$1.5B AI Founder: The ONE Rule for Building an AI Startup in 2026

If you believe AI is going to have a big impact, then you should try to stay close to it. Think about the core things that you are good at and figure out how you could augment those with AI.

This is Christopher Pedregal, CEO and co-founder of Granola, the AI notepad valued at $1.5 billion. He built it in 3 years and turned it into a standout product in a crowded AI market. When it comes to competing with big corporations, is there still opportunity to build something major in 2026?

There are a lot of products out there, a lot of people trying to do things. It's like, can you care more than everyone else. There's so much advertising that's happening that I think that if you don't have a product that itself can like pop out and get noticed and loved, it just feels like a losing proposition.

If you had to start from scratch today, what would be your playbook?

I would definitely build shit.

Welcome to Silicon Valley Girl. I am so excited to be chatting with you because we've been using Granola for, I think, over 6 months now. I know you launched earlier, but when we discovered it and started using it, it's an amazing product for our team. And what I'm going to do, I'm going to launch a task right now so it starts recording so that at the end of the conversation, we'll be able to see how it actually works.

Yeah, perfect.

Back in 2024, you said it's so much easier now to build very specialized workflows for a small group of people versus earlier because now you can use AI. Your bet was that with current AI tools, we can build something for for a small group of people because we don't have to use so many resources, right? Because we can vibe code all this stuff, we can ship faster versus like 10 years ago, if you wanted to build something, you would need a huge team. So serving small group of people wouldn't really make sense. Do you think it's still the case in 2026 or we moved to a world where anyone can vibe code anything so that you don't really need to build a very specialized tool? What is your sense of the market right now?

The one thing I know is that everything's changing and it's hard to predict the future just because people can vibe code things doesn't mean that we're only going to use vibe-coded software all the time. I think vibe coding, building tools from scratch, is incredibly powerful if you're building like an internal tool that your team's going to use. I still think that there are certain areas where you want to have the best possible tool, and that takes tons of time and effort and care and continued investment. So I do think there will be lots of software out there that exists, but it's hard to predict exactly where the line of what will be vibe-coded versus what will we pay for.

Do you feel like it's much harder to build now? Because you have experience building in pre-AI era, comparing that to building Granola.

Harder now?

Yeah. Just because I feel like the market is so crowded. Like you see, because anyone can become an entrepreneur. If 10 years ago, in order to build something, if you were not a coder yourself, you needed to find an engineer. If you were an engineer, you needed kind of to find a product person. So you needed those resources. Feels like now there are so many solo founders. And the market is really crowded.

It's two sides of the same coin, right? I think it's so much easier to build now. You don't need as many people, as many resources, which means more people are trying it. It's kind of like when digital photography became common. It used to be really hard to get a camera, right? And then only a few people had cameras. There were photographers. And then digital photography made it easy for everyone to take photos, right? It doesn't mean that professional photographers aren't still needed and way better than your average person. So I think it's the same thing with with starting products today. There are a lot of products out there, a lot of people trying to do things. A lot of them aren't that good, right? And I think that's what it's all about. It's like, can you, can you care more than everyone else? And can you create something better? Like, what I have seen in terms of the one thing that seems to help an AI company break out from a crowded marketplace is just that the product actually works and is the experience of using it is better than the alternatives. We see people are more willing to switch for slight improvements in products now. So it's like people are very, very attuned to the quality of the products that they're using. And in that sense, I think it's no different than before. It's like you still— like it's just you have to fight for that and you have to be really, really focused on it.

I really like how you said you have to care more. I feel like that applies to any niche where you're competing, anything that you're doing. If you care more than others, And this is kind of, this helps you stand apart. Talk to me about launching a product in the AI era because you didn't do like a public launch. You started with a few users, saw their reaction. If you had to start from scratch today, what would be your playbook?

Yeah, I think the conventional startup wisdom before was launch as soon as possible, get feedback from real users, and then iterate your way to something great. And I think because of precisely what you said before, there's so many people putting out slop, like putting out crappy products, that it is actually a differentiating factor, a differentiating approach if when you launch, when you come out into the world, your product is better. Yeah. And so we didn't have this whole strategy about building an AI. We were just like, our philosophy when we were building Granola was basically we'll do whatever it takes to learn as quickly as possible what we need to do to make our product better. And for the first year, the way we learned the most was literally by sitting next to someone, watching them try to, you know, install it and use it, figure out everything that was wrong, go home, try to fix that, do it again the next day with a new person, and do that over and over and over. And we didn't need to launch publicly to learn what was wrong with the product because every day we would see exactly what was wrong with the product by just watching one or two people use it. After about a year, it got to the point where it was like, it was actually pretty good for those folks. We said, okay, it's now time to launch publicly because then we'll learn at scale what's wrong with it, how we can make it better. That was the approach we took. I think that really, really made the difference. In a world where anybody can make software, the only thing that really matters is how good is the software that you're trying to use. If you said if I was starting it from scratch, I would definitely build in private, or closed beta until I felt really, really secure that the product was meaningfully better than the competition.

