May 27, 2025 • 27 mins
EPISODE 10
Serial CTO and Acompli co-founder JJ Zhuang joins Kevin and Jason to explain why the next generation of great products will be built by tiny, AI-super-powered teams. He unpacks lessons from taking Acompli’s first line of code to a $200 million Microsoft exit, steering Instacart through pandemic hyper-growth, and relearning to code with tools like GitHub Copilot, Cursor and Devin now writing the majority of his team’s code. Along the way JJ outlines the shift from hiring stack specialists to adaptable generalists, the rise of agentic “AI teammates,” and his mantra of “fast software” that lets founders prototype, test and ship at the speed of imagination—while still cutting through the 2025 hype cycle around agents and other buzzwords.
CHAPTERS
00:00 – Shots Fired: Surprising Start & Show Kick-off
01:52 – From Acompli to Outlook: Building All-Star Teams
06:11 – Instacart’s Pandemic Sprint & Early AI Adoption
11:50 – Pair-Programming vs. Agentic Devs: Devin, Cursor & More
22:12 – Fast Software & Filtering the 2025 AI Hype
LINKS
Connect with JJ Zhuang
Stay Connected with Founder Mode
Subscribe to our newsletter: foundermode.kit.com
Connect with Kevin
Connect with Jason
RESOURCES
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
I don't know what happened. You're right, one of the air
tubes pop. Oh, this door.
No way. Like the glass exploded.
I thought somebody was shooting at me.
No, dude, I thought I was like, bro, you took them tired.
That's intense. That's intense.
Like I think that I was too excited.
Yeah, the pressure from your excitement caused the window to
(00:21):
burst. That's amazing dude.
That's a while. And we're back with another
episode of Founder Mode. Yeah, we're digging into how
scrappy founders build world class products with lean AI
powered teams. Kevin, when a founder is staring
(00:43):
at an empty GitHub repo, what's the bigger hurdle?
Hiring that first engineer or trusting AI to write version 1.
That's a tough call. I mean, when Javier, JJ and I
built a complete and we're less than 15 people and it was, you
know, very much about the people, nothing like the AI we
had today. But but it's a balance.
(01:03):
I mean, today the tools and I mean, we've talked about V0 and
Devon and we'll probably get into that later today.
It's incredible what they can do.
And so I think the, you know, the risk for a start up that
hires too fast before they've automated some of these things
is could really be a a game changer or a game ender.
Yeah, exactly. I've seen, I've seen some folks
(01:24):
make that mistake. And I think ultimately the
conversation with JJ will illuminate the right way to
think about this and, and how tohire.
So I'm, I'm pretty stoked about it.
So from zero to a $200 million exit through a pandemic, IPOJJ
knows how to build a win. So let's let's hear how he
thinks about bootstrapping tech teams, where to sprinkle AI 1st,
and why smaller might actually be faster.
(01:52):
JJ, welcome to Founder Mode. Great to have you here.
Happy to be here. Yeah, we're, we're pretty
excited to have you, JJ. You've been building tech for a
long time. Tell us how things are different
at this moment in terms of how you build teams.
OK. Yeah, I think that it's quite
different now because, you know,we used to think that the best
team you wanted like a, you know, certainly always the best
(02:15):
people, right? But how we define best people is
kind of changing in my mind because can I even remember that
when we were at Accompli, we used to sort of like design all
these tests, right? Like, so when we interview,
let's say, engineers, we put through these, put them through
these loops and see like who cancrush all the questions?
(02:39):
I would not interview people this way anymore because we used
to just focus on what people know, you know, what their
expertise, where their expertiseis.
And you know, we hope that we have the, you know, best iOS
guy, the best Android guy, the best is like a database guy.
This guy's the DevOps guy now I think, you know, I just want to
(03:03):
find the best generalists, right?
And that's the best way to builda team because, and we want
people who have good intuition, but more importantly also
they're just, you know, more informed that they they're
always staying on top of everything, learning new things
really, really fast. And you know that that's the
(03:25):
best way to build a team. Take us back to day one at
Accompli. I mean, we're about to build
what, you know, becomes Outlook mobile and, you know, when
Microsoft drops a boatload of cash on us.
And how is that different compared to this sort of like AI
playground we're in now? Well, yeah, I, I don't think
they when we knew that that's the outcome, right.
But it is quite amazing that going from just the first line
(03:50):
of the code all the way to 500 million users in a few years.
But you know, it's day one we were really all about, OK, how
do we put together an All Star team, right?
And you know, in that case we did have, we did find that, you
know, all the best people, right?
(04:11):
There was indeed the best iOS engineer, the best Android guy,
and I was supposedly the best e-mail guy.
And you know, because I built a e-mail server and a e-mail
client before that, right? And then I didn't know a lot of
tricks about like, you know, howdo you squeeze out every ounce
(04:34):
of performance from these standard e-mail protocols
because, you know, otherwise howcan your e-mail app be better
than other people's e-mail app, right?
What differentiates us? And remember that we had to
like, you know, it's even just on the technical side alone,
right? It's like you have to overcome a
(04:55):
lot of different challenges. You know, even though people
think that, OK, mobile, like theiPhone came out, I think that
2007, right? And we were building the mobile
e-mail app in 2013. So that's like, you know, a few
years in between like 6 years isa long time.
