Cursor's Ryo Lu on Designing Living Tools, the Future of Coding , & Creating Soulful Things with AI

In my career as a professional product

designer, the thing I hate the most is

like like people want the design to be

final. Uh where's the final version of

this mock? If you don't have it, I won't

start building it. Like that doesn't

make sense [laughter]

cuz the first mock is never right.

>> Yeah,

>> because AI is really good at composing

parts. I'm actually thinking we need to

like build bricks. Really good bricks.

Put your soul in it. You need to care

about every detail. You you need to not

accept whatever purple gradient the AI

gave you as the end.

>> Like that is just the beginning.

>> Ah yes.

>> You always start with [ __ ] You always

start with slob with AI and then you

refine it. You make

>> the beginning not the end.

>> Yeah. You you just poke at it with

little prompts and then it'll get

better. Before we get into the episode,

I have an announcement. In case you

missed it, I'm going fulltime on

Dialectic thanks to the support of my

new presenting partner, Notion. I

[snorts] guess first and foremost, I'm

just excited and grateful. I'm about a

year into this. I crossed my year

anniversary of starting at the end of

November. Um, and it feels fitting to be

able to fully lean in and consolidate

and focus on something that has just

felt like being in my lane. Um, getting

to amplify people I'm excited about. And

I've been reflecting on this and I think

this ties to notion, too. Like I've been

reflecting like what is what is the

show? What makes it good? What am I

trying to do here? And there's been a

handful of patterns that have become

more obvious over time, things that have

become more legible. I think it's

definitely a show about ideas,

particularly I think I love to talk to

people who make stuff about the ideas

and philosophies that underpin them. But

I was also reflecting on like what what

are the patterns that stand out most and

I think they tie into why notion is such

an ideal partner for me. The first is I

think it's a show about where ideas meet

action. I love introspection and

reflection and thoughtfulness and

philosophy. But I think I also love

people who are able to take those things

and use it to make contact with reality.

This combination of introspection and

agency and action. Ideas are powerful,

but we got to put them to work. The

second pattern is craft. Craft is

aspirational. Craft is when we deploy

our taste. Craft is a human touch. craft

is saying, "I'm just going to push

things a little bit more to make them a

little bit better." And whether my

guests are people who design things or

write or invest or whatever else they

might create, I think there is a deep

amount of craft inside of how they

approach what they make and inside the

things that they make. And the third

pattern is soul or soulfulness.

This word is obviously a little bit hard

to pin down and you might instead say

authenticity or originality or even

aliveness.

But soul is about when somebody line is

lined up I think like in who they are

with the way they're showing up in the

world and maybe even more than that a

willingness to reach deep. And so I

think when I think about what I'm drawn

to and all of the people I admire and

certainly the people I talk to for this

show, it is soul at its core. One of the

things I'm most proud of for this show

is the audience. It feels like it's my

kind of people. Some of my guests are

listeners. Some of the people I've met

through the show have been incredible.

And Aka Kthari, co-founder of Notion, is

a listener. And so, we've gotten to know

each other the last few months. And when

I started to think about what it would

look like to go full-time on Dialectic

and bring on a partner, it was

ultimately a pretty easy choice. I think

it was clear to me that he really got

the maybe even intangible elements that

made the show special to me and to the

people who were listening. But also I

think those those patterns I mentioned

earlier um really do embody notion too

and that's why it made it such a right

fit. Notion makes beautiful tools for

your life's work. I think I'm someone

who's certainly interested in tools.

I've talked to a bunch of tool makers on

this show including Notion's own Jeffrey

Lit. He wasn't at Notion when we spoke

and he is now. But also on those themes

from earlier I mean Notion is a tool for

taking your ideas and turning them into

action. Whether that be tinkering with

them or expanding them or sharing them,

it starts with ideas. With notion, it's

a brand and a tool that despite a long

road and tremendous scale and a great

deal of complexity has embodied craft, I

think, at every step of the way, both as

a brand and as a product. And then

finally, soul. Again, soul might be in

the eye of the beholder, but I think

notion is a tool that cares deeply about

letting its users pour themselves into

the product they use. And I think

Notion's community and templates and

remixing and creative expression are all

evidence of just that, a product that is

full of aliveness. So it ultimately

wasn't a very hard decision to partner

with Ocean and I feel so grateful to

them for helping me embark on this

journey. As for what's to come, I mean I

I think a lot more of the same.

Hopefully people who are inspiring to

you, people you're really excited about

and people who surprise you. I I would

like to keep you guessing. I think too,

a lot more video for those of you who

are listening um or haven't tried. Video

is coming. And more than anything, I I

hope to amplify people who can or have

the ability to shine. Last but not

least, while I'm so grateful to Notion,

I'm even more grateful to those of you

who have listened, watched, read,

whatever, found a way to support me. I

feel so lucky. I hope I am doing you a

service when you spend your time here

listening to these conversations. I hope

you go take your ideas and turn them

into things. I hope you do it with

craft. I hope you do it with soul. With

that, I will I will turn it over to the

episode, but thank you so much and and

I'm so excited to continue to share

dialectic with you. Welcome to Dialectic

with Rio Lou. Rio is the head of design

at Cursor. Prior he was a designer at

Notion working across so many different

projects and features including notion

AI for about 5 years and he was a

designer at Stripe and Auna. He grew up

between China and Montreal and now lives

in San Francisco where he's focused on

building cursor and helping anyone

create software. We talked extensively

about his design philosophy and how he

is constantly moving between simplicity

and complexity, bare material and

abstraction and why in his words so many

of these ideas and these patterns are

all the same thing. We also talk about

how design is changing where in the past

using tools like Figma it felt more like

painting or drawing, now much of Rio's

design feels more like sculpting clay or

finding David in the marble. um so much

of his philosophy is about getting

closer to the material and in the case

of digital things of software that is

working with code and that's why I think

why he's so excited about cursor

the line between vibe coding and real

engineering is also I think everyone's

feeling that it's flattening and there's

no better example of that than Rio's

personal project Rio OS which you can

find on his website which is essentially

a nearly a full-on operating system of

apps and games and simulations you can

talk to Rio's agent you and I've watched

him literally make games and new apps

for real OS in real OS and in some sense

it's entirely vibecoded. Um he's built

it using cursor. Uh and what's I think

so outstanding about it is that it's

quite literally the opposite of AI slop.

It is so deeply personalized. It has so

much soul. It feels so much like Rio. So

we talk about how he is iteratively

designing both his personal projects as

well as all of the design decisions he's

making at Cursor and helping more and

more people across the team work with

him in a range of different ways. This

is definitely a philosophical

discussion. Much of it is about

designing things that feel true or even

inevitable. Um but in many ways I think

Rio is also an amazing example of

somebody who is doing a lot more doing

than thinking. And so I think that

marriage together u makes him so

effective and I hope and I think we we

really dove into that today.

One of my favorite things Rio wrote is

an essay on how to make great things and

we talk extensively about what goes into

that breath versus depth. Uh iteration,

prioritizing doing and learning over

thinking, balancing quality and speed

and more. If you already make things,

especially software, I hope you are

inspired to be all the more uh willing

to try things to be more flexible, be

mind be more dynamic and expand the

boundaries of what you can personally

do. And if you feel like you could be

making more things, I hope you are

inspired not only to try tools like

cursor and make software, but to apply

some of this philosophy to making any

range of things. Um, I just so love

the way Rio thinks about um getting up

close with material and how learning

with material, getting feedback from it

is how we design anything. Um, it's

addictive. It pulls us in. And in the

limit, uh, we end up making things that

other people get to enjoy. I hope you

enjoy the conversation as much as I did.

With that, here's Rio Lou.

>> Rio Lou.

>> Okay, let's go.

>> We're here. Thank you for being here.

This I'm really excited about this.

>> Yes.

>> Um we're going to start with a I guess

what you could call a catchphrase of

yours, which is

>> you love to say it's all the same thing.

>> Yes.

>> What does that mean and what does it

tell us about design? H

it's like when you look at all the apps

you use or even like everything around

you if even looking at ourselves as like

humans as like life forms we are always

built. It's almost like with the same

parts

that are really simple,

>> but when you merge them or combine them,

recombine them,

they give a rise to complexity.

Um, like the most fundamental elements

are the same.

Like a lot of the concepts that we use,

you know, regardless if you call it

like, uh, this is a task management

thing or like a document thing, they're

all just like information organized in

databases. [snorts] Yeah. So there's not

that much difference.

And then there's always

like something at the core that is like

the the simplest form of the thing

itself.

And it's most likely things that you've

seen before or there's like analoges in

nature or like patterns.

When you talk about those simple things,

are they abstract things?

>> Mhm. Like are they as you say are they

patterns or like metaphors or sort of

like ideas or are they can they be also

like very concrete?

>> Oh yeah I think they can be very

concrete and it's like the same thing

manifested at different levels

>> different levels of abstraction.

