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.
Is it Adaptation, Evolution, or Revolution? Plus Shortwave AI.
===
[00:00:00] Intro
---
Nabeel Hyatt: I don't remember
it now.
I'm not kidding. You just looked at this recently,
Fraser Kelton: can you play it back? It was the idea that if you go back to these transitions over time with new technologies, new capabilities,
Nabeel Hyatt: new platforms
That's right. We're talking about, um, Avi Goldfarb who has a great book by the way, called prediction machines. And he talks about at the beginning of any horizontal disruption, you first get point solution changes where you drop in a process to an existing workflow.
And you later you get systematic change where you can change the workflow entirely.
[00:00:32] "AI Week" at Startups
---
Fraser Kelton: ---Hey. Welcome to hallway chat. Welcome. Welcome back.
Nabeel Hyatt: We should start with app of the week.
Can We call them App of the Week? We're not doing that. We should start with a product in AI that we've been using a lot lately. Thank you. A
Fraser Kelton: product that we've explored, that is built with AI. And that was shortwave. Right. We both spent time with that.
Nabeel Hyatt: ShortWave is an I've got an AI email client, like Gmail with AI integrated, that's been around for I don't know, probably like 3 or 4 years now. but they had an AI week and they announced a new AI feature every couple of days. By the way, have you noticed that's like a thing now, somewhere over the last year to 18 months? Companies want, instead of having the big launch day. Video like Arc did last week.
I'm getting a lot of let's do a dribble every single day for a whole week. What do you
Fraser Kelton: think that is? I don't know. Is it rally? The team is it you have a whole bunch of hackathon like projects that you can just. Bring out one after another.
Nabeel Hyatt: I don't know. Yeah, it's happened about Descript of it as well.
My cynical side is it's, the CEO ego of being annoyed that they spent all this time working on a product and they release it on Twitter and it like gets an hour. Of attention. And then it's gone. And so, you were like no, I want, I want a little more attention than that. How do I dribble it out?
Fraser Kelton: Maybe, I mean, it could be morale within the team,
Nabeel Hyatt: everybody gets their moment to shine. Everybody gets their moment to
Fraser Kelton: shine.
They're also like. None of the features are substantial. Yeah. Even a package of them up, it's kind of a collective. Okay. Yeah.
Nabeel Hyatt: But yeah, if it's a list of features of five things.
It feels less it's not version 2. 0. Maybe the nature of AI itself is kind of leading itself to this, which is like, I think Amjad has talked about this at Repl. it, They started with a kind of like co pilot y feature, but they slowly felt like what they don't want it to feel like is you come to Repl.
it, you write in Repl. it, and then you use the AI tool. Instead, you want it to feel like it's breathing AI. It's in 20 places, not one place that's right. And that leads itself to smaller features with interesting workflows that use AI. And so that leads itself to launching those kinds of features.
[00:02:43] Is it Adaption, Evolution, or Revolution
---
Nabeel Hyatt: I think
Fraser Kelton: so not to get ahead of, How I've been thinking about shortwave in this case, but it also feels like, we are living through a period where things are evolutionary and not revolutionary.
If I go back to a framework that you introduced. Uh, last year as to how we might think about AI coming into these products. Yep. And maybe you want to take the opportunity to lay that out and then we can talk about shortwave.
Nabeel Hyatt: I don't remember
it now.
I'm not kidding. You just looked at this recently,
Fraser Kelton: can you play it back? It was the idea that if you go back to these transitions over time with new technologies, new capabilities,
Nabeel Hyatt: new platforms
I was using mobile a lot as the example. That's right. We were talking about the difference between vertical category, disruption and horizontal category disruption. Right. We're talking about, um, Avi Goldfarb who has a great book by the way, called prediction machines. And he talks about at the beginning of any horizontal disruption, you first get point solution changes where you drop in a process to an existing workflow.
And you later you get systematic change where you can change the workflow entirely.
right.
Um, his example was when the steam engine was invented.
They just dropped a steam engine in the middle of a factory floor and then extended belts to all the points of the factory. And it would actually take 20 years for them to change how factories were built to account for the steam engine.
Fraser Kelton: Right. Right.
Nabeel Hyatt: So, so yeah, most disruption is in one category. It's vertical.
There's a new startup that upends FinTech or games or social networking. And most startup playbooks and most venture firms are built for these moments, but horizontal disruption throw out a lot of these rules.
To the phrasing we use at spark. Um, and we were kind of as a firm bill out of the mobile revolution. Which was arguably the last horizontal shift we use the phrase. Adaptation evolution and revolution.
So in the mobile context, adaptation is the New York times.com. Making the New York times mobile app. It's the same users, same behaviors and just slightly better workflows.
It's now like a thing that's right. In your hand, , adaptation usually favors incumbents, as you'd imagine
the evolution is Instagram and mobile games. It's similar users and similar behaviors, but they're building entirely new workflows. Evolution, usually favor startups, but sometimes fast moving incumbents can get there.
And the last category revolution. Is often talked about in the mobile context is a company like Uber.
Um, it's reaching out to a whole new market of users. While inducing an entirely new set of behaviors. And doing it with entirely new workflows. Uh, the difference between evolution and revolution is generally that the first feels kind of like an application and the second feels kind of like a platform. Anyway, where were we going with that.