Yeah. And when it comes to picking out idea, was that your initial idea, smart notes, or did you have to iterate through ideas as well?

Yeah, ideas are tricky, right? Because it's like on one hand you want to be thoughtful from a strategic standpoint. Like if I build in this space, is it a dead end or is there a big opportunity there?

Yeah.

Right? And that's kind of high-level thinking. And then on the other hand, really a lot of building is it's better not to think and it's better to just put something in front of people and, and learn how they react to it and, and kind of, uh, follow that, you know. And, um, and I think when you're— if you're trying to think about a startup idea, you want to do both of those things. You want to make sure you're building a space where there's, uh, there, you know, there's like—

future.

Yeah, there's a future, exactly. But then it's sometimes better not to think too deep. Like, once you make that bet, uh, it's almost better not to think at a high level and an abstract level and really just to like follow the scent in terms of what people like. And, um, in 2022, I, I came across LLMs for time. This was about 8, 9 months before ChatGPT launched. And I was immediately, I was like convinced. I was like, okay, this, I don't know what this new technology is, but it's going to, it's going to change everything. It's going to change all the, the tools we use for work, for productivity. I felt that very strongly. And I knew, I knew that was a space I wanted to build in and be excited about. But then when we had to figure out where to start, that's where we put some prototypes in front of users, in front of people. And they didn't care about most of them, but this idea of like a real-time notepad that would take notes for me and that I could interact with, people's eyes really lit up when we put that in front of them.

Was it like a just word-by-word description of what you want to build, or did you vibe code?

Well, vibe coding wasn't a thing. Yeah, vibe coding wasn't quite a thing, but it was more the vibe code. It was, um, I'm a big believer in prototypes, so like cheap basic prototypes that let people actually mess around with a thing. I feel like you're going to learn a lot from that, so My co-founder and I built a few different prototypes and the notes one was just, it was just some JavaScript on an HTML page that we threw together manually. But it was enough to give you a flavor of what it would be like if it worked properly.

How did you select those first people who were evaluating your idea?

Friends, friends of friends. It was just people we had access to.

Was there any qualification criteria? Because now that I'm thinking about it, if I'm trying to build something, I also wanted to put in front of the right people, like people who are maybe paying for a lot of tools, people who are working in a big corporation, so they have access to, you know, some budget because just randomly asking people.

Yeah, no, that's, that's a good point. Maybe we were a little bit more thoughtful about it than that. Like, we built Granola for ourselves, and by ourselves we mean people who are like knowledge workers, tech savvy, um, are using different types of tools, like live in tools like Slack and Linear and, uh, Superhuman and Gmail. So the folks that we would talk to were oftentimes folks working at startups of different sizes, just because that was kind of the the environment that we were in.

What was your criteria of deciding whether to drop idea or continue working on it? Was just like somebody said yes, or were you tracking something? I know I was talking to Josh Woodward from Gemini, and he said the way they test products at Google, they watch how eyes light up.

Yeah.

When users start testing it. So they don't really have a metric. They're like, they're relying on this intuition that they're seeing this. Kind of surprising for a company like Google, right? Where you expect like a metric after metric.

Yeah, it was very, very intuition and qualitative. And in the early days it was the opposite. It was watching a lot of people being frustrated and unable to actually like use the thing the way we wanted them to.

When it comes to competing with big corporations, right? Because we have Zoom who has AI. Like every product now has AI notes. Can you walk me through your mindset, entrepreneurial mindset? Because When I'm building something and I see a large company releasing something similar, like my first thought is, "Oh no, I'm done." But then I'm like, "Okay, we're going to make it through." Things are moving so quickly and companies are launching things all the time.

And I think now we've all gotten a bit more used to it, but maybe a year, year and a half ago, it just felt like, oh, the world's like just like the sky is falling and the world's changing every 5 minutes. When we launched Granola, AI note-takers had already been around, like the earliest ones had been around for like 7 or 8 years. So there were tons of AI note-takers. The Zooms and the Googles of the world, they already had AI note-takers, not as advanced as the ones they have now, but they already existed. And when we would go and we'd interview people and try and understand, like, were they using them, were they being useful or whatnot, it became really clear that they were only like marginally useful. And they weren't actually kind of doing the job that people wanted from a tool like that. I guess what I'm saying is like, we kind of did it to ourselves because we entered this like crazy saturated space. And I think we were able to break out because even though Zoom or Google create notes and Granola creates notes, the way we've designed Granola, the way we think about it is just very, very different from those tools. And Granola is very much a, it is your personal tool. It's like your personal notepad that you are in control of. And you, you can put notes in there. I can go into Granola and I can basically chat with all my meetings from the past 2 years. And as the AI models get smarter, the level of like insights or the level of conversations I can have across that corpus gets smarter and smarter.

