Granola, Designing AI for User Agency, and how to compete with $$$
45:19 Share

I feel like in a world where most of us are trying to process the entry of this new alien species, this new thing, this LLM and how it's going to work in our life, I think we've mostly processed it in very, very short go-to-market spurts. The world is filled with incubators that are asking you,

You know, in three months, make sure you launch the product and get some traction to show up for demo day. Or it's a hackathon. Let's see what we can make in two days over the weekend. But my personal feeling was like, no, also, like, we should get weirder before it gets better. I think that we have undoubtedly seen in the market that if you go try to get, like, your merry men of researchers together to go, like, gallivant around and start a company, it's

I don't think that usually works out well. And then inevitably in the kind of roller coaster of things, you're going to wake up one day and you know, there's some headline of somebody who's raised a billion dollars at, you know, whatever it is, it's 10 X more than you've got. That's for sure.

Hey everybody, welcome back. This time for real. I'm Fraser. I'm Nabil. Welcome to Hallway Chat. Welcome to Hallway Chat. You have a product for us, Nabil. You came metaphorically running into the Slack channel where we talk about products and you dump something in. And it was the first time in a while where I've seen a lot of us jump on it almost instantaneously. Tell us what it is.

And let's jump in.

and use AI to solve it, which there's a whole category of 20,000 of them in the app store right now. But that's not what we're talking about. He's doing a new company called Granola, just launched this week.

And, you know, it fits loosely into the AI note-taking app. Everyone jumped on it in Slack because we all like to use products. But let's be real. We also do a lot of meetings and this is an AI note-taking app for meetings. So it's also a little bit of like customer fit as well for the team. And some of us really already know and respect Chris. The product really simply is you jump into a Zoom with somebody,

There's a little tag on the right-hand side, the little G for granola that says, join now. And it pulls up a blank sheet of paper for taking notes the same way you would have brought up Apple Notes or something else. I think there are a lot of very interesting product decisions that...

Chris, he's just a product person, did that are counter to the 30 other products that have tried some version of AI note-taking. And I'm thinking about everything from stuff like Otter and Nada and Fireflies to whole screen recording companies like Limitless to a product that we talked about, I think, in maybe February.

six or seven or eight weeks ago called Superpowered, which is the previous AI note-taking product I had been using. And so I think we should just get into it. I think there's some interesting product decisions here with lessons for other AI products. Before we jump in, let's set up the discussion by

by talking around the high-level value proposition, because I think that's why we were all drawn to it initially, is when he shared the announcement, he said, if a bot is writing notes for you, it's thinking for you. And instead, Granola starts with your notes and makes them better. We've talked a lot about whether or not there's an agent or an assistant that is taking control of tasks for you. And we've also then spoken about co-pilots or...

paired programmer-like experiences that are collaboratively building your work product. And this is the first note-taking app, at least that I've used, that's in that second camp. The other ones seem to basically want to take full control and put on autopilot the notes, whereas this one you have, I don't know, for lack of a better word, editorial control over what it is that it's going to key in on based on the notes that you're taking while the conversation unfolds.

The experience for those who haven't tried it yet is you're just writing into a blank notepad that way you normally would. And then at the end, it goes through a transcription and note taking step where it then does the thing that say Otter or Notta would do, which is to give you a structured transcript.

summary of your overall conversation, but it then integrates and is influenced by the notes that you took. So if you had, say, written down a basic outline as you were going through things or a couple of key insights that you really captured or a quote that you really loved, it'll fit into the hierarchy of the way it then does summary controls, kind of merging the two things together. In practice, it feels almost like you wrote down the things that super mattered and you had an intern

fill out all the rest. And I really like that. I think, you know, we've had lots of conversations about agents. We had a long conversation about Devin and cognition and coding agents. These things that might try and go away for a little while and go do a thing. None of these things at this stage of AI are really

or great. I remember when Anthropic came in, they gave the analogy of, we're going to start with an AI that's good enough to be a kind of okay intern and then slowly move your way up the stack in terms of capabilities. And so another way of kind of like inverting that and thinking about a UI cycle is, what are the types of tasks in the world not

Not that can be done entirely by somebody else, like that I would outsource to India to go do, which is like customer service use cases and so forth. But what are the kinds of tasks that normally somebody is already looking over your shoulder, uh, to help with that your manager is already kind of heavily involved with and looking at. Those are probably really prime examples of situations where an AI can be the co-pilot, the assistant, uh,

