0:00

Lovable is your personal. a software engineer,

0:02

you describe an idea and then you

0:04

get a fully working product. The reason

0:06

is to enable those who have had

0:08

like such a hard time finding people

0:10

who are good at creating software, that's

0:12

been their absolute bottleneck, and let them

0:15

take their ideas and their dreams to

0:17

into reality. You guys hit 4 million

0:19

ARR in the first four weeks. You

0:21

had 10 million ARR in the first

0:23

two months with just 15 people. You're

0:25

the fastest growing startup in all of

0:27

Europe. How did you decide on lovableable

0:30

as the name? It's so... So sweet.

0:32

The best word for a great product

0:34

is that it's lovable. A lot of

0:36

jargon that I like to use to

0:39

like emphasize what we should be striving

0:41

for is building a minimum lovable product

0:43

and then building a lovable product and

0:45

then building a lovable product and then

0:48

building an absolutely lovable product. And then

0:50

building an absolutely lovable product. So I

0:52

took that jargon with me in the

0:54

company name. People wonder just what jobs

0:57

will be more important, what skills will

0:59

be less important, as many skill sets

1:01

as possible for each person I hire.

1:03

What have you done that has allowed

1:06

you to grow this fast with so

1:08

few people? People love the product. That's

1:10

the driver of the growth. Today, my

1:12

guest is Anton OcK. Anton is

1:15

co-founder and CEO of Loveable, which

1:17

is essentially an AI engineer that

1:19

takes an English prompt and codes

1:21

a product for you in minutes.

1:23

You can then talk to it

1:25

iterate on the product and then

1:27

launch it launch it to the

1:29

world. It's one of the fastest

1:31

growing products in history, the fastest

1:34

growing startup in Europe ever. And

1:36

as Anton describes, their goal for

1:38

Lovable is for it to be

1:40

the last piece of software that

1:42

anybody has to write, because it'll

1:44

be able to create all future

1:46

products for us. They launched just a

1:48

few months ago in the first four

1:51

weeks hit 4 million ARR in the

1:53

first two months, crossed 10 million ARR,

1:55

all with just 15 people. Absurd. In

1:57

our conversation, we covered a lot of

1:59

ground. including a live demo of Loveable,

2:02

how their team operates, how they hire,

2:04

what is most enable their team to

2:06

scale this quickly with so few people,

2:08

pro tips for using Loveable, how it

2:10

all started, how he recommends you build

2:12

product teams going forward with tools like

2:14

this existing, what skills will matter more

2:16

and less going forward, plus how to

2:18

think about Loveable versus competitors, and so

2:20

much more. If you're trying to wrap

2:22

your head around how product building will

2:24

change with the rise of AI tools,

2:26

this episode is a must watch. If

2:28

you enjoy this podcast, don't forget to

2:30

subscribe and follow it in your

2:32

favorite podcasting app or YouTube. Also,

2:34

if you become a yearly subscriber

2:36

of my newsletter, you now get

2:38

a year free of perplexity and

2:40

notion and superhuman and linear and

2:42

granola. Check it out at Lenny's

2:45

newsletter.com. With that, I bring

2:47

you Anton OcK. This episode is brought

2:49

to you by Cinch, the Customer Communications

2:51

Club. Here's the thing about

2:53

Digital Customer Communications Communications. Whether

2:56

you're sending marketing campaigns, verification

2:58

codes, or account alerts, you

3:00

need them to reach users

3:02

reliably. That's where Cinch comes

3:04

in. Over 150,000 businesses, including

3:07

eight of the top 10

3:09

largest tech companies globally, use

3:11

CINCH's API to build messaging,

3:13

email, and calling into their

3:15

products. And there's something big

3:17

happening in messaging that product

3:20

teams need to know about.

3:22

Rich communication services, or RCS.

3:24

Think of RCS 2. Instead of

3:26

getting text from a random number,

3:28

your users will see your verified

3:30

company name and logo without needing

3:33

to download anything new. It's a

3:35

more secure and branded experience. Plus

3:37

you get features like interactive care

3:39

cells and suggested replies. And here's

3:41

why this matters. U.S. carriers are

3:43

starting to adopt RCS. Cinch is

3:45

already helping major brands send RCS

3:47

messages around the world. And they're

3:50

helping Lenny's podcast listeners get registered

3:52

first before the rush hits the

3:54

U.S. market. Learn more, it

3:56

gets started at cinch.com/Lenny.

3:58

That's S-I-N-C-H-D-D-I-L- This episode is

4:01

brought to you by persona, the adaptable

4:03

identity platform that helps businesses

4:05

fight fraud, meet compliance requirements,

4:07

and build trust. While you're

4:09

listening to this right now,

4:11

how do you know that

4:14

you're really listening to me, Lenny?

4:16

These days, it's easier than ever for

4:18

fraudsters to steal PII, faces,

4:20

and identities. That's where persona

4:23

comes in. Persona helps leading

4:25

companies like LinkedIn. Etsy and

4:27

Twilio, securely verify individuals and

4:29

businesses across the world. What

4:31

sets persona part is its

4:33

configurability. Every company has different

4:35

needs, depending on its industry,

4:37

use cases, risk tolerance, and

4:39

user demographics. That's why persona

4:42

offers flexible building blocks that

4:44

allow you to build tailored

4:46

collection and verification flows that

4:48

maximize conversion while minimizing

4:50

risk. Plus, persona's orchestration

4:53

tools automate your identity

4:55

process so that you

4:57

can fight rapidly shifting

4:59

fraud and meet new

5:02

waves of regulation. Whether

5:04

you're a startup or

5:06

an enterprise business, persona

5:08

has a plan for

5:11

you. Learn more at with

5:13

persona.com/Lenny. Again,

5:16

that's with p-e-r-s-o-n-a.com/Lenny. It's a pleasure

5:18

to talk to you, Lanning. Great

5:20

to be here. I don't know

5:22

how you have time to do

5:24

this podcast. Your life must be

5:26

insane these days with the pace

5:29

at which you guys are scaling,

5:31

just how much is changing in AI

5:33

every day. So I just extra appreciate

5:36

you making time for this. I think

5:38

you said it's 1030 your time is

5:40

when we're doing this. I'm a bit

5:42

tired, yes. Mostly from the crazy pace

5:44

of everything, but yes. We're going to

5:47

be invigorating. I'm sure I'm sure I'm

5:49

sure. Okay, so for folks that are

5:51

maybe a little bit familiar with

5:53

lovable or not at all familiar

5:55

What's just what is lovable? What's

5:57

the simplest way to understand it?

5:59

I'd say lovable is your personal

6:02

AI software engineer. You describe

6:04

an idea and then you

6:06

get a fully working product

6:08

from the AI. And what this

6:10

means is that entrepreneurs actually, today

6:13

they turn their ideas into

6:15

real businesses. We have a lot

6:18

of the signers and product managers

6:20

that create the first version

6:22

of their product ideas to

6:24

show to the teams. And

6:26

some of them become founders

6:28

because of like their empowerment

6:30

from this. But also

6:33

developers themselves, they actually

6:35

writing code or creating

6:37

products much faster. And

6:39

I mean, the reason, it's pretty

6:42

obvious for me, so I'll spell

6:44

it out. The reason why we're

6:46

doing lovable is that I don't

6:48

know about your mom, but like

6:50

my mom doesn't write code

6:52

and not my friends. Almost

6:55

all my friends throughout my life

6:57

reached out for health, like Anton,

6:59

I need to build something, how do

7:01

I find a great software

7:03

engineer? And we're building for this

7:05

99% of the population

7:08

who don't write codes. Currently,

7:10

if you're technically inclined, you

7:12

get much further, but over

7:15

time, naturally, the way to build software

7:17

is by just talking to an

7:19

AI. That's how is it. I love

7:21

the way that you guys describe it

7:23

and you didn't mention it, but I

7:26

think it's like building the last piece

7:28

of software ever. How do you how

7:30

do you phrase that? Yeah, we say

7:32

we're building the last piece of software.

7:34

The last piece of software. Okay,

7:36

we're going to do a live demo, but

7:39

first of all, can you just share some

7:41

stats on the scale of this

7:43

business at this point because it's

7:45

quite absurd? Yeah, so we launched Lovable

7:48

less than three months active

7:50

users and 30 of those 30,000

7:52

of those are actually paying

7:54

and in this growing on this

7:56

at the same rates like you

7:59

just almost only through organic word

8:01

or mouth. Okay, and I'll share a couple

8:03

stats in terms of revenue, just so folks

8:05

know this, and we'll have this in the

8:07

intro too. I think you guys hit 4

8:09

million ARR in the first four weeks. You

8:12

hit 10 million ARR in the first

8:14

two months with just 15 people. You're

8:16

the fastest growing startup in all of

8:18

Europe. And you guys had to rewrite

8:20

your entire code base recently, and you

8:22

couldn't ship any new features for a

8:25

while. Is there right? That's right. Yeah,

8:27

people were saying like, oh, you're shipping

8:29

so fast. And we were all quite

8:31

frustrated because we wrote our service in

8:33

this kind of scripting language. And then

8:35

as we started scaling, we were just

8:37

now we have to throw everything away and

8:40

rewrite it in a more performance way. Okay.