Yeah, I would love to talk to you about that later in this interview because this is like, if a company is not recording their meetings, I think they're losing of 50% of what they can build later with all of these insights they're getting. Because this is their employees' taste. This is the way they make decisions. This is the way they move. And the only way to teach AI how to mimic or enhance that is to record it. Is to give it the context. Yeah, it's just, yeah, it's the context. Exactly. All the data. I want to thank the sponsor of this video, Granola. Granola is one of the apps I use every single day. There is a rule that I made for myself. If a task repeats and it's not the work that actually makes me money, I automate it. Cleaning up meeting notes was one of the first tasks I actually automated with AI. Every call I take— strategy partnerships, team syncs, intro chats— gets recorded and sorted. And because I've been doing this for a long time now, at the start of every new call with the same person, I have a clean list ready— what we agreed on last time, what I still owe them, what they still owe me. It's not a bot that joins your call. Nothing is really visible to the other side. You stay fully in the conversation. And after the meeting, Granola transcribes everything and turns it into a clean summary you can work with. And of course, at the beginning of the call, you disclose that you will be recording this with Granola. So here's my real example. Last week I was in a partnership call. We were going through financials, timelines, deliverables. There were a lot of moving pieces, and my manager was not part of that call, but I really wanted to send her a follow-up email. I felt really engaged in a conversation. I could focus on the person I was talking to without having to take notes of every detail because I knew every number and every date would get captured. After the call, I opened the transcript and I just asked Granola to create that follow-up email, pull a list of deadlines. Drafting the email part took me about 25 seconds and I copied and pasted. That's it. As if my manager was on the same call with us. My team and I have been using this for a few months. We miss fewer things, which really matters because with AI, the number of tasks we're tracking has actually gone up a lot. If you want to try it, use the code MARINA and get 3 months It's free. The link is in the description. And now let's get back to our conversation with Christopher Pedregal. For an entrepreneur who's starting today and thinking, okay, I really want to build this tool, but I'm afraid that a big company is going to release a similar— I don't know if you watched Google I/O, but they released a very similar tool to Whisperflow.

How did they?

It looks the same. The small bar appears, has the— but the difference is while you're talking to it, it also references all the files you have in Google Drive and Gmail. So you can say like, oh, by the way, insert a table using this data and it's going to do it. It's not only transcribing, it's also adding context. And I'm like, I can see how I'm still using Whisperflow because I want just the transcription, but I also see how I'll be using more of that as well. Like, can you give advice to an entrepreneur who's building something, but again, constantly in this era when everyone's competing with everyone?

It's a great question, right? And it's one of those things where it's, I think if anyone had a crystal ball and could say like, okay, there's, there's an extreme world where we are only using like— there's only one tool in the future, right? We use it for everything. And there's a different version of the future where we use even more tools than we have today. I think we'll end up somewhere in the middle, but it's hard to know exactly where we'll be. The way I would think about it is— so it's basically a 2x2 matrix. It's basically how frequent is the use case that you're going after and how important is it for the user. If it's an infrequent use case, then I think it'll be really tough to compete with the larger companies or the larger tools that are more established. I think the— if it's an infrequent use case, people will go to the ChatGPTs or the Claude's most likely, right? In the same way that you didn't see a lot of verticalized search engines in like the 2000s because people are just going to Google and it's easier. They have a habit and that's where they would go. So I think you have to choose a, a use case that's very common. Because if it's common, you have an opportunity to build a habit around it. And there the question is like, is it a common use case that the user doesn't really matter if you do a much better job at? Or is it a use case that's really, really important to people, right? And I think you want to be in that corner where it's basically, it's very important to people where if the product experience is even just like 10% better, like that's reason enough for people to switch to you and use it. And if you're in that quadrant, then I think it goes back to this. If you care more, and that's the one thing you can do over the big companies is like you can just care more because they have to care about a lot of things, right? Then you can build that better product. And I think you can, you can compete.

Do you think we should add a niche to whatever you just said?

Yeah.

Because I feel like if you're just going after a frequent use case for billions of people, then it's a big corporation kind of playfield. Yeah. But if it's a niche, like for your product, it's like people who are fixed on their productivity, want to record, want to be more effective with their notes.

Yeah, or you become a really big company one day. I think that's— Oh, or that. Or that. We definitely want to have billions of people using Granola at some point.