(05:15):
But still, at that time, building the most complex app on
iOS Android was hard. It, you know, the, the, all, you
know, the building blocks, the, the, you know, we take for
granted today didn't exist, right.
So I remember things like, oh, you know, a lot of users have
100,000 emails in their inbox. And I mean, Kevin always has
(05:39):
zero because he's a inbox zero guy.
But like the, the rest of us, wealways have like all the emails
in the inbox read. And so how do you allow people
to kind of like infinitely scroll as fast as they want, but
still like be able to load everything really fast right
over high latent network and, you know, sort of a small screen
(06:02):
and also, you know, just low memory as well as like a, you
know, battery life, right? All these combinations.
How do you overcome these type of things?
So JJ, we were talking about, you know, the headwinds back in
2013, building an e-mail app against all of the like kind of
tech that was required to do best in class mobile development
on iOS back then. So let's Fast forward and let's
(06:25):
talk about AI and your time at Instacart and how you were able
to leverage AI even in the earlier days of people using AI
to really kind of stand out and separate what Instacart was
doing from the competition. To be honest, we didn't really
think that much about the competitions because it's always
about how, how to better serve customers, right?
(06:48):
This is the same, you know, whenwe were going through the Kevin,
you remember like, you know, when we're going through the
crazy growth period during the pandemic, right?
Especially the early stage of pandemic lockdown, that was
crazy. And you know, the, our mind was
always like, how do we ship faster and how do we serve
(07:08):
customers, right? So, you know, after the pandemic
when things such as GitHub hop pilot became a thing, that was
one of the decisions we made very early on, right?
Because you know, it's always about to running faster.
So something like the AI tooling, coding tooling was like
(07:31):
the first glimpse, right? Oh, this will actually help the
team run faster. So we gave access to everyone
immediately. Certainly not not everyone used
it because at the time people were still thinking that OK, I'm
a better coder than this, right.So, but a lot of people use it
and then later on quickly, you know, also gave everyone access
(07:54):
to cursor and you know, these numbers are now a few years old.
So I, I feel comfortable quotingthem.
But you know, I think that in first three months after people
start to adopt these tools, where we see across the team the
number of lines generated by AI was, you know, very quickly 20%.
(08:15):
And by now and then that is think considering at that time
the tools were very limited, notthat great.
So today, you know we're lookinginto really like more than half
of the code being generated, right?
I mean with AI doing all the heavy lifting, do we even need
Rockstar engineers anymore on day one or can you just hire an
average dev and let the machinesdo all the magic?
(08:38):
I I, I don't know, like it really depends on like how you
define what does average dev mean, right?
Because in the end it's about, you know, I have to say that
how, how do I put this? The skill set required is
slightly different, right? Because it's sort of like,
(09:02):
what's a good analogy? It, it's, it's, it's like, you
know, if you're working a factory where you know,
everything's manually done, thenyou have to find the people who
can carry a lot of load or I canmove things around.
But if it's, you know, automatedfactory and there's a lot of
(09:24):
robots deployed and then you probably want to find people who
can keep track of fast moving events in the factory floor,
right? Not like anyone who can lift a
lot of weight. So same thing with, you know,
using AI tools. You just want people who have a,
(09:45):
you know, deep understanding of what how to the methodology of
building software, but not necessarily the people who can
type the fastest or remember allthe details of, you know, the
syntax of the programming language or the tricks of the
programming language. Those things start to get pushed
into the background, right? So the skill set is changing,
(10:07):
but you still need, at the end of the day, the best engineers.
Just how you define them is different now.
I know there's a little fight happening right now between the
two of you about a certain AI coding tool, and I'm really
hopeful that we're going to get into that.
Yeah. So, so that, that's a good one.
I, I think that, you know, everyone can learn this, right?
(10:28):
And the important thing is to learn, you know, to develop the
new intuition. Like how do you harness these
things? And different people come to the
intuition intuition differently,right?
So you know, for, you know, somepeople say, oh, I have this
(10:49):
intuition because I trained my own LLM, right?
So that's great. But it doesn't have to be that
it can be like you just spend a lot of time to build your hobby
companion bar and you became sort of like the best GPT
whisperer, right? So it doesn't matter how you got
(11:10):
there. It's just that like you, you,
you need these skills and you know, training the existing team
totally is doable because it, itdoesn't really take, it's not a
rocket science and you can learnthese things relatively fast.
Yeah, Jason alluded to it, but Ithink I've been loving Devin
lately and JJ saw like the wrathof my Devin use sort of spread
across and sort of infect some of the code bases that we've
(11:32):
been working on together. And I think you're right.
Like I think like it's funny because I think we're, you know,
I'll be honest, I got overextended in some senses is
that Devon works great for things I understand well.
So I've always been like front end ish, sort of like building
websites and doing a lot of the sort of secondary coding stuff,
but having it go in and touch sort of code where, you know,
(11:54):
sort of full stack or deeper back end engineers, I think is
struggled, but it's it's interesting, right?
And so I think, you know, beyondthe obvious, like, what else did
have you seen like where, you know, building text kind of
fundamentally changed in this AIera?