8:42

Before we get to the demo last question,

8:44

you shared there's some companies that

8:46

have started based on Loveable. I

8:48

didn't even know that. So what are

8:51

some examples of companies slash businesses?

8:53

that have launched off of Lovable

8:55

and now are actually companies. I

8:57

mentioned the signers using Lovable and

8:59

one of our early users, Harry,

9:02

he started shipping real web apps to

9:04

his clients instead of just shipping

9:06

the science. And then he went

9:08

on to say, okay, wait, I'm going

9:11

to start an AI startup. And his

9:13

company, he like launched on product hunt

9:15

and everything and making money is

9:17

just like, let's anyone upload their

9:20

photo libraries. And then it's cat

9:22

like, the AI is in process and

9:24

categorizes it. And if you go to launched.lovable. app,

9:26

like this is an app built with

9:28

lovable, which is again of a product,

9:31

product, hand version where you can see

9:33

a lot of businesses or small sales

9:35

in future there. Okay, cool. So we're

9:37

going to come back to some of

9:39

this stuff. But let's get into a,

9:41

let's get into demo. I rarely do

9:43

demos on this podcast, but I'm finding

9:45

that I think it's really important for

9:48

people to see. these products in

9:50

action because in a large part this is

9:52

the future of product building and a lot

9:54

of people hear about, oh yeah, ice coming. And

9:56

I don't think a lot of people actually

9:58

see what the latest tools are. capable

10:00

of. And so I love showing

10:03

these sorts of things on this

10:05

podcast. So Sleni, I was

10:08

thinking, did you ever consider

10:10

making a copy and with

10:12

your own Arbienes? I haven't.

10:14

But go on. How about you

10:17

do that? Let's do it. Okay,

10:19

so we're going to make our

10:21

own Airbnb. Okay. So I just

10:23

put in the first prompt for

10:26

an Arbiend clone. Okay, and what

10:28

was the prompt just for folks

10:30

that aren't watching? Two words, Airbnb

10:33

Clone. That's the problem. I like

10:35

to start simple. And then what

10:37

you get is that the AI

10:40

says, okay, I'm going to go

10:42

through what doesn't beautiful Airbnb Clone

10:44

looks like, and it goes through

10:47

a bit of like design decisions,

10:49

and then I'll zoom out to

10:51

see more of it. We have

10:53

this just UI, that is. I mean it

10:55

has all the nice things you

10:57

would expect from our Airbnb clone

11:00

where you see different categories and

11:02

you can see two listings from

11:04

Airbnb with login buttons and everything.

11:06

So far it doesn't have the

11:09

functionality of Airbnb, it just has

11:11

the UI. I would now ask

11:13

for an improvement on some of

11:16

the functionality, like if I'm switching

11:18

category I want to see different

11:20

listings, let's say. But if you have

11:22

any thoughts on what we should build

11:25

next, let me know. Okay, and

11:27

so you had this preloaded, so you

11:29

didn't see how long it would take,

11:31

but how long would this normally

11:33

take for it to just write all

11:35

this code and have it for you?

11:37

The first prom takes 30 seconds.

11:40

Okay. And it's like a very

11:42

good copy of Airbnb. Yeah. I

11:44

love that you don't have to show

11:46

to design, you just tell it, Airbnb.

11:48

And you just tell it, Airbnb.

11:50

explore buying the place that I look

11:52

at just like is this for sale? So

11:54

what if we see what that would feel like

11:57

if you're just like a way to buy, buy

11:59

a list? Okay, good. So let's, let's,

12:01

how about, we add,

12:03

I mean, prompting is

12:05

important here, so let's

12:08

be specific, but

12:10

we would ask, creating,

12:12

add a button on

12:14

the listing, which has

12:16

purchased this, this

12:18

Airbnb home. Is that

12:20

it? Perfect. Is it add

12:22

a button. And I'll, even

12:24

more specific. It will

12:27

pop up a model. to

12:29

purchase the listing. Perfect. And I

12:31

love, so I think something as you're typing,

12:33

I'm just going to share thoughts as

12:35

you're doing this. So the site that you

12:37

ask this AI engineer to build, like

12:40

it's actually a functioning website that you

12:42

can browse around, it's not just a

12:44

design. The say, obviously there's

12:46

no like actual listings here,

12:48

like there's an actual houses

12:50

here. Say you were trying to like

12:52

actually build Airbeam Bee and

12:54

you wanted to start adding like

12:56

actual homes that plug into this.

12:58

How does that sort of step

13:01

work? So as you say, this is

13:03

just kind of the mock-up UI,

13:06

but it's also interactive.

13:08

If I want to log

13:10

in and add listing management,

13:12

then we will connect something

13:15

called the backend. So where

13:17

data is stored, where users

13:19

log information is stored, and

13:22

I can show you how

13:24

to do that. First, let's

13:26

just try out where we

13:28

got with this short prompt

13:30

on adding the purchase listing.

13:33

And it didn't do exactly

13:35

what I wanted. I said

13:37

add a button, or I didn't

13:39

say what a button should say,

13:41

here, but it says book now.

13:43

And if I click book now, I

13:46

get a booking confirmation.

13:48

So the AI was like, okay. It

13:50

didn't really, it was probably... surprised by

13:52

you wanting to buy the listing since

13:55

it's Airbnb, right? So it still says

13:57

book the listing, but it shows a...

13:59

a pretty model where I can click confirm

14:02

and pay. And then it says, yeah, booking

14:04

confirmed. I'll just say real quick. I love

14:06

that this is actually a really good example of

14:08

why being a good product manager is

14:10

important. A lot of wasted time happens

14:12

when you're not clear about the problem you're

14:14

trying to solve and why you're trying to

14:16

solve it and all that kind of stuff.

14:19

So it's really cool that this is a

14:21

use case where you have to be really

14:23

good at explaining what it is you want.

14:25

And it's interesting, you don't have

14:27

to tell this AI-Y, you know,

14:29

humans want to understand, why is this

14:31

important? Mostly, you need to be

14:33

very clear about what it is you're

14:36

doing. And I love that's a really

14:38

strong PM skill. Yeah, the PM's really

14:40

good at that. So we have to.

14:42

Hey, explaining exactly what you expect and

14:45

what you're not getting, is even more

14:47

important with AI than with humans. But

14:49

so I'm going to hooking up. more

14:51

of the factual functionality.

14:53

But first, I'll actually

14:56

show you something, like what's

14:58

the fastest way to change what

15:00

went wrong. It's created buttons

15:02

that say book now, and I

15:04

want them to say buy now. And

15:07

what I could do is to select

15:09

this item and say change it to

15:11

buy now. But what we just released

15:13

is that you can actually

15:15

edit this. Like this is

15:18

a fully functioning product. But

15:20

you can edit it visually, like you're

15:22

doing it square space and weeks and

15:24

so on. So I'll just change the

15:27

text to buy now, and then it

15:29

instantly changes. It actually changes it deep

15:31

down in the code base, but it's

15:33

very fast to do that. So I

15:35

think people listening to this and seeing

15:38

this, if you're not aware, like this,

15:40

like this is the cutting edge of

15:42

tools like this. No other tool. out

15:44

there lets you generate code from an

15:46

AI engineer and then actually just like

15:48

change a small element of it of every

15:51

other tool that I'm aware of. You have to

15:53

like ask the agent, do this for me and

15:55

then you hope that it does the right thing.

15:57

So this is a huge deal which you just

15:59

showed right up. Now it says by now.

16:01

Okay, that's amazing. And that's something

16:03

you just launched. Correct, you just

16:05

launched this a few days ago.

16:07

But I wanted to go into

16:10

fulfilling the full functionality, but what

16:12

it looks like is that you connect

16:14

an open source backend as a

16:17

service, and that's called Superbase. And

16:19

I have this instance to connect

16:21

to that's completely empty, just like

16:23

one click to set that up. And

16:26

now it's connected to the

16:28

backend. It's just like it.

16:31

automatically, generating and explaining, generating

16:33

some code and explaining what

16:35

I can do next. And what I

16:37

would do now is say, let's let's

16:39

add login, let's say, let's add login.

16:41

And where is it actually hosted

16:44

on the back end? Everything, general.

16:46

Yeah, so everything can be one click

16:49

deployed and then it's running, it's

16:51

hosted by a cloud vendor, which

16:53

is hosting, I think a huge

16:55

chunk of the internet. It's called

16:57

a cloud flirt. And the backend

16:59

is also a good cloud writer,

17:01

which is called Superbased. Amazing. Okay.

17:03

Let's wrap up the demo. That

17:06

was... Unless there's anything else.

17:08

Was there anything else? Was there

17:10

anything else really important that you

17:12

wanted to show? Don't. I mean,

17:14

I'll just explain what I would

17:17

do next. I would say, okay,

17:19

let's add login. Let's make the

17:21

listings editable by the users. So

17:23

users can upload listings. Then this

17:25

is going to take a bit

17:27

more time, but with patience and

17:29

good prompting skills, you're going to

17:31

get to a full working Airbnb.

