Transcript: 'The Secret to Building Sticky AI Products'

The transcript of AI & I with Chris Pedregal is below. Watch on X or YouTube, or listen on Spotify or Apple Podcasts.

Timestamps

Introduction: 00:00:48
How Chris made early product decisions at Granola: 00:09:14
Chris’s philosophy around product development: 00:13:36
When to follow your intuition v. listen to your users: 00:19:24
How to build a product with “soul”: 00:20:40
Chris’s advice on becoming a better product thinker: 00:25:12
The role travel plays in shaping Chris’s intuition: 00:31:17
Why having fewer users is an advantage for AI startups: 00:45:52
Why Chris is bullish on startups building specialized AI tools: 00:52:09
Where Chris sees Granola in the next year: 00:56:52

Transcript

Dan Shipper (00:00:49)

Chris, welcome to the show.

Chris Pedregal (00:00:50)

Hey, Dan. Thanks so much for having me.

Dan Shipper (00:00:51)

Really glad to have you on the show. So, for people who don't know, you are the cofounder and CEO of Granola, which is actually one of my favorite pieces of AI software. Whenever people ask me who's doing it right in the AI consumer business product world, Granola is one of the first things that comes up. I use it for pretty much all my meetings. So, basically, it just sits in the background of my computer. It doesn't join Zoom or anything. It's not like the Fireflies bot, which I find to be the most annoying thing in the world. And it just records everything and then it turns it into a transcript and then has some automated notes, which is really cool. But one of my favorite features is, you've got this thing where I can send the transcript to someone else or the Granola meeting to someone else and then they can ask questions about it. Because we're operating at a scale now, inside of Every, I can't be in every meeting. Other people can't be in every meeting, but there's always questions about, what did this person say? What did that person say? And just being like, don't bother me with the questions, here's the whole thing. That you can ask questions of the actual transcript is really nice. So long-winded way of saying, I think you're doing awesome stuff and I'm really excited to have you on the show.

Chris Pedregal (00:02:05)

Thanks so much, Dan. And you've been giving us feedback for a while, which I appreciated from day one. So, thanks so much.

Dan Shipper (00:02:21)

That's great. And you just raised a huge round—$20 million. Tell me, where are you right now in the business? What's going on? What's in your mind?

Chris Pedregal (00:2:23)

I mean, we're honestly super early. So, Granola launched in May. We were a team of four when it launched. and we ended up signing a term sheet not that long after launch. I guess it's helpful if you build a product that a lot of VCs use because it becomes less of an analytic exercise where it's actually, they can use the product and see if it's a useful thing. So we're super early. We're still very, very much focused on the product—on growing it. We definitely seem to have struck a nerve for some users like yourself—the early adopters. And I just think we need to grow from there.

Dan Shipper (00:03:01)

Who are the people that you think it strikes a nerve with? What are the characteristics of those kinds of people?

Chris Pedregal (00:03:10)

Yeah. So, I think there's two ways to answer that in terms of job title. It's a lot of founders—startup founders, specifically. Both small startups and big startups and a lot of investors right now. And I'd say it's people for whom they need to make high-leverage decisions on the backs of meetings and really care about the quality of their judgment.

Dan Shipper (00:03:47)

That’s really interesting. And one of the things I'm really curious to dig in with you is, you raised a big round. You said you built the first version with a very small team. I assume the team is still fairly small. Tell me about that decision because I'll just lay my cards on the table. One of the things I've been playing with is you can get a lot more done in AI land with much less money in a much smaller team. And I think for me, as a personality, I like that because I haven't really wanted to raise a lot of money—I’ve raised a little bit of money for Every, not a lot. And we're incubating these products and we're kind of seeing how far we can get with only a little bit of money. And I'm always interested in the total opposite perspective. So tell me about that decision and how you think about building a company in this environment.

Chris Pedregal (00:04:35)

Yeah, great question. So, I'd say my natural tendencies are similar to yours, right, which is a small team, don't raise too much money. We found ourselves in an interesting space. So AI is kind of nuts for a few reasons. One is, it's quite expensive to run a lot of these products today, but it won't be expensive to run them in two years. But in two years, the leaders and winners will already be defined. So there's this middle passage, this middle period here where you're going to be running expensive products that will not be so expensive to run in the future, but you need to make it through that period. The other one is just, there's so much volatility. It's really hard to predict the future. Things are moving really quickly and we had an opportunity to raise capital and we decided to take it basically to protect ourselves against that volatility. We don't know what the world's going to look like 12 months from now, 18 months from now, and we otherwise would need to raise again. And also, it's like I said, it's early, but I do think Granola kind of struck a nerve. So I do think we have the opportunity to scale the user base pretty quickly, which will involve money.