[00:05:23] Shortwave AI - Adapting email to AI
---
Nabeel Hyatt: Uh, in the, case of shortwave. It
Fraser Kelton: feels to me like we are living through the adaptation, if you will.
Yeah. Not that it's not valuable, but it is, obvious. But useful. Like let's make that tangible. They have auto-complete and we'll come back and talk about that in some length. Yep. There's this really delightful experience that, that it just has a summarization at the top of your email threads, where it will say. Uh, probably a well tuned prompt because it's very tight and very concise and quite accurate and you look and you have eight replies back and forth and it just nails it super good.
It has, a chat assistant embedded that you can slide out if ChatGPT came into your inbox. Obvious stuff, adaptation. Still useful. The question that I had when I was using it is. Won't Gmail. Just introduce this. in two months, And won't. I have, uh, my preferred email client just deliver that and do I need to switch to this?
Is that enough To switch. to
Nabeel Hyatt: shortwave? Yeah, I think the features they announced were instant summaries. Uh, AI assistant for mobile. AI Autocomplete, Uh, and maybe a couple of other things. Oh, multi select action item stuff.
I used that yesterday and it was great. Where you can select 5 emails I see. and then do a query. Against those emails. Hey, can you summarize the topics of these things, which combined with search is quite good yep. Awesome.
Fraser Kelton: You know I don't mean my observation in a derogatory way. Right. Like I actually found it to be a delightful experience. The first time I went in and I saw that summarization, I'm like, oh, actually that's useful.
Yep. Uh, I triggered the assistant because there was a long email that came in and, uh, I clicked into the assistant and I had a question and I was just all of a sudden interrogating my inbox. Right. I was saying what else has this person sent me on this company? And can you help me summarize it and where the diffs are?
Yep. And it came back and it was great.
I actually think that the natural place for an assistant to enter our lives. Especially, uh in, work is probably through the email inbox. so the email client is a great just because it has
Nabeel Hyatt: so much context. It has so much
Fraser Kelton: context. We live in our inbox already. Uh, it is our to do list it's, where we process the information that we have to consume and then take action on like all of that stuff lends itself. Well to having a differentiated product experience in the, in that last bucket, the revolution one, I think somebody's going to deliver an amazing. Assistant within your inbox. Yeah. And you know, it could be shortwave. There was a, moment of delight that you and I had.
Do you want to talk about that?
[00:08:07] Familiar but personalized
---
Nabeel Hyatt: I was trying to Uh, shortwave autocomplete feature. Of course is mimics a feature that does exist already in Gmail, which is Gmail has had an autocomplete feature prior to LLMs. you know, In a, very different format.
Now it's not tuned against you. Uh, g-mail isn't, it's just a, what's the natural language, you know, thing in English that's supposed to come here. Whereas of course the promise for ShortWave is that it's supposed to be tuned against the email that you're writing. So yeah, we were sitting here and I opened up a founder's email.
You know, we're kind of demoing this and I'm just hitting tab to auto-complete and it's, you know, it actually used words that I would use. Hey, great to hear from you. And then of course it said cc'ing Lauren, which is the person I would CC sometimes. And I was like, oh, that's so. Good. It suggested a time Tuesday, 10 a. m., How about if we meet, the whole structure felt closer to the email I would write and. It felt magical. It made me want to go do that again. Now it turned out that like, it doesn't look like it was actually looking at my calendar Tuesday at 10 a.
m. was completely packed. There's no way.
Yeah, But if they just
Fraser Kelton: launched that on AI release week, give them a couple of months, they're gonna fix that.
Right. Right. And that's going to be amazing when it pulls in from your calendar, these things are going to figure out all of this
Nabeel Hyatt: stuff. Yeah They, certainly are.
I think, well, the thing that after we used it a couple more times, it became obvious is that it's not really Gmail. Auto complete the way Yeah. that auto complete is auto complete. it's closer to actually a UI change to something that already exists, which is a very interesting thing for founders. to Think about like, it basically, looks like under their hood.
I don't know how they're structuring it. we haven't talked to the founder about it, but like, it feels to me like, it's basically, if you ask ChatGPT To write a reply to this email. and you gave them some other emails that I had so you have a pattern of speech of the way that I speak? Then this is the email ChatGPT would write. But if you've ever gone through that process using another product, like superhuman has this AI product built in there's others that do as well.
The chunk that it gives you, Hey, let me just write the entire email. It just turns out, feels much less intuitive. And I feel like I'm editing somebody else's voice. And it doesn't feel like the way I want to write, whereas taking the exact, literally, probably very similar things under the hood and a very subtle UX change, which is instead of giving me the whole email in one chunk, and I'm looking at some email that a robot wrote and I'm trying to edit it instead.
I'm inline writing and tab auto completing. Yep. It's a subtle difference, but it keeps you in flow, I would use it all the time. Yep. Yep.
Fraser Kelton: You, had a visceral reaction
Nabeel Hyatt: well, because when it gets it right when the tab gets it, right, it feels magical instead of looking at three paragraphs and realizing the four things that are wrong. I was like doing a reply back to a friend of mine and it was like, Hey, looking forward to seeing you at GDC as the like, rest of the sentence.