>> Okay. So you can think of maybe like ah

these are my core ideas but then how do

I say visually represent it in like this

constrained 2D space which is like a

screen.

>> Yeah.

>> Like a phone or like you stretch it to

like a window then you have more space.

Then what are the things that should be

shown like what are the relationships

between them? Um what are the more

important bits that you want people to

get in? Like it's almost like

it's like a multifloor apartment.

>> Yeah. Yeah. Yeah.

>> And then you you want people to ah go to

the lobby on the top floor with the best

view. They can kind of see everything.

Ah this is cool. Now let me go to the

place I want. Um that's more like for

the users. But the same thing applies to

say like you're designing UI, you're

designing some flows, you're designing

how the data model works, you're like

conceptualizing how do I, you know, make

this into like a big scalable

distributed distributed system. And when

you're operating on all these layers,

they're still like like just

manifestations of those core concepts or

ideas.

>> Then you keep everything

together and they feel cohesive.

when like a lot of people

maybe they think of these things as

separate things and then they treat them

as like ah I need to do this box first

and then do that box first and then each

people doing the boxes don't talk to

each other then they build something

that's kind of it's like

it wiggles

>> you know [snorts]

yeah it doesn't have the connectedness

you have you have a you have an essay

little essay you wrote

>> about complexity coming before

simplicity Uh one one part you say it's

like a swan serene on the surface but

paddling like hell beneath.

>> Yes.

>> Which is an amazing metaphor. Why does

complexity actually have to come before

simplicity?

>> I do think say conceptually it is

possible to say

uh these are the core building blocks of

my world and that's it. Let's just go.

>> Yeah. Um

but like it needs to survive in the real

world that we live in.

>> Like there's people who like they don't

come here to look at your essay or look

at your academic idea of like

[clears throat] ah these are the ways we

need to like connect these computer

ideas.

>> Yes.

>> They're here to do something.

>> Yes.

>> So they come here

>> they should ideally you know do the

thing they want to do first without

thinking without

>> without thinking too much.

without thinking too much. They can do

it. They can actually like, you know,

slowly master it, configure the thing,

customize it.

>> Then they kind of know what what is in

there.

>> You can do it from both ends and they

kind of are it's like two sides of the

same coin almost.

>> But a lot of people they only see one

side. say like we do a lot of like user

centered design or like you know let's

start with a user problem and then

decompose it or like do some research

look at some numbers uh figure out if

solution A B for this problem one which

one is the best ah a is the best oh

let's just do a and then you keep doing

this a a a b a b and then now you have a

platter of like random choices

and then they don't connect and then

they're all like discrete buttons on

your on your UI.

>> Yeah. Yeah. Yeah. Yeah.

>> And that's kind of crazy

>> when fundamentally maybe all these ideas

are the same ideas or maybe they they

are like better versions of your

original ideas

>> like a remix version of that or like a

reconfigurations of the thing.

>> Yeah. You're sort of seeing both the

swan both aspects of the swan at the

same time. you're seeing the elegance in

the kind of

>> it's like you need to test your model

with real world examples and people

>> and then [clears throat] as you do that

you figure out

this part of the system is a little weak

I need to make it better.

>> Yeah.

>> We're like ah we maybe we really need we

really need to add this new thing

>> then we should probably do it cuz a lot

of people need it.

>> Yeah. But but if you if you're just

conceptualizing yourself and you're kind

of in your own world thinking and you're

just like ideiating,

then you're not really doing anything

like you're not helping anyone. You're

just you're untethered.

>> Yeah. You're just like I don't know

having fun yourself, I guess.

>> Another line from you. You say the

universe is fundamentally modular.

Simple rules endlessly recombining

creating emergent complexity. Design is

the human practice of participating in

that process consciously. We look at the

world, identify the patterns, extract

the rules, and use them to build new

realities. Obviously, much of the sort

of it's all the same thing inside of

that. Uh I'm curious maybe at the most

zoomed out level like what what

initially drew you to what you describe

as design there and what kind of keeps

you coming back? like what is it about

this

um almost like philosophical approach to

the world that's so compelling to you?

>> I did not come here like you know when I

started I did not know the difference

between even like engineering or design

or product or anything.

I just saw these things that were made

by people.

Like I I started playing with like

software when I was a kid. I would get

these like pirated CDs. Um and then they

all they're almost like software

subscription packs monthly. Like they

get

>> you you just load them on your PC and

then you play with all the all the new

apps. And then I started playing with

like all the office tools like all the

fonts, Excel, PowerPoint,

Photoshop,

um video editing things, 3D making

things, um programming tools, um

starting making websites and stuff. And

as you do these things, as you make

things, you start realize

like

the end output of what we do is just

code,

but there's like a lot of different

depth in all the layers.

Um,

and if you're curious enough, you can go

to every layer really deeply. Um

but the more you do these things like uh

make more websites for different kinds

of people or make different apps for

things um you realize

like a lot of it is just the same ideas

and then you also can trace it back to

history

like when you look at people when they

started this or when they were just

again like ideiating they were things

were not real because things weren't

ready.

>> M

>> um but the ideas were there and all

you're doing is like remixing the idea,

repackaging it a little bit and then you

want to find out what is the core

essence things that you know you cannot

remove that will always be there.

>> Yeah.

>> And then you keep making those better.

>> You use the phrase things weren't ready.

>> Yes. Obviously

technology um design applies across

disciplines. Technology is an area where

design you actually are dealing with

that sort of the the rate of progress.

I'm curious especially maybe now since

what you have this great great uh future

site you made for cursor where you're

listing the kind of arc the lineage of

computing. We're in the middle of um an

immense amount of readiness, you could

say, but I'm curious what your

relationship has been like to things

being ready or maybe not ready even

let's say

>> the last two years with AI models and

cursor.

>> Yeah.

Yeah. There's like the technological

level of whether it's ready,

>> right?

>> But there's also the conceptual level of

whether it's ready. It's like

for example notion

even though technologically as like

everything is kind of fully ready like

notion itself is almost like just

databases in the cloud and then you can

do live editing with people. you're just

manipulating like blogs and databases

like that the ideas have existed for a

long time

>> right

>> but then people have not caught up or

people are not familiar with these ideas

>> then it's like still like kind of

foreign to people

>> and then boom AI happened then it's

almost like using this new primitive new

technology we can actually like help

people understand better or like make

translations of ideas

>> yeah it's bridging the conceptual gap

Right. Right. Right. Like you can use

that to like bridge the gap and

basically instead of people making

databases manually or like they have to

learn about you know coding is like

there's so many layers and then so many

dependencies in order for you to do like

a running program. You need to know so

many things.

um you can actually reduce that to like

nothing

but then it's like people can just start

from the other end they get some output

they play they they tweak and as they do

that they learn instead of like

>> backing into it

>> right instead of doing it in the reverse

>> um it's like

I believe in it's like

we are fundamentally the limiting factor

like as humans

>> like our brains can't process too much

information. We can't hold too many

concepts in our heads.

>> Yeah.

>> Um then

like which what we're doing is

you're like simplifying the amount of

information or ideas that you're giving

to people. It used to be like designers

have to do it, the thinkers have to do

it, the inventors have to do it. They're

thinking about what is the simplest

configuration of the thing, what are the

parts, but now it's almost like a lot of

it can be handled by the AI then you can

reach to like lower level primitives or

even connect more things

>> then

>> you can pull in more complexity

>> cuz this but then the the presentation

layer can still be simple and the

simplicity can be more subjective. It is

not designed by the designer. It is

actually like to you the person using

the thing or you're doing this thing the

ideal configuration for that thing. I AI

can kind of

>> do the translation.

>> Yeah. There's we're talking about

simplicity. There's another comment you

made that um is very similar to

something you wrote about making things

true.

>> Mhm.

>> Um and I think truth and simplicity next

to each other seem interesting. You say

>> design is the practice of seeing through

the surface of things to understand

their underlying structure and then

rearranging those elements into new

forms that didn't exist. Design is

philosophy because it forces you to ask

what is this thing really? What are its

central properties? You talked about

that. What can I remove before it stops

being itself? And once I understand that

what new things can I build this is the

work not making things pretty, making

things true.

>> Mhm.

I I think I have a sense and the

listener probably does too, but what is

maybe not what is the difference between

truth and simplicity, but what is it

maybe even what does it feel like when

you're designing and you're you're

approaching tress

>> or truth?

>> Oh, yeah.

Yeah. It's like you

Yeah. The thing is I think

I believe there is actually like a

ultimate solution

>> given say the amount of this space and

the constraints and the things you know.

>> Yeah.

