Subscribe
Share
Share
Embed
Fraser & Nabeel explore what it means to build great products in this new world of AI.
Two former founders, now VCs, have an off-the-cuff conversation with friends about the new AI products that are worth trying, emerging patterns, and how founders are navigating a world that’s changing every week.
Fraser is the former Head of Product at OpenAI, where he managed the teams that shipped ChatGPT and DALL-E, and is now an investor at Spark Capital. Nabeel is a former founder and CEO, now an investor at Spark, and has served on the boards of Discord, Postmates, Cruise, Descript, and Adept.
It's like your weekly dinner party on what's happening in artificial intelligence.
[00:00:00]
Cold Open
---
Fraser: The other way to think about it as well is that it is teaching us, us being society like people around the world, to not fear this new technology either. It's so joyful and joyous and fun and light that it is creating like real wonder and happiness. Welcome back everybody. Welcome to Hallway Chat. I'm Nabeel, by the way.
I've got the
nabeel: name of our episode. Okay. What is it? Chase, the next dragon. That also would be good, but no, that's your VC speak. I obviously, this episode should be called, what is it With dentists and Vibe coating.
Fraser Kelton: Oh yeah. What is, what is up with that? Remember, remember when we met those guys now a year and a half ago?
Yeah. And, and yeah, we got these dentists in Australia who are building custom apps, and then we hear about it again today. . That's so crazy. Yeah. I don't have enough dentists in my life, [00:01:00] obviously for why this is such a, a very common. Market of
Nabeel Hyatt: vibe coding apps, and we'll tell you, by the way, I don't know what your dentist is like, but when I talk to doctors, they are, you know, inside of Kaiser Permanente instead of a huge hospital and the overhead is so heavy , and trying to imagine 'em use technology.
Thankfully they're using a bridge. We're very happy about that. But, but they, I can't, it's, it's hard to imagine my dentist. You know, it's, it's perfect solopreneur kind of vibes, right? it's an interesting mixture where most small businesses don't have very much cash flow. And maybe it's this overlap of you had to be entrepreneurial enough to start a place.
Then they actually make good money. And so you have a little extra time to go mess around and think about how you want to build your own SaaS software for your dentist office using whatever new, vibe putting platform came out that week.
Fraser: Sure. And you can imagine also if you're spending [00:02:00] however many hundreds of dollars on a couple of seats for your SaaS software, that's like pure profit if you can cut it out too.
Yeah. Yeah, yeah. That's right. That's super interesting. They're the, the early adopters, they're like the New Zealand of social apps is people are going to go to find the dentist association as the group data. Right on.
nabeel: Yeah. Always soft launch your vibe coding app with dentists offices. That's the, that's the rule.
Exploring Poke, a New Email Assistant
---
Fraser: That's the rule. So you, you and I have been texting an assistant a little bit over the past couple of weeks. Do we want to talk about poke? Let's talk
nabeel: about poke. I think we should, we frequently talk about how we haven't had a lot of new apps warming their way into our lives. And I don't know if it will stick or not, but there's no question that it is very.
Good launch from them. There's a couple of good lessons with the execution. There's a couple people in my life, not in venture, that seem very attached and addicted to poke. I, [00:03:00] I don't think it'll stick for me for obvious reasons we can chat about in a second, which is mostly the vagaries of what it feels like to be a venture capitalist and the nature of the kind of emails that I'm trying to send.
Right? So Coke is, for anybody listening who hasn't checked it out, it's a. Just, would you call it an email assistant app? It doesn't appear that way at first. It's really supposed to be just your coach and your, whatever. It's supposed to be your friend OO over text, but I think it mostly boils down to a skill on email.
Is that right? Or do you know any anything else that's been good at for you?
Fraser: No, I think that that is right, but maybe that, maybe that's the MVP, like May, maybe they're Yeah, the roadmap goes well beyond that.
nabeel: Yeah, and, and it went viral a couple of weeks ago because of this. Very aggressive attitude, snarky attitude it has with you when you're first onboarding, and the fact that you have to negotiate the price that, that you get at the end.
And it starts usually with some h like ridiculous. Like, you know, well, you're some fancy venture capitalist. We should charge you [00:04:00] $5,000 a month. And then you have to kind of negotiate down to something reasonable. Did you like that whole shtick?
Fraser: Yeah. I, I, I actually thought it was a delightful personality and twist on on, well, it's all yeah.
Text base and it's delivered through text. And so you're pretty constrained on how you can have product design flare that's memorable or, or remarkable. And I think that they, they executed it really well on that. Right. I mean, the fact that we're talking about it now is partly because of how that was crafted.
That launch was. Yeah. Yeah. Yeah.
nabeel: It, it reminds me very much of the like, innovation that Zinga went through early on. That was the real unlock was with Farmville. Who was the lonely cow? Thing, which was basically you have a cow that's wandered off of your property, you can't find it, go ask your friends if the cow has shown up in their area.
Yeah. Which is ostensibly post to your feed, which might sound kind of rote right now, but was very novel at the time. And more importantly, [00:05:00] like understood the MO medium. It was swimming in, which was, instead of this situation where it's like, invite your friends and then please, you know, send a share to your feed.
It was. Yeah. Staying in the narrative of this world, this tiny little world you had made. And it's a thin veneer, but it matters. And here the the cool thing is that because it's text, it makes sense to be. Kind of banter and conversational, I don't think the same thing would've worked the same way on a website.