And I was like, how did it know I wanted to see that guy at GDC? it was just a great little. moment. here's a
[00:11:09] Speed is a feature too
---
Fraser Kelton: case where, uh, Polish, uh, like what is, it that you're polishing when you deliver these products? Especially with AI and I, my biggest issue with it, uh, was not the hallucinations of like here's a time that doesn't work on my calendar. Right. they'll iron that out. The thing that will prevent me from using it in the short term is the same issue that many people don't use code complete when it's not quote unquote polished is the latency was too long. So I have to would pause.
I'd get out of the flow of writing my email. in would come the sentence and I'd say, yeah, that's right. Or I'd wait for it. And I'd say, oh, that's half. Right. And I could have just written the sentence within that time frame. But again like that's easily
Nabeel Hyatt: fixed.
Well, it's easily fixed except that this is like always the nature of what's really hard about product. is you don't know if that's MVP or not. Yeah. It's very possible that it's just slow enough that people don't use it. That's right. And then you misconstrue that for a broken feature, a hundred. percent. When the answer is not that people don't like it it's that no one wants to wait after every sentence for an extra half second to see if it auto completes the thing that you want.
And if you just built a much smaller model on top of it, which ran at three times the speed Oh yeah. and felt fluid. And if it, felt fluid, therefore you stayed in flow, which is important when you're writing, then you would use it 10 times more and it suddenly, you just took a tiny bit of speed, turned a thing that was a failure into a success.
Yep. Yep. And that's
Fraser Kelton: that, that is where I think that Increasingly the, we think of these things as full stack in the sense that you would trade off quality of model. Uh, on some level for improvement in latency or reduced cost. or reduced cost. Yeah. Because if it's right, 50% percent of the time, but you get that, that nudge in flow right.
When you need it.
Nabeel Hyatt: Yeah.
Fraser Kelton: Versus right. 75% percent of the time, but you have to wait. Yeah. You know, 600 milliseconds for it to
Nabeel Hyatt: appear. Well, you mentioned copilot and Replit is a great example of this, right? He said, publicly he spent like 100, 000 on their first model. Yeah. And it certainly wasn't as performant. at a 3B model size, as, say, co-pilot was, but against their own code base and GitHub, it was good enough.
And what he got instead was like very fast speed. And, lower cost, right? Yep. Yep.
If you're trying to like say, well, isn't Gmail eventually going to launch this. It's possible. But as we've seen with perplexity, sometimes large companies executing against what you think is a bullseye is not as easy as you think.
and that is, possibly, as you've said really well, like when Google tries to ship Gmail, AI tools, They may ship exactly copies of things shortwave and other companies are doing. And then therefore, you're just going to stick to Gmail because it's the default. But they also might ship their org chart. And it just might be flubbed.
And I don't, I think sometimes we try to get too smart for ourselves. And instead of kind of just pulling the thread, especially when you have a really high fog of war situation, when you don't know this is a first sediment layer on what is clearly adapation, , let's be obvious. Like in that earlier framework, I talked about, this is closer to New York times than Uber.
So I don't think it's enough. But it's, you don't know what the next sediment layer on top of it. is, So as long as you keep searching for new workflows and for new ideas, I still think there's a world where it feels like you're in a very competitive environment and the big guy's gonna go stomp on your face tomorrow and all the rest of that.
And then it turns out that like for a variety of reasons, all the incumbent players. Do half assed versions of the product. They don't quite get there for years. And meanwhile, you then build the agent on top of the agent on top of the agent, which is just like the email, assistant that you almost never touch.
that's what I really want. I just, I want an email assistant in some number of years where I'm not looking at the email. I like the list view. Of a cacophony of headlines screaming at me. That's not the way I should start my email experience. I don't know what the new experience. should be in that world, but it does feel like the capability of an LLM gives us an opportunity.
to Reinvent the workflow from scratch.
Fraser Kelton: I agree with that. I think that this is going to be the product surface area where a very compelling revolutionary product enters our lives.
Nabeel Hyatt: That's what you're saying, because email has so much context of our life. It's a great. Vector It context
Fraser Kelton: of our life.
it's our communication channel. It's our to do list. if you think about what this technology is already good at it's good at. Chipping away at those things.
[00:15:51] Taking users on a behavior change journey
---
Nabeel Hyatt: Although you can imagine being a founder on the opposite side and you pitch a completely world changing different view. of Email.
And I can imagine that conversation with you Fraser. Like I bring it in and I'm like, Hey, guess what? I've invented an email client where you never even see an email headline and getting that level of behavior change. Is tough, right? Like, The inverse is tough. And so sometimes it's a, sometimes it's like a path along the IDMAs, right.
You start incrementally adding on features. And then eventually as users come to you, that opens up another door. You would have never done this before, but now that you're with us on this journey, here's the next step.
Fraser Kelton: Absolutely.
Yeah. you're down for that. Yeah. I think that they, have a sidebar that comes in and you have your chat assistant and it's like, imagine if ChatGPT came into your inbox i, think that that is a very viable path to go and realize the opportunity here.
Nabeel Hyatt: Yeah. It's the one that I remember that makes me think of this is actually in the very early versions of Descrip when it first launched. You know, descrip is an audio, video editor.
It's an entirely different workflow, right? It In versus what used to be in audacity and all the way back to avid. you would look at a waveform of the sound and you'd edit the waveform. And that's the UI. Yeah. That it existed for 20, some years, 30 some years before descript came on the scene and now is of course becoming more of a default, but when they first came out you'd have the script view that you see.