>> But the problem is you never know

everything. [laughter]

Um and the things always change. So it's

like maybe it is the ultimate solution

for this point in time for this

condition but then

maybe tomorrow it's not true anymore. M

>> but I think you know

there are always like say when you're

doing a product or making software like

a set of things that don't really change

and it is so important to like figure

out what those things are. um those are

almost like your fundamental building

blocks or ideas of the the software the

it's like I see like software as it's

just like a tree of concepts and you

package it up give it a name

>> and then give it a UI put it out

>> are those concepts changing a lot or

they changing very little

>> like

most likely they don't change

>> okay

>> where it is really hard to change them

especially the ones that are core to the

thing. Um, for example, I worked at a

sauna. A sauna is basically projects and

tasks and everything revolves around it.

Every data model is like kind of locked

in there. And then for example, it will

be hard for Asana to expand into like

whatever. But then it is easy for notion

to do that because notion's building

blocks on the in the underlaying like

abstractions are more flexible.

>> Yeah. And then they actually don't

change that much. All you're doing is

like you're fixing some problems with

how they connect to each other or h now

there's like a different kind of data

that we can present better. What are the

better views for that? How do people

like you know combine these things so

that they can

>> do a lot more crazy things? Um how do we

help people like instead of them

building this thing maybe the AI agent

does this thing. Um and say for cursor

is like that common layer is even lower

which is code

>> and it's so generic.

>> Yeah.

>> It means you can actually do anything

>> is truth universality is it the same

thing

>> kind of or like it's like

given this constraint

what is that ultimate answer or what is

that simplest configuration of your

system that does everything?

>> Yeah. the most beautiful

state. [snorts]

>> You have another idea about

inevitability. Uh you say the best

future solutions seem almost

retroactively inevitable. The

philosopher who said that the truth is

what never had to be said. Yeah. Might

as well have been talking about a

product so perfectly aligned with its

context that no competitor can have

propose a simpler alternative.

>> Is that I mean it obviously connects to

the truth and the universality. Um

maybe it maybe maybe really what you're

pointing to there is what you said

earlier which is that there there

actually is some objective

final at least final for right now form.

>> Mhm.

>> How do you design? How do you design

towards inevitability?

>> Yeah. You kind of project. It's like you

always design

say there's a set of fundamentals that

don't change

and then there's like a ideal future

that you want to go to. Mhm.

>> Then [clears throat] you figure out h

what are the deltas between that

>> is that f that future. Sorry to

interrupt you. Um

>> you you could certainly think take

notion example. Um we are going to take

a really really simple set of very

flexible building blocks.

>> Some of that you when you were working

on it 5 years ago or Ivan when he was

working on it 10 years ago may have had

some sort of future conception. I've

seen some of the early decks Ivan had

like he

>> there's crazy stuff in it.

>> It's amazing. But on some level, of

course, he didn't fully know.

>> And so I'm curious how like how

important it is for the specificity of

that inevitable future outcome,

>> right? It's more like

>> it looks retroactively

inevitable, but when you get there, it's

very ambiguous. Like you actually don't

know.

>> Like you start with you actually don't

know. And then you're you're looking at

what do I have? [clears throat]

>> What do I want to do? or like you know

my future state my ideal you can just

imagine like don't limit yourself

>> and then you start thinking h maybe

there are these kind of big changes I

need to do these are the little steps

that I need to take the closer you are

to the present the clearer the the step

is the further out the muddier it is but

then the only way you can start doing or

start going towards it as you do things.

You build you know steps or I kind of

like

say like prototypes or like pieces of it

and then as they

get built, get used, get feedback,

you kind of clarify the thing and you

move forward.

Obviously a lot of this is

philosophical. Um, someone might listen

to this and this this combination of

complexity and simplicity, it's really

appealing. Most designers, most people

making things along a long road, um,

>> are forced to compromise somewhere along

the line. And so it almost feels like

maybe one of the things getting in the

way of getting to truness or

inevitability is practical compromise.

>> You're also very practical. You're

you're sort of just pulling this thread

in many ways. Like how do you

>> how do you sort of fend? I'm sure there

are a million compromises notion could

have made along the way. I'm sure there

will be many compromises cursor is faced

with.

>> Yes.

>> How do you relate to that?

>> Yeah. It's like

I don't want every single thing to be

perfect or like there are certain things

that are like say

they're actually okay to be a little

divergent or like you you kind of let

let it go a little bit, let it roam a

little bit [clears throat]

>> and then see what people feel. see see

how the thing, you know, does.

And then

you're like on this constant loop of

like

re-examining what you have in your

system,

all the things you add,

see how they're perceived, and then

you're trying to maybe now we need to

like unify these things together. Maybe

now we need to like clean this this part

up.

>> Yeah. Yeah. And then once you do that,

then you maybe open up boom,

this amount of like people can use it

now or you make this part of the

experience better. Um,

and it could

like it's not like a feature level thing

anymore. It's more like all these things

together because they make a better

system because the system is more

flexible or extensible

and you also like

increase its capabilities then it can do

a lot more for a lot more people. Um,

and it's not just about like,

uh, let's make this feature A and then

see how it does and then run some

numbers on the, I don't know, like

adoption, retention, whatever.

>> It's it's I it really kind of feels like

it goes back to the swan. It's like um

or or maybe use another metaphor. It's

like

>> you seem to be constantly taking stock

of both like what is this pixel and also

what is the what is the picture of the

>> you need to like go

around these layers of abstraction.

Yeah. If you

>> really want to make something truly

simple,

>> it's like a lot of people also think uh

simplicity is about like removing things

or let's just get rid of all the I don't

know any any feature that gets used less

than 5% by of users and then you're like

removing something that maybe the 0.1%

power user really loves and depends on.

Maybe the better way is to just like

you just like marie condo it like you

just clean it up a little bit or

reorganizes that so that like most

people get the

like the most easy path but there's

still like little pathways for others.

Um you don't have to take things away.

You just tuck tuck them away maybe or

like you build like elevators.

What do you say to It's funny you bring

up Mie Condo. I think like for many

people that's very aspirational. For

other people they're like how

unrealistic. Like she doesn't live in

the real world. She spends all her day

cleaning. Like

>> you've you've written and talked about

minimalism which maybe is a little bit I

think minimalism maybe

>> people take it too far. Gets a bad rap.

>> Yeah.

>> How do you relate? Like you it it

doesn't seem you you you present you're

very refined. you you you um you clearly

care about aesthetics and yet Rio like

it has like a little lived in like a

livedin messiness almost.

>> Uhhuh.

>> Is that I don't know what my question is

there but like do you how do you how do

you have that sort of tidy thoughtful

careful and also like aliveeness

in a system designing it?

I think it's like a lot of people think

these attributes is like you have to

have this or this when you can actually

have both.

>> So like should it be simple or should it

be complex?

Should it be flexible, should it be

rigid?

Um

to me it's almost like

because software is it's like it's

almost like a life form. It's like a

run. It can mutate. It changes itself.

>> You don't have to be like so

opinionated.

Like your opinion is actually taking the

stance of I don't have too much opinion,

but you always make things start really

simple.

Um, and then you let people play with

it. You let people discover what they

want or the the the way to to do things.

what is you know their way to do things

it is not my way like

I don't want to force my like my way of

thinking or ah this is how you do it one

two three onto you

I just kind of give you

like pathways and elevators

>> and the tools to do the thing you want

>> yeah you have a line somewhere you say

no point solutions always spectrums

which I think captures that Yeah. Yeah.

It's like

like fundamentally all these tools are

the same things.

So

like if you're okay with that then you

don't have to really pick like ah do I

want to do this like cursor for um

salespeople or cursor for coding. It

might be the same thing. M I want to

talk about that kind of process of

making and you you started to get out a

little bit. You have this metaphor of of

sort of like sculpting or finding what's

in the stone that I think is really

powerful that's not totally intuitive

for how people think about

>> creating. Um you say there's there's a

quiet almost mystical art to starting

with something so unrefined that you're

unsure if it's mud or marble and

patiently revealing its shape until

others recognize its beauty. In the end,

they'll say, "Of course, it's so

obvious."

>> Yes.

>> Why? Why can't greatness

be why must it be emergent?

>> Because you haven't seen enough. You

haven't tried enough. You think ah this

first idea I have is perfect.

and you throw it out there and you

realize

maybe only I think like that or

maybe people people like it but they

don't really understand the words or the

nuance in there

>> then you need to like keep tweaking and

keep getting input. It's like

you never start with something that's

like the the ultimate answer. You always

start with [ __ ]

>> and then you make it better and better.

>> Is that the case for every medium?

>> I think so.

>> Like um

>> like even when you're painting.

>> Yeah.

>> You start with like the pencil sketches

and then you layer on top like the paint

>> or like you're sculpting, you start with

just like a blob of clay and you're like

making the highle like shapes good

enough and then I start like working on

the details. It's the same thing like

you never you never get the first shot

right. Even more true with like AI.

>> Yeah. Um, but with AI is like or like

say with curs composer one because it's

so fast. It's like

it's a different way to do things now

like you you're building as you're

seeing things as you're thinking,

>> right?

