Product Talk | Chris Pedregal | Granola

I'm looking forward to this

conversation. We have uh Sarah

Drinkwater here who uh runs a solo GP

firm uh called Common Magic. Uh and

we've got Chris Pedro who is of course

our co-founder and CEO. And I will leave

you guys to it. So, thank you very much.

Let's have a round of applause for

Thanks, Jack. Um, hi everyone. I'm

Sarah. I'm really delighted to be having

this conversation um, as a long-term

Granola user and fan. Before we get

started, can I ask you to put your hand

up if you work in product right now?

Yes. Who is a founder right now? You can

be both, of course. Who is an engineer

right now?

Who is everybody else doing cool things?

>> Great.

um

>> who who wants to be a founder.

>> Yes, come and chat to me later. So,

we're going to have about half an hour's

worth of questions and then we're going

to go over to Q&A. So, please be

thinking about things. We really want to

make this an informative, useful

conversation anchored around product and

granola. Um and I want to start by who's

using granola daily now.

>> Great. For those of you that aren't, how

would you describe granola, Chris?

Yeah, Granola is an AI notepad for

people in meetings. So, it's a lot like

Apple Notes. It's an app on your

computer and when you open it up, it'll

listen to what people are saying in the

meeting and it'll take whatever notes

you've written and then at the end of

the meeting enhance it and try to fill

in the things you didn't write down.

>> Yeah. Can we talk a bit about the origin

story? So, this is your second startup.

You sold the last one to Google. How did

you decide to build something

particularly in a space that you could

argue was competitive? Um, you know,

obviously you were based in London

already, but could you tell us a bit

more about the genesis of Granola?

>> Yeah, absolutely. So, uh, my last

startup was called Socratic. It was an

AI tutoring app back before this wave of

AI. So, it was really, really basic

compared to what you can do today. And,

um, after it was acquired by Google. So,

after four years at Google, I quit. I

knew I wanted to do a new startup. And I

didn't know what that would be. Um and I

started exploring and I very quickly

stumbled upon GPT3 at the time. So this

was about seven, eight months before

chat GPT came out and I just started

playing with it and I I everyone here

probably had this moment at some point

in the last couple years. I was just

like blown away that this technology is

different than anything I've seen

before. Um, and I I think as I started

building I just built all these personal

projects on top of um, LLM to get a feel

for it. And as I was doing that, I I

just kind of became convinced that a lot

of the tools we use to do like knowledge

work on our computers were going to get

reinvented on top of LLMs. I didn't know

what that would look like. I still don't

know what that looks like to to be

honest. Um, but I had this like deep

conviction that that was going to be the

case. Um, and I just wanted to be a part

of that. So I started looking for a

co-founder. I was very lucky to find

Sam. Um he was an amazing co-founder.

He's a designer. He can do everything

but he's a designer by by training and

um had come from the tools for thought

world

>> and um we started iterating and you know

one thing led to another and granola.

>> Well and that's a you know you know a

shared passion you and I have is the

tools for thought space. Tell me more

about that for those of you for those of

us in the audience that may not be so

familiar with that kind of sphere.

>> Totally. I So I'm a huge like I love

tools. I you know there's the famous

quote which is you we shape our tools

and thereafter our tools shape us. Um so

like I think they're from a product

building perspective a just super

interesting super fun space to be in. Um

uh I think we like we were just talking

about you invested in Vermont, right?

And like the founder Paul is like super

into um like the history of computing

and like I I think I'm like we're really

inspired by Douglas Engelbart and this

idea of augment augmenting human

intelligence and um like all those ideas

were like floating around when we

started playing with LLMs. I mean it's

it's not surprising. It's kind of like

you you come across this new wave of

this new AI and it feels it's like this

new technology like you discovered fire,

right? And you're like what what are you

going to do with this? It's probably

good. There's probably bad. There's

probably all these applications that are

hard to think about. It's just like what

do you what vision of that do you want

to put into the world?

>> Um and and that's really the the space

where Sam and I started with this. Like

we didn't say,

>> "Oh, we want to build a meeting notes

app." We said, "Oh, we want to let

people like our vision, we're we're far

from it, but we're building towards it.

It's like we want to build tools that

let people do more, let them be smarter,

let them solve problems they couldn't

solve otherwise. Um, like literally

expand like human capabilities and LLMs

and the future generation of AI that is

coming

>> is is like the perfect tool. Like if

>> if the personal computer is like a

bicycle for the mind, you know, to

paraphase paraphrase Steve Jobs, like AI

should be the jetpack for the mind,

right? It really should let us do more.

Um and it should let us do more

collectively. Um but anyway, so that's

the big vision. We're like, oh, where

the heck do we start?

>> That's what I was going to ask you next

is like how you get into the product

process and choices.

>> Yeah. Um and there are two So like if

you're building with AI, AI is only as

useful as the context it has about you.

That's something that I believe very

strongly and if you're we were thinking

okay we're a startup you know we need to

start somewhere how do we get a lot of

context about someone quickly so that we

can be useful and we only really saw two

paths we saw emails right like you know

you could become an email client and

then you have the history of emails you

have a lot of context or conversations

in meetings right because just

>> just one 30-minute meeting is like 40

pages of transcript there's so much

nuance and so much information in there

>> um and out of Those two were like, "O

getting someone to change their email

client seems really, really hard." But

people use all sorts of different tools

to take notes during meetings and it's

usually the thing that's easiest to grab

like five minutes into the meeting when

you're like, "Wait, a that person

said something important like I should

write it down." And um and then from my

previous experience building products,

the thing that I've learned that is

almost impossible to do is to get is to

build a habit in using your product.

like you get someone to try your

product, they say, "I love it." And then

they forget it exists. They never come

back to it. And um

>> beautiful thing beautiful thing about

meetings is that they're in your

calendar and we can send you

notifications. So it seemed like a

really from a habit building perspective

and from a strategic perspective, very

rich place to start because of the

context.