Fraser: No way. You'd be, you'd be where, where's the, how do IX out of this? And just get to the, the, the page. You know, I think the thing to call out though what you just said is, I think a, a really important piece is that the banter nature of it is the piece that I think we can be inspired and learn from the snark and like the specific aspect of the playfulness, I think.
We'll work this one time and then you'll get annoyed and frustrated with it every other time. Right, because it was, it was new, it was novel, it was clever. And you know, I, I worry that people are gonna learn the wrong lessons and you're gonna have snarky [00:06:00] negotiations with the next five apps and you'll roll your eyes and get frustrated.
The experience of hooking up your email male and behind the scenes, it is. It is doing obviously a bunch of different things, off of your email address. It's probably then going, I don't know, in a pitch you'd call it an age agentic type experience to research who you are. Yeah. And then the surprising moments of delight are when it interjects its negotiation with.
Those pieces that is gathered without you giving it any information. Right. And you're like, oh, how'd you know that? Come on. And then you can push back and it, it shapes its view of you and then you get to a, you get, you get into it. It's great. I thought it was awesome.
The Role of Personality in AI Products
---
nabeel: Yeah. The thing it signals, which is really important, is an area of weakness for the chat GBTs of the world, which I think has been called out before here.
We talked about it, which is just that. Where do you sit on the relationship to Utilitarian Spectrum and [00:07:00] because of the nature of the generic kind of overview of what Cha BT is, it's not that they won't try and put a breathy female voice in their voice avatar every once in a while or whatever they're gonna try and do, but it's, it's really hard for.
A product like that to present itself with very strong levels of personality. It, it'll be Google, you know, the, the, the in, in the, in the edge is trying to be a blue logo. Rounded edges, perfect thing. And yeah, understanding that that is a blast radius of where that's gonna be can kind of reveal the topology of where you might navigate.
that'll just feel unique and different and wonderful. It's the same thing. Reason I like things um, there's a company that has that AI for that is like a little alien called the Lins. It just has more personality. Then you could ever really imbue into something that is perfectly utilitarian. Like, like Chacha pt.
Yeah. And there's an edge to that. 'cause there's a place for that in your world.
Fraser: Yep. [00:08:00] I think the, the idiosyncrasies of our job means that after I got into the product, it, it's probably not the product for me is what I would sum it up as. Is it takes, right now, what, how would we describe what the actual features of the product are?
nabeel: It reveals to you very important emails that things you need to respond to. One of the nice things it does is, is just revealing, DocuSign links and stuff like that, pulling them out and putting them into the text. Then more importantly, of course, it says like, oh, I'll draft this for you, and then it tries to draft emails back, and so it ends up being this kind of from afar executive assistant tool, which, yeah.
Yeah. Same for me. the problem is that we're in an environment where. I'm not gonna write an email back to one of the founders I work with, you know, through this service. Like, it's just the, the words matter too much. Like the specific sets of things that need to be said are too critical for 80% of the conversations that I'm having.
Right. And is that, is that what you felt too? Yeah,
Fraser: a hundred percent. [00:09:00] Tab complete in cursor. A couple years ago worked when it was just like a function that you needed to finish writing. That was like very standard. And I think we're at the tab complete era for writing your emails as well.
Right? If you need that library called, that's mm-hmm. Known, right? And you just need to write the boiler text for your code. Like, that was great. Yeah.
nabeel: Yeah. Look, that that's the, that's the important thing as we talk about these things as technologies, like there's an end all be all, but email, , we were having this conversation the other week.
Raising the Floor on AI Email Assistants
---
nabeel: Email is like, you can try. And raise the floor on email so that your random, I don't know, talking back to a vendor and saying, no, I don't wanna do this deal right now, or whatever moves up. And in, in that use case, I think for most people, like 60, 70% of their email is just send the yes or no. Like the floor is low.
Yeah. And, and it's just, it's just rote communication. And [00:10:00] look, if this product moves that floor up a little bit. Then great. Like I, I use a product. it's not launched yet, so I can't give the name or whatever, but I've used a product for passing on candidates for not spark, but for this side gig that I have on the weekends.
And it's just very simple. Pass emails. Nope. Disposition is not for you? No, and, and I need to send them in bulk and it's very much similar to the sales. Automation outbound email companies, right? Kind of stuff? Yeah. Only, it's only in this case, it's one where I, it's it's open world agentic and I can roll my own and I can design it how I want and these kinds of things.
And for that stuff, like it doesn't need to be a plus work. It's just, you have to know what you're trying to be great at and, in that world, being great is not actually getting the email to be perfect.
In that world, it's trying to make you sending more emails, you know, maybe it's slightly AI slop emails, but, but making more emails being sent efficient and easy and good and not making mistakes, right? Yep. [00:11:00] Yep.
The Future of AI Assistants
---
nabeel: I, I, I really wish I saw more products that were trying to raise the ceiling. And this goes back to our conversations about, you know, is it an oracle or is it a muse? And, and a wonderful muse. Yeah. Takes something that's a minus work and is you know.
Sometimes you use this one phrase and it's very impactful. Maybe this is the place you should use that, or, you know, you're a slightly charting partner who says, I, I think you can do better. the things that help you stay in flow, give you coaching feedback, help to drive you to better outcomes.
And, and maybe, you know, maybe we're just not there in the cycle yet, but I, I don't even see many products that try.
Fraser: Yep, yep.
nabeel: Yeah. Have you used the coach Me feature in, in granola.
Fraser: I haven't yet, but I've seen some people commenting on how amazing it's actually, I saw somebody comment on how amazing it was, and then somebody had a reply saying that like, you know, it doesn't pull any punches.