And it's funny, we've had Copilot around for a couple of years now and like everyone tried to move past it really quickly without, I think Granola is a very good example of, oh, there's a lot still to mine there. Somehow we jumped out of that when we did AI note-taking and we went straight to, nope, this whole thing is totally filled out instead of thinking of it

you know, it's pair programming. This is like pair programming applied to pair note-taking, which is, I think, a really interesting metaphor. Yep. I think so. I think the other thing that this highlights at a very high level is the importance of product taste and product decisions. Like the product's super immature and young still. So it has, you know, limitations and rough edges. And we can talk about those if it's interesting, but it is pretty clear those are going to be on the roadmap. But what they shipped is so opinionated versus what is in the market today for note-taking apps.

And thoughtfully done. One of the themes that we've talked about for a handful of weeks has been the idea that these models aren't great and you need to fail gracefully across different use cases and find ways to do that to engender trust.

I'll tell you, there was a meeting where I was using it. I carefully took a note that was actually pretty important. And then the auto summarize, I don't know what you would call it, like the co-pilot piece filled in the blanks and gave me a nice, robust note at the end of it. But it had dropped that one piece that I had taken as critical information.

And I was like really angry, right? I'm like, how dare you? If I write it down, don't do that. Yeah, you're yelling at the intern. Yeah, that's constructive feedback. I was giving constructive feedback. But there's a toggle at the bottom where you can actually just go back to the entirety of just your notes. And so you can basically turn off any of the intern scribbles around the margins and get back to where you want to be. And the other place where they do that

is the entire transcript is there. Right. You can drill down to the transcript, right? Yeah, you can drill down to the transcript. And then... I'm not sure if I agree with that one. I don't know if I like that choice. I can understand why they did it for exposure. But, you know, there's something about staring at a transcript of all the uhs and ums and all the terrible things that feels...

It boasts like a little bit more of an invasion of privacy, a little bit more cringy, which is hilarious, of course, because you know it's recording. But staring at all of the words feels quite different to me than you're only allowed to get a summary. So I'll disagree with you there. I probably would have hidden the transcript. I'm sure Chris and the team have lots of good reasons why they did it.

Here's where I think it bubbles up in useful ways. The intern's notes are referenced or deep linked to the relevant part of the transcript. So, yeah, that's right. Right. If it shows you a bullet point that it's written.

You can say, what was the context of that? You click into it, a little overlay appears, and then you see the transcript. Just the snippet, the relevant snippet. I agree with you. Okay, so that part I agree with you. That's brilliant. Like the, you see a note that says, you know, we discussed, you know, AI eye contacts. And then I click a little button on it, the plus to it, and it actually shows me this short bit of transcript, a quote, basically the inline quote that referenced that. That's brilliant. Yeah.

You can do that without having to show me all of the horrific things I said for an hour and a half as I was rambling off the top. For sure. But I'll like, I agree. I agree with that. Here's one that will make you laugh because I'm just looking at the notes from our call before we jumped on here. And there's a bullet point that just simply says Mike Kelly as a follow up.

And Mike Kelly. You can imagine that when you click in that Mike Kelly is something very different than the name Mike Kelly. I think it's Kevin Kelly. That's a mess up. We'll get to Kevin Kelly in a second. Oh, here's, you want a really awesome granola thing? So take that Mike Kelly. Yeah. Change it to Kevin. Yeah. And now if you click open the transcript quote,

It has taken your correction and it has also corrected it in the transcript. Okay, that's pretty good. Isn't that beautiful? That is beautiful. I think the other theme that is really well done on this product is that they've reduced the friction across the product in dramatic ways. And a couple of examples, it offs into your calendar. And so when you have a meeting coming up,

It's the product that basically gives you the notification that a meeting's coming up and there's a one-click way to launch into the video call.

which feels, I don't know, uninteresting. But what it does is it also then drops you into the note-taking experience perfectly positioned on the screen. You have your video on one side, the note-taking app on the other. The title is already filled in for the meeting. Cursor's already live where you can just start typing. Somebody has cared enough about

the product experience to craft all of those details. That's one. And then the second one is what you pointed out, and that is it takes over the system settings. Look, I think metaphorically, if we think about this, how would I want a note-taking AI to insert my note-taking? You can think of these things as three different buckets. And the first one is things like mem or notion, which

which is you're taking notes exactly the way you always took them, and then we're going to layer a ChatGPT-like experience on top of it, right? I can right-click on it and say summarize. I can right-click on it and say expand. I can give you an outline and say flesh it out. That's kind of like, yeah, it's a ChatGPT for my notes. With notes, the way you previously used it being the primary affordance.