>> Um unfortunately it's an incredibly

saturated space.

>> Tons of players, you know, like

>> like the biggest ones are like eight or

nine years old now. So they've literally

been building for almost a decade. Um so

we we entered it kind of eyes wide open

but we were a bit daunted by that.

>> How do you think about if I compare

granola to say some of those other

competitive products? There were some

foundational product and design choices

that you made

>> you know such as the frictionlessness of

how gr how granola feels on the desktop.

>> Um you know the choice to kind of not

store voice data like how did you come

to these tell me about the process of

kind of coming to these decisions.

>> Absolutely. Um,

it's funny. It's like looking in

hindsight, a lot of these seem like

really smart decisions. At the time it

was like not not so obvious. Um,

it all kind of comes down to this idea

of Granola is a tool for you, right?

It's like a tool for you in a moment

where like the the prototypical user we

have is you're running three minutes

late to a meeting, right? And you need

to get into that meeting and you want to

capt you want to be your best self in

that meeting. Yeah. We want to capture

the most important stuff from that

meeting and you're late and you're

stressed and you're just in another

meeting that was that went over. Um, and

we're building for that moment and that

user and and we want to be a tool for

you and we want to enable you to do

more. Right? So that's the that's the

perspective with which we came to

Granola.

>> And if you have that perspective, a lot

of decisions become kind of simple. So,

one is well, if you're running three

minutes late to a meeting, I can't be

opening up a web browser, figuring out

what tab granola is in. It needs to be

really, really easy to open up. Um, I

can't worry about whether it's going to

work on Zoom but not on my Slack huddle

or actually this is an in-person meeting

and there's no Zoom link. So, I can't be

thinking about that. I just need a tool

that it's like a my goal is to be as

easy to use and that's reliable as that.

>> Uh, right. It's like a pencil and a pad

of paper. um storing the audio like 99%

of your like work meetings you don't

need to go back and listen to the audio

right if you're a journalist it's

different like there are use cases where

it's useful but most of the time for our

target user you just care about the

takeaways the notes so storing the audio

just makes it way more invasive way

heavier makes me wonder oh should I be

using it not whereas if it's just notes

it's much lighter so I think all those

decisions kind of stemmed from that

point of view of it being a tool for the

person and I think if you look at a lot

of the AI meeting bot like companies out

there they don't feel like personal

tools they feel like

>> meeting like what

>> impersonal tools

>> yeah I would say that they feel like

meeting artifact like capture systems

you know it's like here's a I log in

it's like here's a recording of all

these meetings from my team it it feels

very different from like this is

something that I'm using for myself to

do to achieve my goals

How did you come to the original? Like I

remember doing a a user feedback session

with you and Sam a long time ago.

>> How did you come to your first persona

when you're thinking about that person

who's 3 minutes late for meetings which

those of you in the audience that have

done meetings with me may recognize

occasionally.

>> Yeah.

>> Like how did you come to that persona?

Uh

so I think it's like like the the short

answer is we are we started building

granola for ourselves and it's a lot

easier to build a product that you want

or that you want to use. Um and then we

kind of found the most extreme version

of that in terms of like who has the

strongest painoint right and it there

are just so many people in backto-back

meetings. Um and and if you it's kind of

um you know OXO the the right it's like

the the company which basically is like

>> for people who don't know the story it's

like they would build um kitchen

utensils for people with arthritis right

and because it's like can be really

painful like peel a potato or something

if you're arthritic and then turns out

they're just more comfortable for

everybody so they're actually like

really nice tools for everybody to use.

It's a little bit like that.

On that note, my I've got a son with one

hand and his question for you. What

speaking of tools that work in

particular use cases because lots of

products built for people like him are

better for all of us. Um he asked me why

did you call the company granola?

>> I mean that's Sam's brilliance. Like he

so so Sam's last project was called

Custard. So I feel like

>> um the story he told me was like oh I

named the git repo this morning and I'm

like oh you call it granola. It's like I

was eating granola at the time. Um but I

think I think the interesting question

is why' we kept the name, right? And

like I think we fell in love with the

name very quickly. It

>> a lot of productivity tools are like

super this like AI that. Um

>> and something like granola is just kind

of you know it's like a pleasant part of

your everyday life. It's kind of like

friendly but healthy you know and

>> it's a good habit.

>> Yeah. I like Exactly. It's like I I I've

also eaten granola like every day since

we started the company because staying

on brand. Um we actually at at the

beginning I we didn't want to use the

word AI anywhere.

>> We were like oh AI is not the point,

right? The point is the product and um

we did all these user interviews on our

landing page and it would they would

like literally stare at it for three

minutes and it would you know it's like

meeting notes blah blah blah and they'd

be like how is it doing it?

And then they'd be like wait is it using

AI? And I was like, "Ah, god." And then

it's like actually it's very useful to

use AI. So our domains now, granola.ai.

>> I love it. But how do you think about

models? You know, from what I'm

understanding, you're kind of multimodel

or you're using the best model at the

time. How did you come to that decision?

>> Um,

>> does it matter what model, you know, if

we're thinking about loyalty in the

model?

>> Right. Right. Right. So I I guess like

we started a couple years ago and when

we started a couple years ago the big

question was like train your own model

fine-tune your model like how like

>> I think the the narrative at the time

was you have no moat if you don't have a

technical moat right a GPD rapper was

dirty word um

>> and we kind of again from the we just

took a very user centric view and we

said okay what's what's stopping this

product from actually being really

useful to people and the answer was

never the model it was a lot of the

other stuff around it. Um, we also took

a view that the model like I think it's

there's a lot to talk about, but like I

think a huge thing if you're building an

AI is what's the time frame that you're

building for, right? Because I think if

you're saying I'm building this I'm

building this product for five years

from now or seven at some point you're

like we've crossed the AGI line. Who

knows like robots might be running the

world. It's like really hard to build

for that future world. And then if you

build for today, like you kind of miss

the boat because you want to build for

where the puck is going, not where it is

today. And things are moving so quickly.