And that's the reason why I've held out on doing it. I think I [00:12:00] need to be on my own with a, with a glass of whiskey on a late Saturday night before I hit Coach Me.
nabeel: I need to be mentally prepared. I understand. I understand. Yeah. I mean, it's all based on Macari. who's a, a pretty famous CCE coach, CEO coach.
They first built it internally as a. Prompt without his permission and then decided before using all of his knowledge? They would, they would, Chris would reach out and, and, and get an okay. And, and thankfully Mike was down for it. But it's quite good and it's a good example of that is not a product feature that's designed to help you send more slop.
That is introspective feature that's trying to help you be better. Uh, it's trying to help you get from a to a plus. Yeah. Both should exist in the world, but not everything needs to be efficiency oriented. Some of this is about us doing,
Fraser: you know, better work. Yeah, absolutely. I use ChatGPT and Claude as a muse to raise the ceiling, as you just said in mm-hmm.
Almost everything that I write, including, [00:13:00] including I'd say maybe 20% of important emails, which is a huge change for me. Like huge. And it's It's not the fear of what education has. 'cause I'm sure there are people who are Hey, write me this essay, and then they submit whatever comes out. I, I use it I.
The muse is a great one because I'm oh, I'm, I'm stuck on, on this analogy that I'm trying to pull together between these three thoughts to get this one paragraph to flow naturally. Yeah. And it, it just riffs with me and yeah, it totally raises the ceiling. And, the, I think the, a question for me is, will I.
Just naturally be going to these broad assistance, or will I want raising of the ceiling features integrated directly into Gmail? am I going to be like, let's riff on this right now?
nabeel: Oh, I, I, I think part of my contention is that the efficiency oriented plays, you know, send 10 more emails are excellent early in the market, but easy to copy and hard to build, lock in, and so they [00:14:00] get immediate revenue.
It's harder to build durable businesses. Not, not, not that you can't be done and there will be useful, impossible. but not impossible, but, but it, it's a different set of challenges. Early revenue's not the challenge there. Right? It's sticking out because you're really also sending a commodity product.
So things like efficiency and cost and all the rest of this stuff, you're building slightly faster. Horses. This is, it's so, it's just about your horse thoroughbred going faster, which is what we see a lot in the market. The flip side, if you have a partner who's now earned your trust to try and make you.
Better at writing this one email. That's super critical to you. Better at talking to your wife about a difficult problem better at whatever it is, but the thing, you are trying to be excellent at making a piece of art, making a piece of music, those things, there's this secondary emotional attachment where it's now.
You've now gone back and forth with it a thousand times on the tiny nuances of a thing, which give it so much context on how you wanna work and how you want to be great and how you
Customer-Centric Product Development
---
nabeel: want to be good. That whether it's actually really good at actually giving you, internalizing [00:15:00] that no model and giving it back to you.
There's emotional attachment from your end. Sure. You know, this is, I think there's real stickiness there and the shapes of those two types of companies will be very different. You know, this reminds me. Lemme pull up. Emmett Shear had a tweet a little bit ago that I will pull up now that kind of got to this idea.
So Emmett said, the way that ChatGPT BT is trying to be the portal for the entire internet reminds me a good deal of the first internet boom in Yahoo. And then I think the ultimate winners will wind up specializing just like the last time shopping, entertainment, search, et cetera. I think the broader point here is as a.
Company. Our startups need to figure out what they want to be excellent in. And I think there's a convergence theory versus divergence theory where it's like, are all these things gonna become one portal? Is Chacha PT just gonna eat every single portion of everything? And I think, yes, the inverse of what happens, I think that that [00:16:00] history of these markets have been that they start feeling relatively muddy.
Then as customers pull them in certain ways, they start to feel very different. Few years in.
Fraser: Yeah. I thought you were going to pull up a completely different comment from Emmett, so I totally agree with you, but now I gotta pull up this other one. 'cause it was also excellent. I love that. We should pull Emmett on clearly.
Okay. Increasingly, ChatGPT is much better at doing what I tell it to do, but I can actually collaborate with Claude. The difference keeps getting more stark a tool versus a being. And I don't know why I read that so dramatically, but I felt like it was apropos it. I don't know if you use both. It certainly feels that way, doesn't it?
Yeah. They feel. They feel different. Yeah. This gets back to what you were saying earlier about the push to utility is like the push to a tool, and I guess if you have 700 million weekly active users and growing, that's the gravity takes you there. [00:17:00]
nabeel: I mean, not to talk about your background, but this is literally something you called out two to three years ago, when, when you left OpenAI and we were having conversations about what was gonna happen with OpenAI over time you say, I think you were very clear.
With me, you're just like, look, I, you know, without revealing any secret information or this is just my opinion, but I remember at that point very, you're like the gravity of what ChatGPT is inside of that org. Those customers will pull them in a direction no matter what, say and where anybody else wants to do at some point.
Yeah. There's just so much gravity that num number of users that you end up trying to feed the beast and that will pull you in a direction. Sure. Yeah, for sure. Yeah. I, it, it, it's, you know it, when I first joined Spark. We had this early mobile social era and we had, you know, Twitter was rising up at the time and Tumblr was rising up at the time.
All these social networks are rising, and I think it's a great analogy to what, not just Chacha, pettier, Claude, but all of the vibe coding platforms, [00:18:00] all of the image models. I feel the same thing over and over again, which is the Pinterest and Twitter. In the early days, you know. Felt kind of close in a way that they do not feel like they're in the same universe today.