Devon or Cursor or the all thesecoding tools, right You can see
that there are very quickly emerging that all these
(12:14):
different kind of mindsets, right.
So if you think about things like Cursor and Winsurf, which
is like another version of Cursor or our friends company
augment, right? And all these are sort of like a
copilot style in your IDE and kind of like a helper tools,
(12:40):
right? So that the main driver is still
the engineer and they, they, they use these tools.
But the emerging tools such as Devon and now open eye codecs,
I, I think they released it onlya few days ago, right?
That's start to shift to a slightly different paradigm.
So that paradigm is like a, you treat AI now as your, your
(13:04):
server tool, that who, who does the pair programming with you.
You treat them as another dev onthe team, right?
And then, you know, they push PR, you review them and so on.
So I think that we are in this weird moment when AI is like
really, really useful, extremelyuseful to almost too useful to
be dangerous, right? But at the same time not very
(13:26):
reliable. So it cannot really do things
fully autonomously. So the the paradigm between sort
of like a copilot paradigm and the the independent, another dev
on the team paradigm, we are quickly moving toward that, but
not there yet. This is why Devon may not work
(13:47):
the best today yet for a lot of tasks, but I'm sure that you
know in a period of time everyone will be on the other
side. Yeah, you mentioned Augment.
So Scott Dietzen's our friend and he's the CEO of Augment.
And one of the quotes I just sawhim post was that, you know, a
few months ago, 100% of their LLM calls were going from chat,
(14:07):
meaning, you know, interactive developer coaching it.
Now they're seeing 3% from chat and 97% from Magentic.
So basically their agent model, which is sort of similar to
Devon. And so the idea that I think a
lot of developers are trying this, I think it also, you know,
sort of feeds the LLMS and the companies that are charging on a
per use basis because clearly like having lots of agentic
(14:28):
calls over and over and over. What's the dumbest mistakes?
I mean, switching gears a littlebit that you've seen like
founders make, you know, when they try to like jam AI into
their startup. I wouldn't say dumb mistakes,
but everyone makes mistakes. But I think that, you know,
first of all. It's you using Devon for
everything, Kevin. Yeah, I'll answer the mistake.
(14:48):
That's the dumbest mistake. No, I'm just.
Yeah. Well, Devon is Kevin's brother.
Devon and Kevin. That's right.
Yeah. But you know, I think that the
the first thing is for start from the first of all, it's all
about like what problem he's trying to solve, right?
AI is just a tool at the end of the day.
And you know, other when people talk to me about their business
(15:12):
ideas and then very often I get excited if it is not AI and you
know, so it's all about what problem to focus on, right?
But the second piece is small team, especially if it's, it's
not that the one AI mistake you make.
(15:32):
It's like if you're not using AIfor everything, that's a mistake
because that's the, you know, leveraging all the tools, no
matter how imperfect they are, there's always a best of breed.
And that you, if you pay enough attention, you can use it for
everything right from like cut coding to writing your pitch
(15:53):
deck to running the team right, just about everything.
And so whoever's the most inventive and can leverage the
tool the best, we'll run fast and we'll win.
Yeah, about 90% of my conversations with with Kevin
are geeking out of the differentways that the two of us are
using AI in every aspect of our life.
(16:15):
So we are you are preaching to the converted.
I think. You know, one of the things I
noticed is going on with you right now is you you've talked
about kind of taking some time for, for learning and discovery,
which is a little corporate speaky.
So like, tell us, what are you actually working on?
What are you digging into right now and and how is it changing
how you view the world? I love the contrarian take by
(16:37):
the way of like, if it's not AI,I'm really interested, but like
realistically, what is? What does it mean?
What does learning in discovery mode mean for you right now?
What? What are you doing?
First of all, back to coding, because I in my career, I took
break from coding twice, right? The first time I, I took a five
year break and then Kevin and I start and I was third Co founder
(17:01):
Javier Sottero, the three of us,we started a company and day one
I was like, well, I need to codeagain.
I haven't coded for five years. And that was really, really hard
to, to get to the muscle memory back to be able to type again,
right? Because it felt like almost like
an idiot, right? And was very scary.
(17:23):
But the second time is like, youknow, last year I started to get
back into it. And this time I haven't quoted
for like, you know, not exaggerating, almost 10 years,
right? And this time it's so much
easier. It's not because I, you know,
don't have the muscle memory anymore, right?
It, it, it's not because I stillhave the muscle memory.
(17:44):
It is gone. But The funny thing is the
muscle memory is the part of that AI can actually replace the
best. And so we got into that.
And then I think that's, you know, taking some time off from
my regular work and focusing on like learning these things,
right? Is, you know, you can certainly
say that we can all have intuition about like how this
(18:07):
thing is evolving and how fast things are changing, Like all of
us have that, right. But when you actually day in and
day out to writing, coded using these tools and see how these
tools evolve, you just, you know, have better intuition
like, you know, and I really enjoy that part.
It's funny, I randomly was reviewing APR before we got onto
(18:30):
this and I happened to go onto my GitHub profile and I have
more green boxes on my GitHub profile in the last month then
in the last 10 years. And it's crazy.
Like partly it's all Devin, let's be honest.