So we early on we kind of made this

decision. We said we're going to assume

models will continue to get better,

faster, smarter, cheaper uh quickly and

we're going to build for that world in

like one to two years.

>> Um and in one to two years when the

models are smarter and faster, what's

going to be holding back the product?

And that's like the long list of things

are like um are the notifications

popping up and like seamless enough. Is

it like in granola you can take your

head put your headphones on and take

them off and it just works but there's

all this stuff that we have to do behind

the scenes around echo cancellation for

that to be the case. So um we made like

a very strong decision to not uh try to

innovate at the technical level at the

model level um until it was absolutely

necessary. So yeah, the the byproduct of

that is we use the best model on the

market at any given point. Um we

flip-flop ideally as quickly as

possible. Um and we use the most

expensive models because we think most

of our users are in the future and we

basically want to be building for

tomorrow and the best way to build for

tomorrow is by using cost prohibitive

models today that will get cheaper and

and more accessible uh tomorrow.

Like relatedly, how do you think about

when you think about your product

roadmap and building for people that

live in the future?

>> Yeah.

>> How do you think about sequencing? So,

for example, you know, I was really

intrigued by Granola launching like I'm

now using the mobile app very often when

I'm having an inerson conversation and I

would not have predicted that in my

behavior a year ago. like how do you

think about choice and kind of and kind

of choosing your version of Granola's

future in such a kind of uncertain

moment in time.

>> So I think the I think the really hard

thing when you think about road map and

is in this space is uh how much do you

build for the local maximum of today and

how much do you build for whatever the

global maximum or the the local maximum

of of tomorrow

>> and it that's an incredibly hard

question. I It's like we take it day by

day. It's it's hard to know. Um I I I

can tell you I I do believe that if we

go back to the origin story, this idea

is this um this tool that augments your

abilities like that's something that's

way beyond meeting notes, right? And if

we don't invest and build in that today,

like we might never get to that. And

that is I think extremely important

because the industry is moving very

quickly.

>> Um and I also think that like that is

the big prize. Like I I think this idea

of software that can assist augment you

across all your work tasks using your

personal context like that's kind of why

we get up in the morning and that's what

we're excited about building. Um that

said there's like a million bugs in our

current product today that we want to

fix and you you kind of need to do both.

>> Um and what's hard about that it's it

takes a very different mindset. So the

>> the mobile app is um

>> it's it's you know you could justify it

in both ways. You could justify it that

oh people have thousands of meetings in

Granola. They're on the go. they need to

access that information. They need to

include, you know, um you as a companion

app for what what's currently happening.

But then there's this view of oh

actually in the future if granola like

you're going to want all your

conversation context in one place so

that you can query it so that you can

run agents on top of it and from that

perspective if all your in-person

meetings are just invisible that's a

huge problem and we need to start

investing in that now.

How do you think about the evolution of

product inside the company? So I heard

earlier you're hiring obviously you're a

product leader so you've been running

product so far but I heard earlier

you're hiring one of your first PM roles

right now.

>> How do you think about what time it is

in terms of

>> you're hiring PM?

>> This is a critically important point

made to me earlier. Granola is hiring.

>> Yeah.

>> How do you think about like what time it

is and like what you'd be looking for in

this first person given that we have a

room full of many PMs.

>> Yeah. So I I guess first off we've we've

tried to do a bit of a linear model

where um every engineer we hire, we

expect them to have um strong product

sensibilities. Uh we actually interview

them for that. So it's a little bit like

making granola a great product is

everyone's job. It's not one

department's job. Um and then um my my

my co-founder is a designer. So it's

like product design is kind of in the

the central DNA of the company. Um I I

think the

the reality is our ambitions are

extremely large and we just need to be

working on a lot of really big things in

parallel and I believe that it's

impossible to build good product if you

can only give it fractional attention

and um each of these big things we're

working on require like full attention

from a person to really really push it

forward and and so it's like up until

now I felt like we could get away

without it. And now we we do need those

product folks. And um we need we need a

product person f well you can call the

the the titles we're hiring for right

now are AI um growth and consumer but

really across the board there's just so

much we need to do. So

>> how does it differ building a company

now versus your first company? Like what

are there what's similar? What's

different?

>> Yeah. So there are a lot of differences

that are specific to my two companies. I

mean my first one was an education. No

one wanted to invest in education

products. It was we had to really really

fight to get attention. It was like a

slow burn. It's we it took us five years

to really break out until we got lots

and lots of users. Um whereas granola is

a very different story. I I do think I

think building products in the age of AI

is different. like if you are building

AI native products or products that are

primarily based on L models there's just

like a different set of characteristics

there

>> how do you think about um when we're

thinking about other AI native products

you love what would you pull out how

optimistic are you AI native

>> so I I uh I always struggle with this

question which then sends me into some

like like deep existential crisis of why

I don't have a good answer to it And

here's what I think. I think we are I

think we're in the stone age of AI

products. And I think in the stone age

of AI products, it's all about technical

capability, right? So,

and I think that will change over time,

but right now it's like, oh, if this

model now has can now it can now um take

in context from the web, right? And it

can include search, like that's a huge

capability. I'm going to go and I'm

going to use that. or this model does

the same quality response but does it

three times faster like I'm just going

to do that and sure that's speed and

quality are huge parts of product but it

really is driven by the technical

innovation at the base layer so

>> all my like favorite products are really

like when deep research came out that

was a big deal but to me that's like a

technical innovation at the at the

technical layer that's not so much at

the product layer and we could debate

this and I'm sure we're having drinks

afterwards so please come up and and

debate this with

Um, I feel like we have seen remarkably

little uh UI innovation and user

experience innovation in AI and it it

befuddles me. Like I I think the only

explanation I have is that things are

moving so quickly technically that

everyone is just running to keep up with

that and they don't have the

time head space to and and to be fair

like product like UI innovation is a

time inensive and um low chance of

success endeavor, right? So, it makes

sense that you're not going to lock

yourself in a room and spend six months

trying to come up with a new UI on top

of text if you there's all this

lowhanging fruit um of other things you

can be doing. That said though, I feel

like you see this glut of products that

all look very similar

>> um and often times don't work that well.