17:33

That was a really good piece

17:35

to add. So basically, this is

17:37

getting to a place where it

17:39

actually is not so different from

17:41

actual Airbnb. People can log in, they

17:43

can add their home. You can add internal

17:46

tools to add listings for your

17:48

say sales team, ops team. Basically,

17:50

it just will allow you to build

17:52

a marketplace. That looks a lot

17:55

like Airbnb. Amazing. Okay, thank you

17:57

for the demo. I think for a

17:59

lot of people. They're like, yeah, I've

18:01

seen this kind of stuff. For

18:03

most people, like, holy shit, it's

18:05

unreal. Like, it's almost like we're

18:07

taking for granted now. You can

18:09

ask an app to build you

18:11

a whole website. And that costs

18:13

probably like a few pennies. It

18:16

took like five minutes versus like,

18:18

it would have been tens of

18:20

thousands and like weeks and weeks

18:22

and months even built just a

18:24

prototype. And when these tools, as

18:26

we see here, they're already good

18:28

as well. But mainly, I would

18:30

say, they're getting better very, very

18:32

fast. And I'd say, like, one

18:34

of the bigger bottlenecks is now,

18:36

they're not integrated into the current

18:38

way that you have your existing

18:40

products and so on. But since

18:42

it's getting better so fast, I

18:44

think the best thing for people

18:46

who are interested in this or

18:48

like interested in just being part

18:50

of the future economies. Get your

18:52

hands very dirty with these tools

18:55

because being in the top 10%

18:57

in using them is going to

18:59

be to absolutely set you apart

19:01

in the coming months and years

19:03

So let me follow that threat

19:05

to say You are magically able

19:07

to sit next to Everybody that

19:09

is using lovable for the first

19:11

time and you could just whisper

19:13

a tip in their ear to

19:15

be successful with lovable. What would

19:17

that tip be? It takes a

19:19

lot to master using tools like

19:21

lovable and being very curious and

19:23

patient. And we have something called

19:25

chat mode where you can just

19:27

ask and like to understand, like

19:29

how does this work? I'm not

19:32

getting what I want here. Am

19:34

I missing something? What should I

19:36

do? Is the best way to

19:38

be productive? Is also one of

19:40

the best ways to just learn

19:42

about how software engineering works, which

19:44

is... And you don't have to

19:46

write the code anymore, but it

19:48

is useful to understand how software

19:50

and how building products works. So

19:52

I think that's the patience and

19:54

curiosity is super useful. That's it.

19:56

second part that we spoke about

19:58

is that being, if I would

20:00

sit next to you, I would

20:02

probably say like, hey, you're not

20:04

being super clear here. Like for

20:06

example, don't say it doesn't work.

20:08

Just explain exactly what you're expecting

20:11

and which parts are working and

20:13

which parts are not working. And

20:15

that's a lot of, that's something

20:17

that a lot of people don't

20:19

do naturally. I love that like

20:21

when you have an engineer you're

20:23

working with, that is a very

20:25

expensive. mistake to miscommunicate something to

20:27

just forget about a future to

20:29

forget about a requirement and here

20:31

it's you do that and then

20:33

like 30 seconds later you're like

20:35

oh okay sorry that was wrong

20:37

and then you could just try

20:39

again that's true it might it

20:41

might be more costly with humans

20:43

okay and the first step so

20:45

the first tip is chat most

20:48

you could just so your advice

20:50

is chat with the what do

20:52

you call it you call an

20:54

agent you got what's like the

20:56

the term for the thing that

20:58

you were talking with you were

21:00

talking with is a name. Just

21:02

lovable. Okay, so you're talking about

21:04

lovable. By the way, where did

21:06

you, how did you decide on

21:08

lovable is the name? It's so

21:10

sweet. I think it's all about

21:12

building, I mean, a great product.

21:14

That's what I want more people

21:16

to be able to do. And

21:18

the best word for a great

21:20

product is that it's lovable. A

21:22

lot of jargon that I like

21:25

to use to like emphasize what

21:27

we should be striving for is

21:29

building a minimum lovable product. and

21:31

then building a lovable product and

21:33

they building an absolutely lovable product.

21:35

So I took that jargon with

21:37

me in the company name. That

21:39

is great, absolute level product, ALP,

21:41

is the new MVP. Okay, so

21:43

we talked about this, the scale

21:45

you guys have hit at this

21:47

point. I imagine it's far beyond

21:49

10 million AR. Do you share

21:51

that at this point or are

21:53

you keeping that private? We don't

21:55

think on the numbers, but I

21:57

mean, I could probably do a

21:59

two X tweet about this quite

22:01

soon, yes. Okay, so it's far

22:04

beyond 10 million error. at this

22:06

point. It's one of the fastest

22:08

growing startups in history, the fastest

22:10

growing startups in Europe. I want

22:12

to zoom is back to the

22:14

beginning. What is the origin story?

22:16

Oblubable. How did it all begin?

22:18

What was the journey to today?

22:20

I think I was not impressed

22:22

by what people were doing with

22:24

the large language models when after

22:26

it. I was using them way

22:28

back, but when Chachjeviti came out.

22:30

they were starting to get really

22:32

good at taking a human instruction

22:34

and spitting out code. And then

22:36

people in my team, I was

22:38

the city over a why-c startup,

22:41

they felt like, oh, anton, you're

22:43

exaggerating, this is not going to

22:45

change anything in the coming years.

22:47

So I wanted to prove a

22:49

point, and I created an open

22:51

source tool called Jupiter Engineer, where

22:53

you could write something like, create

22:55

a snake game. And then it

22:57

speeds out a lot of code,

22:59

a lot of different files, and

23:01

then opens the snake game. And

23:03

then I tweeted a video about

23:05

that. And GPT engineer is to

23:07

date the most popular open source

23:09

tool to showcase the ability for

23:11

large language walls to create applications.

23:13

And it's like 50, something thousand,

23:15

these top stores. and like Dawson

23:18

of academic references. And I know

23:20

that I'll just add that it

23:22

like GitHub shut you down because

23:24

I thought it was some kind

23:26

of attack the like how many

23:28

stars you're getting how many people

23:30

were using it. Right. Yeah so

23:32

that was that that came later

23:34

that that's a lovable. Loveable earlier

23:36

was always creating new projects on

23:38

GitHub when someone used lovable and

23:40

it was that we asked them

23:42

is it fine? Like how was

23:44

the limits here? There's other no

23:46

limits. But once we started creating

23:48

15,000 big projects per day, so

23:50

there were a lot of usage,

23:52

then some engineer, when it was

23:54

on call, maybe they woke up

23:57

in the... night and they saw

23:59

their servers were taking too much

24:01

load because of us. So then

24:03

they shut off down completely and

24:05

we got this email that said,

24:07

oh you broke some kind of

24:09

rules and we didn't know what

24:11

was going on. That's similar to

24:13

a story I heard when ChatGPT

24:15

was originally being trained. Microsoft servers

24:17

were blocked it because they thought

24:19

it was some crawler and it

24:21

was just actually like the very

24:23

first version ChatGBT being trained on

24:25

data. Anyway, keep going. So I

24:27

built this tool called the Jupiter

24:29

Engineer and I was thinking about,

24:31

we're seeing the biggest change humanity

24:34

we'll ever see, I think, where

24:36

like before you had manual labor

24:38

being taken over by machines, but

24:40

now it's actually cognitive labor being

24:42

done better than humans by machines.

24:44

And what's the best way to

24:46

have some kind of positive impact

24:48

here? It's not to make engineers

24:50

more productive, which there's a lot

24:52

of companies using it to make

24:54

engineers more productive. Microsoft to build

24:56

co-pilot and so on. But it

24:58

is to enable those who have

25:00

such a hard time finding people

25:02

who are good at creating software,

25:04

that's been their absolute bottleneck, and

25:06

let them take their ideas and

25:08

their dreams to reality. enabling more

25:10

entrepreneurship in the innovation by building

25:13

the AI software engineer for anyone.

25:15

And then I grabbed a previous

25:17

colleague of mine. I was also

25:19

been a founder, Fabian, and I

25:21

said, we should build something like

25:23

GPK engineer, but it has to

25:25

be for the people who don't

25:27

write code. And that's the source.

25:29

Okay, and then that became lovable.

25:31

There's like the shift from open

25:33

source into a product that anyone

25:35

can use, but also pay for.

25:37

Makes makes sense. Okay, so from

25:39

that point, I saw stat they

25:41

started making a million dollars in

25:43

ARR per week. And once you

25:45

launch and lovable, is that true?

25:47

Yeah, so we launched, so we

25:50

actually called the first version of

25:52

the product, like GPT Engineer App,

25:54

and that was, it was very

25:56

different in some ways, and we

25:58

launched that under a wait list,

26:00

and so like, oh, we have

26:02

this wait list, and we got

26:04

a lot of feedback and iterated.