And you're moving into B2B. You started as B2C company and now you moved into B2B. How is that shift?

Well, yeah, it's a good point. Our strategy was to mimic like a Slack or Dropbox, like those types of bottom-up companies. So it's basically a product-led growth. So the idea that someone inside of a company discovers Granola they fall in love with it, they tell their colleagues, we kind of grow organically inside of the company. And then at some point, someone in a position of authority, maybe it's like the founder, maybe it's the Chief Compliance Officer, what have you, legal officer, security officer, says, "Whoa, everyone's using this software. We should probably like pay for it, have control over it, make sure we know where our data is going and all that stuff." And that was always the plan. We always knew that we would be selling to companies. But at the beginning, we were just worried about— not worried, we just focused on just trying to build something people actually wanted and liked. And that worked. So Granola did spread like virally, organically throughout companies. And now we have some very, very large companies who are on enterprise plans with Granola. But it all started either bottoms up where it kind of, it spread through the company virally, or the founder or CEO was using, like heard about it and was using it themselves and found it valuable and said, actually everybody should. Should be using it.

I think it's a great B2B marketing plan when you start with a consumer. How did you get to those initial customers?

We posted on Twitter. Like, that was basically—

By yourself? Just like the founding team, or?

Yeah, yeah. I think we posted it from my account, and I didn't really have a Twitter following at all. And I think we just, we got really lucky. It was this idea, the way Granola works is like, it looks like Apple Notes. It's a notepad. And then at the end of the meeting, it'll take whatever notes you wrote, and it'll flesh it out. And there's this really nice animation where you see your notes get like filled in. And we had a GIF of that. And at the time, I think a lot of the startup founders or leaders out there were really interested in new UIs or interactions around AI. And so we had a few famous— like Guillermo from Vercel retweeted my tweet. And then Nat Friedman also retweeted. So it's basically there somehow. And I don't— just like the universe made this happen. It just caught a few people's eyes and they tweeted about us. Then we just started growing. The first day, I think we got 500 installs.

That's pretty decent.

It's not bad. It's more than I expected, but it's also a drop in the bucket. Then it just started growing little by little by little because we weren't doing any marketing. The crazy thing about Granola is all the old-school AI note-takers, they're all super optimized for like growth hacking. They'll look like at the end of the meeting, they'll send notes to everybody who is in the meeting, whether they want it or not, whether you want it or not. And Granola doesn't do anything like that. It's like, Granola is like, our only job is to serve the user and give the user wings. The fact that Granola was entering a space that was super crowded and had zero growth loops built into it, and it still grew virally, organically, and kind of was able to to pop out and become really visible in that space, I think is a really, there's something going on there. I think it's a really strong testament that people are hungry for just better software and better software experiences. Good products. Yeah.

Yeah. Interesting. So basically all your marketing is based on people loving it. Great product.

Yeah.

Exactly.

The whole company is based on that. Yeah.

That's amazing. So would it be your advice for any entrepreneur building something? Don't think about marketing yet. Just think about the product and people sharing it?

I think so, because I think, um, to your point earlier, it's so noisy out there right now. There's so many people doing so many things and there's so much investment, right? So there's so much, um, advertising that's happening that I think that if you don't have a product that itself can like pop out and get noticed and loved, it just feels like a losing proposition. By default, I always think about very user-facing products. I think it's very different if you're going after customer support there. It's, I think, all about having the right sales motion and marketing is a part of that. But generally, I think if you don't have a good product, you're like, it's the one thing that you can go and make better with a small team, right? And I think you should do that upfront rather than doing, do that later.

I have one final follow-up question here. How many initial users did you have? So how much feedback were you collecting before pushing it out?

Yeah, we had about 150 active users after that.

That's your friends and that's like your inner circle.

Yeah, yeah, yeah. Friends and friends of friends.

What were you tracking when you gave it out? Like, because you couldn't see their eyes, right? But who were you tracking? The frequency of use?