Uh, and I think there's this like the

shiny ball of because everything is now

technically possible or feels possible,

it's very easy to put out a first

version, but it's very hard to put out

like a great version of a product. So, I

think we're kind of like in this uncanny

valley middle ground. Um,

>> so another way to say is I think the

great AI products um are yet to come.

>> Um, outside of coding, I I think coding

is the tip of the spear. I feel like the

coding products are probably like 12

months ahead of any other function that

I've seen.

>> Yeah, it kind of goes back to the

timeline piece we discussed earlier in

terms of building an AI requires trying

to hold in your head the tension between

>> building for two years ahead

>> but not AGI because that's too

intimidating. And I and I think what we

discussed a little bit earlier was

around, you know, we're seeing these

companies rise and fall really fast,

building AR really fast, but the

products themselves feel very

lightweight. If you're someone that

really likes great product, like how do

you think I feel like product in its

early embriionic years kind of often

owned the sort of user interface, often

owned the kind of design portion of of

the product and I think now how do you

think the field of product has evolved?

>> Has it evolved?

>> Great. I mean, yeah. Uh, great question.

I think

>> here are certain realities of building

an AI startup today.

We maybe talk about the product field

and definitions but one is

>> I think in a world where you have chatbt

and claude out there you have like very

good general purpose tools that means

that you need to build a product that is

really focused and way better than those

and that's like a that's a hard thing to

do right um it also means that in a

world where there are lots of copycats

product that is 20% better like that's

that's meaningful and people will notice

that and people will change their

behavior because of that.

>> Um I also think so I we can go through

my history but I was a PM at at Google

for a while and um I think my old

product hat kind of training makes me

think of like historically I'd be like

like you said the UI is the product the

content is not. Right. It's like if I'm

YouTube I'm like the UI of YouTube is

the product. Sure there's a ranking

algorithm but like I don't control the

videos. That's that's something else. In

a world where a model is generating the

content, um well, you control the

content, right? And a lot of AI

products, if you look at the percentage

of the pixels on the screen, a lot of it

is model output, right? And um it

doesn't really matter where it's coming

from. If you really care about building

a great product, like you need to make

sure that is great. So all the tools

that go into making great um AI

generated output now is part of the I

think the product job description,

right? And that goes from understanding

models, understanding prompting. That

also goes to understanding what context

to give those models, right? And

I think this should be the golden age

for product managers. Like like if you

look at like the jobs to be done today,

there's a lot if you you know like in

the ethos right now, there's like a lot

of the idea of taste. There's the idea

of like um to get good content, to get

good outputs out of a model, you need to

give the model the right context, right?

And then there's a question, well,

what's the right context? And at the end

of the day, those questions all end up

being the same kind of product questions

of like if I can put myself in a user's

shoes and I can think about like what

what would make me achieve my goal, feel

happy, feel good. A lot of the same

process that you would apply to like a

UI or what buttons you'd have, you can

apply to context selection, you can

apply to model selection. It's all the

same tool set. Um, I don't know if I'm

like I actually don't know. It would be

interesting to see are most founders out

there do a lot of them have a strong

product background versus an engineering

background and how is that changing in

the age of AI like it'd be an

interesting question.

>> Yeah. I think what's really interesting

is the blend of backgrounds. So like

Malvika from Schmood one of my founders

who's over here from New York as a

designer who can code and I think um

what I observe is that there are so many

more allrounders and I just personally

like I'm curious when you think about

the granola team you talked earlier

about when you hire engineers you want

them to have product sensibility.

What is the archetype of a granola hire?

Like what does that kind of person like?

What do they care about, think about

across the board in terms of function?

Sorry, ma could call you out. I just

thought you're a great example of an

all-rounder.

>> Yeah, I mean it it's more

what makes a granola person I think is

more cultural, right? So there are three

things basically. One is um

they people who join Granola should be

friendly is maybe the wrong way to say

it but basically like push come to shove

you should be willing to put the team

ahead of yourself and I think uh this

collaborative

um

e like easy to work with excited to

invest in your relationship to other

people like that's super important. Um

the second was people should be really

self-motivated and ambitious. Like we

think it's like this incredibly exciting

time to be building. Um and we want

people who want to come and do the best

work of their life at Granola and who um

think they can level up and who are

self-motivated. Um and then the last one

is that they really they they they

really care about product and they

really are excited about what can be

what can be built in this moment in time

in product.

>> Um that's cross function, right? Like I

think we want that to be true uh of any

anybody who joins. Um and then ideally

functionally you're just

>> uh excellent or um have the ambition to

be excellent and are willing to learn

really quickly.

>> I love that. I'm going to come to the

audience really soon for Q&A. Please be

thinking of questions. Um, for all of

the early career product folks in the

room, what advice would you have them in

in the quest to become excellent, which

all of us should want to?

>> Yeah.

>> What what advice for early

>> Yeah. How would you think about if if

you were them right now, particularly in

2025?

>> I I'm I every day I need to overcome

intuitions that were formed preai,

right? And then and I'm I'm very

intentional about it and I still I still

fall see myself falling into like these

old intuitions, these old old ways of

doing things. So I if I were early in my

career in product, I would what would I

be doing? I would be immersing myself

into what would like what does building

an AI mean? What does working with AI

mean? And I'd be leaning into that

really heavily. And then I would um it

almost doesn't matter what you choose

but I would I would choose something uh

some problem for someone and I would

just see how well can I solve that

myself right like no no nobody else like

no engineers if you're an engineer great

but basically the idea of like

oftentimes what not not and not

necessarily an idea that's like oh this

is going to be a great startup idea like

if you want to build great like like

product muscle the we didn't talk about

this but A a huge thing about building

product at Granola is um how we how

often we talk to users and how that

informs our our product building

process. So maybe we should talk about

that.