>> And as you're designing and it's all

together.

>> Yeah. I wonder like you you referred to

software earlier as almost like an

organism. Um,

>> and maybe that's something that's true

about software inherently, but it feels

especially true with AI now.

>> Um,

>> and you, one of the things you said to

me when we met, you you talked about

sort of how you used to work being much

more like painting or drawing and now it

feeling much more sculpting or finding

something into a stone.

>> Yeah.

>> I don't know that that way of thinking

is intuitive to people.

>> Um, even people who make software. And

so maybe one question I'd have would be

like, have you started to think about it

in a fundamentally different way with AI

or is this actually just a continuation?

>> I think it's almost like going

backwards.

It's like I started building things

myself and designing everything.

A lot of times I did not use like pixel

tools. I just coded it. And then

um I became like a professional product

designer.

>> Yeah. Capital D designer.

>> Yeah. Yeah. And then Oh, interesting.

Now I just make mocks and fancy animated

prototypes and then I'll drop that mock

into my P PM's PRD

>> and I'll wait for things to happen and

things don't happen. [laughter]

Um, and then now it's like going

backwards, meaning like I have an idea.

I'll just prototype it out.

>> Yeah. Like a kid or like a kid with a

piece of glass.

>> Oh, yeah. Or, oh, there's a bug. Okay,

I'll just like

make a screenshot and then circle the

thing. Ah, add cursor fix this and it'll

get fixed. It's like instead of waiting,

instead of getting stuck in pictures or

words, you actually make the thing

where you use software or use code as a

tool to communicate your ideas better.

And because we're software makers, the

best tool is code.

There's a I interviewed early on I

interviewed a couple of designers like

industrial designers, physical designers

um Seway and Taylor and one of the

things that they feel really strongly

about is like they hate renders.

>> It's like make make the prototype.

>> Oh yeah.

>> And I almost feel like this is the

digital version of that is like get it

down in the metal code.

>> Exactly. You you have to play with the

material.

>> Like our material as software makers is

never the pixels. It is the code itself

that renders the pixels.

Yeah.

>> Yeah. You have a line I love. You say,

uh, "But it existed and because it

existed, it could be improved."

>> Mhm.

>> Which so captures the like power of

working with actual material.

>> Mhm.

>> Uh, it I I do wonder like

you you we we were when we first

talking, you said um I use Figma when I

want to go into my my old way of

thinking. Yeah.

>> Which obviously relates to what you just

said.

I'm curious today like and maybe part of

it is that you're designing cursor which

is especially uh conducive to it's it's

less about the pixels already.

>> Mhm.

>> But when do you find yourself sort of

like tempted towards the old way of

thinking and like is it a yo-yo? Is it a

like will you be using Figma at all in a

year?

>> Oh yeah.

It's like there are just tools and

like sometimes we think in words,

sometimes we think in pictures.

>> On podcast we definitely think words.

>> Yeah. Or like I don't know making videos

too. Some people do that.

>> Yeah.

>> Um or like slides or whatever.

Like those are just you know

different artifacts or like forms to

help us think. And I I think

like I don't want to take them away like

different people have their preform

preferred form to think. [clears throat]

>> Maybe some people are more like linear.

They just write text.

>> Yeah.

>> Um I like bullets. I I think I got the

disease from notion. It's like like all

I do now is like I go go out and then I

walk. I have ideas. I'll open a notion

dock and I put in a list

and then once I'm done with my walk,

I'll go go back. Huh. Maybe now draw

some pictures and maybe I'll do Figma

cuz it's so like

cuz I've been doing this for so long. is

like water to me.

>> Like I don't I don't think Yeah. when I

make more artboards or when I do the

Figma like shortcuts. So when they

change shortcuts or like they move

around my things, I get mad. [laughter]

>> They keep doing

>> I saw you you were really mad that they

had changed the check box.

>> Oh to box. Yeah. Made the quarters.

>> That's for another thing. That's okay.

That's more for like it's like I feel

like

like every piece of software is almost

like a person.

>> It has [clears throat] a style.

>> It has like a history. It has some

character

>> essence.

>> Like you don't want to lose that.

>> Yeah.

>> You don't want to order water everything

down to like a border radius for pixels.

[laughter]

>> Um like sometimes it's good to keep

that.

>> Yeah. M

>> keep a lineage and keep a thing that's

maybe a little weird

>> but is so like characteristic.

>> Yeah.

>> Yeah.

>> On the on the note of sort of your

thinking time and you talked about

thinking and

>> using different tools you're thinking

using Figma.

>> Um you you've you've talked about your

walking and like the the value of the

sort of like idle time, the space

between

>> um

>> thinking time isn't wasted time.

Are you and and maybe this is running

against what you just said about it

feeling like water, but are are those

like different modes like when you

>> watching you use cursor at least on your

phone when we were hanging out?

>> It didn't seem like you were doing very

much thinking. You were just like you

were just throwing like paint at the

canvas,

>> right?

>> Um and then when you write about your

walks or like that that feels like a

very structured is that maybe a template

for how

>> that's more for the longer term things.

Ah,

>> yeah. We're like vague ideas, ambiguous,

>> dreaming.

>> We're like, huh, maybe we should do

this. I'm not sure.

>> Maybe we should do it this way. What are

the like the components in there? How do

I like break it down? Um, what are the

things people care about?

>> Whereas when you're using Figma, you're

using cursor. or

>> those are more for maybe like Figma it's

like

there's still some say like difficulty

where it is just like it just takes more

time to say build a really crazy

prototype in like code

>> ah

>> so

if you want just communicate ideas in 2D

space really quickly draw some pictures

that's fine and then when the thing gets

to the state where I think I know what

it this

um I want to figure out how they fit

together, how they work together, what

are the

you know,

especially with like building AI stuff,

there's like so many

like both like procedural and like

non-deterministic things that you need

to think about. It is really like really

hard to simulate in Figma or like in

static pictures.

>> Yeah. And you're not with the material.

you're not up close to the material

>> like you actually need to glue it up and

then see how they fit together. See how

the states transition? Uh if I get this

like error, what happens? Or uh if the

the the return gets too long, what

happens? D like you never get that in

Figma.

>> I want to talk a little bit about Rio

OS. Mhm.

>> Um, both because I know you're you're

very obsessed with it and it it does

feel like the perfect embodiment of this

sort of working with clay.

>> Um,

and I I think it's I would strongly

encourage people listening or watching

to go to poke around with it. Um, as I

understand it, Rio started as a

soundboard app you made for your friends

when you were leaving Notion. And it

sort of feels like it's this just

infinite thread you keep pulling or this

piece of clay you just kind of keep

turning over in your hand.

>> Yeah.

>> Um

>> for people's contact I we we when we

first met you had your phone out and you

were like we were just ch talking and

you were literally making apps as we sat

there and talked.

>> What have you learned about

making things and maybe even about

yourself from this crazy project?

>> Uhhuh.

I learned that, oh [ __ ] I can do all of

this.

I think that's the biggest thing. And

it's like

it's all like little ideas piling up on

each other.

Um, you start with like

something simple, small.

Um,

and you just keep building and building

and building and building and see it see

it grow

and then when it grows to like a size

where it's like

you know there's some constraints. I

actually started the thing in Vzero not

cursor

>> like the soundboard thing.

>> Yeah. Yeah. Yeah.

>> Um like I ran into some like errors that

I'm like I need to do it in cursor.

>> [snorts]

>> And you use cursor routes prior to that?

>> Not really. I tried three times. I

turned three times.

>> Oh, interesting. Why?

>> Yeah. It's like the first time I was

like, "Oh, cool. New new code editor.

Let me try it out."

>> Ah, I typed some lines, it completes

like five lines of code instead of one

line of code versus like GitHub Copilot.

>> Then I tried

>> because you felt like it was trying to

do too much.

>> No, it's like it's just completing code.

Oh,

>> with more lines. Yeah. And then second

time it was the chat.

>> Yeah.

>> Um it's like chat GPT next to your code.

>> Yeah.

>> And it can read the code. It can answer

some questions but it can't do much. So

I turned and the third time it was like

discovering the agent.

>> This is post using VZ or pre

>> after. Okay. It's like

I needed some tool that can let me do

anything.

Then I found cursor

and I'm like hooked. Yeah. And you you

start from like simple things and then

you just ask some maybe a little crazier

idea and then you see it getting built.

And

see now with plan mode you actually see

how the models think and you can change

you can be part of every you know step

but it's still your clay but it's like

the model now handles all the parts that

I don't really care about.

I actually studied like computer

science. Okay.

>> Because I love

computers and software,

but I hated writing code or

>> like all the algorithms and stuff we

learned is like kind of useless. And

>> what I care more about is like

>> like what are the ideas?