And then but it's but it is to your point, like it's not only
fun, but it's way easy again, interms of like, you know, just
doing things and seeing. And it reminds me back to when I
(18:51):
got out of college and it was like I was editing websites or
PHP. And it's like you check
something in, you save it, you refresh the browser and it
works. I think that it's not one thing,
but it's mainly the mindset, right?
It's like for there are there are people who already have
found their product market fair,they're working on something in
(19:13):
that sense sense, then it's execution, right?
It is all about how to leverage,you know, the best team weighs
the best tools to run as fast aspossible.
But for, you know, founders who are still exploring and we do,
you know, Kevin, you and I talk to a lot of people in that stage
(19:34):
as well, right? I find that, you know, we used
to feel that there's a lot of limitation.
Everything takes a lot of time, a lot of effort to build
anything. And in order to figure out, OK,
where's the product market? If I have a good idea, I think I
have a good idea, right? To even get to a stage that I
(19:55):
have a demo to show customers and to understand that is that
going to be something valuable, right?
Takes a lot of, you know, effortand therefore capital.
But now it becomes like, you know, you can go from idea to
something that's working and putin front of customers in almost
no time in the very low effort, right?
(20:17):
So it, it kind of reminds me of this phrase, like we, we say,
what's the word fast fashion, right?
It's like, oh, quick fashion, fast fashion.
And you know, where does that come from?
It's like, you know, social media gives you the trend.
Then you have globalized labour,you have supply chain
(20:40):
optimization, all these things culminating to this moment.
You can't finally have fast fashion, right?
So it feels like AI is now giving us fast software.
So you can really explore so many different ideas in very
short period of time if you havethem and you can realize all of
them, right? And then just put in front of
(21:02):
people and you can almost like be all things to all people very
easily, right? And then, you know, we were
talking about things like Devon,things like codecs, right?
Eventually when these things become really mature, then you
know, what is the role of a engineer or product person or
(21:24):
the entrepreneur, right? Eventually it becomes like, you
know, it's almost like the artist who is, you know,
creating art but just giving instructions right to the
canvas. And then, you know, it's like
that may be a little bit abstract.
And so it's more like, you know,you're a movie director, right?
(21:48):
On the set, you're basically telling the actors what to do
and how to actually, you know, act better or, you know, bring
more emotion into it, right? So we will soon get into that
world. It's like today, it still feels
a little bit like a fantasy, butwe'll get there.
And everyone needs to become a movie director all.
(22:11):
Right. Final question, What you know
2025 AI trend is total BS and you know how should founders
kind of cut through the hype? This certainly a lot of hype,
right, because we kind of live in this social media EE chamber.
So so everything get like a justamplified everywhere.
So you know, probably the biggest buzzword right now is
(22:36):
agent. Everyone just keeps talking
about agent even though I don't even know if they all mean the
same thing, but everyone's like all multi agent agentic.
By the way, not sure if you guyssaw that.
Was it yesterday or the day before Microsoft announced this
thing called the Agentic web? I mean, I actually like the the
(23:00):
word because you know, you do need a word for that kind of
word or like when everything is M CP/M PC and so on.
And you know, I think that even though there's a lot of this
(23:20):
kind of hype, right, we tend to just like everything, we tend to
over hype of something in the short term, because in the short
term, all these things don't really work well yet, right?
You have to put in a lot of workto still make them reliable.
But at the same time, I think that to all of us make this
(23:42):
mistake, right, including ourselves, which is to
underestimate the hype today, because these things will become
really magical, will be really change our lives.
And you know, we're just not very good at, you know, non
(24:02):
linear thinking, right. So in I'm, I'm pretty sure that
in three to five years, the way we build software, the way we
build tech will be very different.
All right, where can folks find you and and follow your journey
on all this stuff, JJ? On this podcast.
Love it. That's good.
That's the right answer. Kevin actually said that he's
(24:23):
referred to you as Yoda a few times.
So you didn't disappoint in thatregard.
Learn, learn interesting things about AI.
We will so so any any links thatyou can share your what's the
best place for folks to connect with you and and and and get in
touch. I am on LinkedIn.
I have AX account but I'm I'm more a reader on X than than
(24:45):
publisher. Kevin is different.
Thanks, JJ Appreciate the honesty and the Jeff.
Thank you so much. Bye.
That was a master class. JJ's rule of thumb?
Fast software. It's blueprint for early stage
founders. Yeah, I love the analogy that
like the global supply chain andlabor force is to fashion what
AI is to software. It's just, you know, from fast
fashion to fast software, it's awesome.
(25:07):
Like it. It really helps crystallize for
me how to think about it. Kevin, what were the key things
to take away from this one if you're a founder listening?
Yeah, a bunch of things. I think the the the first one
was just this specialist roles now to like really looking for
people that are adaptable generalists just in terms of
fitting out an early AI team. And then we talked a bunch
about, you know, our sort of challenges that accompli and you
know, how that sort of shaped a large scale mobile app that, you
(25:30):
know, became Outlook and then AIcoding tools.
I mean, we jumped into my, my friend, my fan favorite Devin
Copilot cursor. And, you know, now these are
quite frankly generating more code than not, right?