26:06

And finally, when we thought the

26:08

product was really good, we said,

26:10

okay, now we have a lovable

26:12

product. And it was mainly on

26:14

the AI that we did a

26:16

lot of improvements. Once we launched

26:18

that, that was 21st of November.

26:20

So that's almost three months ago.

26:22

We just hit like 1 million

26:24

error in a week and then

26:27

it kept growing at that pace.

26:29

It's still growing even faster than

26:31

that pace. Faster than that pace.

26:33

Faster than 1 million error per

26:35

week. Holy shit. Okay. That sounds

26:37

like product market fit to me.

26:39

You said that you did a

26:41

lot of work on the back

26:43

end. I said you tweet about

26:45

this that. you guys figured out

26:47

some kind of unlock on scalability,

26:49

like a new scaling law that

26:51

allowed you to build something like

26:53

this. What can you talk about

26:55

there that kind of on the

26:57

technical element, allowed you to build

26:59

something new and the successful? There

27:01

are many scaling laws, I would

27:03

say, when you build AI systems.

27:06

And this one in particular is

27:08

about when you put in more

27:10

work, the product reliably gets better

27:12

and better. And what you've seen

27:14

generally... when you have AI building

27:16

something is that it can get

27:18

stuck in some place. It starts,

27:20

it's super good in the beginning

27:22

and then it gets stuck. What

27:24

we did was to painstakingly identify

27:26

places where it goes stuck and

27:28

there's different approaches but address like

27:30

different ways how we do it,

27:32

but address the places where it

27:34

gets like tuned and power system

27:36

quantitatively and having a very fast

27:38

feedback loop to improve it in

27:40

the areas where it got stuck,

27:43

the most important areas. It still

27:45

does get stuck sometimes, but that's

27:47

the scaling law and... We're still

27:49

early in that scaling now I

27:51

would say. And so when you

27:53

talk about things getting stuck it's

27:55

like the AI agent just saying

27:57

like I don't know what to

27:59

do from this point and or

28:01

like they introduce some kind of

28:03

bug is that is an example

28:05

of getting stuck? It introduces some

28:07

kind of bug and then it's

28:09

not smart enough to figure out

28:11

how to get out of that

28:13

bug. I see and this is

28:15

a common problem people have with

28:17

tools like this as they like

28:20

get to a certain point and

28:22

then it's like well I don't

28:24

know what to do I'm not

28:26

an engineer. I'm not an engineer.

28:28

I'm not an engineer. Like here's

28:30

a bug it's running into where

28:32

the infrastructure is built the wrong

28:34

way. And so it sounds like

28:36

one of the paths to solving

28:38

that is what you're describing is

28:40

you make the AI smarter to

28:42

get to avoid more and more

28:44

of these places they get stuck.

28:46

Another is people just learning how

28:48

to get AI unstuck. This is

28:50

something when we had Amjod on

28:52

the podcast from replete he said

28:54

that this is like the main

28:56

skill that he thinks people need

28:59

to learn is how to unstuck

29:01

AI when it runs into a

29:03

run into a problem. Just starts

29:05

there, I don't know, anything along

29:07

those lines come up as I

29:09

say that. This is something that

29:11

is a problem today. And the

29:13

frontier of where this is a

29:15

problem is very rapidly receding back.

29:17

So what we did was we

29:19

identified the most important areas, like

29:21

so specifically adding login, creating data

29:23

persistence, adding... payment with strike. Like

29:25

those are the things that we

29:27

make sure it doesn't get stuck

29:29

on, for example, and the places

29:31

where it gets stuck today is

29:33

currently something that where you can

29:36

use being very good at understanding

29:38

and getting unstack, but in the

29:40

future it won't be so important.

29:42

This is just going to not

29:44

get stuck. And I know you're

29:46

not talking super in super in

29:48

depth about this because this is

29:50

one of your unfair advantages, this

29:52

kind of stuff you figured out,

29:54

so I'm not going to push

29:56

too far. I don't know, I

29:58

know you want not everyone to

30:00

do exactly the same stuff. So

30:02

I want to zoom back to.

30:04

the pace of growth that you

30:06

guys have seen. One of the

30:08

big stories, everyone's always looking at

30:10

you guys, have like 15 people,

30:12

10 million ARR in two months,

30:15

that's absurd. It's something, I don't

30:17

know if it's ever been done

30:19

in history, if so it's maybe

30:21

a couple other AI startups recently.

30:23

How have you been able to

30:25

do this? What have you done

30:27

that has allowed you to grow

30:29

this fast with so few people?

30:31

I'd like to take credit of

30:33

having done everything end to end

30:35

in the product, but we were

30:37

building on top of the oil

30:39

here, which we have discovered oil,

30:41

which are the foundation models, right?

30:43

And then what we've done is

30:45

that we've obsessed about what's the

30:47

right way to present this to

30:49

a user, what's the interface for

30:52

the human, to get as much

30:54

out of this as possible, packaging.

30:56

together. I showed you in the

30:58

demo that you how you can

31:00

add authentication and making this work

31:02

seamlessly together as a whole. That's

31:04

what we've done. And then people

31:06

love the product. That's what that's

31:08

the driver of the growth. For

31:10

getting awareness, we mainly been posting

31:12

what we've shipped on social media.

31:14

That's that's how people know about

31:16

us. So building in public is

31:18

how people usually describe that. So

31:20

it's like... I think it's like,

31:22

you guys have the advantage of

31:24

the demos or just like, holy

31:26

shit, you can do that. And

31:29

then you guys share the numbers

31:31

that you guys are growing at,

31:33

so it's neatly interesting and shareable.

31:35

But I imagine most people have

31:37

something interesting to share. I guess

31:39

is there anything that you think

31:41

you did that other companies maybe

31:43

haven't done that make the product

31:45

so lovable? I just give a

31:47

give a big shout out to

31:49

the notice within the code I

31:51

had written the code recently I

31:53

would say and the I mean

31:55

You want people who can ship

31:57

really fast and have good taste

31:59

for like what is simple, what's

32:01

the right abstractions, and I think

32:03

that's what we've done differently and

32:05

have this obsession for us making

32:08

it better and better and better.

32:10

This episode is brought to you

32:12

by the Fund Rise Flagship Fund.

32:14

Full disclosure, real estate investing is

32:16

boring. Prediction markets are exciting, meme

32:18

coins are a thrill ride, even

32:20

the stock market can swing wildly

32:22

on a headline. Hello, Deep Seek,

32:24

but with real estate investing, there's

32:26

no drama or adrenaline or excuses

32:28

to refresh your portfolio every few

32:30

minutes. Just bland and boring stuff

32:32

like diversification and dividends. So you

32:34

won't be surprised to learn that

32:36

the Fun rice flagship real estate

32:38

fund is a complete snooze fest.

32:40

The fund holds $1.1 billion worth

32:42

of institutional caliber real estate managed

32:45

by a team of pros focused

32:47

on steadily growing your net worth

32:49

for decades to come. See, boring.

32:51

That's the point. You can start

32:53

investing in minutes and with as

32:55

little as $10 by visiting fundrise.com/money.

32:57

Carefully consider the investment objectives, risks,

32:59

charges, and expenses of the fun

33:01

rice flagship fund before investing. Find

33:03

this information and more in the

33:05

fund's prospectus at fundrise.com/flagship. This is

33:07

a paid ad. Okay, I want

33:09

to come back to the team

33:11

because I know you have a

33:13

lot of thoughts there. In terms

33:15

of writing code, how much you

33:17

guys actually use AI to write

33:19

the code that is building lovable?

33:22

How does that work on your

33:24

team? We have set up lovable

33:26

so that we can change lovable

33:28

with itself. We have done that.

33:30

There is a lot of hyper

33:32

specific things in terms of running

33:34

a separate, like we spin up

33:36

a dedicated computer for each user.

33:38

It doesn't do everything. Love will

33:40

doesn't do everything. So we use

33:42

the tools that are for developers.

33:44

not for the 99% most of

33:46

the time. And everyone uses AI

33:48

all the time in writing code.

33:50

It's also in great course for

33:52

experimentations. And other tools like cursor

33:54

and stuff like that, like in

33:56

the tools you can change. I

33:58

think cursor is the one that

34:01

almost everyone uses in the team.

34:03

Yeah, okay, cool. I did a

34:05

survey recently on tools that my

34:07

listeners and readers use in cursor.