Now, if you open up Descript, but at the bottom of the tray, they would have the waveform that You could still see if you wanted to do such a view. And then slowly over time, now you look at it and like, you can still, by the way, double click a word and still get to the waveform. But it's slowly been buried, deeper and deeper. And I'm curious if they had started on the inverse, if they had started with it completely gone. If the average this American life editor would have been like this. is So alien. I want control. Where's my control. So you kind of got to take them on a journey.
Fraser Kelton: I agree. I think it is a great example of how quickly AI is going to impact all of our lives. And then, so you asked the question of, will I stick with it? I will stick with it until I don't, because it is compelling enough today, even just that summarization at the top of the thread.
Yeah. Where it gives a little bit of context. So you can just dive in is useful.
Nabeel Hyatt: Being able to. Isn't it so funny that like you made this massive sea change? Of technology. That's a horizontal shift in GPTs that have now been copied elsewhere. It's all over the place. And I gotta tell ya, like, It's still like summarization is gotta be like 75% percent of the use cases. Well,
Fraser Kelton: People like convenience, which is the polite way of saying people are lazy.
TLDR is a product. TLDR.
Nabeel Hyatt: let's
[00:18:37] What are Agent workflows we should default to
---
Nabeel Hyatt: segue. I want to talk about an area that's that there is no really fixed workflow yet. We have been talking about. Really all of these like app. Builder no code frameworks. And this is everything from LangChain to respell. You know, you could call it like Zapier is a category, uh, Lindy there's like there's a lot of these things that are really trying to say, how do I build a unique workflow in AI?
A starter question, which is what do you when you're going to fire up one of these products? You already talked last week about how you didn't. really believe that. agents booking plane tickets should be the canonical thing that we all try and strive for as our default. And yet they keep launching with hey, here, We can book the like send you slack message every three days when somebody else tries to do the book, the plane ticket. What. Is a good example of a thing. Like when you, if you're gonna fire up respell tomorrow, What is the test case that you would use to try and integrate into your life? Just like, realistically, what do you,
Fraser Kelton: yeah.
Listen, on, these, I've tried a couple of different things where I've tried to do help it schedule so I can CC in the assistant and it will stitch together the workflow to help the third party schedule time on a calendar. Yep. It's too brittle, the non deterministic nature of it means that it fails once in a while.
Yeah. And you just can't tolerate that in that workflow, or I can't Right. The other thing is summarizing. I think I talked recently about summarizing. For the weekly email and synthesizing across the documents. and then sending that out. It. just didn't reliably create. The output that I wanted and it was finicky. And so I was updating the, pipes behind the scenes to try to like, get it to work.
And I just realized it's easier for me to just do it myself. Yeah. But those are some of the things. How about you?
Nabeel Hyatt: Yeah, I kind of want to come up with a standard. Set of at least base workflows that like. I would test every time I use one of these. How does it work in X? I certainly try, we have to write a weekly email.
Sometimes it feels like a chore. It'd be great. If there was a thing that gathered data. but that is hard, right? What that really means is look at my calendar. Look at every meeting I had this last week. Then look at the prior notes I have about that founder or that company. then it needs to write it.
In my tone. It's a, maybe a little past where we are technologically right now. GPT 5 maybe gets us there or 6, or whatever. But like, We're not quite there yet. I hate the workflow of book, a calendar appointment. and I specifically hate it. Not because it doesn't work, but because the act of it working is not, it kind of goes back to the kind of comment I come back to, which is like what I really want LLMs to be good at. Is being creative with us. Whereas, actually this feels like a rules based problem. Looking at my calendar and figuring out that I have 10am and 11am and 12pm. And then maybe the one creative language part of it is like drafting the email to the person.
But it just doesn't actually feel like that interesting of a problem set and you can't solve it. So that doesn't feel right.
Fraser Kelton: My observation after playing around with a lot of the web agents and realizing that I never wanted to book my trip. Is that, I think the real opportunity in the short term, based on where technology is and what users want is around. Research and synthesis uh, or research and summarization, and I think that's what perplexity is doing.
If you think about it, right. Is Google shows you the kind of nudges you into corners of the web, where it's like, Hey, we think the answer. Is in these five links or these ten links. Yep. And, perplexity goes and actually does the research and then synthesizes it, and gives you the, answer that you want.
And I think that's where the agents are good enough today. So then my question is what in our life. Can we automate that is predominantly a research and synthesis
Nabeel Hyatt: task.
I think the one that comes to mind is , be my super human memory. The reason that comes to mind is because one, it is shipped. In prior to LLM versions, There's the there's always been going back to, like, there used to be, start ups that would sit as a Gmail plug in, Reportive is a good example of this that would, that sold to LinkedIn. That would look at the LinkedIn of the person. Before you have the meeting, so you get a little bit of a bio, right? There's a pre LLM version of this. And there's lots of very bad LLM versions of this. But it's a good example for me.
If I'm about to go meet uh, CEO I've been working with for four years, there's actually a scale of zero to a hundred about what that output of that product could be. And it's one where synthesis and summarization is really important. And the, we've talked a lot about context where the context is derivable, right?
There's everything from the very base case version that's bad is, you know, the stick, the kind of like fixed rule based ones. what's the title of the person from LinkedIn. What's the date of the last time I met, like that's the base case. That is not good enough.