Dan Shipper (00:06:02)

That makes sense. And what do you think about— There's this double-edged sword. We actually think we have that at Every, because we always build products that we kind of want ourselves. And there's a double-edged sword to that because you can get immediate adoption with the founder-VC crowd, but then expanding beyond that is tricky. You know, a good comp is maybe Superhuman, where I think they were the hot SaaS thing for any VC or any founder to use for a while. And I think they've had some success breaking into enterprise, but it's certainly taken a long time. How do you think about moving from the cohort of people that seem to love this today to whatever the next step is?

Chris Pedregal (00:06:47)

Yeah, that's a great question. I think there are two parts there. I think what you want to be careful of— Also, I don't have all the right answers. I'll just tell you what I think, right? And what we're trying to do—

Dan Shipper (00:07:00)

That's what we're doing here. No, no one has all the answers.

Chris Pedregal (00:07:03)

Okay. Cool. Proceed at your own risk. I feel like the real danger is to get pigeonholed in a vertical that you don't want to get pigeonholed in. So it's interesting. It's like, building products for VCs is notoriously a bad signal because, for investors, they say the TAM is so small, right? Sure, they have willingness to spend, but there aren't that many of them. So you can't build a really big, big business there. So I think it's really important that you protect against that. So, if you don't want to build a product for a specific vertical, make sure you diversify your user base early so you get signals. Otherwise, if you go too far down that path too early, it'll be really hard to switch, I think. By focusing on early adopters, I think you can do that for a while because two things will happen. I think one, if you get early adopters to really love you, they will evangelize you to their friends and their teams. And I also think that with some of these tools that just might be like a changing of the guard. I remember when Notion came out. And it's not like big companies were switching to Notion. I think what happened is that they got all the startups, all the scale-ups using it, and some of those companies have become really, really big. And then over time they've become established and it's easier to get companies to switch. But I don't think you need to. I think it would be a mistake to try to go after a mass-market product too early because you lose what makes you special, if you do that.

Dan Shipper (00:08:38)

That makes total sense. I totally agree with that. I'm wondering— The place where my mind goes, I just think you're such a great product thinker. And the reason I can tell is because you managed to make an AI product that’s not just the shiny bells-and-whistles thing. It's actually useful. It's a useful product that includes AI. And it's not a bolt-on—it does something that would only uniquely be possible in the AI world, but it's not this weird hype-y demo that you use once and then you just never look at again. So, how did you think about making those decisions? How did you get into Granola and how do you think about the initial product decisions?

Chris Pedregal (00:09:17)

Yeah, I guess the short answer is that it took a while. We launched in May and, I think we've gotten a lot of attention since launch, but we had been working on it for a year before then with 100–150 people using it and giving us tons of feedback. I can't remember. You might've been one of them. I actually can't remember if you were in the beta group or maybe you started using it after launch.

Dan Shipper (00:09:48)

I don't think so. I think I tweeted you and then you didn't respond and I was like, fuck this guy.

Chris Pedregal (00:09:52)

It sounds like me. I'm not great at social media. Alright, so you weren't part of the original 150. I guess a few things. One is it took us a really long time to figure out what the core interaction would be in Granola. So the core interaction right now is basically Granola looks like Apple Notes. For those who haven't used it looks like Apple Notes during the meeting. You type into it like a normal text editor. It's transcribing real time. It's listening to the conversation. And then the moment the meeting ends, you see this bar go down and it transforms your notes into really nice notes. It fleshes stuff out. It adds points you didn't write down. It took us a long time to get to that. That's not at all what we thought the product was going to look like at the beginning. The core interaction was real-time during the meeting, you'd be invoking the AI and typing keywords and it'd be writing notes. And that interface, that interaction was really cool— it looked cool, it made for a great demo, it made it look completely different than anything that came before. It just turned out to be really distracting in actual meetings because what would happen is if you have any kind of AI augmentation that happens real-time, you can't help but read what the AI is writing. And if it doesn't get it exactly right, then you can't help but want to fix it. And then you realize you've stopped paying attention to the person you're talking to. So, actually, it took us a long time. We went through a few different interaction paradigms, until we finally settled on the one that's in Granola now. And once we did that, we cut out 50 percent of the features we had built, not because they aren't stuff we'll build one day, just because they weren't good enough. But with AI, it's very, very easy to build a demo version of a feature, right? The happy path. But it's actually still a lot of work to build a great feature that works consistently and reliably and that people love. So, that was a big part of it, I think, is we cut out a lot of the stuff that didn't work, and we just tried to do one thing really well, which is hard in AI. It takes a good amount of discipline because you can do so much so easily.