34:09

17% of all people that read

34:11

my newsletter use cursor already, which

34:13

is absurd. And you guys are

34:15

in there too. Okay, so kind

34:17

of along these lines, there's obviously

34:19

other competitors and companies in this

34:21

space, so everyone's always wondering. You

34:23

bold, replet, cursor is a different

34:25

kind of thing. What's the simplest

34:27

way to understand maybe how lovable

34:29

might be different from, say, bolt

34:31

and replet, which I think are

34:33

probably the closest? The packaging for

34:35

non-technical people is what we aim

34:38

for. And... I showed you in

34:40

the demo that you can edit

34:42

the text, like you can use

34:44

to change the colors and so

34:46

on instantly without having to go

34:48

into like a code editor and

34:50

without having to wait about 30

34:52

seconds for the AI to the

34:54

full change. So that's the big

34:56

way that we think about packaging

34:58

it. And then for, you know,

35:00

making sure that this can be

35:02

used as productively as possible in

35:04

a larger team. Something that's different

35:06

from I think all the other

35:08

tools is that it's it is

35:10

synchronized with a GitHub and that

35:12

means that you can use cursor

35:14

if you're or the people in

35:17

your team that want to be

35:19

more low level they can use

35:21

cursor and while the people who

35:23

don't want to mess and set

35:25

up their local file system and

35:27

commit to GitHub and so on

35:29

they can use slappables. Not getting

35:31

stuck is I think the most

35:33

important thing for people and that's

35:35

why. We came, we kind of

35:37

entered the space late, we haven't

35:39

done the same type of marketing

35:41

as many others and we're still...

35:43

from the people that I talked

35:45

to ranked as the one that

35:47

works most reliably. I love it.

35:49

Okay, so this point about how

35:51

you can just use Lovable to

35:54

build a lot of it for

35:56

you and then get into cursor

35:58

to edit and tweak is a

36:00

really big point. And you're saying

36:02

other companies aren't as good at

36:04

that. I don't know any other

36:06

dust that. I don't let you

36:08

do that. Amazing. Okay. And then

36:10

what's kind of like the vision

36:12

for Lovable. Like. What's the end

36:14

state of this? Is this everybody

36:16

can build anything they want sort

36:18

of thing? What's the simplest way

36:20

to understand where you're going in

36:22

the next 10 years? I mean,

36:24

I have to say, so we're

36:26

building the last piece of software

36:28

and it is inherently very hard

36:31

to predict how the world looks

36:33

like in five years this day.

36:35

It's very hard. But the last

36:37

piece of software, how I see

36:39

that is that it's almost instant

36:41

to go from what you want

36:43

to change in a product or

36:45

what product you want you want

36:47

to build you want to build.

36:49

to having it fully working and

36:51

then integrated with any of your

36:53

existing systems or integrated with the

36:55

very powerful third-party providers. Already today

36:57

you can just ask ad and

36:59

a shot with open AI and

37:01

then you get a shot with

37:03

open AI in your product. But

37:05

that's like just working perfectly is

37:07

the something that's coming in the

37:10

coming two years I would say.

37:12

And then after that. There is

37:14

a lot of things in building

37:16

a product that is not just

37:18

the engineering side, right? And I

37:20

think an AI can be very

37:22

useful in aggregating and understanding your

37:24

users. So if you use the

37:26

analytics tools, you know that there

37:28

is something quite common, which is

37:30

to see how users have interacted

37:32

with the product. AIs can do

37:34

that on an absolutely massive scale.

37:36

and propose changes to a human

37:38

to see like, oh yeah, that

37:40

sounds like a good change to

37:42

make it a bit more. intuitive

37:44

and it can also automatically run

37:47

spin out AB tests so that

37:49

you can see with data or

37:51

these improvements to the product. So

37:53

I think that's on the horizon

37:55

as well quite since. Like what's

37:57

interesting about this in one way

37:59

is people wonder just what jobs

38:01

will be more important, what skills

38:03

will be less important. Let me

38:05

share a thought I have and

38:07

then I want to get your

38:09

take and see where you go

38:11

to this. It feels like what

38:13

is getting more valuable is being

38:15

good at figuring out what to

38:17

build, and then knowing if the

38:19

thing you had built is correct

38:21

and good and ready. So it's

38:24

like discovery, ideation, idea, part of

38:26

the step of launching a product,

38:28

and then it's like taste and

38:30

craft, just like, is this the

38:32

thing? Is this gonna solve people's

38:34

problems? Because the building now is

38:36

being done more and more, and

38:38

it's interesting, it used to be

38:40

the reverse engineering, was the hardest,

38:42

most valuable skill, and now it's

38:44

like. figure out what to build.

38:46

You could sit there and you

38:48

could just tell what to build.

38:50

And a lot of people get

38:52

to your screen, I'm sure, and

38:54

they're like, I don't know what

38:56

to build, I don't know what

38:58

people want. And it's like, that's

39:00

the thing now. So I just,

39:03

I've reactions to that and thoughts

39:05

on what skills will matter more

39:07

and less. I mean, if you're

39:09

a founder or you want to

39:11

build something, yeah, I totally agree

39:13

that figuring out what else, what

39:15

are pain, what are pain, what

39:17

are pain points, pain points, pain

39:19

points, pain points, pain points, and

39:21

seeing, pain points, and seeing, like.

39:23

There are often currently solutions to

39:25

every, some kind of solutions to

39:27

everything. What is the, and how

39:29

can you make this 10x better?

39:31

So somehow, like figuring that out

39:33

is super important. When you have

39:35

an existing product, then I think

39:37

taste, and I could find, tasting

39:40

what is good is even more

39:42

of the important part. The, like

39:44

the engineer skills set is still

39:46

going to be important because that.

39:48

helps you understand what are the

39:50

constraints or what you can build.

39:52

And I just think a lot

39:54

of software engineers are probably a

39:56

bit scared now. Like, okay, I

39:58

want... Am I out of a

40:00

job? What's going to happen? But

40:02

they should see themselves as people

40:04

who translate the problems that are

40:06

stated by a human probably to

40:08

technical solutions. But they do have

40:10

to abstract themselves up a few

40:12

steps, not just like looking at

40:14

the in their tech stack, like,

40:16

oh, I can just do the

40:19

front and changes. The engineers or

40:21

technical people are very good at

40:23

understanding what are the constraints technically,

40:25

and they should see themselves as

40:27

that. translators. Is there like a

40:29

like is it almost like you

40:31

want to be learn the engine

40:33

manager skill of overseeing engineers versus

40:35

like the actual engineering skill or

40:37

is you think it's still going

40:39

to be really important to learn

40:41

how to code and be really

40:43

good at that? I mean doing

40:45

a bit of everything being in

40:47

general is I think much more

40:49

important than it used to be

40:51

and if I'm putting together a

40:53

product team today I would re-obsess

40:56

about getting as much of as

40:58

many skill sets as possible for

41:00

each person I hire. They should

41:02

know how architecting a system works

41:04

preferably. They should know design. They

41:06

should have product taste. They should

41:08

know how to talk to users.

41:10

I think everyone should be able

41:12

to know a bit of what

41:14

of that preferably. Easier said than

41:16

done. It's hard to find people

41:18

that know all these things. So

41:20

let's segue to hiring and how

41:22

you hire. How many people do

41:24

you have at this point? Is

41:26

that some you sure? Yeah, now

41:28

we're at 18. 18. Okay, wow.

41:30

So I love that you're, it

41:33

sounded like you're about to say,

41:35

oh, we have 100 people now,

41:37

no, 18, okay, so you went

41:39

from 15 to 18. Okay, great.

41:41

So what do you look for

41:43

when you're hiring people? The way

41:45

I saw you describe it on

41:47

Twitter is you look for cracked,

41:49

engineers, the best crack team in

41:51

Europe, things like I guess, just

41:53

specifically, what are you looking for

41:55

when you're hiring? They're not just

41:57

like, oh, I'm here for a

41:59

job, I'm here for... being as

42:01

a passenger on this journey, but

42:03

everyone should really care about

42:06

the product, the users, and

42:08

care a ton about the team,

42:10

how the team works together,

42:12

and that you're always

42:14

contributing to making the

42:17

team work more productive

42:19

together. And that care

42:21

or preferably obsession gets

42:23

you a very long ways. And then...

42:26

You do often want to

42:28

have like absolute absolute superpower

42:30

in some dimension. To be

42:32

able to understand and do

42:34

as many things as possible,

42:36

like have this generalist brain

42:38

that quickly learns any skill,

42:40

but be super super good in

42:43

one dimension. And that's for

42:45

us, that's mostly cramming as

42:47

much out of AI, out

42:49

of the large language models,

42:51

understanding the... then entire parameter space of

42:53

what you can change to make our

42:55

product to perform better. So how do

42:57

you actually test for these things? You

43:00

know, like some of these things describe

43:02

it, I think everyone's looking for it,

43:04

like they care about the user, they want

43:06

to collaborate well. Just like when

43:08

you're, because like you're 18 people

43:10

building in a company that's growing

43:13

more than a million AR every week,

43:15

like that's an absurd scale, and the

43:17

people you've found are clearly world class.

43:19

And I think a lot of people are

43:21

going to like want to hire the type

43:23

of people you're hiring. So

43:25

when you're actually interviewing, how

43:27

do you suss out some of these

43:30

things like their AI cramming skills, their

43:32

team building collaboration? What do you

43:34

actually do? I ask people what

43:36

they've done before and these people

43:38

that I'm describing, they have often

43:41

done something where they care a lot

43:43

about what they've done before and dig

43:45

into this about the technical things

43:47

that they did. And then, I mean, we

43:49

do the normal thing of giving, showing

43:52

a very hard problem, that is a

43:54

bit unorthodox, that someone hasn't seen before,

43:56

preferably, and see how they think through

43:59

the thinking reason. through that. Then

44:01

something that I think is more

44:03

uncommon is that we do, I

44:05

pretty much always have people join

44:07

the work simulation for at least

44:10

a day, often or four weeks.