>> I was gonna ask you about that next.

>> Yeah. Yeah. Yeah. Let's

>> um but yeah, so to answer the previous

question though, uh just choose

something and then just see how far you

can take it, right? And and like the

fundamentals of building product are

often times you look at something, you

try to figure out from your intuitions

what's good or bad about that and then

you put that in front of other people

and you figure out what's good and bad

bad about that from what other people

say and then you just try to make it

better and then you just do that like 50

times, right? And as you do that 50

times, your intuitions should be getting

better and better. Um, and you should be

able to do that instead of 50

iterations, you should be able to do

that in 10 iterations. It'll never be

zero iterations. It'll be never be one,

but it might be like five or 10 instead

of 50. Um, and

what the beautiful thing about this

moment is like nobody really knows how

to do this. So, there aren't people to

go out there to learn from. Like you you

can do this from you'll have to like

teach yourself from first principles and

and if you do that, you it'll be a lot

easier for you to go further than me

because I have all this other stuff I

need to forget, which was true, I think,

in the PreAI world. I'm I'm 100% serious

here. Me, too. Um, and I think that like

you see people who are very young, um,

not everybody, but very young in their

careers who kind of like grew up with

these tools

>> and then use them to an extent that is

like like very surprising for uh uh for

folks like me and I think that that will

just compound right um and that those

intuitions those learnings over time

will compound and there'll be a delta

and then there'll be very few people out

there who are like have built AI native

products s that feel great and it could

be a prototype. Like it doesn't matter

if it's like commercially viable, if

it's something people are using. It's

like you should be able to send someone

a link and say try this, right? And it

could just be a really complicated

prompt. Like that's fine. Like it

doesn't matter what's under the hood.

It's what's the output, what's the

experience that matters. And I would

just I would just try to build something

I'm really freaking proud of. Um and

once once I'm really freaking proud of

it, I think that's like then email it to

me and come work here, you know? like

that's like or email it to anybody and

go work there because it's like a very

very rare skill uh today.

>> Yeah, I love that. And I feel like the

bar to build stuff now is like to your

point also had a whole career pre AAI

and so I think losing all of the ideas

around building MVPs takes time. I think

it's always surprising to me now when I

meet somebody looking for a job and

they're not already building stuff and

using the stuff they're building as a

way to get a job because to me that

feels so obvious.

>> Um

>> going back to your point around user

feedback,

>> you know, I'm an exjournalist. you've

got a masters in journalism. The old

rubric of product was always the tension

between listening to users deeply and

synthesizing that information and kind

of having intuition and you know kind of

like developing your own gut, your own

taste versus being purely driven by

>> what you're hearing. Like how do you

think about that here? What have you

>> Yeah, I like our our view on it is is is

very simple. Um so I think there's

so one end of the spectrum is basically

you just you talk to users and you build

exactly what users ask you to do. The

other end of the spectrum is you say

like I'm a designer like I I'm you know

people don't know what they want.

They'll ask for faster horses. I'll just

figure out the perfect thing. Um and I I

think it's completely it would be

completely impossible for me to build a

good experience on either end of those

spectrum. So the question then becomes

how do you blend it? And um it can be

very very hard to blend because if you

talk to a lot of users, they'll ask for

a lots of different things and just just

categorizing and organizing that

information uh becomes very hard because

what you end up doing is you end up

listing here are the feature requests,

right? And feature requests are usually

the wrong way to talk about building in

product. The right way to do that is I

think it's like it's it's pain points

like what are like needs, jobs to be

done. Um, so basically what we do is we

talk to a ton of users

and then we let that context inform our

decisions. But like so like we we have a

an opinion of what Granola should be and

uh the future we're building for and how

it should work, right? And I think we

have it's very important for product

builders to have an opinion. Otherwise,

you don't you're probably not going to

build something great. like it's got to

be something you believe should be in

the world and you should believe you you

want it in the world because it should

work a certain way or it should have

certain characteristics. Um it's just

very easy to go off the deep end if

you're just like talking to yourself or

thinking about what's really good for

you. So you just need to immerse

yourself in user feedback. So like um an

example like a something that really

helped at my last company Socratic was

like we were kind of wandering in the

product wilderness for a couple years

until we set up this thing where we were

building for high school students and

this is like presumed so it was like

really hard to talk to them where on

Tuesdays and Thursdays we had like 10

high school students in our office every

Tuesday and Thursday and sometimes we

had nothing to show them and it didn't

matter. we would pay them and they would

go off and they'd be super happy and

other times we would show them tons of

stuff, right? And the moment we changed

the process so we had easy access to

students, we could ask for their opinion

on all kinds of things, whether it's the

copy on this thing or this feature or

like how do you feel when your parents

see you using Socratic or whatever it

may be. It it it made our intuitions 10

times better. So like the way I think

about it is like uh user feedback should

sharpen your intuitions and you should

build based off your intuitions.

I love that.

>> Okay, final question, then we're gonna

hand over to you.

>> Um, a product that you love. You said to

me that products should have soul.

>> It doesn't have to be AI native. We've

already said that they're they're not

great yet. It can be a physical product.

It can be a digital product. What's a

product that you really love?