Dan Shipper (00:12:09)

I guess the thing that I imagine people are—and I'm sort of wondering about it too—is okay, so you had the original thing and it's doing the real-time transcription, which is a really cool—wow—thing. And then you did a bunch of iterations and you realized that this much simpler interaction pattern, that's maybe a slightly less of a wow, especially while you're in the meeting, it turns into a wow at the end. What's really funny, actually, before I go there is that I don't use it that way at all. I don't ever take notes in it. Sorry! I just record and then get the notes. So, it's really interesting, and I think that actually is getting to my question, which is, I think a lot of people, myself included, but, I think I've gotten a little better over time. A lot of people are like, okay, but how do you know who to listen to? How are you getting that feedback and then sorting through it? You have that UI, you have that user interaction. And when you say it's distracting, are you paying attention to yourself using it? Are you just watching lots of different users and listening to some group of them? Because I imagine there's some group of people that really maybe liked the distraction. I don't know. I'm just making stuff up. So, how are you sorting through all that stuff?

Chris Pedregal (00:13:30)

I don't know. I don't know if you get it perfectly right. My product philosophy is, I think you need to make product decisions based on your own intuition. And you want to inform or give your intuition as much of the relevant context as possible. So to me, it's less about, okay, you talk to a user and you write down exactly what they say, right? And then you go and do exactly what they say. I think we're lucky with Granola where we actually use Granola ourselves. We don't have the same experiences like an investor would or, let's say, like a salesperson directly, but we do do sales calls. So we can kind of empathize there. I think the most important thing is basically to build your internal mental intuition model of what feels good or bad and then the best way to do that is to get— I FaceTime with users. Written feedback is a data point, but I think it's when you talk to the human and you can kind of see the human in their full context that it becomes a lot easier. So what's worked best for us, and it's harder as you scale is to have a few people you talk to a lot. And we do that basically with over Slack, and then with periodic calls like this. Ideally, you go into their office and shadow them for a day, which is logistically really hard to pull off. But whenever we've done it's been super valuable.

Dan Shipper (00:15:08)

Yeah, I love this. So, I mean, I think you're saying there isn't some list of rules or set of processes. There is some data gathering, but ultimately it's about having a perspective and that perspective or that sense of taste is implicit rather than explicit. And it's sort of about trusting that to some degree, and informing it and having it evolve over time. So what does that feel like? What does it feel like when you're making a decision and you're letting the— Because intuition is a hard thing, some intuitions are bad, some intuitions are good. What does it feel like to you when you're kind of aligned and you kind of get into the intuition of, this is what I want to do.

Chris Pedregal (00:15:59)

This is generic, but deep down, you need to believe in something. And it's easiest when you're building for yourself, because you can be like, okay, this is what I want, or this is what feels right to me. And I think there should always be that anchoring, because otherwise there are lots of products you could build in the world, right? The one that you're anchored on is the one that you care about. There's another school of thought, which is don't even talk to users, just pure design, follow your truth. I don't subscribe to that. And the reason I don’t subscribe to that is because it is unfathomably hard to put yourself in another person's shoes. If you try to do it in an abstract way, you're like, okay, what does this person think or feel? And I think perhaps anyone who's been in a marriage or a long-term relationship can kind of attest to that, which is it doesn't matter how well you know the person, it's so freaking hard to put yourself in someone else's shoes. What I find is that I love user tests, I love putting a piece or a feature in front of users and seeing how they react and usually it's not the small things that you notice along the way. Usually very quickly you're like, oh yeah, my mental model was right, or it's completely wrong. And you feel that very quickly, right? There are lots of different people out there. People have different preferences. So there's a lot of noise, but usually you're like, oh, I had this implicit assumption I didn't realize I had, and that thing is wrong. And therefore everything downstream is going to be a piece of crap. Or, okay, maybe it didn't hit exactly the bar, but it's generally there. We're getting there.

Dan Shipper (00:17:47)

I mean, I really agree with that. I mean, I do think you can make products for people that are not like you and sometimes that's necessary, but there's something about that building mindset that I think people miss or people misunderstand or misapply, which is, and I've done this too— There's that whole, over the last 10 years, it's the lean startup and you want to treat business like a science or whatever. And so you want to objectively study your users and your customers and sort of find out what are their needs and what are their problems so you can solve them or whatever. And I think what people don't realize about that is if you go in without a perspective you evoke that lack of perspective in the other person. They don't have the answers. They don't know what their problems are. And to some degree, in your mutual interaction, you are creating a world that potentially has a problem or doesn't have a problem. And there is some reality there, but you have to evoke it in the right way. And so when you come in with a product or a specific perspective, it's much more valuable for figuring out what's working or sort of training that intuition vs. going in with just so what kind of product do you want? Or what problems do you have or whatever? You just are going to get blankness because you're presenting blankness. And that's a very subtle distinction that I think is really interesting.