The second is the inverse of that. It's not a note-taking app. The AI is the thing that takes the notes, and then you can do something with them differently. And that's the Otter, the Nada, and the Fireflies of the world, which drive me crazy. I've used all of them. I've tried all of them. They are okay for...

some specific use cases but especially the ones that don't let you edit the prompts themselves so it's some sales template thing that somebody at one of these companies decided is the only way you're supposed to get your summarize meeting is just just not interesting the one I do use constantly is called superpowered which is quite useful I record lots of you know meetings when I need to and they have a custom prompt framework so I can have it output the outputs that I really want and summarize it the way I want it's great actually I really like it that's why we talked about it before

But this is something different, which is just, it's not that the AI is the first input, and it's not that my old view of the world, the way I would think of Notion, is the old input. It's this, again, it's this kind of co-pilot approach

that I now, I immediately came away from granola trying to think about lots of other consumer categories and enterprise categories and saying, oh, well, what would you do if it was that? And so I love it when a product is so inventive in its product surface area that it makes you kind of reevaluate whatever.

And I won't list all the other categories or anything like that because we'll have time to go find those founders or whatever. But it certainly made me think about what are the implications for other things. And normally, I mean, the last thing I'll say is it just also reminded me how much we have not really been experimenting with interfaces. You know, we're a little ways into ChatGPT and there is just not enough people

trying weird things with how AI is supposed to surface in our life. I was at a dinner last night with a bunch of people talking about agents and AI and, you know, UIs came up and somebody brought up

the mother of all demos, which if you don't know what it is, we'll drop it in the show notes. You know, it was a presentation in, I think it was like early 1969, late 1968. It was a presentation from Engelbart at the Augmented Research Center, ARC, where he basically laid out

the next 40 years of computing interfaces. Like, he had a device connected between Palo Alto at SRC up to, I think it was San Francisco. Alan Kay is in there. Like, Bob Sproul's in the audience. Like, it's a, you know, Hall of Fame group. Thankfully, thankfully, I think it was like, Stuart Brand happened to be in the audience with a camera. And so he was recording it. It's like insane. And he's, you know, it's like,

Let's show you what a mouse might be like. Let's show you these dual screens and how you might communicate over a computer. Like stuff, frankly, that's in there that we still haven't gotten to. And I feel like in a world where most of us are trying to process things

the entry of this new alien species, this new thing, this LLM and how it's going to work in our life. I think we've mostly processed it in very, very short go-to-market spurts, right? We are now, you know, 50, 67 years later, the world is filled with incubators that are asking you, you know, in three months, make sure you launch the product and get some traction to show up for demo day, or it's a hackathon. Let's see what we can make in two days over the weekend. And

And, you know, it was kind of counter to the conversation we had at dinner, to be honest, because a lot of the conversation with dinner is like, wow, a lot of these agentic companies are producing demos that kind of like 80% work or 70% work, but they don't really work enough to deploy into production. And that is also true. But my personal feeling was like, no, also...

We should get weirder before it gets better. I don't mean hackathon weird. I mean deeply ambitious alien weird. We need to explore the design space a lot more to kind of figure out what is the GUI of this world. We're sitting staring at MS-DOS prompts and no one's made Windows yet. And I think...

you know, I don't even know what the Xerox PARC or the MIT Media Lab should be for this era. It probably is not going to be, you know, Google R&D. All the incentives nowadays are very differently aligned. And so I don't think that's going to happen there. It's probably not going to be the Media Lab. So I actually don't know where those experiments are going to come from. They probably have to come from founders, right, doing the work. I would think so. It's interesting to frame it like that. When you

When you told me this initially, my response was, this happened in the labs, right? You come in and Aditya shows you a tennis ball thing that's pixelated and you go, what's going on with this? But if you squint, you're like, okay, we're a couple of cranks away from having Dolly, right? And-

There's so many weird, lovely things that are percolating there. We just haven't seen it yet. We see it after it scales up and they've polished it or we don't. But you're saying at a level even above that. Yeah, I would push back at that. It was OpenAI, frankly, like when you joined OpenAI, right? But after the chat GP and before, right? When they're making Rubik's Cubes robots and things like that, I think...