You know, over 50% of the code that we're we're building today
is, is from these and you know, we're seeing that across the
industry and just this notion ofa gentic AI as a full teammate
versus like, Hey, it's like you're Co piloting, right?
(25:51):
And it's kind of funny, you know, pilot analogy there of
like, you know, who's flying theplane on this one?
Is the code flying or is it, youknow, is it on autopilot or is
it on copilot mode? And then you know, of course,
fast software, you know, that whole mindset of like, how do
you exploit AI everywhere? And then you really got to like
be careful and sift through the agent hype and make sure that
you apply it, you know, sort of when it's ready.
Yeah, if this sparks some ideas,fire it off to a founder who is
(26:14):
hustling with a tiny crew. Yeah, Smash.
Subscribe, drop a quick review and help more builders find the
show. Links to J, JS, LinkedIn and X
will be in every tool that we mentioned will be in the the
show notes in the description. And next time, we're jumping
from building lean teams to thinking about how commerce and
direct consumer has changed across the logistics side of the
(26:37):
business and D to C businesses. Yeah, we'll breakdown guerrilla
growth tactics and kind of why they still matter in 2025.
Remember, when you're short on headcount, the boldest move is
to let AI take the night shift so your team can build smarter.
I don't know what happened. You're right, one of the air
tubes pop. Oh, this door.
No way. Like the glass exploded.
I thought somebody was shooting at me.
No, dude, I thought I was like, bro, you took them tired.
That's intense. That's intense.
Like I think that I was too excited.
Yeah, the pressure from your excitement caused the window to
(00:21):
burst. That's amazing dude.
That's a while. And we're back with another
episode of Founder Mode. Yeah, we're digging into how
scrappy founders build world class products with lean AI
powered teams. Kevin, when a founder is staring
(00:43):
at an empty GitHub repo, what's the bigger hurdle?
Hiring that first engineer or trusting AI to write version 1.
That's a tough call. I mean, when Javier, JJ and I
built a complete and we're less than 15 people and it was, you
know, very much about the people, nothing like the AI we
had today. But but it's a balance.
(01:03):
I mean, today the tools and I mean, we've talked about V0 and
Devon and we'll probably get into that later today.
It's incredible what they can do.
And so I think the, you know, the risk for a start up that
hires too fast before they've automated some of these things
is could really be a a game changer or a game ender.
Yeah, exactly. I've seen, I've seen some folks
(01:24):
make that mistake. And I think ultimately the
conversation with JJ will illuminate the right way to
think about this and, and how tohire.
So I'm, I'm pretty stoked about it.
So from zero to a $200 million exit through a pandemic, IPOJJ
knows how to build a win. So let's let's hear how he
thinks about bootstrapping tech teams, where to sprinkle AI 1st,
and why smaller might actually be faster.
(01:52):
JJ, welcome to Founder Mode. Great to have you here.
Happy to be here. Yeah, we're, we're pretty
excited to have you, JJ. You've been building tech for a
long time. Tell us how things are different
at this moment in terms of how you build teams.
OK. Yeah, I think that it's quite
different now because, you know,we used to think that the best
team you wanted like a, you know, certainly always the best
(02:15):
people, right? But how we define best people is
kind of changing in my mind because can I even remember that
when we were at Accompli, we used to sort of like design all
these tests, right? Like, so when we interview,
let's say, engineers, we put through these, put them through
these loops and see like who cancrush all the questions?
(02:39):
I would not interview people this way anymore because we used
to just focus on what people know, you know, what their
expertise, where their expertiseis.
And you know, we hope that we have the, you know, best iOS
guy, the best Android guy, the best is like a database guy.
This guy's the DevOps guy now I think, you know, I just want to
(03:03):
find the best generalists, right?
And that's the best way to builda team because, and we want
people who have good intuition, but more importantly also
they're just, you know, more informed that they they're
always staying on top of everything, learning new things
really, really fast. And you know that that's the
(03:25):
best way to build a team. Take us back to day one at
Accompli. I mean, we're about to build
what, you know, becomes Outlook mobile and, you know, when
Microsoft drops a boatload of cash on us.
And how is that different compared to this sort of like AI
playground we're in now? Well, yeah, I, I don't think
they when we knew that that's the outcome, right.
But it is quite amazing that going from just the first line
(03:50):
of the code all the way to 500 million users in a few years.
But you know, it's day one we were really all about, OK, how
do we put together an All Star team, right?
And you know, in that case we did have, we did find that, you
know, all the best people, right?
(04:11):
There was indeed the best iOS engineer, the best Android guy,
and I was supposedly the best e-mail guy.
And you know, because I built a e-mail server and a e-mail
client before that, right? And then I didn't know a lot of
tricks about like, you know, howdo you squeeze out every ounce
(04:34):
of performance from these standard e-mail protocols
because, you know, otherwise howcan your e-mail app be better
than other people's e-mail app, right?
What differentiates us? And remember that we had to
like, you know, it's even just on the technical side alone,
right? It's like you have to overcome a
(04:55):
lot of different challenges. You know, even though people
think that, OK, mobile, like theiPhone came out, I think that
2007, right? And we were building the mobile
e-mail app in 2013. So that's like, you know, a few
years in between like 6 years isa long time.