44:12

Awesome. Okay, so work trial, that's

44:14

awesome. So basically they work with

44:16

the team for at least a

44:18

day. You said, like sometimes a

44:21

week. Yeah. And I love this

44:23

point you made about they show.

44:25

They cared deeply about something they

44:27

previously worked on and you look

44:30

for, just like, obsession with the

44:32

thing that they built last or

44:34

something they worked on. Like what

44:36

percentage are engineers of these 18?

44:38

So 12 at least, write code

44:41

in at least part-time? 12 at

44:43

18? Okay, cool. When we were

44:45

setting up, you're like, oh, our

44:47

engineers creating content now? I think

44:50

that's a cool example of how

44:52

people do a lot of different

44:54

things. Yep. Also, okay, so I

44:56

have your job posting that you

44:59

shared once of like, their actual

45:01

job description. I'm going to read

45:03

a few lines from it. It's

45:05

very inspired by Shackleton, right? Would

45:07

you agree? Cool, I love it.

45:10

By the way, did you write

45:12

this or did you have AI

45:14

write this job description where you

45:16

like, create an engineering job description

45:19

track? Let me read it. Long

45:21

hours, high pace, candidates must thrive

45:23

under a high urgency, under AGI

45:25

timelines approaching. Difficult mission ahead, honor

45:27

and recognition in case of success,

45:30

those seeking comfortable work need not

45:32

apply. And then there's a few

45:34

other things. Collaboration or other exceptional

45:36

minds, purpose larger than any normal

45:39

engineering role. Generous share in the

45:41

venture success. Amazing. Thank you. Thoughts.

45:43

Yeah, so I did I did

45:45

get some up with the the

45:48

formatting of this but then I

45:50

was mostly me doing the exact

45:52

pricing of the different sentences so

45:54

good and I love that you

45:56

know so some people is giving

45:59

like whole shit I'm not signing

46:01

up for this but to a

46:03

lot of people the people you

46:05

want is like yes this is

46:08

exactly what I want to be

46:10

doing great amazing yeah okay cool

46:12

so so it feels like one

46:14

of the elements of hiring here

46:16

is create a really good filter

46:19

to be clear about just how

46:21

intense this is so that the

46:23

people that want that are the

46:25

ones drawn to you okay and

46:28

then you're also you're in Sweden

46:30

fastest growing startup in Europe ever

46:32

Thoughts on building in Europe slash

46:34

Sweden versus the US slash San

46:37

Francisco? Yeah, so this this ambition

46:39

level that you're talking about in

46:41

the job ad is more uncommon

46:43

in Sweden. And I think that

46:45

is the like the biggest unlock

46:48

that someone like me who says

46:50

that this is the like the

46:52

time in human history when you

46:54

have the most impact per worked

46:57

hour. And that's why we have

46:59

to be super ambitious, like just

47:01

up the ambition level. And then

47:03

we can maybe retire and have

47:05

AI take care of most things

47:08

in society. And inspiring people to

47:10

be this ambitious in a place

47:12

where the average ambition is lower,

47:14

but the talent, the role talent

47:17

is much more available, is a

47:19

great recipe. I think that's a

47:21

great recipe. So that's what's. I

47:23

think it's some kind of advantage

47:26

there. It's a bit of a

47:28

double-edged sword, but it's some kind

47:30

of advantage. So I'm hearing is,

47:32

like there's incredible people in Europe,

47:34

they're just not, they're harder to

47:37

find in what I'm hearing is

47:39

like, the key is, how do

47:41

you suss them out and get

47:43

them to want to talk to

47:46

you? Yeah. Most... people in Europe,

47:48

they haven't thought that, oh, do

47:50

it going on an extremely ambitious

47:52

mission is what I want to

47:54

do. So that figuring out who

47:57

those are is... is a big

47:59

part of it. Awesome. Okay. I

48:01

want to talk about prioritization. I

48:03

imagine all these things that I

48:06

just shared about just like how

48:08

ambitious this mission is, how much

48:10

you're doing, the last piece of

48:12

software. You must have a bazillion

48:15

things that people ask you to

48:17

build that you want to build.

48:19

What's your approach to deciding what

48:21

to prioritize and actually build? Just

48:23

top line, I think, identifying what

48:26

is the... biggest problem, I got

48:28

the biggest problem and iterating fast

48:30

on saying, okay, this is the

48:32

biggest problem, let's really resolve that

48:35

problem and then pick me the

48:37

next one. And not overthinking, not

48:39

like dreaming out a long road

48:41

map. That's my fault. There's a

48:44

very very simple algorithm. Understanding what

48:46

is the biggest problem is not

48:48

always a simple problem. I think,

48:50

yeah, so we spend time as

48:52

one should on... talking to users,

48:55

the list, reading up what people

48:57

are writing, we have the feature

48:59

board for, or people do a

49:01

lot of requests as you say,

49:04

and then when we pick one

49:06

of the problems, we're quite engineering

49:08

led. Like for a product like

49:10

ours, it's hard to be, like

49:12

have product managers that are not

49:15

engineered near, say, oh, this is

49:17

what we should do now because

49:19

the right solution. to the problem

49:21

might be entangled in things that

49:24

are technical details. They might be

49:26

entangled in technical details. So like,

49:28

okay, yes, this is the biggest

49:30

problem, but we should have this

49:33

larger technical initiative that's going to

49:35

solve all of these problems. So

49:37

it's quite engineering lab compared to

49:39

many other product companies. As it

49:41

should, I'd be worried if you

49:44

guys had a product manager at

49:46

this point and make so that

49:48

would not, that wouldn't make no

49:50

sense right now. I imagine the

49:53

answer is it's chaos and there's

49:55

no actual defying process. But just

49:57

like, what does it look like

49:59

generally, like what's kind of the

50:01

cadence you guys operate on? How

50:04

do you take an idea to

50:06

like build it, spec it, launch

50:08

it? Just like, what does that

50:10

look like if you have something?

50:13

If you look back like three

50:15

months, we mainly said, okay, let's

50:17

do this weekly planning. And we

50:19

have like a big jam board

50:22

where we have all the main

50:24

problems and then we have kind

50:26

of ranked them, which has to

50:28

be focused, when we're focused on

50:30

next or this week or this

50:33

week. And then we have a

50:35

demo on where we say like,

50:37

okay, are these the things we

50:39

ship this week? So to get

50:42

everyone on the same page. And

50:44

we do have a bit more

50:46

of a roadmap now. And where

50:48

we say, like here are we

50:50

going to make sure you can

50:53

support custom domains next. They're going

50:55

to add collaboration after that. And

50:57

like the biggest problem now or

50:59

the biggest initiative now that solves.

51:02

the biggest problem is making the

51:04

system more agantic. And that has

51:06

a bit of a longer roadmap,

51:08

but we still do the cadence

51:11

of weekly planning. These are the

51:13

things we're focusing on this week.

51:15

It's mostly, there's a good word

51:17

for this that I would want

51:19

you help with, but Polish, like,

51:22

fixing the bags and Polish this

51:24

week. And that was the planning

51:26

on Monday. That was actually this

51:28

week was Polish, Polish week. I

51:31

love that. How far is this

51:33

roadmap that you are now having?

51:35

I mean, it's clear over the

51:37

coming months, but it stretches up

51:39

three months and then, but in

51:42

one month, it's probably going to

51:44

look a bit different. Okay, and

51:46

then what are the tools used

51:48

just for folks that want to

51:51

understand, like the latest tools? So

51:53

you said, Big Jam, what else

51:55

is in that stack of tools?

51:57

I mean, we do so many

52:00

things in our company in linear,

52:02

because it's an amazing product. We

52:04

do talent application tracking in linear.

52:06

Yeah, well. And after going through

52:08

and this signal of the other

52:11

two. made custom made tools for

52:13

that linear and then fake jam.

52:15

So simple. How soon until one

52:17

of your engineers is an agent

52:20

engineer, an AI engineer, do you

52:22

think? Do you have a sense?

52:24

I love to dig into what

52:26

does that question actually mean? I

52:28

think we've been talking about like,

52:31

oh, AI, that would require more.

52:33

something playing chess, that's AI, like

52:35

if you even if a computer

52:37

can play chess, that's AI, and

52:40

now that's like, oh no, that's

52:42

a chess program, and we're always

52:44

shifting this forward and forward. I

52:46

think anything that a human doesn't

52:49

do is just a smart computer

52:51

system, right? So like what isn't...

52:53

When is a software engineer and

52:55

agent, I think it's always going

52:57

to be just, we're building and

53:00

lovable is just an interface that

53:02

humans interact with to create the

53:04

software that they want. And then

53:06

how we solve that is going

53:09

to be an agent under some

53:11

definition? Yeah, sure, I think so.