Once the Chachapiti gun goes off, a lot of it is, it's not that there isn't advanced research and not that people aren't doing amazing, crazy, interesting things, but it's all in service of something that's relatively deterministic. And the problem with the corporate labs, the large ones like Google, is we've had this conversation before.

I'm trying to get my paper published. And that's also a very, very different incentive than if you think about what's going on inside of Xerox PARC at that point in time. Yeah. You don't think there's the same crazy shenanigans happening and we just haven't seen it? There's not the equivalent of Sora 3 that you can squint and see, but we're not going to see it in the world for five years? It's a good challenge. I would say...

There's probably a little bit happening at the edges of Google R&D at least a year ago. I'm often giving Google a hard time

for maybe even being the Xerox PARC of this era in that they like invented all this technology and then we'll see if they ever effectively bring any of it to market. But look, in the world of consolidation where everything's moving together at Google, I kind of want them to consolidate so that they can hopefully execute really well so that we can actually see Google as a good citizen in this set of product surface areas and, and,

invent the future of AI with everybody else instead of just kind of making demos. But the downside of that is, and we've kind of had this experience. We interview people that are coming out of Google. What kind of research did you do there since you're joining XYZ New Startup?

It is thinning in its aperture, not widening. Right. Right. It's narrowing in its aperture. Right. It's not widening that Google has targets now. That's fair. That they're going after. And so they're going to do less of that work over time. And so this is me just kind of like raising a hand and being like, I understand granola is not like the most mind blowing, crazy, you know, it's not Sora. Yeah.

But that's kind of my point. I think there's still a crazy amount of innovation to happen, even just at the UI and UX and product layer, not at the deep research algo layer. There's nothing crazy technical about inventing the GUI, the graphical user interface. The first time you ever saw a window, you could move around with a mouse.

There's actually nothing crazy technical about the mouse when it was first invented. It is thinking about the product experience and then developing technology that maps to the product experience that we think we want to have with this new technology that's entered our lives. I think we're going to do great at new fancy algorithms. I'm less certain that we have the right pockets in technology right now to...

innovate on new product experiences because everybody's chasing growth and we have been organized to think about customer development and MVPs and up into the right so I can raise my billion dollar round tomorrow and all the rest of that stuff. Yep, yep, yep. And I'm not against growth. Like, eventually we're in the venture scale business but we're just early and we're

It's just, there's the room for weirdness is important. I get all of that. I was just sitting with the founder that I work with last week and he's exceptionally early in a market that's early and he's iterating, you know, toward a product that's resonating. And my feedback to him was like, that sounds good, but

But at this moment in time, I would make sure that you're taking large swings adjacent to where you're currently going to make sure that you're not marching up a local maxima. Right. I agree. There's so much stuff going on. I don't know. I feel like we have to give permission to be creative at times. We as in...

The tech industry. Yeah. And we have to be patient. I think that's a big thing, though, is if you were to see a demo of that type of stuff now, think about the world's response. Yeah. Well, there's a world where some of those people are trying to do some of those demos and maybe the world just says this is fake. Maybe the world says this isn't shippable and, you know, this founder is full of it.

Yeah. Listen, we've seen that exceptionally loudly on Twitter. You shipped a wait list. You shipped a demo. You ship. There's a cacophony of voices screaming at you and you have to be extra. Twitter's not one. Twitter's not reality. Right. To all we all you really ask when you're shipping a demo or talking about a concept is that you be transparent. It's a concept.

There's nothing wrong with talking about a crazy idea. Fair. The problem with talking about a crazy idea is presenting it as shipping in 30 days. Just give me your hundred dollars. Right. And it's ready. Right. Right. That's the problem. There's nothing wrong with making claims and being highfalutin and being, you know, you can be Walt Disney. Just make sure everybody understands that you're dreaming at this moment.