(05:15):
But still, at that time, building the most complex app on
iOS Android was hard. It, you know, the, the, all, you
know, the building blocks, the, the, you know, we take for
granted today didn't exist, right.
So I remember things like, oh, you know, a lot of users have
100,000 emails in their inbox. And I mean, Kevin always has
(05:39):
zero because he's a inbox zero guy.
But like the, the rest of us, wealways have like all the emails
in the inbox read. And so how do you allow people
to kind of like infinitely scroll as fast as they want, but
still like be able to load everything really fast right
over high latent network and, you know, sort of a small screen
(06:02):
and also, you know, just low memory as well as like a, you
know, battery life, right? All these combinations.
How do you overcome these type of things?
So JJ, we were talking about, you know, the headwinds back in
2013, building an e-mail app against all of the like kind of
tech that was required to do best in class mobile development
on iOS back then. So let's Fast forward and let's
(06:25):
talk about AI and your time at Instacart and how you were able
to leverage AI even in the earlier days of people using AI
to really kind of stand out and separate what Instacart was
doing from the competition. To be honest, we didn't really
think that much about the competitions because it's always
about how, how to better serve customers, right?
(06:48):
This is the same, you know, whenwe were going through the Kevin,
you remember like, you know, when we're going through the
crazy growth period during the pandemic, right?
Especially the early stage of pandemic lockdown, that was
crazy. And you know, the, our mind was
always like, how do we ship faster and how do we serve
(07:08):
customers, right? So, you know, after the pandemic
when things such as GitHub hop pilot became a thing, that was
one of the decisions we made very early on, right?
Because you know, it's always about to running faster.
So something like the AI tooling, coding tooling was like
(07:31):
the first glimpse, right? Oh, this will actually help the
team run faster. So we gave access to everyone
immediately. Certainly not not everyone used
it because at the time people were still thinking that OK, I'm
a better coder than this, right.So, but a lot of people use it
and then later on quickly, you know, also gave everyone access
(07:54):
to cursor and you know, these numbers are now a few years old.
So I, I feel comfortable quotingthem.
But you know, I think that in first three months after people
start to adopt these tools, where we see across the team the
number of lines generated by AI was, you know, very quickly 20%.
(08:15):
And by now and then that is think considering at that time
the tools were very limited, notthat great.
So today, you know we're lookinginto really like more than half
of the code being generated, right?
I mean with AI doing all the heavy lifting, do we even need
Rockstar engineers anymore on day one or can you just hire an
average dev and let the machinesdo all the magic?
(08:38):
I I, I don't know, like it really depends on like how you
define what does average dev mean, right?
Because in the end it's about, you know, I have to say that
how, how do I put this? The skill set required is
slightly different, right? Because it's sort of like,
(09:02):
what's a good analogy? It, it's, it's, it's like, you
know, if you're working a factory where you know,
everything's manually done, thenyou have to find the people who
can carry a lot of load or I canmove things around.
But if it's, you know, automatedfactory and there's a lot of
(09:24):
robots deployed and then you probably want to find people who
can keep track of fast moving events in the factory floor,
right? Not like anyone who can lift a
lot of weight. So same thing with, you know,
using AI tools. You just want people who have a,
(09:45):
you know, deep understanding of what how to the methodology of
building software, but not necessarily the people who can
type the fastest or remember allthe details of, you know, the
syntax of the programming language or the tricks of the
programming language. Those things start to get pushed
into the background, right? So the skill set is changing,
(10:07):
but you still need, at the end of the day, the best engineers.
Just how you define them is different now.
I know there's a little fight happening right now between the
two of you about a certain AI coding tool, and I'm really
hopeful that we're going to get into that.
Yeah. So, so that, that's a good one.
I, I think that, you know, everyone can learn this, right?
(10:28):
And the important thing is to learn, you know, to develop the
new intuition. Like how do you harness these
things? And different people come to the
intuition intuition differently,right?
So you know, for, you know, somepeople say, oh, I have this
(10:49):
intuition because I trained my own LLM, right?
So that's great. But it doesn't have to be that
it can be like you just spend a lot of time to build your hobby
companion bar and you became sort of like the best GPT
whisperer, right? So it doesn't matter how you got
(11:10):
there. It's just that like you, you,
you need these skills and you know, training the existing team
totally is doable because it, itdoesn't really take, it's not a
rocket science and you can learnthese things relatively fast.
Yeah, Jason alluded to it, but Ithink I've been loving Devin
lately and JJ saw like the wrathof my Devin use sort of spread
across and sort of infect some of the code bases that we've
(11:32):
been working on together. And I think you're right.
Like I think like it's funny because I think we're, you know,
I'll be honest, I got overextended in some senses is
that Devon works great for things I understand well.
So I've always been like front end ish, sort of like building
websites and doing a lot of the sort of secondary coding stuff,
but having it go in and touch sort of code where, you know,
(11:54):
sort of full stack or deeper back end engineers, I think is
struggled, but it's it's interesting, right?
And so I think, you know, beyondthe obvious, like, what else did
have you seen like where, you know, building text kind of
fundamentally changed in this AIera?
Devon or Cursor or the all thesecoding tools, right You can see
that there are very quickly emerging that all these
(12:14):
different kind of mindsets, right.