They're both public companies that are big. Yeah. You know, and they're both social networking companies really. And they are, could not be, you know, more different. And so sometimes these things feel muddy in the middle.
Fraser: Like, remember there was the really heightened moment where Twitter and Instagram were fighting over the social graph and who could share photos and how photos rendered when they were shared, they all went nuclear because they thought they were building the same product.
And Yeah. Right now, I mean, it would be laughable to even think of that. Right. And so your, your point here is, is these small little differences that we're seeing today. You feel they might be the same product, but the small differences in product design and how you're using them [00:19:00] on the arc of 10 years with continuous innovation and and growth and everything else takes you to a very different place.
nabeel: I had a really, you know, us one of those core memory moments as a founder years ago where. The product's not important, so I'll leave it out. But I, I had launched a product, we got into a million users pretty quickly, but it was in kind of a new market. who knows what this thing is gonna be. And I was on a walk in New York with an old friend who was running a company that was much larger.
Um, it's actually Alex Opolis, the, um, who we ended up investing in later on, but the, the maker of Guitar Hero on rock band at Harmonics. And, um, he had exited. Harmonics the year before for whatever it ended up being a couple billion, and we're going on a walk and I'm talking about, about this product and I, I was obsessed with what the competitors were doing.
I was they just launched a thing and oh, I wanna like an engineer. And he was just like, listen, it's very possible that all of you are [00:20:00] just bumping into walls and not finding avenues or alleys. He's like, what do your customers want? Like you're much more likely to find. The answer by not figuring out what your customers want at some abstract, high level, but really getting deep into where they're leading you.
'cause the truth is, it's possible, especially in a new market. None of you have the right idea at all yet. And so, I, I think that's great. There's a Steve Jobs quote. You have to start with the customer experience and work backwards to the technology, which is a simple, clean, repeatable. Mantra, almost like a, you know, like a guy at the gym telling you to do five more pushups because the thing you know is true and it's a truism, but also we don't think enough maybe about what that means.
For instance, I think a mid journey customer and a crea or ideogram customer are different customers. Even though they're all image models, they're just, they have different needs and uses and [00:21:00] they will lead to very different things. And I think that's probably true in, in an SDR platform. That is probably true in your vibe coding platform.
You're actually more horizontal than we think. And I think differentiation by listening to whoever's attaching to them is probably better than trying to chase the mistakes your friend is making last week.
Fraser: I think so. Totally aside, but I just got lost in, in the thought. As you quoted, Steve, what's the second order effect of our current tech heroes not being cut from Steve's cloth in the sense of like Steve was the soulful product.
Humanist who, what you just said is so amazing, right? And he just had so many of those always about the end user, always about caring, about the quality of the product experience. We don't, we don't have that. our tech heroes now are either exceptionally ambitious and great deal makers, or they're people who can summon AI gods from Common [00:22:00] Crawl.
It's just so different. I think
nabeel: my expectation is that, you know, five years ago. Was not the Steve Jobs era. You know, we get the leaders we need. And I don't think the ecosystem at the end of the Red Ocean of B2B SaaS was that kind of an era. And consequently, on the flip side, I really do think this is that era.
if you just think about the kinds of decisions that need to be made in a fog of war. It's just different and No, it's interesting. I, I think it goes back to some of the conversations we're having about the nature of what venture capital is today and the way these firms are shaped and who they will promote or not promote.
I had a pretty ridiculous conversation with a friend of mine who just became a VC not too long ago about the incentive structure he sees inside of his venture firm, which I won't share here, but, you know, it, it, it really was this kind of. [00:23:00] Did the person go to Harvard or Stanford? You know, almost like buying IBM was the way people talked about it in the eighties.
Like the decisions I won't get fired for. Yeah. When I think it's so you're just less likely to fund the Steve Jobs. And by the way, that's not to say that Steve Jobs was soft, but that, that he's hard. Like you still He is. Oh, for sure. He was a killer. Yeah. Absolute killer. But it's, are you, you know, what are you being a killer to what end.
And, yeah, yeah, yeah. And no, I think this is a, this is that time period where I think you need those kinds of founders. I just think the ecosystem will make it harder for them to punch through. And if you ask me in five years. Who are we gonna regale as our heroes or think about as our heroes? I suspect it will be not all of those people, but I suspect that will be the next generation.
Fraser: Yeah, sure. That's very inspiring. I hope so. I hope so. 'cause it's now a, a meme to dunk on is the intersection of liberal [00:24:00] arts and technology or whatever that like silly thing is that he had and like I I, it's old and it's, it's dated. That mentality is the same reason why we, like it's said, the reason why we have the iPhone.
Right? It's like the, it's the reason that, that these beautiful products have been birthed. Um, and I miss it
nabeel: by the way. I think we're seeing the inklings of that, you know, to start without calling out companies. For instance, I think David Holtz at Midjourney is. A hard charging founder who wants what he wants, but is absolutely a humanist and is a deep technologist as well.
I think Chris at granola is a hundred percent cut from that cloth. I could keep going. I think there's a set of folks, not all these companies will work out. If I could give you five more, but you're often, you're in the midst of a generational change or a seismic change and you can't feel it. And I, I really do believe we get the leaders that we.[00:25:00]
Need. And so I suspect that when we fast forward five years, the texture of the folks that are running public companies will map to the times, which will be a little different than what it was five years ago.
Fraser: Going back to the, the point that Tumblr and, and Twitter and Instagram all felt like they were perhaps the same product for a moment in time.