53:13

But that's less important to me.

53:15

Okay. I like that. Let

53:18

me ask this. You guys are

53:20

moving super fast scaling like crazy.

53:22

You described a little bit about

53:25

your process, weekly planning, big jam

53:27

board of ideas, and now there's

53:29

a roadmap that you're kind of

53:31

thinking out in the future. Is

53:33

there anything else that you found

53:36

helps you move this fast? That

53:38

gives you a lot of leverage

53:40

over the small team you have

53:42

to ship quickly and move fast

53:45

that you haven't already mentioned? We

53:47

work from the office most of

53:49

the time. I think it's pretty

53:51

nice. Then you can. So like,

53:53

hey, I think we're thinking wrong

53:56

about this thing, or shouldn't we

53:58

actually do this other thing? And

54:00

especially, I think lunch, like eating

54:02

lunch together, is a pretty productive

54:05

hour, where you cross pollinating. I

54:07

mean, people are constantly thinking subconsciously

54:09

as well about how to solve

54:11

these different problems and which the

54:13

most important ones are and then

54:16

being in office and has this

54:18

like focus most of the time

54:20

usually focused, but you also have

54:22

this like high bandwidth where everyone

54:25

has a bit unstructured communication. I

54:27

love that. The answer to the

54:29

CEO of a company that's one

54:31

of the most advanced AI tools

54:33

in the world is one of

54:36

your answers to how to move

54:38

fast is like lunch together. I

54:40

love that. That's so human and

54:42

so it makes all the sense

54:45

in the world, but I love

54:47

that that's still a part of

54:49

this. Yeah. Okay, you talked about

54:51

this kind of on the same

54:53

thread. You talked about if you

54:56

were to start in a team,

54:58

like a new product team today,

55:00

say you were head of product

55:02

somewhere, or head of VP of

55:05

product somewhere, building a new product

55:07

team, scaling a product team. What

55:09

would you do going forward that's

55:11

different from? what people have done

55:13

in the past in terms of

55:16

who you're hiring, how you're structuring

55:18

them, that kind of thing. Just

55:20

like, what do you think people

55:22

should be thinking as they build

55:25

product teams going forward, knowing tools

55:27

like Lovell will exist and all

55:29

the other stuff that's going on?

55:31

I mean, everyone should be excited

55:33

about using AI. I think that's

55:36

a pretty big one. And then...

55:38

And the team working well together

55:40

is the lunch, you have to

55:42

sit down and solve problems together.

55:45

You should, at the bottleneck for

55:47

most products these days, it's not

55:49

going to be as much on

55:51

engineering, but having good taste, good

55:53

intuition about your users and... that

55:56

I mean engineers and everyone preferably

55:58

in the team should have that

56:00

like willingness at least to want

56:02

to go through that motion and

56:04

listen to the users and truly

56:07

understand what they care about. What's

56:09

kind of like the background of

56:11

most of the engineers and people

56:13

you've hired are they like Is

56:16

there anything like in common? Are

56:18

they just like super? Impressive humans

56:20

generally like you know champions of

56:22

programming contest stuff like I don't

56:24

know like what are some attributes

56:27

of the folks you've hired so

56:29

far? I think raw cognitive capabilities

56:31

that's strongest like diamond the strongest

56:33

correlate of being a lovable lovable

56:36

lovable There is this start-up mindset

56:38

that I think is also very

56:40

strong. Being a bit more being

56:42

much more interested in moving very

56:44

fast and iterating fast than having

56:47

like a lot of structure, a

56:49

lot of process and thinking about

56:51

the business as a whole more

56:53

than thinking about my specific profession,

56:56

my specific craft that I'm seeing

56:58

myself like wanting to dig into

57:00

on me. Amazing, okay, so smart,

57:02

like very smart, entrepreneurial acts like

57:04

an owner, doesn't just, isn't just

57:07

like, this isn't just a job,

57:09

but they feel like they actually

57:11

have agency. Okay, this is great.

57:13

There's something you said, kind of

57:16

along these lines, that I think

57:18

is important, that one of the

57:20

things that gets you excited about

57:22

what you're building is giving people

57:24

superpowers, and especially people that don't

57:27

add a code, basically 99% of

57:29

people, is there anything along those

57:31

lines that you think is important

57:33

to share to share? It's very

57:36

clear to most people who have

57:38

been engineers or been founders that

57:40

there's so many that have failed

57:42

in their endeavors because they didn't

57:44

have someone that knows how to

57:47

solve the technical parts. And now

57:49

that we're close to having people

57:51

know that this does exist and

57:53

they solve everything. And it's going

57:56

to be a campaign explosion. entrepreneurship

57:58

and better software product, we're not

58:00

going to settle. for all the

58:02

annoying bad technology that we use

58:04

today. And everyone who has an

58:07

idea is going to say like,

58:09

okay, I'm gonna build this

58:11

thing and show you that

58:13

this is the best version

58:15

of the product or what our

58:17

company should be doing instead

58:20

of having long meetings

58:22

or like writing up

58:24

documents. So it's. going to

58:26

be empowering across a lot

58:28

of different professions and places

58:30

in the world. What's next for

58:33

lovable? What's kind of like the

58:35

next few things they might launch

58:37

as this episode comes out?

58:39

I mentioned this agentic behavior. And

58:41

when I say agentic, what it

58:43

means is that you give more

58:46

freedom to the system to decide

58:48

what happens next. It might want

58:50

to write a test, run those

58:52

tests, and say, like, let's fix

58:54

those. So that's... one of the

58:56

big unlocks for getting further

58:59

faster. And on, then there's

59:01

some more like obvious things

59:03

that you want to do,

59:05

you know, to go all the way to,

59:07

easily go all the way

59:09

to making money with Lovable.

59:11

And that's like, how do you

59:13

set up so that it's hosted

59:15

on your specific domain? How do

59:18

you collaborate seamlessly with your team?

59:20

I was just going to say

59:22

that that's just obvious things. And

59:24

something we're thinking about is to

59:26

help us founders succeed after they

59:28

build their first version. And like,

59:30

how do they get more users?

59:32

How do they get feedback? How

59:34

do they get the word out

59:37

if they build something useful? I

59:39

was just going to say that. That's

59:41

exactly where my mind went

59:43

is, like, everyone's going to be

59:45

building all these things. No one's ever

59:48

going to. get any traction with these tools

59:50

because no one knows how to find users,

59:52

get anyone to basically go to market and

59:54

growth is like a whole different skill. So

59:57

that is so cool that you're thinking about

59:59

that. How do we... run some paid ads

1:00:01

for you, how do we think

1:00:03

about a CEO, how do we

1:00:05

think about word of mouth, reality

1:00:07

referrals, that is very cool, okay?

1:00:09

We already have some playbooks that

1:00:11

we have today, people building with,

1:00:13

how do you do those things

1:00:15

that you can find up on

1:00:17

the blog? Interestingly, this makes me

1:00:19

want to buy some meta-stock because

1:00:21

all these apps that everyone's building,

1:00:24

they're going to be running paid

1:00:26

ads on Facebook and Google. Oh

1:00:28

my God, what a good business

1:00:30

those other guys get. I want

1:00:32

to come back to, you said

1:00:34

that you can work on your

1:00:36

existing code base. This is actually

1:00:38

a big question for a lot

1:00:40

of people. They see all these

1:00:42

tools. They're all like amazing for

1:00:44

prototypes and concepting. You talked about

1:00:46

how you can actually do this

1:00:48

within your existing code base, use

1:00:50

Lovable. Let me correct you there.

1:00:53

You cannot use it on any

1:00:55

existing code base. Got it. We

1:00:57

kind of have a research preview

1:00:59

of importing your code base, but

1:01:01

what you can do is if

1:01:03

you start in Lovable. engineers editing

1:01:05

it how in whatever tool they

1:01:07

want to use for editing it

1:01:09

Okay, cool. That's great clarification So

1:01:11

I guess just for people because

1:01:13

a lot of like most listeners

1:01:15

here are not building something brand

1:01:17

new They're working within an existing

1:01:19

product. So you're saying that that

1:01:21

is coming You can use level

1:01:24

in the future in some form

1:01:26

with your existing app and product

1:01:28

Great Wow, that's huge. Okay, because

1:01:30

that's basically the most most most

1:01:32

people so that's gonna be a

1:01:34

big deal. Okay A final question.

1:01:36

We have the segment on this

1:01:38

podcast called Failure Corner. Okay. Where

1:01:40

most people come in this podcast,

1:01:42

they show all these stories of

1:01:44

success and everything's going great and

1:01:46

here's all the things always winning.

1:01:48

You guys, this is a good

1:01:50

example. Just up and to the

1:01:53

right, the fastest growing product ever.

1:01:55

What's an example when something totally

1:01:57

failed in the course of your

1:01:59

career and what did you learn

1:02:01

from that? I'm a bit hard-pressed

1:02:03

to find something that... totally failed,

1:02:05

but I think there's a bit

1:02:07

of a product lesson where I

1:02:09

was the first employee at an

1:02:11

AI store up here in Stockholm.

