounds simple. It's not simple. They added an LED to the front that actually explains what it's doing with these wonderful tactical knobs that feel awesome to turn back and forth. They actually even look nice to touch.

They are, and it's a great- There's a good texture on it. Yep, great choice. They lifted up the place where the grounds come out by like a half an inch so that less grounds go out. And then they added a little drip tray which hides some of the grounds so that you, so the whole thing looks cleaner. And then there's one or two other things. They're all great,

I probably half of you didn't listen to half those things. All I can say is that if you were in their Discord channel and you were listening to the user feedback, this was like incredibly good refined product choices that are perfect for this kind of product. I own the first version of it. Yeah. It is probably the product that has brought me most happiness before my Roborock. Yep. Very long time. Yep. And you just ticked off all of the little minor annoyances that I have with what is already the best product.

In my life. Yeah, well done. Yeah. But again, this is a good example of like understanding where you are in product development, drives different. And I didn't know where we were going to get there, but we somehow linked it all together. But that doesn't make sense. Like adding a drip tray, if you were at like the bottom of the S-curve would be insufferable. Yes. Right? But like they've already gotten way up there. If it's the very first coffee machine, you don't then spend the next six months thinking about drip tray design. This knob has nice tactile appeal to it.

Yeah, which is hard because we all love a good knob. Let's be clear. Yeah, but that's true. But the coffee is great. Yeah, highly recommend. It's an expensive machine. I think if we're riffing on that type of stuff, maybe it's a sign that we need to be done. Yep. All right. Thanks, everybody. As always, if you see an AI product worth chatting about, I'll let us know. Talk to you later. See ya.