Mm-hmm. And now we look backwards and they're not, is there going to be one vibe coating product that wins the market? Or are we looking at a bunch of great companies that are going to be the Instagram and the Twitter in, in 10 years? Even more broadly is how, how to think about the divergence and the convergence of these technologies.
nabeel: Look, I think it's hard as a founder to listen to your customer. I think we should acknowledge that. You might find that your solopreneur, dentists, or whatever are your best customer and you wanna listen to [00:26:00] them. And then us VCs of the class, not you and me specifically, but this VC class is screaming at you to be a deck acorn.
Or larger. And so why would you build this thing that feels narrow, why you should build a horizontal thing that does all things for all people. And so I think the inputs into their feedback cycles are a little off. And because it's hard to realize that that often the way to something incredibly, incredibly large is focused, it is understanding your specific customer.
It is understanding product differentiation and market differentiation. You know, we were having a. Breakfast this week with a founder who, you know, it's a, it's a live deal, so we won't talk about it here, but you know, having breakfast with a founder that is in a very, very horizontal space, but you talk about how they think about their competition and why somebody picks them versus somebody else, and they're not afraid to make courageous choices about what they're bad at, [00:27:00] and that will.
Knock out a big chunk of customers. And I don't know, there are probably some VCs who look at that and are well, you're just not being ambitious enough. And it's like, no, I wanna win. And like I, I can get to those guys. I can get, I can go eat the rest of my lunch after I, I get the first part done. But I have to take, I have to be amazing at something.
You mentioned vibe coating. I think the thing below vibe coating is just that. This is gonna feel pedantic, but these models are really good at coding, period. So before we get to coding, like the root of this is that it was really good at chat and now it's really good at coding and it's just a new enabling technology.
And yes, the most horizontal version of that is to sit in a command line interface and have it code for you. But I, I, I think, I think that in many ways, the most, that's the most reductive, worst way of thinking about this market for sure. It's okay. You can write a bunch of codes for free for your customer.
What's that good for? And that's like, that's like a thousand companies, not like one company. [00:28:00]
The Evolution of Coding AI
---
Fraser: Oh, yeah. Yeah. Do you know what I mean? A hundred percent. Like there, there will be, um, you'll have your retirement planning AI that's going to write code, like, rather than putting it into a spreadsheet that your financial advisor or whomever can then walk you through.
It's going to write code for your specific use case and probably go and get all your financial information from all your different sources like age agently and come back, write bespoke code to do the calculations to show you like your situation. And so I I picked that only because it's hard to think like, that's so far afield from command line code writing.
Yeah. And there will be a, a lot of code written in a financial planning app.
nabeel: So, so if we take the idea that Instruct helped GPT models get, go to chat, you launch a thing called chat bt everybody has this new affordance. And they're ah, what can we use chat for now? And they run around and they try and figure [00:29:00] out what chat's gonna be used for across the market.
And we're two, three years into that cycle. And code, I would argue some people would peg code to a year, year and a half ago, probably with you know. With Claude, but I might even peg it to more recently actually, where it's really reliable, where you, where you're starting to literally just now last few months, get to the point where I can abstract myself out of the code.
I'm not constantly bug fixing. It just does good work. That's pretty recent actually. What are the lessons, having gone through this on the operator side, what are the lessons that you think the chat arc might teach us about the code arc and how it'll play out? Or maybe put more succinctly, what were the mistakes that you felt like startups made three years ago when the gun went off on chatting with LLMs as an affordance that we can, or the positives
Fraser: here?
Here's a positive. I think that [00:30:00] we have seen a bunch of really interesting companies, specialized chat. Into different markets and and are building meaningful products and meaningful companies because of that. I saw this awesome demo of a legal tool. I don't think they would care if I share. It's like gc.ai.
Specialized AI Applications
---
Fraser: Mm-hmm. And you'd look at it and you could dismiss it. 'cause it's a chat box that you start with and then you interact with it. No different from the ChatGPT or the Claude chat box and then. You watch carefully and they, they have written their own parser that gives you references and deep links into your PDFs, like with like sentence level precision.
Yeah. And so it's just a very small difference in a feature that the others haven't built and probably won't built this as well as you can for law. As this group and you're like, oh yeah. Like of course. Of course. This is a [00:31:00] beautiful technology for a whole bunch of things involving words and then customizing it on the margin to be even better within your workflow as a lawyer is amazing.
And I think we've seen that in a bunch of different, uh, like we've seen it in law, we've seen it in medicine, you know, and on and on. I think three years ago, people would've dismissed that saying, you know, oh listen, that's just going to be gobbled up by the main chat assistants.
nabeel: Yeah. I've been talking a lot recently about the rise of what Andrew Mason called Open World Agents, and you can almost see if you, once you start squinting a little bit, you can almost see it everywhere.
gamma 3.0 just came out recently, which is a AI presentation tool, and. It has instead of this chat bot interface that just can write text for you, now you can do sweeping edits across the entire presentation, right? Make it more visual, more charts, right? And it articulates a whole bunch of different things.
The Impact of Open World Agents
---
nabeel: producer [00:32:00] AI and music and de script as usual. They're early in everything. We're one of the first to launch this. An open world agent notion 3.0 is getting really close to this. They just launched as well. And the kind of root this, once I say this, I, I, I suspect anybody listening will start to see launches and start to you know, most of the most progressive organizations started about six months ago to four months ago on, on seeing this happening and, and releasing it are now are getting to releasing.
I suspect we'll have a avalanche of them between now and the end of the year that kind of most progressive orgs. And I think users will be attracted to that kind of work. And I think the, the reason why open world agents. Work is because inevitably behind most of those is one tool use and two code that.