>> How do people, you know, feel

um

>> how quickly can I make this thing? I

thought of

>> Mhm. Exactly. It's like the thing, the

idea, the concepts. I want to play with

the concepts.

>> You mentioned it like real OS.

It doesn't really seem like something

like that could be should be able to be

built by just throwing more paint at the

canvas. Like it feels like the type of

thing that should have needed to be more

plants.

>> There's a lot of things that say like

it's not just throwing.

>> Okay. So it's almost like it's a

constant throwing things and cleaning up

[ __ ]

>> Okay, same more.

>> It also happens there in Rio.

>> What is the cleaning up? That's that's

what we're not seeing, I think.

>> Yeah, you don't see that, but you can

see in my commit logs,

>> the maintenance.

>> Yeah, it's like

>> the more things you add, the more things

you realize. H it's the same same thing

that I just talked like earlier. is like

[snorts] ha these all these apps need

say some AI endpoint and some O and like

they need to store their states

>> uh they they need to write or read into

the file system D like

maybe I started you know doing the file

system part from the text edit app but

then now I want you know all the other

ones that that can use the same ideas

um

to use the thing then I need to re kind

of abstract the system like put that

part out or unify some you know state

management things. Um,

and then you need to kind of refactor

your original things. Even though maybe

to the user it looks exactly the same.

>> Um,

>> that part of it though I think is where

like for lack of more precise language

people get stuck.

>> Yeah.

>> It's like again I I watched you use

cursor. It's like you're literally it's

like you're just nudging the model and

and your prompts are not Yeah. It's that

it's that demeanor um for for the

listeners. You're just poking at it.

Yeah,

>> it's not these long specs. I'm watching

you just be like, can you come up with

an app idea?

>> Like your language is really casual and

so I think to the average person using

Vzero

>> um uh or the person who tries cursor and

turnurning, I think we'll talk about it

later. You're very clearly focused with

cursor on building for the hardcore

user. Mhm.

>> But for someone who has somewhat of a

computer science background, hadn't

written a lot of code, it maybe what I

wonder about is like in the poking

process, you're getting more invested

that you care enough to do the hard

maintenance part.

>> Oh yeah. I learned a lot by building

real

like before

even like since I became a professional

product designer, I would [snorts] have

little projects I do Yeah. on the side

like the first few years I kept doing

those

>> and then I got busier or something and

then I stopped

>> and then every time I tried to go back

oh [ __ ] I need to learn like React 18

when tell CSS whatever all of this like

new things and then it takes a long time

I have to read all the docs I need to

understand how people do things now um

but it's

Now with the agent, you don't have to do

that, but you're still doing that. It's

like the agent maybe helps you do the

research. It comes up with some, huh,

here are how people do it now. And then

maybe gives you some alternative

options. Maybe you know certain things,

you also don't know certain things. Um

but the agent can kind of help you find

your way and then you can say ah now

just do this. It will write the code.

You can look at the code still you can

learn from its output

how things work. Um

>> yeah you're getting deeper into the

complexity.

>> Yeah. Yeah. Yeah. Like

>> by it's what intentionally or otherwise.

>> Yeah. Yeah. Like it's almost like just

by reading

like a lot of users say this too. It's

like they love reading how the models

think.

They actually want to expand everything

and then they want to look at look at

all the output

>> because it helps them understand

what the model is doing, gain trust from

it and learn, you know, especially when

they're starting to code.

>> Yeah. It's a it's it's might be a

strange comparison, but uh somebody I

interviewed, he was talking about

reading with his like seven or

eight-year-old daughter

>> Mhm.

>> and how reading with her this these

these books that were actually far

beyond her sort of ability level

>> it pulled her in and it and now she's

reading whatever I don't know if she's

reading an but like she's reading well

beyond her level and there is something

about sort of like

>> being exposed to someone else's

thinking.

>> Yeah.

>> Even if it's GBT5 codeex or composer or

whatever.

>> Yeah. It's like

most of the as you said it's like most

of the prompts that I did in real is

like just really short simple things.

>> Yeah.

this it's like

theoretically we are ready like you can

actually build a lot of things and you

just vibe but there is like you know I'm

a little cheating too cuz I know things

before

>> right

>> so I know like when when the AI gets

stuck how to like get get it unstuck

or like

um as I play more

like my full-time job is to play with

all these models and use cursor.

So I kind of develop like some intuition

on how these say different models behave

as I make it. Yeah. Or like what are

their limits? Maybe this one's faster,

this one's slower, this one's smarter at

certain things. um that a lot of people

like

they don't know they don't really know

what to do yet.

So that helps me like put this back to

the to the tool. On that last note, um,

>> when is it your job as the design or

maybe a better way of asking, when is it

cursor's job to try to solve those

things versus the models improvements

job to solve those things?

>> I think it's both.

>> Okay.

>> Um, the models can kind of raise in

capabilities or like say now the models

are getting better at say using terminal

commands,

uh, clicking around in a browser, stuff

like that. Like as they get better,

like

you still need a way to kind of unlock

those capabilities.

So you need to fit them back to the tool

itself, package them up. Um make make

them

just really obvious. Um so people can

just

play with them.

um they don't have to think too much

like how do I

I don't know trigger it or get it out or

use this crazy like script or MCP thing

to do something.

>> Yeah.

>> Um like you start simplifying

making things that are possible more

obvious.

>> Yeah. For more people.

>> Ah [sighs] that's an interesting way of

thinking about it. Yeah. Making things

more obvious. making the next step more

obvious.

>> Yeah. It's like you're constantly

simplifying, unifying,

figuring out like, uh, now that we I

have this and this and this, now how do

I like clean it up even better? It it

feels like it relates a little bit to

the like readiness thing we talked about

talked about earlier which is like

>> it feels like maybe the model's job is

the technical readiness and your job at

cursor is the cognitive readiness.

>> Yes.

Like again humans are kind of

we're like kind of singlethreaded. You

know we've been trying a lot with like

multi- aent or like parallelization of

like agents.

>> Yeah.

And like nobody has really solved it yet

because most people are still thinking

about no like let's just give you 15

agents. Here you go. 15 agents are like

have done all these changes like 2,000

lines of changes.

>> Here you go.

>> It's like all horsepower, no steering

wheel.

>> Yeah. Yeah. So, we need to like

figure out,

you know, these like

I'm not even sure if there will be new

patterns, but it's like better framings

or

packaging

or interfaces

for people to just get out get utility

out of these things. um

without

breaking their minds or like changing

too much or feeling

overwhelmed.

>> Yeah.

>> Yeah.

>> Yeah. You um you've obviously worked on

a lot of different types of systems. Um

and you're sort of drawn to almost like

this container

>> type tool or product or something. Um,

and certainly at least with cursor and

notion, you have you have a line where

you say systems thinking is essential

because the only path to building

products that scale not just technically

but cognitively along the lines of what

we were just saying.

>> Yeah.

>> Um,

what are the is the goal when you're

designing a tool like that um to allow

the user to stay as singlethreaded as

possible and like like is that

essentially what you're designing for?

>> No.

>> No. Well, it's like it's up to you.

>> Ah,

>> um it's like you need to design the zero

state, the one state and the end state.

>> Ah, [clears throat]

>> for everything and then see how they

melt together.

>> This is the simplicity complexity.

>> Yeah. Yeah.

>> Like when you have n * n * n, it will be

kind of crazy.

>> But if you really want to be there, so

be it.

>> Yeah. You you should meet the user where

they're at.

>> Yeah. Yeah. Yeah. Like if you are

actually like someone I don't know you

you have ADHD or something like you want

like eight different windows all like

running so be it.

>> Yeah.

>> Yeah. Like

>> the average person probably wants,

>> right?

>> Maybe average people just want one main

thread and then

>> it's like how we're thinking right now

is like

>> instead of having you like you need to

review changes from these 15 different

agents, maybe we help you like kind of

cluster them a little bit, organize them

semantically. Maybe you instead of

talking to each of them separately, you

just talk to one person or like one

agent and then it's almost like your PM

or like your assistant and then it's

going to figure out ah these these guys

are blocked. Do you want to like approve

the terminal command? Uh these changes I

think they're pretty good. This is bad.

You should look at it.

>> There's a very small subset of users

want Starcraft and most people want

Candy Crush or whatever.

>> Right. It's actually like I'm fine with

both. Yeah,

>> we can actually do like both like a

I don't know a Tik Tok and a Starcraft

>> because of AI.

>> Yeah.

>> Yeah. Um there's an idea that I think is

really interesting that I think is

connected here which is

>> about slack

>> in systems. You say the best systems

have slack in them. Redundancy isn't

always waste. It's optionality. Multiple

paths mean you can explore without

breaking everything. The core remains

simple

>> uh while layering itself into more

complex permutations. Controlled chaos

means you're able you're stable enough

to not collapse but loose enough to to

evolve.

>> Mhm.

>> I think that's such a powerful metaphor.