Yeah. So what we would do is we would set up a first call. We do it in person if we could. Otherwise, we do a video call where we would ask them to share their screen and then we would watch them try to install Granola and try to use it without us saying anything. And then we would schedule a call in 3 days, again, share their screen and walk through the meetings they use Granola for and talk about what was good or not. And that That was the highest signal. That's like the super qualitative. That's where you learn the most. But then once the product started getting good enough that people would actually use it, then we track usage. And there's this thing I had never heard about it before Granola, but one of our mentors told us about it. And it's a thing called a dot plot. And a dot plot, basically, it's, think of it like a spreadsheet and every row is a user and every column is a day. So the default dot plot we had would show the last 30 days. And then in each cell, you basically put, for our case, how many meetings did they use Granola for on that day. And then you change the color of the cell. So if they used it for 10 meetings, you should make it dark green. If they use it for 0, you should make it white. And then you can, at a glance, very easily see the patterns. And the idea is that you start with a dot plot when the product's not very good. And then you iterate, and you iterate, and you iterate. And what you should see happen. You should see it light up. Because normally when you do analytics, you group all the usage together and you just get like a usage graph, right? Which you're like cumulatively, are people doing more meetings or not? But that's not actually very helpful in teaching you what's wrong with your product and is it working. And then a dot plot, you can see things where it's like, oh, okay, like this person was using it a lot and then they stopped using it and then they like clearly remembered it existed and started using it again. Or like maybe they went on vacation or you could be like, Oh, actually there's a few people who started a little bit and then they had like one day where they did like 5 meetings and from then on it became a habit, became hooked. You can be like, oh, how do we get people to get to have that kind of day? So it becomes like a very easy visual way to get a, to stay on top of the pulse of what's happening with your users.

Amazing. All right, talk to me about your AI stack. What are you using apart from Granola?

I struggle with this question because I try to use Granola for as many things as possible. And I use— we haven't launched it yet, but I just got this Apple Watch and that is a— that's a really nice feeling because it's just here all the time.

Yeah, this is how I take my notes when I go to conferences. Either I write—

What do you use?

Just voice notes and they go to my phone and then I use whatever we're using to transcribe. So it's a journey.

It's a few steps.

From my notes.

Yeah. But in terms of form factor, I think the Apple Watch is 'cause it just feels, it feels very natural.

Oh, 100%.

Yeah.

It, it should be the form factor for all the conferences and everything. Yeah.

Yeah. Yeah.

Yeah.

Um, and then, uh, next, I mean, I use Claude, right? Claude's probably my second one. Um, and then one of the engineers, uh, at Granola set up this internal agent. We call it Nacho. I actually don't know why we call it Nacho. Um, but it has like a little nacho as the icon. And, um, we've connected basically all of our internal tools to this one agent. So literally every single data source that we have. Is accessible to this agent. And there's like an internal portal, but we also interact with it in Slack. And that one's really interesting. So for example, I will oftentimes, what will happen is like, I will notice something kind of weird in the product because, you know, that's my job and I'll post about it in Slack. And then someone will ask Nacho to like, be like, hey, can you look at the analytics for the last couple months and see if that supports like Chris's, you know, annoyance or whatnot. And then That'll come back and then someone will be like, okay, here, what if we change the way this worked and put a button here instead? Then you ask Nacho and Nacho goes and talks to Cursor and prepares a change. It still goes off the rails all the time. We have to be like, no, Nacho, that's not what I wanted, or think harder, you made some assumptions here that aren't right. There's still a ton of human back and forth. But it definitely changed the way we've worked internally.

So it's basically I'm trying to describe this role. Well, what is— he's not like a C, like a chief of staff. He's more like goes to— does he help him with strategic decisions or is mostly like, pull me data?

It's a lot of like, pull me data, do this thing that would have been like 30 clicks before, you know, or like opening up 3 tools and saving data into a file and uploading it somewhere else. Just do all of that for me. So it's, it's in some ways it's like maybe an intern would be the right, you know, it's like we're not, we're not outsourcing big decisions. We're going to be like, hey, go pull up the data, go pull up this thing, look at how those two connect. Okay, here's what we want to do. So it's very much the ideas are coming from us, not, not from Nacho, but Nacho's executing on it.

Did you use a tool for that or did, was it like built from scratch? I wonder like, because if you, you can totally build this with Perplexity Computer?

Yeah, we didn't. So I actually have to ask. We didn't use anything like that. It's something like ClaudeBot, but it's not ClaudeBot. I can't remember what it's called. And we run it ourselves. So that's why we're comfortable with all that data going through this agent, because we run it and control it.

Okay. What haven't you delegated to AI yet, or what are you doing without AI?

I think a lot of building great product, it's all about how does this make me feel?

Mm-hmm.

Right. And a lot of it is human intuition-based. It's trying to put myself in the shoes of another person and imagining how they'd experience that. I just don't use AI for that.

And you can't. What you're describing is something so uniquely human.

We have some really young people on the team and they, they just naturally default to using AI for everything. It's just like their default behavior. And more often than not, I look at that and I'm like, oh, that's clever. I wouldn't have done that, but that's actually really smart and I should do it. I think the product stuff is probably one of the last things, at least in our immediate work, that I think we'll get. I don't actually know if AI will ever fully get in there because I think, I don't know, the lived human experience is actually the one thing that we have.

Totally.