That is the manifestation. It's not that the consumer inside of D Script or, or Gamma is writing That's right. Is writing code. It's that it's using code as a tool to serve the user's needs. Yep. And so it can do [00:33:00] things it couldn't do before.
Fraser: Yep. And then, and then to go full circle. The, the big argument that you've been making all day today, it uses those in service of those specific users' needs for that person's job, right.
And we are seeing that great companies that are, you know, targeting a vertical but are still pretty horizontal, uh, yeah. Are able to create a lot of space for themselves.
nabeel: Yeah, I mean look, you could just do this for the email category, right? We have poke we talked about earlier. I have a mobile app I really like using called Net for answering emails.
with ai there's superhuman Notion. Perplexity just launched an email client we're doing this, forget five coding. we're doing this in an email literally right now. And you can imagine just these two use cases. Are you trying to draft. An okay email quickly to lots of people or, you know, and that's, that's when I'm doing that.
I want, I want mass use. I wanna send one emails at [00:34:00] once. I wanna think of them not as an individual email, but I wanna think of them as baskets. And so I want you to be my, you know, my EA who's drafting a whole bunch of stuff and I can deal with it at a more macro level. Everything about that user interface, what you need to be good, everything about, that's different then.
I am trying to write to my boss about getting a raise, or I am trying to give feedback to my coworker about a difficult thing and yeah, you know, in one slap is okay. In fact, flop is the point is to kind of move quickly through things. In the other one, it's just a wildly different thing, and of course it's gonna end up as a different kind of product, right.
Fraser: It's gonna be amazing though. I, got a lot of faith that creative people are going to help solve these problems. I think we we're pretty good at. Here's a basket of things that I just want to have a whole bunch of, like tokens chucked at. And then I wanna send it all at once. I think the, the raising of the ceiling and the, and the craft one, going back to code, I, I read this thing on Twitter, which [00:35:00] I thought was lovely.
He used Claude to analyze all of a competitor's websites relative to his own. And then he took that output and he put it into Rept and then had it vibe code a website based on the feedback that Claude. Provided against like the counter position in the market. You know, I think we're at tab complete for these email clients, right, where it's just super stupid.
Let it finish the library and, and automate me having to write this boiler plate out. But soon you're gonna have something where it's it will identify that it's an important email. It will just throw tokens at it in a very thoughtful way. Like it will draft it and critique it and draft it base, a revision all behind the scenes.
while it's, while it's processing, right? It will critique it and prove it, iterate it, it will show you the final output, but it might be revision 10 that it's gone through.
nabeel: With knowledge of who the person you're emailing is. Yeah. And how they've reacted to previous conversations with that person.
And you know, the whole, that's like turns a phrase that seemed to work on that person and work on you and things you [00:36:00] love saying and just the whole thing. There's you know, for sure terminable depth of wonderful humanistic product features that you can already see it you know.
Fraser: In poke, like the fact that poke behind the scenes goes and gets who Nabeel is.
And so it's negotiating you with this insight, but it doesn't tell you that it has that until it reveals it as part of the negotiation. It's like that same mechanism is going to be working to draft your emails behind the scenes and it's, but we're just on, we're on the first turn of the cycle right now.
nabeel: Look, I think the other reason to talk about focus and to kind of bring this back to customer focus is that it's not lost on any of us when we talk to founders. And go through this journey that the bar keeps going up on execution. I, I remember to go, go back to that Facebook example, you know, Farmville was launched in two weeks and got to 40 million revenue in a handful of weeks and it was crazy.
Raising the Bar in AI Innovation
---
nabeel: And then. A year later, about a year after Farmville came out, CityVille came out and I think CityVille took nine months to make. [00:37:00] And it's the same thing in any market. Like the anti end goes up, the the what do consumers expect, what a businesses expect goes up and you're standing on the shoulder of somebody elses who's now been running.
You know, with customers for a year and so what might have been a three month sprint to get something quick. Now might take six months or nine months or a year, a year and a half 'cause you're just catching up to all the features that your competitor has done. Maybe we're in the scrums so quickly that, that it's our, maybe our job to help remind founders sometimes that that's gonna happen and the way, the way to not fall into a situation where you are gonna need months and months and months just to catch up with the baseline feature set that is now the new anti end to your market is to say, no, no, no.
We're just like, here's the thing. We're doing unbelievably well and this is why you're gonna use us. It's focus. I
Fraser: had a realization that's been brewing for a while. I think Dolly to nano banana Sora to [00:38:00] I think all these things are so important to society in this moment in time, and it, it has really irked me that we've, we've lot latched onto this meme of it being AI slop and everything else like that.
We're starting to see since GPT five Pro High, whatever it's called, comes out that it's able to advance science and math. That only the brightest minds in the world are able to appreciate. Like I read that, mm-hmm. It, it contributed to an advancement in quantum complexity theory. There's a thousand people in the world who can appreciate like what that actually means, and like to use that as, as a compass for where they're at.
But Sora too, uh, and especially because it becomes this global phenomenon, this mimetic moment for all of us. It resets expectations as to where we are on the curve of AI's frontier. And importantly, it helps us get a, a sense of the rate of change. And I think that that's so important for governments as we wrestle with [00:39:00] regulation and how to regulate it and how to even think about it is so important for enterprises and companies as they think about how quickly they should race to embrace this type of technology.