And you can imagine that the superhuman version of that is drawing context from all the notes I've taken, it's picking up over the last two years, what are the themes that come up a lot? Right. You can imagine an unbelievable version of that. Maybe it's a, I just yeah, this is just coming up with this today, maybe that's a good example of, it's also one where the more pipes you have in. Uh huh. You're gonna have very different. Outputs.
Like if you can get notes. Great more context. I can produce better output. If you can get transcriptions of the last four phone conversations we had or zooms, great more context. If you can get, who are our friends of friends and then what did we all talk about? Like that's great more context and all of that would lead to something that probably feels more and more magical.
Fraser Kelton: Yep. Yeah I, that makes sense. Uh, And the thing that's also nice about it is that it probably meets the technology where it is today. In the sense that you probably can get value by go to your email archive, go to your calendar, go to your notes and just. Rag that into the LLM and give you a summary. right?
And then you could imagine,
Nabeel Hyatt: And you know, we could do that in many of these products. And, it would be exactly what you expect, which is like kind of B minus work. And then, there's this feeling of a really well integrated system with the right sets of context, that's looking for insights and setting up patterns and maybe it knows me as well.
And what I need, or maybe it knows my job. Like me as an investor is a very different set of context needed. If, I'm a therapist or a professional coach or a doctor. Like, Like it's a broad, horizontal set of things to solve. that Inside of that, you might get wildly different workflows. Yep. Yep, with similar outputs. Yeah.
Fraser Kelton: And actually what you care about in that summary? And synthesis is going to be different than what I care. I would hope so. Yeah. Right, right. And so that is maybe another trait of when build your own. Productivity or workflow tool from these horizontal players really makes sense, right? Because you and I aren't going to go and use this, ideally we're not going to get value from the same, like SaaS software that does this for us. Yep. you're going to say, I actually care about this, that, and the other thing, and, you know, you care about. the heavy iPhone app equivalent of mail in this use case.
[00:26:04] TLdraw & Notion, over Pipes, as the core UX for AI
---
Nabeel Hyatt: This is actually a great way to bring up the thing I've been thinking about all week, which is that. Most of the interfaces into these no code workflow, AI products. Are, you know, basically derivatives of a Yahoo pipes. View of the world.
Yahoo pipes was, an early. Go Google it. Yahoo Pipes was a, You know, Katerina fake built like, a, you can import APIs and then it's a really like engineering view of the world. It's if then statements. You know, in a GUI. in a GUI. Yeah. I, arrange little boxes with little prompts inside of them, And then I, stretch them out and. And look, it's the way I often think about problems, but the thing I've been thinking about this week is like, that's such an engineer's view. of a problem set or the world.
And it's just like, not how. People. Problem solve. Not generally. It's not how engineers Problem solve. It's not how generally people workflow out things. If I sit down with you. And we are going to design a webpage together. The first version of that. It's even if it's an application, even if I'm designing Twitter for the very first time. I don't start from little boxes of if then statements, Like what do you do, right.
Fraser Kelton: You sketch the UI and you describe it. That's right. Yeah. Here's I'm picturing this. And then if you do this, then this happens and
Nabeel Hyatt: here's the other. thing. And then you Work backwards into the workflow. Yeah. Right. You start from, what does the output look like? And, it's very interesting that we don't have many examples of that.
All of my workflows seem to be falling back into notion. And I was trying to figure out why. And I think this is why I think they're falling back into notion. Not because notion has like the best AI tool. It's not doing the best contextual search or the best summarization, or anything like that.
It's not even that crazy good at it. Is that if you think about notion, it's a little bit like Excel, notion is starting from the end and working backwards. Right. I'm looking at a notion page, And I'm trying to lay out what I want, the final spreadsheet or database or page to look like. And then I have to hit, edit, prompt, and go under the hood and try and figure out the workflows to get it there. And it's still very primitive and you can't do very complicated things inside of notion. But at the very least, it's starting from, I think the UI layer or the display layer and working back
So if we go back to the use case of your. Getting ready for your next meeting.
Okay. Use case. Uh huh. And you were saying earlier, like Fraser, what Fraser wants when he's going into his next meeting is going to be. As a meeting brief is different than what Nabeel wants. Like there's two ways of trying to attack that problem. One is, Fraser, Great. What are all the, if then statements to generate the like briefing email for you, right?
That's a kind of, you can already imagine, like that's kind of Byzantine and it's problematic. And so forth. The other way is Fraser, Why don't you just write the perfect briefing email? Of how you would want to be briefed, right. And then can we work with. You know, can we work with an agent to figure out how that's generated and go Backwards Can you
Fraser Kelton: yeah. I get it.
So rather than saying look at my calendar. Find the name, go to my email. take the content.
Nabeel Hyatt: Da step 1A. 2, And then if then go step 2A. B. you're saying
Fraser Kelton: I can write what I'd like to see the output. And then basically say, go figure out how to do
Nabeel Hyatt: this. I think That's the way people think.