>> Great question. Um,

until you said the word soul, I I I knew

what I was going to say and then

>> without soul and with soul. Well,

there's a when when uh when I first

interviewed at Google, I was asked like,

"What's your favorite product?" And um

>> I was too

>> I was uh I entered a like I had a

Moleskin notebook which I had recently

discovered at like the age of whatever

21 or something and I just gushed about

this thing. But there was something

really um incredible about simple

physical products that um

get out of the way but give you an

emotional reaction. That's what you

want. Um, I weirdly came across this

Nespresso machine a couple weeks ago and

fell in love with it, which I never

thought I would say because I like

historically have always hated them. But

this this Nespress there was like there

was not one unnecessary thing on this

machine and that I thought was

incredible. It had two buttons. There

wasn't an on button. There's literally

like a button for like one shot or a two

shot and that was it. And like I

literally took it apart and I was like

could I what could I get rid of? and I

couldn't figure out anything to it's a

bit like um you only understand the

genius of Hemingway when you try to take

away like a word from his a paragraph of

his books and you it's very very hard to

do that. So I don't know I I found I was

like wow whoever whoever is responsible

for this thing like yeah

>> good good on you. Um yeah I don't know

if those are the most inspiring.

>> No I love that. That was amazing.

>> Okay please don't be shy. We've got

about 25 minutes

>> questions. Okay over there. That was the

first hand.

>> And please say your name at the start

and what you working on if you want.

>> Okay.

>> Or if you're looking to be hired, that's

also fine.

>> My name is So I I am just software

engineer right now.

You talk about a lot of things like

being a dirty word and you think about

things very user focused

um and for example one building like

whatever someone tells you I know

exactly what it is.

>> Yeah.

>> I feel like a very mainstream opinion is

that the model provides most of the

value for companies like

like what is important for an AI product

and it feels very confusing because is

it the model is it the distribution

is it you really understand the workflow

very well so I'm curious if we can use

granola as a specific

>> yeah I'll uh I'll tell you how I think

about this can can the microphone can

everyone hear that and is do I need to

say the question back or

like what's the moat for AI products or

specifically like model wrapper products

and what is it for granola specifically

right um

I think about this a little bit like uh

when products move to the cloud right so

think of like uh Google Docs moving to

the cloud and a lot of stuff had to be

re rewritten for that so that wasn't

trivial but um

the cloud was kind of accessible to

everybody, right? Uh no one was like,

"Oh, my moat is the cloud, right?" Uh

people would say like, "Well, maybe the

moat for Google Docs is collaboration

because and the cloud enables that." But

it's not like the cloud itself was going

to bring the moat because everyone can

move to the cloud. Um and and that's how

I think about things in AI. Basically, I

think the moes are the same that existed

before AI. I think for whatever reason

we thought there w it would there's this

expectation that it would be different

post AI but I think what's happened this

wasn't clear like three years ago is

that uh so far and I believe this will

continue because of the amazing

competition between the leading model

providers and the open source community

um this like incredible intelligence is

available to everybody equally right and

that's and amazing for product builders

um but it what it what it means is like

access to the model I don't think is is

a moat. I don't think fine-tuning models

I don't think data is a moat. Like I I

like if anything I think the importance

like you need less and less data because

the base models are getting smarter and

smarter to like get fine-tuned results.

So I think the answers are the same as

before. So switching costs, right? If

you have um if there's like a a social

graph where the value is is a

consequence of how many people are on

there. Um and I think I think for

granola it's two things. I think the

more meetings you do in granola, the

more context we have about you, the more

useful we should be able to be. Um, and

the second one is like if context is

important to be useful, then like

switching that to another uh provider is

just a pain, right? And that's like the

same type of thing that would keep like

um like this this idea of switching

costs or integration, switching cost is

the same thing that like makes Zapier a

pain in the butt to change even though

it's totally doable, it's just like kind

of hard to do. So that's what I believe.

I think there's this other thing which

is uh AI is AI space is moving so

quickly that if you're not innovating

and staying ahead then it's very easy to

fall behind and if you fall behind then

like then everyone's going to leave. So

it's like assuming you can stay at the

forefront then I think it's all the

traditional like product like modes.

>> And there was one Yeah.

Imagine

a scenario

>> where okay so the question is like if

only work on LLM features or non LLM

features when will that

Um

so I guess the

uh I don't like the way the question is

formed is is the is the answer uh seems

like a false like here's my thinking

like we should work on what's going to

deliver the most value for the user

right or if whatever is going to deliver

the most value for the business like if

we can build a thing that will help us

grow five times faster we should we

should also go that um I think

likely today a lot of that is going to

come from like new capabilities unlocked

by AI. Uh that said, a lot of what made

Granola good had nothing to do with AI,

right? It was a lot of like polishing

paper cuts. Uh I think now that we kind

of have I think we do a decent job with

meeting notes and a lot of those paper

cuts have been fixed. It's a lot about

like oh now we have all this context

about you. How can we make your life 10

times easier? Right? And like that's the

next forefront for us. And that's all

LL1 based. Like that's all AI based.

That's that's even like that's stuff

that hasn't landed yet because there's

you know we have so much data. There's

like no way to process that amount of

data and the speed and quality that we

need today but will be possible within

the next six months.

>> Lucy was next I think.

>> Please.

Um

so

uh oh the question is like do I have any

um non-work habits that will help with

clarity of thought like uh meditation or

whatnot? Um, so I think at some point I

said to be able to build great product I

think you need to be able to give it

like undivided attention and lots of it.

And I think uh what's been very

challenging for me to be quite a uh

honest is that Granola has scaled

significantly over the last year. Like

we launched a bit over a year ago and um

with that there's a tremendous amount of

complexity. Like we were four people

when we launched Granola. We're like 28

people now. Um, and as there are more

and more and more things happening

and that fall on my plate or take up my

head space, it becomes a lot harder to

do the product focus. Um, I don't come

into the office for the first few hours

every morning. I go somewhere literally

people can't find me. I um I try not and

then depending on and that's like

another Hemingway reference, but it's

kind of like do the hardest thing you

need to do first thing in the morning.