Right. Yep. I do think it is really hard to be patient though, Frazier. I agree. It's very hard to keep iterating on a thing to really get to new things, especially when all of your competitors are

are probably chasing a little bit of traction and are probably raising a bunch of money. And so you're trying to iterate on refining the product and getting it perfect. And you're rethinking, you're like throwing out the UX for the third time to really make it work. And then you wake up one morning and you open up TechCrunch and somebody else is doing something orthogonal. It's just raised $150 million or whatever the number is. We as a venture community are part and parcel of

in trying to solve our own problems on a monthly basis and get our, you know, whoever principal at some firm is trying to get a promotion. So he's trying to jump on the hot thing that week is, is certainly makes it harder to build really good product and make really good decisions. No question. Right.

I also think that personal incentives are incredibly important in everything, like full stop. And I think the structure of performance reviews and yearly perf packages that people pull together, I think it would be almost impossible in that framework to find the structure that rewards that level of exploration.

Right. Like it's probably probably doable, but it would take even at the org level some real careful consideration and sculpting. Yeah. The you're the old adage from Charlie Munger. Show me the incentive and I'll show you the outcome. Oh, for sure. 100 percent. Right. Yeah. Yeah. Yeah. You know, talking about waking up and seeing the headline that that your competitors raised hundreds of millions of dollars. That is a topic that has been on my mind for decades.

a while now, and I'm sure it's been on yours too. - Yeah, it's impossible to not be operating in the AI market and trying to be either an employee in a startup, a CEO startup, an investor in a startup, and not wake up on both sides of this equation at some point in your life. You're gonna wake up, and you're gonna feel great about having raised a bunch more money than somebody else, and you're trying to figure out how to use it, and we certainly have those companies, and then inevitably, in the kind of rollercoaster of things, you're gonna wake up one day

And, you know, there's some headline of somebody who's raised a billion dollars at $10 billion or $100 billion at, you know, whatever it is. It's 10x more than you've got, that's for sure. And then they call you, Fraser. I know you've had this happen. I've had this happen. What do you say to somebody who calls and says...

man, they just raised that much money. Often they're the same size as us or even smaller. How did that happen? What do you say to a founder and how do you think a team, not even just the founder, how do you think a team should internalize when there's money flowing around in that kind of like haphazard way that may or may not feel like it just took you out that day? I think any good piece of discussion around startups, there's way too much nuance for us to have a tight conversation

solid piece of wisdom. It's not three headlines. It's not the three steps to the epiphany. Because listen, we went through an era where SoftBank thought that they could overcapitalize every company and crown the market winner. And I think that we've seen that that experiment didn't work out. You know, like all things, it didn't work in 20 places except for the place where it worked. Like it really worked for DoorDash. Right. Like that, it

Absolutely worked. And then there was like, you know, and probably maybe one other place. But the capital creates the winner was a graveyard otherwise. Right. Right. Right. It just incinerated a bunch of capital and an awful lot of startups. Yeah. And I feel that in my bones. Like one thing that I think is is like the cardinal sin of a founder. And I felt it when I was a founder was money.

overcapitalizing and celebrating the amount that you raised or your valuation. Because I think that, I don't know, it's probably not causation, but it's pretty darn close to that, right? It feels, I can't think of one example where that's the case.

You have on one end of the ledger a situation where overcapitalization kills companies. The idea that capital can crown the king also really didn't work out. And yet I feel like in many markets right now,

You will lose your edge if you're not at least in the same ballpark. If you're a company that's raised $20 million and you have competitors that have raised $150, $250, $400 million, that's

If you're trying to compete for the same large enterprise customer, if you're trying to compete for the same researcher who's been at DeepMind for five, six years, I do think that you're going to be perceived very differently in the market with that type of a split. And then I think that over time, it's going to have a real effect, a negative effect on the company. Yeah.

I think that's somewhat true. I think it's worth digging in and giving a little bit of nuance back to a team that's trying to process this. Like, when you're thinking about larger competitors, I just think it's important to keep your customer in mind. And it all roots back to your customer. Your customer is going to make your success, not a VC and not your bank account. Right. And so sometimes the money matters because your customer matters. Sometimes the money matters because it matters to your customer. Uh-huh. It's often that, not because it is a situation where...

it made the next engineering hire happen or not. I started a company a long time ago called Conduit Labs. About nine months into its life, it pivoted into Facebook games as the Facebook app platform opened up and that led to the acquisition by Zynga. But early on, we were building virtual worlds in this kind of post-second life world. And I just remember the product we were working on, MTV...

funded and built a competitor to that product. And I was freaking out. And I remember talking to Alex Rogopoulos, who's the CEO of Harmonix, guys who made guitar here in Rock Band at the time. And I just remember the advice he was giving me, which is just right now you are all defaulted. I don't really care how much money comes in. The default process is that whatever they just come up with is just not good enough.