Totally. What we'll do though is, so we'll get lots of feedback from users, right? And then grouping, classifying that, basically making that feedback, putting it into a form that's really easy for us to like build intuitions on top of and making decisions, super useful for that. But then actually what do you do with those intuitions? What changes do you want to make? That's still very, very manual.

And it's a very founder-driven thing because you're like the soul of the product, has your vibes, so it has to have your feelings. I don't know if you can even put it into a product, but I love that. Do you have any, I don't know, I call them magic prompts that totally change how you interact with AI? For example, I just asked my Granola, can you identify bottlenecks in my company? And it went and analyzed my conversations like, number one, and you know it, you're the bottleneck. That was number one. And then it came up with a few few more things that we're currently fixing. Do you have any other prompts that anyone can use with their AI that's going to change their work?

It all comes down to the AI needs to have enough context. So if you use Granola in all your meetings, then it does, and then you can ask it some pretty incredible things. So let's, let's just assume that the person's doing that. Um, the, the things that really opened up my eyes and where I was surprised at how good they were, were, um, coaching-level things. Like, that really— there's but there's one recipe in Granola which is called Coach Me Matt. What's kind of great about coaching is that harsh— if you ask an AI for feedback, an AI can give you harsh feedback and there's no person, like, it's not worried about hurting your feelings, right? So that— and it's— an AI can say something to me and I think I can hear it better than if, like, let's say my wife said something to me. I might be a little bit more defensive, if that makes sense. So, uh, anything around, like, deep coaching— hey, what are these like patterns you observe and how I do things that maybe I'm not aware of that are not helping or that I could improve. That's a really big one. Oftentimes I'll go into other tools. If I use ChatGPT or Claude, they feel quite dumb to me compared to Granola because they don't have all that context baked in.

But you can connect now. I mean, you know.

No, no, I can. But what I mean by that is, so I have 2,000 meetings in Granola or 2,500 meetings, right? So.

Wow.

When I ask Claude a question, it doesn't read 2,500 meetings, right? It'll, it'll read 10, and it'll try to form a picture about me from those 10 meetings, right? So I have a recipe in Granola which basically says, look at my last month of meetings and write me 5 pages about who I am, what Granola is, what's the Granola product.

Can we try that? Can you give me my Granola? Let's do it. Yeah, I really like this. Let me, let me see what it tells me.

Perfect. So if I go here And then we just say, I'm going to use ChatGPT to do some work and I want ChatGPT to understand who I am, what I'm working on and what I try to— what I'm trying to achieve. So it'll have better context about me. So please look at all my meetings from the last month and write 3 pages that I can paste into ChatGPT. So I'll have all the context on what I'm trying to achieve.

Wow. Oh, nice. It even extracted some stats. Pushing cadence. Yeah. Nice.

So what I find is now if you take this and you can go to any AI out there, you can go to ChatGPT, you can go to Claude, you can go to any, anything. And if you, if you just say, here's some context about me and you paste this in and then you ask whatever you're going to ask, the AI will do such a better job answering your questions because it understands so much more about you. Exactly.

So because it's connected to my Claude.

Yeah.

How can I ask Claude to self-update using this? To like add context to all my projects?

Yeah. Well, I mean, there's probably some way where you could set a trigger where it does it every day or something like that. Or you could just wait for us to launch that soon. Because it would— that would be kind of cool, right? If this thing— basically we're building a version of this where it will auto-update every day. And then you could just use that context anywhere.

Yeah, that's— this is fascinating. So I feel like you are building something like a virtual chief of staff based on—

moving towards—

based on this data that you have. I also write a newsletter where I go deeper on AI tools that I use, career strategies, and things I can't fit into a 30-minute podcast. It's free. Link is in the description. We've been talking for almost an hour, right? And I'm remembering some things. But there are some details that I might be missing that are important to me. How does Granola work in terms of picking out those details? How does it decide what to surface?

In the notes, you mean?

Yeah, yeah, yeah.

Yeah, so—

Oh, nice. It gave me product strategy and crowded markets. I really like it. Building in the AI era, market dynamics, product strategy.

What we realized early on is that what are good notes for you would be very different than what are good notes for me, right? So like the point of notes is really dependent on who the person is and what they're trying to achieve. I think we might have been the first to do this. It's basically notes are generated for each person and they're different for each person. And what we do is we take as much about the person into account as possible. So I don't know if this was a calendar event, but let's say you join a Zoom meeting and use Granola. Granola will go and try to do research and figure out who everyone in that meeting is and what their roles are. And then we'll use that to figure out what the meeting's about and like what should be highlighted in those notes.

If I tell Granola my goal for the next few calls is to, I don't know, make sure we follow up with everyone if we had agreed on a to-do list.

Yeah.