Um and maybe slack is that like

willingness to go as complex as I want

to. Yeah.

>> Um,

>> but yeah, I wonder about like you

somewhere else you talk about that sort

of chaos and order together.

>> It's like you you let diversions happen

>> and you let things evolve. It's like

evolution.

>> Yeah.

>> It's like the na like nature is

constantly

like making more, you know, permutations

of the same thing a little different.

see which one works better.

>> How do you give a tool more slack?

>> It's

what does it

>> mean to add slack to cursor,

>> right? It's like

>> it's a little complicated, but also it's

like

sometimes you just kind of, you know,

all designers or people we're like kind

of perfectionists.

uh we want like things to be exactly

what we wanted, but

sometimes you just allow this ugly thing

to pop up or this random button someone

else added and then I kind of keep a

blind eye on it. [laughter]

Um

and you let it simmer a little bit. You

let people play with it more like our

internal group of people. Um, and then

as you do that or like maybe people, you

know, threw the first bucket of paint

and then

now that it's there, you can see it, you

can play with it, you can think about it

more, understand it better then

>> versus sort of roping off the the

canvas.

>> Yeah. Yeah. Then it's like ah now I know

how this thing fits with the other

things or like ah this thing is actually

like a start of something much bigger.

Mhm.

>> Um

then it's almost like this constant,

you know, chaos convergence

thing

and it gets into like a equilibrium and

then you want that thing to be like

almost at the edge of like the maximum

chaos you can allow

>> for the thing. Your [clears throat] job

as a designer is almost

>> like you're trying to help people like

uh here is the line don't cross it.

>> Yeah.

>> And then you're also helping people like

bring this like

>> just like reducing like entropy

>> like just just

>> tame it a little bit back. Um we're like

ah these you you should talk together

and then make this thing actually the

same thing.

>> Uh or like ah you're making a new thing.

Cool. think about these four things that

we have.

>> Yeah.

>> And that's it. I'll just let them think

about how does this new thing relates to

the four things and then

>> ideally they come back with a good

answer.

>> You're almost like um you're

[clears throat] like the game maker or

you're like you're the agent of

evolution that sort of like setting the

rules of a little bit of what is

tolerated, but critically you're not

snuffing things out too early.

>> Yeah. And it's like you're you're mostly

like an observer

>> or like

I'm not dictating how things should

happen.

I just tell you like uh given all the

things I know, here's probably how we do

it.

And this is also maybe why it seems like

you're very attuned to not just the

different ideas for cursor inside of the

company but

>> all over like all over Twitter different

stakeholders,

>> students, whatever.

>> Yeah.

>> Cuz you're almost trying to like broaden

the aperture of what is allowed in,

>> right? Cuz fundamentally it is the same

thing. All the vibe coding tools,

cursor, all the CLI agents,

it's all the same thing. But only like

cursor kind of tries to bridge all of

them.

>> And like I try to give people like their

ideal form.

And I think like one big like a big

reason

cursor got popular is because it looks

exactly like VS Code at least before.

Um but as we kind of noticed like people

changed their patterns of usage, people

kind of moved from like manual coding at

like reviewing every line

to do more agents.

>> Yeah, you have to move with them.

>> Then we just flip,

>> right?

>> Like our defaults change as the world

moves and as the product evolves, but

fundamentally it's still the same thing.

>> What is cursor? H

>> obviously cursor is a plugin or a skin

of VS Code on some like not just that

>> of course of course not just that and

it's changing every day like um

>> again at least when we spoke first like

you talked about cursor like it like at

least the way you seem to relate to

cursor is almost like it's your little

butler that just does things for you.

It's your hand. Um, and we talked about

code being the universal language. Like

in many ways it almost feels like cursor

is just this medium to work with code

with computer,

>> right?

>> And so

>> I'm kind of asking about what cursor

will be when I ask what cursor is, but

like do you have a conceptual do you

have a metaphor? You like it is a tool

but it's sort of this is it just the

agent.

I see it as as like we

we started from like one slice of like

making software which is you're just

actively coding

when you're sitting on the computer.

>> Yeah.

>> We put an AI next to it so that I can

help you write the code.

And now it's like like I want cursor to

be

it's like one place where you can do

everything about making software and

that is not just writing code and it's

not just the developers

there's like the PMs thinking about what

to do how to measure things aggregate

all the data synthesize it figure out

like what are the problems to fix. Um,

breaking it down into tasks. There is

the designer.

Maybe they're trying to kind of, you

know, explore in in 2D space higher

level abstractions.

There's the

engineers writing the code, but also

they need to like review. They need to

test whether it worked.

um once you put it out,

you need to like gather feedback and

input from the market and people using

it.

Like all of this is making software

uh especially in like a team or like a

company.

Um and now people's

people's workflows and tools and the

metaphors they use the artifacts are all

scattered and disjoint.

>> Yes.

Whereas I think cursor can actually help

everyone

put everything together again

and then using the agent it's the same

agent to help you translate between so

your form of thinking your preferred

artifact into the code itself.

Um

then it's almost like anyone who wants

to build software or any team they can

just

be closer together.

>> Yeah.

>> And then the agent kind of helps them.

It's like solving a lot of the issues

that we have today that were kind of

created by all the tools that that we've

made in the last

>> Yeah. We just need one more tool.

You need you need a thing that kind of

melts them fully.

>> What about cursor shape though?

>> Yeah.

>> Every people have been trying to build

the final tool forever,

>> right?

>> What about cursor shape,

>> right? [clears throat]

>> Makes it what you're describing

theoretically possible acknowledging

still currently serving mainly deps.

>> Yeah. I think it's like like people joke

about like cursor is like a fork of VS

Code and it's just code editor. But if

you look at VS Code like deeply, there

is actually like really good low-level

primitives.

For example,

like in VS Code, there's a concept of

editors like you can open different

files in different kinds of editors.

Some of them might be looking like, you

know, the code editor. Maybe there's

like a diff viewer. Maybe there's like a

markdown preview. Maybe there's a

browser. D.

Like just having this allows me to just

present different things to people

differently

even though you know underneath it's

still the same code.

>> Is that because it works with files or

>> Yeah. Yeah.

>> Yeah. So that's another thing is like in

VS Code there's a concept of workspace

which is just like folders and files.

Maybe they're tied to a repo. It's like

a lot of these low-level ideas

again it's like they don't have to

change and I don't intend to change them

though like I don't know if we will ever

detach from VS Code at some points maybe

once we kind of you know go fully agent

>> um

>> or at least a lot of the people using

>> code exactly right

>> but I think it's still like

like the challenge for me is to come up

with a way to

so you're tying all of these different

workflows and people's preferences

together into one thing

and you're trying to come up with like

different reconfigurations of that

thing.

>> Yeah.

>> How they transition between these states

um for these different people? What do

they each see by default?

>> How do they like customize it? How do

they actually talk together? That's a

really complex problem.

>> How do we move from like cursor from

like a single player thing to like a

multiplayer thing?

>> Not sure.

>> You got your work cut out for you.

>> Yeah.

>> Um on the on the note of like literally

using cursor, we talked about the way

you kind of poke it at least when you're

using Rios.

>> Yeah.

>> Um you had given me like your advice was

like treat it as someone who's like a

little dumb.

>> Yeah. [clears throat]

>> Composing things it's seen before.

>> Yeah.

>> Don't expect to come up with full

components. you shared a list of 12

rules or tips for using cursor back in

April.

>> Yeah.

>> Um so those are almost like two slice

timestamps of of advice around cursor

like

>> uh one of those I think that stood out

to me is if the code is wrong just write

it yourself. Cursor learns faster when

from edits than explanations. Obviously

that's

>> that works for someone with a coding

ability not without a coding ability.

>> Um

>> how often is advice like this changing?

>> Oh yeah it changed a lot.

>> Okay.

>> I would say a lot of the things I said

in April don't apply. Okay.

>> For example, like the agents now are so

good at finding stuff

that you don't have to say like at the

exact file anymore.

>> Back then it was like if you don't

include the right context, the agent

will just come up with something random

or it will make some mistake. What is

the what is the is there anything that

stands out as long as you've been

working on cursor that has been

um true consistently or even like the

type of person who consistently remains

good at like what is staying the same I

guess is what I'm asking

>> not much

>> not much you got to be surfing the new

wave every

>> yeah things are constantly changing

>> even the things that appear the same

might be replaced under the hood

>> that's both exciting but also back to

the you were talking about you u I don't

know it was notion or something else

like some you you have a tool you're

used to and they change a little Jeffrey

Lit has this metaphorical they change

your chef knives

>> that's hard to

>> yeah I guess there are things that don't

change say

the agent

it used to be like you know before I

joined cursor

there were like five things like there

was command K tab chat composer composer

agent.

Uh the first thing I did was to merge

the agent. So chat composer composer

agent became agent with like you know

more specific modes if you want you know

more specific behaviors.