Um, and that's where I try to do my deep

product thinking. Uh, I try

I try to open as little technology as

possible and sometimes that's you know

not possible sometimes you know you need

to go into the analytics or whatnot but

I find that the moment you kind of open

that floodgate there are more and more

things crying for attention it becomes

much harder to like my dream would

actually be if I previously printed out

the 10 things I need to look at the day

before and then I just looked at that

analog and did my writing analog or did

my thinking analog and then and then

moved into digital world um

but yeah hyperscaling startups It's like

a It's a new class of hard things. Um,

and I think I'm invest I'm like I'm

investing in trying to learn how to

manage

myself and how to like build the team

around me to to scale. Anyway, this is

like what every every founder ever has

said when they their company started

growing and things started breaking.

They're like, "Oh, I don't know how to

do this." Um, but it's like an active

it's a thing I'm like actively investing

in because otherwise

I do that at the back.

>> So, hi, my name is

So, I have two questions, but I'll ask

one. So, you mentioned a lot about

content.

>> Yeah.

>> How have you made that?

You wouldn't consider it to be cont.

>> Yeah.

>> What?

>> Happy.

>> Yeah. Um,

so yeah, I I think I think in the in the

age of LLMs, like context I don't know

if context design is a thing yet, but it

will be if we don't use that term like

like we use it internally. Um, context

design is a really big deal. Um, I think

there's there's two kind of questions.

One is like what are the sources of

context? And we said like right now

granola is primarily notes, not

exclusively. We also pull in information

about people and companies. So, like one

of the I think one of the things that

like blew people's minds when Granola

came out was that if a founder and an

investor were both using Granola in the

same meeting, their notes looked

completely different, right? And they

look completely different because I

mean, we know like you probably don't

care so much about what you said in the

meeting. You probably care more about

what the other person said. Um, but we

also know that like, hey, if you're a

founder, if you're an investor, it's

pretty clear if it's like a pitch, you

care about like what's the idea, like

what's the traction, what's the market

size, like how much have you raised?

like it's it's like a clear set of

things. Whereas if you're a founder,

it's like what did they seem not

convinced by, right? Or like what kind

of check sizes do they write? Or like

where in the process are we like the

questions are very different and we had

enough context to and to design our

prompts and our models so that those

notes look completely different. Um

we're missing tons of sources of

context. We're going to add email. We're

going to add a ton of stuff from the

web. We're going to add like let you add

documents. Um, so we're g we're we've

started that. Then the next question

becomes like out of all this context you

have access to, what's the right context

for this specific job, right? And how do

you uh expose that to the user and how

do you let the user manipulate that?

Right? And those are completely unsolved

questions today. I I I think the the UIs

I've seen that allow you to do this are

incredibly rudimentary, right? We have

no great way to visualize context. um

seeing thumbnails for like 30 different

like files is of text files is like not

helpful. Um and then the the UI

interaction for controlling or

manipulating that is like non-existent.

So like anyone here has like great ideas

or wants to go and invent the future of

that like come talk to us because that's

like I think very relevant to granola

but also very relevant to AI work in the

future in general because like a lot of

the human's job like our job is going to

be what context do I need to bring

together so that I can then collaborate

with an AI. It's a little bit like um uh

I don't know like how do I set up my IDE

so that it like actually it's like set

up in a way where I can do my job

easily. It's like the same type of like

thinking.

>> I think there was one here. Next.

>> My question is about product

and um it's more about like like

a lot of people are back.

>> Yeah.

>> And then how do you like like I'm a

marketer. I really how do you like how

do you like

get into that like how do you figure out

the persona like type of product like

a lot of people uses it. How do you

figure out which one to focus like

focus?

>> Yeah.

>> And when do you focus it?

>> Yeah. Yeah. Yeah. Um

Yeah. So, so, so our our history here

was we always wanted Granola to be a a

product that lots of different user

types could use, right? So, we wanted it

to be a horizontal product. Horizontal

products are hard. Uh, when Sam and I

started off, we were building it. And

then we realized we needed to just

choose one ICP because otherwise we'd be

interviewing 20 people and they would

all tell us 20 slightly different things

and it was hard to narrow it in on

narrow in. And um, our initial one we

chose were were investors, right? And uh

which everyone says don't do. They're

like don't like no people say never

choose uh VCs as your target user

because uh there aren't that many of

them. They're weird like as they don't

act the way other people act. Uh for

example, they're willing to pay for

things. They're price incentive in a way

that very few other uh rules are price

insensitive. Um but that's the best

thing that we ever did honestly. Like

one is like we they're very willing to

try the product and give us feedback.

Um, and two,

they have a lot of money and like when

you need to raise money, like it's great

if they're using your product and they

like they like it. So, um, I think and

then the moment we launched publicly, we

switched. We said, "Okay, we no longer

care about investors. We now care about

founders is what we said, right?" Um,

and we chose founders because

they felt like um they felt like a like

a company and a person. It's like a

founder might do a sales call, might do

um a legal call, might do a you know

marketing meeting. Um and it's really

hard but we thought okay if we just try

to build a great product for founders

then we'll build a decent product for

everyone else and then we can expand it

out.

What?

>> Hi, I'm

so

absolutely love it. And I'll be honest,

we used

I met him

when I spoke to about switching to the

paid plan, they essentially said there's

not much incentive. You know, we get

enough team

etc.

I just want to know a bit about the rale

behind.

>> Yeah. Um

uh

so the short of it is um we have been

really focused on building a good

product, right? And we see granola as

it's like

15 maybe 20% of what it's going to be,

right? And

we think so much of the value that we're

going to provide to people and companies

lies in the future. So we haven't been

very precious about putting lots of

stuff behind the pay wall. We basically

give away most functionality for free

right now. Um and we haven't spent a

whole lot of time thinking about pay

wall or

I can tell you though that in companies

where

it or somebody like switched off granola

um and there's like 30 40 people in the

company using granola they usually they

usually uh kick and scream pretty loudly

um and that forces some conversation. So

like I'm not um I we'll see what company

you work for. Let's see if we switch it

off. Like like

what happens? Like I think I I might get

an email from the founder. Like I'm not

We'll see.