Would it be highlighting that for me in every meeting? Does it have like a universal memory of how I want my notes to be presented?

Not an automatic one yet. Yeah. So that's something you can go and set up a template in Granola. You can basically say, You can have different templates and so you can kind of say, I want notes in this structure. During a call or during a meeting, you can be like, hey, Granola, make sure to say include this in the notes and it'll do that. But it won't, it doesn't have like a memory about you said this in the last call, so I'm gonna do it in the next call. Which is, you have to be careful with memory, I think. Like memory's super powerful, but with explicit instructions like that, the reality is like we underestimate how much things change, right? What you don't want is you don't want an instruction that like you said—

From last month. You said something last month and like Granola still thinks it's really important.

Important.

Like, my ChatGPT still thinks I want to be an actress. Yeah, just like 3 years ago.

There's a guy on my team which was like, I mentioned, uh, muffins, you know, like he, like he had one question about muffins that ChatGPT wants, and like now, like, ChatGPT just keeps bringing up muffins all the time. It's like, as a muffin connoisseur, you know, it's like, no, I just, I just asked about muffins. And same, that's why you have to be a little bit careful with— and that's the difference if you use Granola a lot The thing is, is like there's just so much richness and context in our conversations. It's a little bit like, I don't know, think about your best friend and think how many hours you've talked to your best friend and like how well they know you. It's like very, very different than like if you're just chatting with something like ChatGPT or Claude. It's like a very, very superficial—

Very granular. Yeah. Yeah. But once we fix that, if we can make this dynamic memory based on like asking AI to identify my priorities on a certain day, then this can become my chief of staff.

Yeah.

If it can just pull those things like, oh, now Marina's focused on that.

Yeah, yeah.

I'm going to help her in this meeting by suggesting these questions. I'm going to identify this process that's broken in her team clearly because I've heard it in other calls within her company.

Yeah.

Because for me, like recording my calls This is the way— it's a way to build a virtual chief of staff, which we're trying to achieve.

Yeah, I think almost what everybody in AI is trying to achieve, really, right? Yeah. Like, that's, I think, one of the things.

Some level of autonomy. Because for now, I feel like AI has made us much more productive, but it only means we're working more.

Yeah.

Because we see all this productivity gains, we see how much better it is, and we just work more. What I want the next step to be is like, give us some more free time. It's summer. I want to take a few weeks off. I can't. I think we kind of do that.

We do that to ourselves though, a little bit. Oh, true.

But this is our nature.

Yeah.

And it's interesting whether we're going to cross this period in time where AI is helping us with strategic decisions, so we intentionally take more time off.

Yeah, yeah.

I don't see this happening now.

On the point you were talking about a second ago, which is there's this interesting question of how are you going to interact with this chief of staff or this AI? How directive are you going to be? Basically, are you going to be like, "Oh, always do this," or give it instructions and it always follows that? Or I think there's a different model, which is like the AI is almost a little bit invisible and it just observes what you do and then tries to infer from that what it should be doing.

Yeah, exactly. That's what I wanted to do.

Exactly.

What will Marina bring up in the next meeting based on her previous meetings?

Exactly. Yeah. A good example here, months ago I tried building a version of Granola generating follow-up emails. So like you'll connect your Gmail and it will just learn from your previous messages.

With that person? Or in general?

Yeah, no, both. Both. And so, so like an example there that's really important is, for example, let's say oftentimes people will need to send a link to like an important doc. Like for example, before this you sent me a doc saying here's some instructions, right? That doc might change. Like next month you might decide to use a different doc.

Yeah.

And if you had to tell Granola that you changed the doc, you might forget. Whereas if it has access to your emails and it notices that, oh, you now use this new doc, I'm gonna start using this new doc. You don't have to think about it at all. So I think there's, I think a great model for AI is one where the best designed things become invisible, right? Exactly. And I think the, like the best AI is gonna be stuff that you don't even realizes there.

Self-learning, self-updating, learning from what's changing. This is exactly what you're describing.

We try to get everyone at Granola to think about product in the same way. And when someone new joins the company, I basically paint them this picture where I want Granola to feel like a handrail. You know, when you have stairs, there's that railing. And people always look at me like, what do you mean by that? It's like such a weird thing. And I'm saying, well, handrails are basically invisible. Right? They're on every staircase. You never notice them, right? You don't pay attention to them until you trip, right? And then your hand shoots out and it needs to be like right there.

Yeah.

And it needs to like hold your weight. Exactly. And it's a really, really important thing and it needs to be like super intuitive. But then you go back to living your life and going up and down the stairs. And, and that's, that's how I want Granola to feel like. I want Granola to have your back, to to in any moment of need, like if you're tired, if you're tripping or whatever, it's like we're right there for you. But otherwise, you're the star of the show. You know, you're out there doing things. You're, you're living your life.