>> And then the agent the idea is they're

all the same. They're just like apply

configurations on top of the agent.

Maybe for this agent it has some custom

prompts. It has a specific model set to

it. Maybe it has like some tools that I

can use or cannot.

That's it. And you give it a name.

Um,

and then these agents all, you know,

operate on different models. Those don't

change. They need context that don't

change. And then you need to show

something with the editors that don't

change.

>> Yeah. But all those things are changing.

>> But all of these things are changing.

Yeah. It's like all the things inside

are changing. I guess your bet is that

somebody

>> know the art changing,

>> right? So if your bet too is that if

somebody's playing with the clay,

they're okay with change because they

are living with the material in a way

that

>> you have to or like

>> I think

like in my career as a professional

product designer, the thing I hate the

most is like h like people want the

design to be final.

Uh where's the final version of this

mock? If you don't have it, I won't

start building it.

Like that doesn't make sense [laughter]

>> cuz the first mock is never right.

>> Yeah.

>> Like you have to keep building it.

>> Yeah.

>> Like now it's almost like the reverse

happens at cursor which is kind of

chaotic but I'm actually okay with it.

It's like our engineers or like some of

our like enterprise PMs, they start like

vibe coding

and then some weird patterns emerge or

>> you need to clean it up again.

>> You need to like wrangle it back. And

then now it's like

because AI is really good at composing

parts.

I'm actually thinking we need to like

build bricks. Really good bricks. It's

like

from all the things that we have as like

kind of suck all the patterns the the

core bricks.

>> This is something that seems like you

guys did a really good job in notion

which is that you're like pretty

principled about what the bricks were

going to be.

>> Yeah. Notion does it more like on the

conceptual level.

>> Oh, you mean like tangible feature

bricks almost?

>> More like I don't know. It's like

low-level components up to like patterns

that people can just reuse.

>> Yeah.

>> That are not just you know every

dialogue is different or list view is

different.

>> [snorts]

>> you you um

>> like you start helping people create

these

patterns that just work and just fit

together that both humans and agents

can, you know,

>> Yeah.

>> make things a little better by not

reinventing the wheels every time

cuz the agents when they're like lacking

guidance, they have a tendency to do

that.

We talked a bit about like I think

you're clearly designing for hardcore

users. Um even if people are vibe coding

with cursor like maybe the lines are

thinning.

>> There was I think a line from you

somewhere that I found where you

>> or maybe I made this up but

>> um I think you talked about like

designing for power like to give the

user power.

>> What does that look like maybe in the

context of cursor or more broadly?

>> Yeah, I think a lot of people

So, I don't see your users as like

they're dumb. They're not. They can

figure things out. They don't have to be

like babysitted. They can

It's like

I want to make things the simplest that

you can when you start,

but as you go,

you get all the, you know, depth that

you want. Um, like as a beginner, you

get the same tools as what the pros use,

just maybe packed a little differently.

>> Yeah.

>> Um,

>> yeah. You don't have 18.

>> You don't see everything yet, but

maybe this this thing that you get can

do like 80 90% of what you wanted. Maybe

on the other side like currently there's

I think most people's my my intuition

would be that most engineers

relationship is like there's five coding

and then there's real engineering

obviously that's

>> that's the same

>> challenged yeah

>> what is how what does it look like to

design for power and for serious

hardcore users on the like vibe coding

dimension

>> and part of that is conceptual right

because it's like they have to be

willing to say I'm going to I'm going to

give up the wheel

>> or not the wheel Maybe, but I'm going to

let the engine be.

>> Yeah. Yeah. Yeah.

We do like little nudges and we change

our defaults sometimes. And I think

those are probably the most powerful

tools that you can do as a as a product

or like a piece of software.

>> Yeah.

>> And then you want to introduce them in a

way that like people can still get out

of it if they want, but you want to show

them that, huh, now here's the new

world. Here's how you do it. Um, if you

don't want it, you can get out, but

it's almost like again

the same thing but reconfigured or like

slightly more optimized for the new way

of doing things.

>> There's a little trust there too, right?

Like it's like actually if you if you

trust us for a minute, let us show you

how much the agent can do.

>> Yeah.

Yeah. people like a lot of people

haven't felt it yet or maybe they've

tried it before but it didn't work and

then they kind of lost their trust

>> right

>> and then they never

>> they can turn three times like you

>> yeah and it's like so it's

I would say for now

you can probably do something pretty

impressive even on the first shot but

even say like for for a month ago it's

not the

So maybe the first time you tried cursor

it didn't work or it got blocked or it

did something stupid

and now you're like I don't want it. Um

it's like we need to figure out how to

like get the new people in

um without too much thinking and setup.

They can do stuff.

get the existing users,

you know, onto like better ways to do

things that are more like up to date

>> without feeling like they're behind.

>> It's like you want to kind of carry them

over instead of like teleporting them to

the new world and then they're like,

"Ah, what the [ __ ] is [laughter] this?"

>> Yeah. And then there's like getting the

people who maybe tried cursive before

that thought

it was not good to come back cuz it's

good now.

Um

yeah, there's like work for us to do

there.

>> Solvable problems.

>> Yeah,

>> many many problems to solve. Uh some

questions about kind of process and

>> some some other stuff that relate you

have this amazing essay about about

creating something great. Um so a few

things in this

>> broader vein.

>> First, like I guess we kind of talked

about this and maybe this is silly, but

is design kind of just writing now? Like

it seems like most of the design you're

doing

>> Mhm.

>> you have your walks you go on and then

you go to cursor and you write.

>> Yeah. Yeah. Yeah.

>> Maybe you write a longer spec sheet

>> as cursor improves.

>> Yeah, I do write docs and stuff.

I think it is just about like

communicating your idea and all the

details that you can think of

in a way digestible for your peers

>> and the agent

>> and the agent critically.

>> Yeah. Um

so depending on like who I work with

even I will change the way I make these

things.

>> So like I work with an engineer his name

is Ian. He loves mocks. He loves

pictures. Like when I do like live code

prototypes, he doesn't like it. He just

want Figma mocks with all the like every

detail in one picture.

>> Yeah.

>> So I just do that with him. Uh or if I,

you know, talk about something more

vague, people have like also vague

ideas, then I keep it more like maybe

they're just bullets, maybe they're like

simple writing. And then maybe when we

want to do something like it's going to

be like a multi-month stage thing that's

a little bigger.

>> Yeah.

>> Then I'll write a big RFC.

>> Yeah. It's all like kind of inherited

from the way we do it at notion

the writing part. M [clears throat]

>> but with cursor it's like now there's

like also

you just kind of

ah I have this idea I'll add it to my

prototype and then ho ho look at this

should we do it yeah let's do it

>> I suspect those two modes together are

quite powerful

>> like you get from like the most high

level like ab ab ab ab ab ab ab ab ab ab

ab ab ab ab ab ab ab ab ab ab ab ab ab

ab ab ab ab ab ab ab ab ab ab ab ab ab

ab ab ab ab abstract level to the most

detail

>> writing when when you say the abstract

level you in long writing

>> or like even just high level bullets or

what are the ideas and the constraint

>> are are is a is a really detailed spec

doc and a actual prototype two forms of

like two almost different trees of

detail.

>> It's like the same thing but

>> visualized differently at different

levels.

>> Totally.

>> Yeah.

>> What is on that note maybe like what

does a week what does your time look

like over I think cursor has like one

meeting a week.

>> Mhm. You're going on walks and thinking

you're proddding rios whatever uh you're

>> in Figma sometimes like what is that

like pie chart of time

>> that's kind of random

>> every week's different

>> yeah very different

>> yeah we also like jam with people

at the office people are always like

there

um

not much meetings

um

>> but a lot of talking Sounds like not

scheduled meetings, but a lot of

>> Yeah. chatting and talking and jamming

and

um yeah, drawing pictures, finding

people to help

join us. Yeah. And

>> podcasting sometimes.

>> Oh, yeah.

Getting designers to to turn into

coders.

>> You're a big ring leader for that.

>> Yeah. I want to make it happen. What do

you say to the average designer

currently who's feeling stressed out?

>> You're ready.

>> You're ready.

>> Yeah. Like it's it's time. Just start

building.

>> Just start pulling the thread. Get it

get in there with the clay

>> and then send me all the feedback and if

you don't like what you're seeing, we'll

fix it. Maybe on that note, although

this could apply to engineers or any

maker too, um I think one intuition

people have around AI, maybe the average

creative or artist, nontechnical person

especially,

>> is that vibe coding or AI or whatever

can make slop, but it can't make soulful

things.

>> You have made one of the most you've

certainly made the most soulful vibe

coded seen. If if that's um

>> right, you just

>> you need to put your soul in this. You

need to care about every detail. You you

need to not accept whatever great uh

purple gradient the AI gave you as the

end.

>> Like that is just the beginning.