Fraser Kelton: it's the way that a lot of people think
Nabeel Hyatt: for sure. It's the like,
[00:29:43] TLDraw as metaphor for "show me the output" application building
---
Nabeel Hyatt: TL draw. Make it. Version. Oh, Nice. Yeah. Of AI application development. share what TL draw is. ,
TL draw. If you were on Twitter last summer, you probably saw it go crazy viral. It started out as like a sketch, like prototyping tool where you could draw in a browser and then they released the AI tool You can do what we were talking about earlier. You can draw an application framework that looks kind of like Twitter and say, make Twitter right now. And it will try to write code underneath the hood to generate that output.
Obviously it only works for very simple applications and it's more of a cool awesome demo right now than a fully fledged thing. But it's very evocative of something that feels exactly like this, which is just like, make this.
Fraser Kelton: So you, go into like a canvas and you. Draw some squares, some circles and you add some buttons
Nabeel Hyatt: Yeah, in order to make Twitter in this environment, instead of starting from code and starting to write out everything in Python or whatever you're doing, you're opening up a canvas and you're just almost like you're in Figma or Adobe illustrator. You're drawing the input submit box and you're drawing the header, and you're drawing a scroll bar on the right hand side, and then you're pointing at all those things on a layer on top of that and saying, this is a submit box you type Into this box and then hit make it, and it will try and infer what you're trying to make out of this product and produce the code.
Huh. And.
Fraser Kelton: Can you muck around with the code after the fact, do you get to see the code and edit the code?
Nabeel Hyatt: Yeah, you get to see what's produced. Yeah, Nice.
[00:31:13] History of Web Development as analogy for AI no code
---
Fraser Kelton: Nice. Yeah, I can get behind that. You know, The other way to think about this is you. Tightened it up by saying, that the product experience for most of the no code, low code efforts that we're seeing is still not accessible enough, right.
It still requires some level of programmatic. But then you observe something that I thought was really interesting is that it doesn't give enough flexibility and power to people who are actual capable software
Nabeel Hyatt: engineers. Right. Yeah, it's In the muddy middle. It's kind of in the muddy middle. Yeah.
It's like not enough detail and control for somebody. Who's really a coder. Uh huh. And it's not high enough levels of abstraction for somebody who's not. a coder. Yeah. A good analogy of this is where website development evolved over time. For a long time website development was trying to find a way to lay out products that both engineers and non engineers could use.
But if you look at where web development is over the last four or five years, what ended up happening even in no code, was you had a little bit of separation. You have Webflow, which is a GUI interface, But is a very technical interface that is really built mostly for people who work in design agencies who have coding backgrounds and have been trained on it to use.
It is not for Jimmy up the street to go build a website. Yes or Squarespace. Right? So. It's like it Barbelled even in a kind of like no code GUI framework, it barbelled to a completely templatized, very little control approach Which is the Squarespace version and then Webflow, which really is, if I had looked at that five years ago, I would have assumed that was just, that's not going to be the product that takes that's too technical.
Right. I would imagine that it might evolve here, which is what is the AI application framework, or what is the. AI workflow tool for somebody who is a completely non technical person. Which probably feels closer to Squarespace, or I would argue closer to Excel or notion where you're actually dealing with it at the display layer first and going backwards.
And then an entirely different likely winner and product is aimed at the engineer who still wants to use. Tools that will make them faster And is happy to use GUI but can get under the hood really quickly. Yep. I can dig that.
[00:33:37] Is it worth talking to VCs when you aren't raising
---
Nabeel Hyatt: I have one last thing I want to bring up today because it just, came up and I don't know what we're going to say about this, except like, I just saw this tweet and it like hit me right in the feels. Brand new minted VC a day old, two days old. Nikanj Kothari who just joined, the ranks of the dark side.
Said like, I'm just going to read this tweet for a minute and I don't know what we're going to say about. this because it just like really got to me, like he says, I'm overwhelmed with all the support.
Thank you. But, you know what the saddest part about announcing that you are now a full time investor is, he says it's story time for as long as I can remember, I spent my nights and weekends trying out new products, testing out user flows. Testing out the user flows, taking screen recordings and sharing feedback with the founders that I love.
And then he goes on to basically say that his hit rate with those founders was like 90%. percent Before Like, he read a thoughtful thing, you send a loom, and they respond. And then you get into a nice interaction. Sometimes that led to an angel investment or a job, or just a relationship. And he said basically in the last 24 hours, a lot of those emails have turned from founders have turned into. Hey, let's chat later, but we're not looking for investments right now.
I'll see you next month. We're heads down that, kind of stuff. And I don't know, like, it just, I really felt that, I used to mentor and talk to a lot of my peers, they would mentor and talk to me.
And, the moment I became a VC, like it was crazy, but I had this magical power. And all of my friends companies went from struggling to like up and to the right. Like, suddenly all of their problems disappeared, and they were doing really, great.
Fraser Kelton: What a crazy month that was. Everybody's killing it.
Uh, yeah, listen, I feel for that guy, right? Like, you're a curious tinkerer who loves this
Nabeel Hyatt: stuff. And look, that's not most investors, right? Yeah.
Fraser Kelton: Oh, yeah, right? Well I'm, empathizing with them because you just want to, you want to show that you like what they're doing, and it's as pure as that, and then all of a sudden you're put in the bucket and you can't break through.
I don't know, I've been a founder, my inbox Exploded on any moment, right? Where it started with real estate agents, but then it became real estate agents and recruiters and lawyers.