So if you think about things like Cursor and Winsurf, which
is like another version of Cursor or our friends company
augment, right? And all these are sort of like a
copilot style in your IDE and kind of like a helper tools,
(12:40):
right? So that the main driver is still
the engineer and they, they, they use these tools.
But the emerging tools such as Devon and now open eye codecs,
I, I think they released it onlya few days ago, right?
That's start to shift to a slightly different paradigm.
So that paradigm is like a, you treat AI now as your, your
(13:04):
server tool, that who, who does the pair programming with you.
You treat them as another dev onthe team, right?
And then, you know, they push PR, you review them and so on.
So I think that we are in this weird moment when AI is like
really, really useful, extremelyuseful to almost too useful to
be dangerous, right? But at the same time not very
(13:26):
reliable. So it cannot really do things
fully autonomously. So the the paradigm between sort
of like a copilot paradigm and the the independent, another dev
on the team paradigm, we are quickly moving toward that, but
not there yet. This is why Devon may not work
(13:47):
the best today yet for a lot of tasks, but I'm sure that you
know in a period of time everyone will be on the other
side. Yeah, you mentioned Augment.
So Scott Dietzen's our friend and he's the CEO of Augment.
And one of the quotes I just sawhim post was that, you know, a
few months ago, 100% of their LLM calls were going from chat,
(14:07):
meaning, you know, interactive developer coaching it.
Now they're seeing 3% from chat and 97% from Magentic.
So basically their agent model, which is sort of similar to
Devon. And so the idea that I think a
lot of developers are trying this, I think it also, you know,
sort of feeds the LLMS and the companies that are charging on a
per use basis because clearly like having lots of agentic
(14:28):
calls over and over and over. What's the dumbest mistakes?
I mean, switching gears a littlebit that you've seen like
founders make, you know, when they try to like jam AI into
their startup. I wouldn't say dumb mistakes,
but everyone makes mistakes. But I think that, you know,
first of all. It's you using Devon for
everything, Kevin. Yeah, I'll answer the mistake.
(14:48):
That's the dumbest mistake. No, I'm just.
Yeah. Well, Devon is Kevin's brother.
Devon and Kevin. That's right.
Yeah. But you know, I think that the
the first thing is for start from the first of all, it's all
about like what problem he's trying to solve, right?
AI is just a tool at the end of the day.
And you know, other when people talk to me about their business
(15:12):
ideas and then very often I get excited if it is not AI and you
know, so it's all about what problem to focus on, right?
But the second piece is small team, especially if it's, it's
not that the one AI mistake you make.
(15:32):
It's like if you're not using AIfor everything, that's a mistake
because that's the, you know, leveraging all the tools, no
matter how imperfect they are, there's always a best of breed.
And that you, if you pay enough attention, you can use it for
everything right from like cut coding to writing your pitch
(15:53):
deck to running the team right, just about everything.
And so whoever's the most inventive and can leverage the
tool the best, we'll run fast and we'll win.
Yeah, about 90% of my conversations with with Kevin
are geeking out of the differentways that the two of us are
using AI in every aspect of our life.
(16:15):
So we are you are preaching to the converted.
I think. You know, one of the things I
noticed is going on with you right now is you you've talked
about kind of taking some time for, for learning and discovery,
which is a little corporate speaky.
So like, tell us, what are you actually working on?
What are you digging into right now and and how is it changing
how you view the world? I love the contrarian take by
(16:37):
the way of like, if it's not AI,I'm really interested, but like
realistically, what is? What does it mean?
What does learning in discovery mode mean for you right now?
What? What are you doing?
First of all, back to coding, because I in my career, I took
break from coding twice, right? The first time I, I took a five
year break and then Kevin and I start and I was third Co founder
(17:01):
Javier Sottero, the three of us,we started a company and day one
I was like, well, I need to codeagain.
I haven't coded for five years. And that was really, really hard
to, to get to the muscle memory back to be able to type again,
right? Because it felt like almost like
an idiot, right? And was very scary.
(17:23):
But the second time is like, youknow, last year I started to get
back into it. And this time I haven't quoted
for like, you know, not exaggerating, almost 10 years,
right? And this time it's so much
easier. It's not because I, you know,
don't have the muscle memory anymore, right?
It, it, it's not because I stillhave the muscle memory.
(17:44):
It is gone. But The funny thing is the
muscle memory is the part of that AI can actually replace the
best. And so we got into that.
And then I think that's, you know, taking some time off from
my regular work and focusing on like learning these things,
right? Is, you know, you can certainly
say that we can all have intuition about like how this
(18:07):
thing is evolving and how fast things are changing, Like all of
us have that, right. But when you actually day in and
day out to writing, coded using these tools and see how these
tools evolve, you just, you know, have better intuition
like, you know, and I really enjoy that part.
It's funny, I randomly was reviewing APR before we got onto
(18:30):
this and I happened to go onto my GitHub profile and I have
more green boxes on my GitHub profile in the last month then
in the last 10 years. And it's crazy.
Like partly it's all Devin, let's be honest.
And then but it's but it is to your point, like it's not only
fun, but it's way easy again, interms of like, you know, just
doing things and seeing. And it reminds me back to when I
(18:51):
got out of college and it was like I was editing websites or
PHP. And it's like you check
something in, you save it, you refresh the browser and it
works. I think that it's not one thing,
but it's mainly the mindset, right?