>> Ah yes.

>> You always start with [ __ ] You always

start with slob with AI

>> and then you refine it. You make

>> the beginning not the end.

>> Yeah. You you just poke at it with

little prompts

>> and then it'll get better.

>> It'll take some turns.

You say uh in the age of AI the question

everyone's asking is will I be replaced.

The real question is do you know

yourself well enough to become

irreplaceable?

>> I don't think we're through with

technique and skill and craft and

mastery. Um I am curious if there are

any of those that you think are worth

mastering now. But it seems to me that

it's actually more about what you might

call intuition or sensibility.

>> Yeah. Yeah. Yeah.

>> What can you talk about that? Like what

what goes into that? Because that feels

like the it's not the end, it's the

beginning.

>> The beginning feels like I don't like

the purple slop.

>> Uhhuh.

>> That's like I know what I like and I

know incrementally what I like.

>> Right.

>> True. It's like the AI models are

trained on all the public knowledge

information and the code that I can see.

And you are trained on the same thing

like all the books you've read, all the

fonts that you know, all the artists

that you admire,

the world around you, and you build that

intuition or taste or whatever, and you

start forming an opinion about how you

want to shape the world.

And

you

you express it by building

Yeah, that's what it is. Like

>> not by thinking by the way.

>> Yeah. Not by thinking. Not just

thinking.

Then it's like you have to keep making

things and keep looking at things.

>> Yeah. One of the things that get missed

in the when people talk about taste is

taste is eating food.

>> Yes.

>> Stop thinking about food.

>> Yeah. Yeah. Yeah. You need to keep uh

eating and making [ __ ] Yeah.

>> And then make the [ __ ] better and

better.

>> M.

You critique design as aesthetics, I

think, a lot, but you're also like very

attuned to aesthetics.

>> Yeah.

>> Rios is like the most specific thing

ever. Like you've perfectly handcrafted,

recreated Aqua among many other things.

Like

what is the maybe it's back to this

taste thing, but like what is your

relationship to

sort of like not holding aesthetics too

tightly, but also still clearly really

putting a ton of time and effort and

energy and thought into

>> I think it's like

it's like how you present things

visually will always be there

and

like I don't really think about it

anymore. You just start noticing like

this feels off, this feels wrong.

[clears throat]

>> And

once you you have almost like a a set of

patterns, then you don't really think

about it anymore.

Um, unless it's like something new that

you want to stress on or

you want to like put a little bit more

flare into it. Um, but it's like all the

foundational bricks,

they need to fit perfectly even in the

visual space. It's like the visual

space, the the the bricks are it's like

the color, the spacing, the layout, the

grid,

the different like type type scale, font

sizes, and

all of that.

>> It's sort of part of the It's part of

the big picture.

>> Yeah. It's part of it. It's more like

one layer of it. Yeah. Yeah.

>> But it's like ideally the thing

is also constructed in a way

that it's like

like it's almost like the simplest form

for the low-level ideas that you want to

convey.

>> Yeah. I like that. It's a it's a it's

they're compressed.

>> Yeah. Yeah. Yeah. It's like compression

to pixels. What are they?

>> Yeah.

>> Yeah. Yeah. [snorts] Um in the

>> so you you still think about it but you

don't think about it too much.

>> Yeah. [clears throat]

>> Once you're over

>> it has its hierarchy. It has it role in

the hierarchy.

>> Right. And I also dislike how people

think of them separately sometimes. It's

like so at Google they have like

interaction designers and visual

designers. They're split [snorts] and

that's bad.

Then you create a world where the visual

designers only think about how the

button looks

and then they fight

>> not what it looks like to press the

button

>> or um

>> or feels like I should say

>> like how should the buttons be fit

fitted together. Why is there so many

buttons?

>> Yeah. Yeah. You're always backing into

this. You need to have the cohes the

cohesion in mind when you're in the

micro.

>> Yeah.

It's like um I don't know you in that in

that greatness piece you wrote about

focus and breath like we're taught to

focus early, choose what's important,

discard what's peripheral, the genesis

of a thing that might be great. Um

strict focus is a ruse. The treasure

lies in expansive searching and

stitching together tapestry of

interrelated issues. Later, once you

roam far enough, clarity will guide you

toward the right edges. Until then, let

curiosity roam. And it almost feels like

that is going in two axes, which is

>> the axes of like incremental new thing

and the axes of like hierarchy and

cohesion.

>> Yeah.

>> You do that at the same time

>> and that's why it's chaotic.

>> Yeah.

>> And ambiguous.

>> Yeah. Yeah. And you have to rein it in

with the order and

>> Yeah. Yeah. Like when people try to put

this into like a linear process or

order, they just [ __ ] it up.

>> Yeah.

because there's no more like emergence.

>> Do you think that

one view just says that like Google

doesn't have

>> realoo or whatever and pick your

favorite designer? Another view that

says the people at Google are talented

and actually like they are their system

is failing them. Yes.

>> Seems like you think the latter.

>> I think the latter

>> and I think say a tool like cursor or

its ideal form can help with this. M

>> meaning like people with different roles

or they're kind of stuck in boxes right

now.

>> Yeah.

>> You just break the box

>> and let them build the thing they want.

>> Yeah.

>> Uh another part of that essay on

greatness uh pursue agility and quality

in equal measure.

>> Mhm.

>> The myth says you must choose move

quickly and break things or move slowly

and ensure elegance. But genuine

excellence emerges from a dance between

speed and depth, agility and quality. I

love this. Like a skilled musician who

can improvise yet still maintain

impeccable technique. Yes.

>> You must learn to adapt fluidly without

compromising the integrity of the final

piece.

>> Yeah.

>> I'm curious how this this dance it makes

sense to me that it could happen working

solo on a short-term project without

that much of a plan, maybe Rio. How does

that happen maybe at other modalities

either with wide collaboration or let's

say you're working on cursor 2.0 and

it's this big long-term project. How do

you how do you embody that in that type

of context?

>> It's kind of like the

you let chaos be and you wrangle it at

the same time or

it's like you're you're

You don't pick size,

you find like a equilibrium.

>> Yeah. Between the complexity and the

simplicity.

>> And same thing with like how much fast

you want to go versus like how much

thinking do you want to do.

>> And I think especially in this age,

it's actually so easy to just try try

things out.

>> Maybe it starts with so much. In so many

of your answers, it starts with just

saying like it doesn't have to be a

choice. Like you're allowed to do both.

Oh, yeah. They're the same thing.

>> Yeah. Yeah. Yeah. It's like

people get stuck thinking like they need

to pick sides or they need to make these

hard tradeoffs when

all of these are just like variables and

you can

>> add a little bit here, lower a little

bit here.

>> Um, it's all dynamic. You want to be

more flexible to the situation you're in

and the change that's coming.

Um, you don't want your system to be

stale or stuck in like a form that you

can't get out as the world is changing.

Um,

you want to keep the essence clean and

simple. you won't create like a space

for people to play with ideas so they

can ship really fast, but maybe it

doesn't disrupt like the rest of the

system as much. And then once you have

say more more room or even like you're

constantly doing this like ah let's

wrangle things back, let's like unify

things, then you keep the core system

better as you add more things or as you

experiment with more things.

Yeah. It's like a complex system can

actually be

quite high quality and fast if its parts

are simple.

>> Yes.

>> Yeah.

>> Yet we build all this complexity and

scaffolding and arbit bureaucracy

whatever all these things that all these

shoulds,

>> right? And ideally you'd actually get

rid of all that all of that crap that is

not even part of the system

>> the software itself.

>> Yeah. It's bloke.

>> Yeah. It's like everything around it,

the processes,

a lot of it just don't make sense

or they slow things down. They slow this

loop down. Like

>> you have an idea to see it real to you

test it out and then you iterate on it.

>> How does this How many people are cursor

now?

>> 300.

>> And you were obviously in notion for a

long period of that growth like

>> Yeah.

>> How does when cursor is 3,000 people

>> Uhhuh. How does this not happen? You

guys like you you don't really have that

much of a road map.

>> The the planner agent will be ready by

then. [laughter]

And then multiplayer cursor will be

there.

>> Yeah,

>> fair enough.

>> Then people can be still like pretty

like I think how cursor does it really

fast and pretty good is like a lot of

people we hire, they're just really high

agency people. They were like founders

before that have made stuff before. They

just want to build. They don't want to

think too much.

>> Sure. But that maybe that works with

that definitely works with 30. Maybe

that works with 300.

>> Yeah. Yeah.

>> All wisdom would say that doesn't work

with 3,000. Even if you had 3,000 Steve

Jobs, it would actually be a disaster.

>> Yeah. I'm not sure. I think that is

actually one part of the

it's like a part of the questions we

need to answer which is like in this new

world of building with AI, how do teams

work?

And I think it won't be that like it

won't be

too close to what we had before like

layers of management and linear

processes.