And so don't stare at them with their money and be covetous. Stare at their product and be covetous if they've come up with the right thing.

And, you know, this similar thing happened. Ambient Devices, a previous company I started that was a spin-off from the MIT Media Lab. I remember LG, after we'd had extensive conversations about licensing agreements, a couple, six months later, they launched their ambient computing initiative at LG. And we were like, oh, man, big guy. You know, it just turned out the whole category wasn't a category. It wasn't a massive category. So, it's fine. You know, I think...

Cruise is another great example where, man, that was a capital war. You had Google with Waymo budgets after Cruise raised, not long later, Aurora raised. Chris Aronson became the CEO. He headed the self-driving car team at Alphabet for seven years. And then it turned out, by the way, that—

I don't know, Cruise was the first to market. We'll see what GM does with it now. Waymo is another viable player. Aurora, we'll see what happens. But certainly, even with the war chest that they built, didn't immediately make them winners in the market. That's for sure. So I don't know. I think capital, I think of it three ways. One is if it's day zero and you're an overfunded pre-product market company,

Capital is oxygen, and you could absolutely die by oxygenization. Right, yes. Good. And I remember when Thalmic Labs raised a $100 million round years ago there, an early frontier tech startup we backed. You know, I remember showing at that point in time the list of other $100 million pre-launch companies that had existed, like literally every single one that had existed. And it was an unbelievable, Quibi-like company.

graveyard of companies that did not work out. There's a bunch of reasons we need to go into now about why I believe that is true, but it's not categorically true, right? But it's pretty rare and hard

To raise that kind of capital and still keep a disciplined, fast-moving, execution-oriented startup. 100%. Yeah. And keep up with everybody else. So sometimes a bunch of money into somebody else can be a gift. The second bit, which you were getting to, I'd say, on a large raise or somebody who's doing it, is who is your customer? If your customer is a consumer... Mm-hmm.

I don't know, $100 million, $500 million in the bank. It doesn't change your retention curves. Nobody cares. At all. That's right. Your customers barely pay attention. They don't really know. I wouldn't worry too much about it. If anything, again, it might handicap other people. If your customer is, I don't know, some old...

money, water company in France that wants to buy the safest thing in the planet, then you should worry. Yeah. Right? If your customer is very safety-oriented, late adopter type who just wants safety, then them looking at a startup that has a 10 times higher bank account than you, it will affect your business. And you better find a way to fight the capital game. Right. Or at least, as you said earlier, at least be within an order of magnitude.

Right. Right. Yeah. Same universe. In the same universe. You don't have to play a capital war, but you better have the same number of zeros as the other person or else again, it's, but again, that's not really about the money in the bank. That is about the amount of capital you have is, is,

a feature of your company when you're trying to sell to a large organization. For sure. It's a part of your product, strangely enough. For sure. But that's very rare. Most of the times, most AI companies are not trying to sell to late adopters

Kind of by necessity. If you're selling to a fear-induced late adopter, first of all, thank you, elephant hunting, whale hunting startup, it's hard. Right, impossible. And it's going to be slow and it's going to be painful. You should probably see if you have early adopters you can sell to first. That's the nature of things. Right. But yeah, and then there will be the few situations, just like there's no playbook, there's no three headlines to get you through, but there will be some situations where...

you are in a capital war. That's just the bottom line. Yeah. The other thing that we've heard from very capable founders who tell us that they know better is that the types of talent that can make a company in this moment, and I'm talking about applied researchers or product-focused researchers, are valuing total comp in the moment rather than

the opportunity for appreciation. And we've met a couple of really, really strong founders that I respect who then say, listen, I know, I know, I know, but to hire the three types of people that I need to hire over the next 18 months to make the company, they're going to evaluate it on that vector. And so it's important that I'm again in the universe of valuation. Yeah. There was a tweet

A couple of weeks ago, that was like, if you're interviewing at...

open AI and you're interviewing at a startup, then the startup should just stop interviewing you immediately or something along those lines. It was if you're a startup and you're interviewing somebody who's also interviewing at big tech, you should just stop. Yeah, that's right. They're like a tourist who's going on a little side journey to understand how interesting it might be in a startup, but they're never going to come to you. I will push back slightly on that. And I did in that conversation, which is that