And now we are in that bucket as well. Yeah, that's fair. And there are teams, swarms of teams, uh, who are just tasked with
Nabeel Hyatt: outbound. Especially at the large crossover funds. Of course. Yeah, right? And now they're generating ChatGPT emails to ping everybody on a regular basis. That's right. It's overwhelming.
That's right. Like,
Fraser Kelton: they're outbound sales, really. Right? And how do you, I think the challenge right now is there's an awful lot of great advice for founders, and you need to, like everything, interrogate from first principles when you're going to always follow the median advice and when you're not and this guy probably should have people engaging with him some of the time, but he's put simply into the binary bucket of, Uh, I'm not raising, therefore I'm not going to speak to this person because that's not what I do right now.
Because he is one of dozens of emails coming into
Nabeel Hyatt: this person's inbox. Yeah. Most advice is right most of the time, and then you have to have some level of understanding of when that advice might not be right, and maybe, look, maybe the founder he reached out to just didn't connect with him and want to talk to him, and that's fine, but at the same time, there's a big difference between, hey, I saw your announcement, you know, I'm doing a survey of the space, let's chat, and I don't know, I don't know, and Uh, three paragraphs.
Like, I had this happen two weeks ago. I sent three paragraphs of thoughts on a product I had been using a lot that I just was really interested in and curious about. And we went back and forth a couple times on email with the founder. He did not immediately blow me off. We just engaged on the product.
And it was like, hey, do you want to get together? And then he had a maybe we should do that in three or four months, not sure, I'm raising, you should know that I'm not trying to raise right now, blah, blah, blah, and like, I'm not trying to preempt a round, I'm just literally trying to have a conversation about product and not be transaction oriented, and if it turns into a situation where we invest over time, like fine, if it doesn't, fine, it's like, not trying to do the. end result thing first, the disservice I think founders do with treating fundraising as purely transactional is that's the best version that they're going to get of the relationship.
That's right.
Fraser Kelton: Yeah, that the
Nabeel Hyatt: it doesn't mean they have to be, spend time with everybody and if they don't have a connection with me or you or whatever then don't. I don't know, I think of fundraising as BizDev and I Recruiting, not sales.
Like, it's not that you have a window for a week and you're going to go out and see what the market price is for the thing. That's right. You're trying to recruit somebody to your board for the next two to twenty years. You want to know them well before you go through the term sheet dance, you know?
These things
Fraser Kelton: are super messy. They're never up and to the right. Even when they are up and to the right, there's catastrophes happening all over the place. Don't you know it. Yeah.
You want somebody that you trust, that you respect, that you have good rapport with.
Nabeel Hyatt: But I get it, like the flip side is like, easy for us to say when we're trying to get attention from the founder. Of course. The reverse side that we both felt is the They, yeah, there's like just too many of you.
I'm sorry. I can't, I gotta actually build a product here. I need to
Fraser Kelton: actually build, I need to talk to customers, and then it's annoying because you get onto a call and it turns out that you're just, uh, you're educating somebody to become a, logo in a market map so that they can then do their content marketing the next week.
I get all of that. I have empathy for all of that. I was as quick as anybody just to like ignore a lot of it. But I think the blanket advice of don't engage at all. is not optimal and that, like anything, that you need to be a little bit judicious in who you're going to engage with and why, and with intentions.
But then, do it. And then you should do it. And you should build rapport with the, small subset of people that you think could be that great partner for, the, like, ten years that it's going to be.
Nabeel Hyatt: Yeah, the other reason that I think you do that is because, you know eh, It's very hard to get a read on a VC in the first meeting.
And we get, I give this advice to founders all the time. We're usually early stage investors, which means we spend more time with a founder helping them raise their subsequent rounds than there were rounds before we've invested. So I've just, you know, spent a lot of time with a founder trying to walk them into, okay, so now how do we think about the Series B and the Series C and who should you talk to?
And the one thing that you gotta be really careful about as a founder is like, everybody gives a first meeting that is great. If you're a VC and you can't give good first meeting, like, what are you doing? Like, they do first meetings all week long. And the question is, can you give good second meeting?
Can you have real kismet about this product and this company that lasts once you're outside of the scripted first, 30 minute Zoom call where they're like, Oh, that guy is impressive. Oh, he had a good hot take about my company. Can you really get depth? Because that's honestly what the relationship is going to be like, you know, going forward.
Yeah,
Fraser Kelton: I think I'm like a B on first meeting. You're still new at this, man. Just wait. But that's okay. From my new perspective, I don't want to be with a founder that is running a shotgun auction. And it's like you have 24 hours and 2 meetings to make up your mind.
Because I'm going to do a very small handful of these every single year, and then be absolutely committed for a decade. However long the course takes to run. Yep. And I want to have the rapport and the respect and the trust that comes from a good second meeting, a better third meeting, and a great fourth meeting.
Yeah, although
Nabeel Hyatt: to flip hats, again, it's easy for us to say, and a founder is like, Yeah, that's because you just want to be slow, and you want to take your time and wait and see if any other terms should show up. Or all the other muckety muck that investors do all the time. I've
Fraser Kelton: never once thought of that.
I know, but
Nabeel Hyatt: I,
But this is why the market's developed the way it's developed over time. Of course. There's enough ambulance chasing Uh, I want to figure out whether XYZ Firm is in the deal first, who's leading, bull, uh, that it's hard, you know, I hate to do an analogy to relationships, but it's relationships.