1:02:13

Thana Labs. And the premise was

1:02:15

just, okay, so humans learn in

1:02:17

different ways. If you personalize, then

1:02:19

you get two standard deviations, more

1:02:21

effective learning. So there are a

1:02:24

lot of products, like education software

1:02:26

that helps you learn, that is

1:02:28

not personalized. And we were building

1:02:30

an API to personalized learning. I

1:02:32

mean, the AI and so on,

1:02:34

it was pretty good, but the

1:02:36

thing that we were doing in

1:02:38

the end was to say, like,

1:02:40

okay, here's this product, here, someone

1:02:42

has to build a product or

1:02:44

some way to learn, or be

1:02:46

it like English, thing do lingo.

1:02:48

And then the people that have

1:02:50

the product have to use this

1:02:53

advanced AI API to start making

1:02:55

it personalized. It was very hard,

1:02:57

like retrofitting, like, oh, you have

1:02:59

to switch out the engine and

1:03:01

put in this AI, and it's,

1:03:03

well, the big learning here is

1:03:05

that it didn't work very well

1:03:07

for the company. I mean, the

1:03:09

company wasn't super successful in this.

1:03:11

The big learning is that you

1:03:13

have to start with, like, how

1:03:15

is this product working end to

1:03:17

end? And then add AI, or

1:03:19

think, where should we add AI?

1:03:22

So that was a big learning

1:03:24

for me that. You really want

1:03:26

to see how the, how, what

1:03:28

is the big picture of the

1:03:30

user, what's the big picture of

1:03:32

how do you think the user

1:03:34

experience should be, and then add

1:03:36

something with AI to solve specific

1:03:38

problems. And now some of the

1:03:40

labs is doing great, but it's

1:03:42

not on top of that product

1:03:44

specifically. I think it's a lot

1:03:46

of people hear this now, like

1:03:48

of course, but I think it's

1:03:50

so. Hard to actually remember this

1:03:53

point when you're have some cool

1:03:55

tack and you're like holy shit

1:03:57

everyone needs to try this they're

1:03:59

gonna love it and then you

1:04:01

don't realize like no one actually

1:04:03

cares if it's not solving a

1:04:05

problem for them. Yeah, there's like

1:04:07

a lot of novelty products that

1:04:09

like everyone want to use for

1:04:11

a little bit and then like

1:04:13

forget it. I don't actually need

1:04:15

this often. And so I like

1:04:17

what this makes me think about

1:04:19

is there's all these product lessons

1:04:22

for what is likely to help

1:04:24

your product be successful. And an

1:04:26

app like a tool like Loveable

1:04:28

can help you do this because

1:04:30

if someone is building something You

1:04:32

can guide them. Okay, what's the

1:04:34

problem you're solving for somebody? How

1:04:36

many people have this problem? How

1:04:38

much does this matter to them?

1:04:40

Maybe we should add like the

1:04:42

Lenny mode. It activates, in lovable,

1:04:44

it activates like this product product

1:04:46

coach. That would... You can ask

1:04:48

any questions. You're like, no, wait,

1:04:50

hold on, why are you doing

1:04:53

this? Why? Let's take a step

1:04:55

back. Yeah, exactly. What's your experiment

1:04:57

plan? Yeah, what's your experiment plan?

1:04:59

That's actually, I think there's actually

1:05:01

a big opportunity there to say

1:05:03

people, because, you know, there's like

1:05:05

a play around with this thing

1:05:07

and then there's like, okay, but

1:05:09

really, is this anything people actually

1:05:11

want? I love it. Can we

1:05:13

call it Lenny mode? Is that

1:05:15

a fine with you? 100 percent.

1:05:17

Awesome. Let's do it. Anything you

1:05:19

want to leave listeners with? before

1:05:22

I let you go and go

1:05:24

to sleep. I think again the

1:05:26

world is changing quickly and it's

1:05:28

very fun. You should see that

1:05:30

I have fun in all of

1:05:32

this change and the best thing

1:05:34

you can do for your current

1:05:36

profession or if you want to

1:05:38

have a new job is to

1:05:40

be in the top 1% in

1:05:42

knowing how to use AI tools.

1:05:44

So go out there, use lovable,

1:05:46

use other AI tools and become...

1:05:48

Make sure to understand or try

1:05:50

to understand as much as possible

1:05:53

in how to use them productively.

1:05:55

That's something I tell all my

1:05:57

friends in the generally and I

1:05:59

love the audience to know as

1:06:01

well. Okay, well I got to

1:06:03

try to make this even more

1:06:05

specific for people. How do you

1:06:07

know if you're in the top

1:06:09

1% like what's like a heuristic

1:06:11

almost of like slash how do

1:06:13

you get there? Is it just

1:06:15

use it 100 times a day?

1:06:17

What else what can you recommend?

1:06:19

Yeah, I think if you spend

1:06:22

a full week on trying to

1:06:24

reach an outcome, the best way

1:06:26

to learn is like I want

1:06:28

to do this thing. And then

1:06:30

I want to use AI to

1:06:32

do that thing. And then you

1:06:34

spend a full week, you're in

1:06:36

the top 1% in the global

1:06:38

population. If you have friends that

1:06:40

you surround yourself with friends who

1:06:42

have this obsession or they also

1:06:44

care a lot about this, then

1:06:46

you'd be quickly in the top

1:06:48

0.1% percent. So what I'm hearing

1:06:50

is like find a problem that

1:06:53

can be solved, like find a

1:06:55

problem, a pain point for yourself

1:06:57

or someone, and then end-to-end, like

1:06:59

fully solve that problem, spend a

1:07:01

week, getting from getting from idea

1:07:03

to like a thing that was

1:07:05

actually if somebody's actually using. Yeah.

1:07:07

And you're in the top 1%?

1:07:09

Yeah, I think a little top

1:07:11

1% by just spending a full

1:07:13

week and making, like, asking AI

1:07:15

if you don't understand. So making

1:07:17

sure that you can understand. Yeah,

1:07:19

like, that's the thing people forget,

1:07:22

you just ask, like, would you

1:07:24

ask the chat feature of lovable

1:07:26

in this case, or would you

1:07:28

go to cloud a chat or

1:07:30

chat chat you BT to ask

1:07:32

for advice? use Lovell to build

1:07:34

software and learn that's the AI

1:07:36

tool. And then you should use

1:07:38

chat mode. And chat mode, I

1:07:40

have to add, is something you

1:07:42

activate in your user profile. It's

1:07:44

not launched in the main trouble

1:07:46

products. So it's in labs. But

1:07:48

if you add that flag, then

1:07:50

you can use chat mode. If

1:07:53

you want to learn some other

1:07:55

AI tool, then... you should, I

1:07:57

mean, ask that tool or ask

1:07:59

Claude, I was such a bit

1:08:01

about how that topic that domain

1:08:03

works. Okay, amazing. Where can people

1:08:05

find you? Where can they find

1:08:07

Lovable and how can listeners be

1:08:09

useful to you? Lovable posts updates

1:08:11

and means on Lovable underscore Dev

1:08:13

on Twitter. We post things on

1:08:15

LinkedIn as well and there are

1:08:17

a lot of things coming out

1:08:19

and changing in how we will

1:08:22

software. So you can follow Lovable

1:08:24

underschool Dev and you can follow

1:08:26

me at Anton Osika at Twitter.

1:08:28

I love more feedback on what

1:08:30

people, like where people see this

1:08:32

is a huge change for them.

1:08:34

There are a lot of people

1:08:36

posting about that on Twitter, but

1:08:38

we have a discord where you

1:08:40

can share like, oh, this is

1:08:42

how I use Lovebable and it

1:08:44

was super useful to me. And

1:08:46

feedback. Loveable. Dev. You can ask

1:08:48

for. new features. There's a lot

1:08:51

of people asking enough of both

1:08:53

things, what features you want. Thanks.

1:08:55

And that's a reasonable, that's the

1:08:57

most important thing for us. We

1:08:59

just want to solve people's problems.

1:09:01

Amazing. Anton, you're doing incredible work.

1:09:03

What a journey. I'm excited to

1:09:05

have you back some day when

1:09:07

we see more chapters of this

1:09:09

journey. I have a lot more

1:09:11

to learn. As do we all,

1:09:13

that's why people listen to this

1:09:15

podcast. Anton, thank you so much

1:09:17

for being here. Thank you so

1:09:19

much, Lenny. Thank

1:09:23

you so much for listening. If you

1:09:25

found this valuable, you can subscribe to

1:09:27

the show on Apple Podcast, Spotify, or

1:09:30

your favorite podcast app. Also, please consider

1:09:32

giving us a rating or leaving a

1:09:34

review, as that really helps other listeners

1:09:36

find the podcast. You can find all

1:09:39

past episodes or learn more about the

1:09:41

show at Lenny's podcast.com. See you in

1:09:43

the next episode.