There are some people who actually don't really know the difference between joining OpenAI or joining Google and joining a startup. Like there are some people that are just earlier in their educational journey about what that choice really means and the pluses and minuses of the choices of being at big tech or being in a startup environment. And so sometimes it's worth explaining to people. If you haven't explained to people the appreciation of stock and the amount of charge

change and responsibility and excitement and agency that comes with being in a startup versus working in a lab at Google, then shame on you as a founder. And there are plenty of people, especially if you haven't grown up in Silicon Valley, there are still plenty of people on the planet, even those who've been working for many years, who don't really understand startup culture. They understand technology culture, but they don't really understand startup culture.

So I think you have to explain to them what it means to join a startup and what that means from a compensation standpoint. And also that, frankly, what it means from a working and responsibility standpoint. And then, yeah, if at this point you're haggling over those kinds of choices between open AI or a startup, they should probably just first pick big company or startup before they even get to the name of the thing. Right. Because they're just wildly different paths. Like it's.

you know, it is like different careers in a way. Listen, I get all of that. The market's telling us something different right now. The market might be irrational in the moment. We've seen that. But these are, these are like seasoned, capable founders who are telling us that

They need to play that game to make the couple of hires that are going to make the company. No, I just disagree. Like, I just disagree. Like, that person, because the problem is that person is coin operated. That person wants whatever the highest comp package for that year is. And what that means is that they're going to join you and you're going to go through all this integration and they're going to get work done. And then six months later, somebody is going to give them a signing bonus and they're going to be gone anyway. Everything in my bones tells me you're right. Yeah. Right? Yeah.

Everything in my bones tells me you're right. Yeah. No, I think that the challenge that I'm wrestling with is I've

I've seen firsthand how one person, one person can breathe life into something in this moment that is ethereal. Like it's not like you're hiring a 10X engineer who's going to come in and you're fighting for that type of incremental efficiency on your engineering team and you're going to ship your backend data pipelines a little bit faster. I think we're sitting in a moment where there are- It's like signs of life is what your point is. Well, signs of life from-

a very scarce number of people who can control the magic and grant it to your company. If you are really going to push the frontier of AI research in a domain and in an applied setting,

With that research, I don't know, like I've seen Alec Radford, like this guy is amazing. You probably should, you know, put yourself in a position where you can get the N of four in the world into your company. I think the challenge then is if that's true, people misconstrue that and do that same playbook for the backend engineer where it really is immaterial. The trend I have felt in the last three or four or five months by folks who've been in AI for a long time is people

I only want one of those deep researchers at most. And me as the founder or the CTO is that one. And I thought I needed 10. And the truth is, now I'm just going to train some really good ML people internally. And I will get 99% of the way there on every single thing I want to do.

And frankly, they'll be more product-oriented, more consumer-oriented. Two years ago and a year ago, the kind of prevailing thought process was, I want a really hardcore business

world-class research team on top of a world-class engine product team. And I need to build both of those and keep those dualities together. And we've both seen that the trend for the best performing companies now has been incredibly slim to almost no research team. Maybe, again, one to two people, often the founders. And so...

Again, unless you're anthropic. Anthropic has a reason to have a massive research team to be in a state-of-the-art large language model. Yeah, this is where so much nuance matters on the decision because I think I agree with you for when you're starting out. I think that we have undoubtedly seen in the market that if you go try to get your merry men of researchers together to go like gallivant around and start a company, I

I don't think that usually works out well. You have a researcher, usually the founder, founding CTO, who's pushing forward with a lot of product motivation behind it.

There is a time, though, when that company is no longer three people and just founding. They have a product to market or they have the scaffolding of an organization around them and they've started to scale a team. You can basically, I don't know, is like the equivalent of a two pizza team again. You can replicate that.

that playbook multiple times, depending on where you are in the arc of your company. I would take multiple people like that and build a team around them and let them run rampant. I think there's a moment in time where that can happen.

So let's bring it back is a founder comes, they've raised, let's keep it at tens of millions. They say, hey, listen, this company just raised a billion dollars. They just launched the company. They haven't even launched the product. They've introduced the company with the billion dollar round. And actually, all the people on the team are not the founder ethos. They're, you know, gray haired, grizzled industry veterans, right?