>> I mean this is user feedback right here.

What I'm hearing is they should be

charging you for it. Um okay. I think we

haven't taken enough from the back.

Anyone there?

>> Uh gentleman here.

>> Hi my name is

hypothetical question.

You mentioned how long.

So question number one, how long do you

build granola for? And on the back of

that, do you ever envision another

surface of interaction with so obviously

the first open having

some kind of device? Hypothetically

speaking, would you ever imagine

granola?

>> Granola bot.

>> Granola bar.

>> Granola bar.

>> Bar.

Okay.

>> Edible.

>> Got it. Got it. Okay. Uh so so first

question. I think the um the the time

horizon we choose to build for is like

like 12 18 months from now, right? Um

and I think like you can't there are a

lot of start like if you go to SF a lot

of startups are kind of have a longer

time horizon that they're building for,

right? And and to me the metaphor there

is kind be the future becomes so

unpredictable and it's so hard to get

feedback like if you're if you're

assuming that agents can go off and do

like tons of stuff without any human

input. It's like what does that world

look like? Right? And I'm sure there's

lots of opportunity if you build for

that world. It's just making decisions

there is incredibly hard. So I think

there's a lot of like like an American

football reference is kind of like okay

you know the quarterback's going to

throw the ball somewhere but you're not

allowed to look. you just have to run

and like hope you happen to be in the

right place at the right time when the

technology hits. Like that's kind of my

mental model there. So I think that's

really hard. So whereas if you build for

like ah one to two years from now, sure

there might be like a fast takeoff, but

you can kind of like squint and be like,

okay, what are the labs like releasing

or like where what are the trends that

we're seeing on whatever speed context

window like thinking whatnot. Um in

terms of like the the device um

everything's possible, right? Like my

view here is like I think the social

norms in work and the social norms uh in

like our our social sphere are very

different and I think people a lot of

the visions I hear is like there's a de

a device I'm going to be wearing all the

time. It's going to listen to everything

and it's possible but I feel like the

norm I think it's

>> people are going to be comfortable with

conversations being recorded or

transcribed in the work context I think

much sooner than they are if you go to a

party and you're like I don't know who

you are but I'm recording you. I'll have

it forever. Like I think that may never

change or will take a long time to

change. I think in the work context the

question will stop being uh should this

meeting be transcribed or not? It'll be

who has access to the transcript, right?

So it's like is it just the two of us?

Is it my team? Is it the whole company?

Is it my manager? And I think that's

where the conversation will will shift

to. Um and we care about the work

context. And in the work context there

are devices everywhere. So like sure

maybe they'll there'll be a granola

specific device but like there's no like

literally everyone has a phone, they

have a laptop, there might be like a

like a Zoom machine in the room. So it's

like there no shortage of like context

capture devices. Um so I think like for

a small startup like us it'd be really

distracting uh to to do that. There's so

much low hanging fruit on top of the

infra infrastructure that exists.

>> So we only have time for two more

questions I'm afraid.

>> Hi

Right.

>> Um, so you talked a bit about like this

idea that like you have these kids that

are like picking up tools and sort of

using the capability of what anyone ever

imagined.

>> What's been a surprising way that people

use

>> granola?

>> Yeah.

>> Um, yeah. So that's a that's a

something that people did in the early

days before we had we any shared

functionality. There these two founders

who basically they wanted everything in

granola. They wanted every every

conversation they ever had between each

other, every brainstorm and they I think

they actually shared one account. So

this is before we had sharing. So they

both logged into the same account so

that it could all have the shared

context so that they could then speak to

like all of those ideas in in one place.

Um which is kind of I mean that was I

was like wow okay that's pretty eye

opening. There's like a lot to be said

about having centralized context. Um and

then I mean I just said we care about

work stuff but like the number of people

are like oh I use this for my therapy

sessions. I use this for my medical uh

whenever I go see a doctor. Whenever I

go talk to my uh kids teacher and they

tell me how they're doing in school,

like there's like all these kind of

conversations out there that are not

workrelated

that granola works really well for

because there's a bunch of like specific

information you want to look back on.

>> Um yeah, those are those are two that

come to mind.

>> Okay, so this is our last question. Um

but I think we're able to stay here for

drinks afterwards and I think Chris will

have some time for some more questions

we don't get to on stage. Is that okay

for me to say that?

>> Yeah. Yeah, 100%.

>> Just volunteering him. Okay, final

question. I think I saw your hand first.

>> Yes.

>> Sorry.

>> Do you say voice agents for GP triage?

>> Wow. Okay.

>> Yeah.

questionramework

surprised you might fall back.

Sure. Sure. Sure. Um talked about

product frameworks. I've had to learn.

It's not necessarily product frameworks.

There are some of those, but it's it's

also um just how I work generally. Like

I don't I don't use AI enough. I should

turn to AI more than I do. Um I don't

invest enough time in creating

um the AI workspaces or like the context

buckets that I shoot.

Oh, you asked me a question about how

people use like AI. There's this one

user we we interviewed and he had taught

himself how himself how to do sales and

um I forget the name of the guy, but

there's like this famous sales person.

there's like this method, right? And

there are these like eight stages you

might be in with a customer. Um, and he

created uh these context buckets for

each of these stages where it's like

here's all the stuff from the book,

here's all the stuff from his company,

here are all the like questions. And

then he would drop Granola meetings into

these different buckets and then ask it

to like generate the next set of action

items based on this contextual bucket

that he had created.

I talk about context selection, right?

Like he took that to this level that I

like I would never intuitively like

think about doing.

That's why

my default is like, oh, how much faster

is this going to make me now? Right? And

it's like, h maybe it slows me down a

little bit, maybe a little bit faster.

What I don't think about is the

compounding effects like over time. And

and I think that's where

>> Thanks everyone for joining us.

Thank you.

>> Thank you so much.

>> That was fun. Right, let's pop these

back here. Really?