It's a gift, right? Like it, it resets. Everybody's global expectations in a way that you can't argue with. Like, you just feel it, you, here's the rate of change. I can see it because Sam's running around in a meme video. Right. Like that just, that's my, that's my rant, my observation for this week.
nabeel: Oh, you, I, I was trying to figure out how you're gonna get to, from Quantum theory to nano banana, and I, I think what you're saying is, are you saying that it's important because it's a visual manifestation?
To people in a way that they can rock and understand how much is actually moving.
Fraser: Oh, that's good. That's accessible. It's approachable. It makes it real. Like if you look at Dolly two versus nano banana versus Sora two, like you're just like, holy wow, that was whatever it was, two or three years. And [00:40:00] I think if you look at chat GBT upon release and then GPT five.
I don't know. There's literally a thousand people who, who can probably appreciate the edge of the frontier and how it's been moving. Yeah. And I do like, I, I think that that having a shared understanding is pretty important.
nabeel: They're, they're at the, they're at the front end of the PR war trying to explain the, in, in the old days, this would've been like a.
You know, this is what a journalist would do, or, or a famous author, they would try to sense, make the world, uh, of what's going on in some corner. And here you're just saying like, look, the technology is doing it itself. It's helping us understand. You can look at the guy with, with five fingers. You know, a year ago, and then think about it today, and you're like, look at this change.
Fraser: Immediate diffusion. Immediate diffusion of where the state of the art in the technology is. And I actually think that there's a direct line between nano banana and why we're not seeing like AI deep fakes. You know, cause all sorts of havoc around the world. And it's because the diffusion is not slow.
It's instantaneously because of these imaging [00:41:00] video models. Like everybody all at once understands where we are. And then you learn. to be skeptical, to be concerned, to be worried to be on the lookout. Versus if a thousand people knew about voice cloning and the state of AI image generation, like only a thousand people know about GPT five, five pro's ability to do this thing.
Like in quantum complexity theory, the world would be in like a much worse place. It would be a much worse place.
nabeel: That's interesting. I think about that. I think I, I understand what you're saying and I think it's actually a, a very important point. The other thing that I think these visual mediums and musical mediums and other kind of creative, popular culture mediums are doing, as we watch these models get so much better.
The other thing they're doing, if you just think about what the effect of it is, is they are teaching the world that technology and innovation gets better. That when something is broken, yeah. The answer isn't, oh, well that's never gonna work. The [00:42:00] answer. And this is the, of course, a meme, which is this is the worst it'll ever be.
And, and that might sound in the midst of the Silicon Valley world, that sounds almost like a truism. And what do we care? But I actually look, we live in a world where politically in the, in the US I think there's. We're trying to make sure that the next generation understands why capitalism is good, why innovation is good, you know?
Yeah. We're fighting on the flip side, Luddites and, and if you buy into all of that theory, then you buy into a different economic theory of how the US should work. You buy into a kind of almost fatalistic, different belief system. the pie isn't growing. It's it's a different belief system.
And so I, I think the net of what you're talking about is also, it's hard. You can just, oh, yeah. Literally see it in your feed. The innovation for sure. And you can see it in your feed, the betterment. And that might teach the average person way more about why we love doing the things that we [00:43:00] do in mass than having some conversation about how, I don't know, the efficiency of making tires has gone up two to 2.5 x in the last 15 years.
That's right. Uh, and that's, that's contributed monsters from gallon and so and so forth. It's so abstract.
Fraser: Yep. I I It's the same point that you're making, but the other way to think about it as well is that it is teaching us, us being society like people around the world to not fear this new technology either.
It's so joyful and joyous and fun and light that it is creating. I think real wonder, right? real wonder and happiness as we have these collective moments where new, new, new things are enabled that are entertainment because of these image video music type models.
nabeel: The flip side of it, of course, it makes people freak out about all the artists losing their jobs.
and so it actually kind of inflames the letter to a little bit as well in a way that you don't really get when you're just talking about these AI models are gonna cure cancer and so on. [00:44:00] What is your response when somebody. Talks about that. Do, do you have a way to talk to people about what? About artists and their jobs?
Fraser: I think there's two different things there. One, one of the reasons I was really disappointed, for the past four or five years of the extreme doom argument that this is going to end to the downfall of humanity was, I think it really distracted from. Questions around how do we think about ip, both for training and generation, how do we think about jobs in the short term?
I think it comes back to. In the simplest way for me, back to what you said at the start around, you know, you can either raise the floor or raise the ceiling and it can do both.
I think we're going to see. These become new tools that can do both, and so creative people are going to be able to use it in new ways to do like even more beautiful, creative things and itself will be viewed in time as an act of creation and, and artistry, and it will introduce new people to [00:45:00] being able to create as well, which is all good.
Nabeel Hyatt: This feels actually very related to the pre previous conversation in ways I don't think I would. Have really pulled out, which is my son is taking an AI and ethics class in his high school right now, and it's kind of great.
I've been reading along with him and so they're using this as a way to describe, I'm doing all the work, and they're using this way to describe, first of all, utilitarianism versus deontology versus social contract theory and virtue ethics and so and so forth. So they're just kind of giving ethical frameworks to think about all this, which is really a wonderful process, you know, and, and, and an excuse to read more.
About Kant my age, is a wonderful thing. But you know, one of the conversations I've had with them, and actually his teacher when I was visiting the school a little bit ago, is just some of these are abstract, theoretical, philosophical arguments, and some of these are actually just much, much, much simpler than we make them.
Teaching AI Ethics
---
Nabeel Hyatt: Is it cheating or not? Is it robbing us of thinking or not? Just think of it as if it was a person and if you told somebody that you had a [00:46:00] geography essay. Then you said, can you please write that and give it back to me? They have robbed you of your thinking. Yeah. Right. And so that, that's a bad thing for society writ large.