100% in that case, I'm like, you should just ignore it. Like you don't even pay it any mindshare. They've just killed the company and they don't even appreciate it. I think the thing to be worried about is the company or at least

inquisitive about is the company that's in your market, in a market where your customers care about things and or there's a small number of people who can wield this magic today and care about outsized comp rather than appreciation and have raised many multiples of you. Because I think that your... You said product earlier. I think that's a reflection of your brand in the market. And I think that people...

A lot of markets buy on brand. Yeah, that's right. I think they do buy on brand. I don't think consumers buy on brand. No. I think even mid-market to small business, if you're, don't buy, the people who buy on brand and safety are middle to late adopter. Yeah. Large, large elephant hunting enterprise customers, which is actually a vanishingly small number of startups actually have that customer base. That's right. And for everybody else, like we're glossing over a bunch of things. Of course, money matters, right?

but there are a million different things that are going to kill you as a startup. And frankly, most startups are going to die by suicide, not homicide. It is the staring out your window and everything else is just distracting you from the fact that it is get your culture, right? Get your product shipping, right? Make product market fit, right? Take some weirder choices on, on UI until you find the right things, push your engineering and your, your,

research teams to see around the next corner, like all of those things are going to matter a lot more than capital, to be honest. And the interesting thing, of course, is that those sorts of things do tend to beget capital. And I wish we had some charts or graphs of the number of times that over the years where, you know, some of this is just like once you've lived through it, like five or six or seven or eight times as a founder and then as a VC, you just

It's like old Chinese proverb of the guy with the horse where you're like, you know, the thing that looks like a gift, you're not sure if it's actually a gift and we'll just turn another page. And I don't know. And the number of times that the most well-funded player has won, I don't even know if it's correlated. No. At least in the early days. At some point, like IPO and beyond, like it's a different world. Sure. But in these early, you know, startup malleable, we'll see who wins, you know,

person who did well this week, person who won the month, person who won the quarter. It's a long game. I think you can overstuff a duck and end up with foie gras. I think you do that with a startup and you kill it. The way that you framed it to a founder once that I think has really resonated with me is if you know what you want to do with the capital, all of that capital, go and do it. And if you don't know what you're going to do with the capital, just think of it as going to be a negative for your company.

Yeah, that's right. Is it fuel? And I mean, fuel in the real way, you know what to do with it, not just more runway. If so, it's possible it will help you get ahead. Right. That's right. Let's be done for this week, man. It's great chatting with you again. As always, I'll leave with one little thing, which is at a dinner last night, my friend Michael Galpert said is is WebSIMs.

the geocities of the llm era and so if you haven't tried web sims we're not going to talk about it here but i would i would suggest going and playing with it it is a crazy fun little demo and as long as we're talking about new uis and crazy things i think you should go take a look okay i gotta do it didn't you have a product you were going to roll out with me or talk about as well i'm just going to drop it and then we'll hang up i tried this new amazing product called superhuman finally i

I love Gmail. Yeah. I have lived in Gmail for now 20 years. I don't want to adapt a new workflow when I download a new email client. I just want to do email. And Superhuman has made the decision that you can just do email in a beautifully designed UI if you want to do it.

They've introduced a handful of AI features. And I think the thing that I take away from that is that they're feathered in. That's the way that I would describe it. It's not ham-fisted. It's not trying to turn it into a marketing event where it's like, hey, here's this new AI feature that we're going to cram down you. The summarization of threads is really nice. And then you can actually click it and it expands to show you a

a different like depth of the summary, which is really nice. They have instant replies that are nice. And then again, like thoughtfully done where they're there if you want it and not if you don't. I will say though, I have so indoctrinated into the superhuman workflows, but it's interesting to ask what's the granola copilot version of superhuman. And for me, I would want to log in in the morning and

and have the most common emails I would reply to have drafts of the replies already there.

That would be the granola way of doing this. And I think that would be awesome. And I could delete them. I could say, no, I didn't want to say that. But right now it's a very notion view of how to influence AI, right? It is a little command key that I can type in, finish this email or so on and so forth. And I'd love something a little bit more copilot-y but assertive. But before that works, I think there's probably a hurdle on the writing style getting better as well, but it's a thought to end on.

Cool. See you later, Fraser. See ya.