You know, whenever I post, like, I'm excited, this company just launched this and I've been using this company for so long now I can do this. Oh, why are you happy? AI is stealing your data. Oh, AI is going to replace you in 3 months. Oh, like corporations are just eating, eating us, whatever. What would you say to those people?

I think the world's going to change a lot over the next couple years, and I think whenever there's a period of a lot of change, there's going to be turbulence. That is just a reality, and I don't think we're— I don't think anyone knows exactly where we're going to end up. I'm excited about AI as a tool that augments us and enables us to do more and better things than ever before, um, and I think I think there's a lot of areas where that's the case, where actually AI is not going to replace people. It's actually going to let people do more. And there are all these examples in history where all of a sudden if something becomes more accessible, the demand for it goes up because now people can use it. Jevons paradox, I think it's called. Yeah. It's not going to be everywhere though, right? Like there's definitely going to be pockets of society where it's going to be very disruptive, right? And I think that's happened lots of times in history as well. That it's like change can— changes can be exciting, but it can also be really hard. I think it's important to hold the excitement but also the, the reality of the downsides, uh, in our minds at the same time. Um, because I think that's what's going to happen.

What do you tell yourself when you have fears about AI, if you ever have them? I—

so my, my view there is I think about what I can control. Generally, this is my like philosophy in life. I think about what I can control and things I can't control. And I don't worry about the things I can't control. And I think about the things you can control is if you believe AI is going to have a big impact, then you should try to stay close to it. And by that, I think you should, you should try to use it. And I think that's really the only thing you can do.

Totally, 100%.

And I think that, um, and I've seen this, I've seen— because I think what's happening with, uh, engineering is it's like you can kind of see what happens in engineering, and the same thing that's going to happen with coding and engineering is going to happen in other sectors years later. And I see, again, we have a guy on our team, he's 20, and he's, uh, the way he uses AI is incredible. And he's just able to do all kinds of incredible things that I never would have expected. And so, like, I think the only advice I have to people is, like, don't shy away from it and lean into it. And that doesn't mean you need to— there's a lot of, like, AI theater, productivity theater. It's like, I think there's a lot of people, there's almost, like, more talk about how AI has helped them than it's actually helping them be more productive, right? I think we're, I think we're in productivity, AI productivity theater phase. But I think we're gonna come out on the other end of that where it just really does augment your productivity tremendously. But it doesn't mean you have to spend 24/7, you know, like following every single launch. Like I, that's, that's not what I mean. What I mean is think about the core things that you are good at that you need to achieve in your job and figure out how you could augment those with, with AI. And like we were talking about for product, for me, it's not, oh, how do I get ChatGPT to make product decisions? But it's maybe how do I get AI to—

Get all the data.

Get all the data so I can better inform to make better product decisions. So that's how I would think about it. I have a 6-year-old and an 8-year-old, and I think about what's the world going to look like when they're older? And I don't know, right? But it's perhaps going to be a little bit easier for them because is the world's going to change a lot over the next few years. So, well, I don't, I mean, again, I don't know, but it's, they're going to, they're already growing up in a world where ChatGPT is normal.

Yeah.

Right. Whereas I think if you are maybe in your mid-20s right now, like early in your career, and now there's all this change that's happening, I think that's perhaps a harder time. But again, maybe it's easier than if you're in your 40s. Maybe it's not your style. I don't know. You know, like it's hard to tell.

Yeah. Okay. And last advice for founders building. In the AI era? What should they be avoiding?

I think this has always been the case, but it's so much more extreme with AI. There's so much noise. There's so much FOMO. There's so much imposter syndrome. Like, if you were just— if you just look at Twitter, you'd assume that everything is just solved.

Companies are run by agents.

Exactly. And I think the reality is very far from that. And I think ultimately the thing you can do, again, what can you control is you can understand a problem and a user better than anybody else in the world if you really wanted to. And you can just care more about building a really great solution for those folks. And you can have a peripheral awareness of, of other stuff that's happening. I think it's good to understand directionally where things are going, but do not let it mess with your head because it's so easy to obsess and to look at those things and to assume that they have it figured out. And the shiny objects, and it's like the— what's the, the fashion of this week versus that week, or what have you. But the underlying problem that you're trying to solve, that probably hasn't changed at all in the last like 2 weeks, right? Or even the last 2 years, probably. And so like, that's, that's what you need to work on, right? That's your job. It's exciting, but it's also— you have to, you have to manage that mentally, because otherwise you'll you'll, you'll, you'll be too distracted and you have to care more about your particular problem.

Love it. Thank you so much.

Thank you so much for having me.

Thank you.