If you're going speed dating Monday, Wednesday, and Friday, and it's year two of speed dating, it's really hard to show up on a Tuesday and be like, Is this person, here for a one night stand or is this maybe the person that I'm going to be with for a long time? I don't think I've
Fraser Kelton: told you this, one of my, one of my rules that I have, uh, uh, actually stuck to is I have never once asked, who else are you speaking to or who else
Nabeel Hyatt: is, uh, circling around.
Because who cares? Who
Fraser Kelton: cares? Who cares? I'm going to make up my decision from doing my work and the facts will speak for themselves and
Nabeel Hyatt: as you said, who cares? Well also, I mean once you've done this for any length of time and you know any sets of other investors at any firm, even frankly partners that I respect.
Like, it's just not a liquid market. There's not a set price. It's somebody fell in love with a thing at some random firm. That's why they're leading it. And then I just had too many times where I've unpacked the logic. You know, you go get a whiskey with the partner later, and you're like, oh, why'd you lead that deal?
And sometimes the reasoning doesn't hold up, right? Everybody has a week where they make a mistake. Even the smartest person has a decision that they made that just is a bad mistake. And so I'm not trying to invest after somebody else's bad mistake, even if I think they're smarter than me.
They could have made a dumb reason. I'll
[00:43:24] Building relationships, Bizdev vs Sales
---
Fraser Kelton: tell you I so yes, I love all of that. But then this has all made me think of, uh, some of the most capable, accomplished founders. That I've had the pleasure of meeting over the past handful of months, do a very opposite take. Yes. Right? So, I was in a call two weeks ago, and I won't name a name, uh,
super successful.
Yep. And he goes, you know, we're not raising now. But here are the three things we really need solved in the next six months, and we really hope that the right partner is going to emerge. Put
Nabeel Hyatt: you to work! A hundred percent.
Fraser Kelton: Very smart. And I walked away from it and I thought, you know what, am I very interested in this person, this opportunity?
Right. And if so, uh, he has now shown the on ramp to have the To a relationship. To a relationship. Total self serving, right? So he has taken the opportunity to open the on ramp to say, you know, yeah, sure. Yep. But I'm not going to teach you about your market map. I'm not going to do all this other mumbo jumbo.
He put the onus on me.
Nabeel Hyatt: Can you help me solve these three problems is very good. The other one I've really noticed from veteran founders that doesn't come from early founders, but second and third time founders, a lot is.
One, if they realize it's a relationship and they start relationship building over time, they treat each of the meetings as a way to basically try and problem solve for something they're working through in the time, like that week. I feel like it's pretty clear that all they're really doing is working through something that they're trying to talk through themselves that they're about to go talk to their team tomorrow about.
And they're happy to see if I have insight into whatever, but mostly they just need to talk it through out loud. And so it's kind of not as much of a waste of a time because it's just a practical problem that they're
Fraser Kelton: having. If you're a smart person who has seen a lot of the market, can they help me sharpen my thinking on this?
Nabeel Hyatt: Yeah, like I think, I used to get quarterly coffees. With Jason Citron at Discord for the year leading up to when we invested and every time we got together It was not here's a new pitch. Here's a new pitch. It was Clearly like the things he was writing his team about that morning Yeah, that he was trying to problem solve and he would just treat it as like, okay Here's another human in the world That I can just sit down and try and problem solve with for an hour and I will at least come through with some level of All thinking.
Oh, I well
Fraser Kelton: Listen, uh, my guess is that he's also debugging you. Oh, of course. Right? Because he's going to text you if you have the luxury of investing in his company. He is going to want to have the person who is actually good at, Working through those things with
Nabeel Hyatt: him.
Well, I mean, the benefit of that is that, is a proxy for what the relationship is with the board member anyway, later on. That's right, Like, are you
Fraser Kelton: willing to, like, tinker with a hard problem that is not going to be relevant in three weeks? Yeah. Because he just needs that, that the, somebody to, bounce
Nabeel Hyatt: it around.
So why don't more founders do, given that it's not a waste of time, why don't more founders do that when they're younger in their career?
Fraser Kelton: I think there's two things at play. I think that there is so much inbound now from investors. Because of the industrialization.
Nabeel Hyatt: You can't get brunch with all of them. You
Fraser Kelton: can't. You can't. Why would you want to spend that
Nabeel Hyatt: much time with a VC?
Fraser Kelton: No, that's actually the biggest failure state, right? And so the standard advice is not to do that. And then they don't discern and with intention break that rule where and when it makes sense.
Because you also shouldn't have that conversation that Jason had with you, with everybody.
Nabeel Hyatt: No, you can't, yeah, you can't be at brunch every morning with a different VC. You're not going to go to the
Fraser Kelton: junior associate at the crossover firm who has no ability to do that and build that rapport with them.
Yeah, that's right. And so you have to be intentional and have, thoughtfulness as to who you're going to engage with that. Otherwise, it will
Nabeel Hyatt: be a time suck. Yeah.
That's fair. Alright, should we wrap up? Let's be done. Okay, thank you. Thanks everybody. If you have any thoughts on stuff we should be trying or thinking about or stuff you want to talk through, drop us a line. Try out Shortwave.
Fraser Kelton: It is a beautiful way to see how AI is going to come into our email client.
Yeah. Give it a try. Take care, bye bye.