If you the student explain the concepts and the core arguments you're trying to get across to somebody else and they write the essay, that is call cheating in education. If the job is to try and get you to expand on your arguments and write in other environments, that's just called management.
That, that that's what a manager does. And I think we're all gonna become unbelievably good managers, quite frankly, like our next generation, my kids and kids'. Kids are gonna be so good at describing to people what they want because they will be, have been describing the models since they were tiny little babies.
Yeah. what they want for sure. And then, and then the kind of last bit is if you write the article or the geography essay and then. Some tutor or your dad or your teacher gives you feedback on the problems. Yeah.
Fraser: Yeah.
Nabeel Hyatt: Like this is [00:47:00] the same thing as the email conversation we were having earlier, right?
That is not, that's it. Cheating . That is gonna make me smarter. That is me getting better. And look, we don't want that in everything in life, but all of these things are gonna exist in society and they will just be like most technologies, all three of these things will be massively accelerated. Because of what we do.
Yeah. And if you're looking for opportunities as a startup, of course the low level cheaty thing is the first thing that happens in a market. Of course it does. Yeah. And then you get to stage two and you get to stage three.
Fraser: Yep. that's a beautiful, beautiful summary of, of what I very much believe and my journey has gone from maybe your second bucket to the third bucket over the past year, year and a bit.
And it's been, it's been marvelous. It's also why the Anthropic marketing campaign right now is. Brilliant. Uh, got so,
nabeel: so brilliant,
Fraser: so brilliant. Like everything about it I just love and I think I mostly love it because it's captured in a way that resonates [00:48:00] my interaction with these models right now, where I think I am arguably thinking more, thinking better because I have this and fewer things,
nabeel: you know, like to be clear,
Fraser: right?
Yeah, probably on the stuff that matters more though. Like on That's what I need. Yes. On the stuff that matters more. Increasingly, it feels like I have a great colleague who's got all the time and patience and we're co-creating this together.
One last thought on the AI and art piece because Yeah. Seth, that producer AI. Made an argument to us, which I've been sitting with, and I I really, really like it. And you know, everybody's oh, we're gonna have, Spotify where you just listen to an endless stream of AI generated pieces and it will be tailored to your specific needs and mood and everything else like that.
And his argument was. You're still gonna have named artists and you're still going to have shared experiences around the things that they create. They'll just do it via ai. And so there will be, I don't know, [00:49:00] Pasquale mm-hmm. As an artist who, you share their music, you love their music, you return to the music, you have all of the same relationships as you have with that artist's creation . Like that's a place where I've just completely updated.
nabeel: Music is a great example of all three of these industries existing. There is music as background slop. That is what Muzak was in the old days. Music. Yeah.
The Future of AI in Creative Industries
---
nabeel: That is what Study Beats is like. Of course there is music at that. Then there is music as as narrative. Attachment to another human being there is Taylor Swift released an album. Everybody's listening to it together collectively. you're in that narrative story, whether she came up with that guitar riff or not is kind of actually not the point.
In fact, used producers and other friends and it's collective and so on and so forth. And you know, there's a lot of work that a lot of the renaissance painters actually didn't do themselves. They had students. It's, it's a part of the brand, you know, of course that exists. And then the third thing that's true, [00:50:00] and music's a great example.
People just do it 'cause they love it. Like the nature of lots of music is right. They go to Guitar Center and they get a guitar and they build a whole studio set up in their backyard with their buddies and they play and it's karaoke singing with your friends. And you do it because it's an enjoyable activity in and of its intrinsic self.
Yeah. And a a, all of those things are true, frankly, for a lot more than just music. We just don't see it that way because it's too hard to make or so and so forth. And as those things get easier to make, more fun, now you see it in games. I could have literally had the same conversation about the games business.
There's games that you play because your brain is off and it's the end of the night. And you're loosely going through whatever mindless activity. Right, right. And then I could just keep going. And then there's the other two camps as well. So it'll be fun to watch all of those things go from loose meth to probably very specific and separate companies over, over the next little age.
Fraser: I was just That's exactly right. I was just thinking that is the, [00:51:00] the, the things that all feel like they're image AI or like music ai, there will probably be very different companies of different levels of success in each of those three buckets for all of these different mediums. Yeah. Amazing. Wow.
Listen on that one. I think we should be done.
You're like, that fell. That's great. Like. Close. Let's, let's let it close. Yeah. That, let's let it, let's let it close. Actually, let me, let me end on another very inspiring, upbeat piece of technology. I saw a founder yesterday, Stanford grad. He was born without hearing, and it's been his life's work to help solve hearing loss and mm-hmm.
He doesn't have better hearing aids, he doesn't have this or the other thing. He has totally normal. Looking glasses, you know, no cameras or anything on it. And then through a waveguide little thing in, in one of the lenses and then beam form, microphones. And [00:52:00] on device AI model, he does closed captions for the real world.
And so streaming in your eyesight is the. Caption of who's talking and it, it was one of the most, it's like, it's why you gotta love technology? 'cause it's so, it's going to be, there's no stigma. You don't have the, the hearing aids in there. They're even better experience in many of the circumstances I spent yesterday evening, just like so, so excited.
AI and Accessibility
---
nabeel: So
Fraser: excited
nabeel: we lose track of the rate of change. We should be looking more at at net banana, apparently.
Fraser: Yeah. And then, and then the good thing. Alright, see ya. See ya. Take care.