It's like for there are there are people who already have
found their product market fair,they're working on something in
(19:13):
that sense sense, then it's execution, right?
It is all about how to leverage,you know, the best team weighs
the best tools to run as fast aspossible.
But for, you know, founders who are still exploring and we do,
you know, Kevin, you and I talk to a lot of people in that stage
(19:34):
as well, right? I find that, you know, we used
to feel that there's a lot of limitation.
Everything takes a lot of time, a lot of effort to build
anything. And in order to figure out, OK,
where's the product market? If I have a good idea, I think I
have a good idea, right? To even get to a stage that I
(19:55):
have a demo to show customers and to understand that is that
going to be something valuable, right?
Takes a lot of, you know, effortand therefore capital.
But now it becomes like, you know, you can go from idea to
something that's working and putin front of customers in almost
no time in the very low effort, right?
(20:17):
So it, it kind of reminds me of this phrase, like we, we say,
what's the word fast fashion, right?
It's like, oh, quick fashion, fast fashion.
And you know, where does that come from?
It's like, you know, social media gives you the trend.
Then you have globalized labour,you have supply chain
(20:40):
optimization, all these things culminating to this moment.
You can't finally have fast fashion, right?
So it feels like AI is now giving us fast software.
So you can really explore so many different ideas in very
short period of time if you havethem and you can realize all of
them, right? And then just put in front of
(21:02):
people and you can almost like be all things to all people very
easily, right? And then, you know, we were
talking about things like Devon,things like codecs, right?
Eventually when these things become really mature, then you
know, what is the role of a engineer or product person or
(21:24):
the entrepreneur, right? Eventually it becomes like, you
know, it's almost like the artist who is, you know,
creating art but just giving instructions right to the
canvas. And then, you know, it's like
that may be a little bit abstract.
And so it's more like, you know,you're a movie director, right?
(21:48):
On the set, you're basically telling the actors what to do
and how to actually, you know, act better or, you know, bring
more emotion into it, right? So we will soon get into that
world. It's like today, it still feels
a little bit like a fantasy, butwe'll get there.
And everyone needs to become a movie director all.
(22:11):
Right. Final question, What you know
2025 AI trend is total BS and you know how should founders
kind of cut through the hype? This certainly a lot of hype,
right, because we kind of live in this social media EE chamber.
So so everything get like a justamplified everywhere.
So you know, probably the biggest buzzword right now is
(22:36):
agent. Everyone just keeps talking
about agent even though I don't even know if they all mean the
same thing, but everyone's like all multi agent agentic.
By the way, not sure if you guyssaw that.
Was it yesterday or the day before Microsoft announced this
thing called the Agentic web? I mean, I actually like the the
(23:00):
word because you know, you do need a word for that kind of
word or like when everything is M CP/M PC and so on.
And you know, I think that even though there's a lot of this
(23:20):
kind of hype, right, we tend to just like everything, we tend to
over hype of something in the short term, because in the short
term, all these things don't really work well yet, right?
You have to put in a lot of workto still make them reliable.
But at the same time, I think that to all of us make this
(23:42):
mistake, right, including ourselves, which is to
underestimate the hype today, because these things will become
really magical, will be really change our lives.
And you know, we're just not very good at, you know, non
(24:02):
linear thinking, right. So in I'm, I'm pretty sure that
in three to five years, the way we build software, the way we
build tech will be very different.
All right, where can folks find you and and follow your journey
on all this stuff, JJ? On this podcast.
Love it. That's good.
That's the right answer. Kevin actually said that he's
(24:23):
referred to you as Yoda a few times.
So you didn't disappoint in thatregard.
Learn, learn interesting things about AI.
We will so so any any links thatyou can share your what's the
best place for folks to connect with you and and and and get in
touch. I am on LinkedIn.
I have AX account but I'm I'm more a reader on X than than
(24:45):
publisher. Kevin is different.
Thanks, JJ Appreciate the honesty and the Jeff.
Thank you so much. Bye.
That was a master class. JJ's rule of thumb?
Fast software. It's blueprint for early stage
founders. Yeah, I love the analogy that
like the global supply chain andlabor force is to fashion what
AI is to software. It's just, you know, from fast
fashion to fast software, it's awesome.
(25:07):
Like it. It really helps crystallize for
me how to think about it. Kevin, what were the key things
to take away from this one if you're a founder listening?
Yeah, a bunch of things. I think the the the first one
was just this specialist roles now to like really looking for
people that are adaptable generalists just in terms of
fitting out an early AI team. And then we talked a bunch
about, you know, our sort of challenges that accompli and you
know, how that sort of shaped a large scale mobile app that, you
(25:30):
know, became Outlook and then AIcoding tools.
I mean, we jumped into my, my friend, my fan favorite Devin
Copilot cursor. And, you know, now these are
quite frankly generating more code than not, right?
You know, over 50% of the code that we're we're building today
is, is from these and you know, we're seeing that across the
industry and just this notion ofa gentic AI as a full teammate
versus like, Hey, it's like you're Co piloting, right?
(25:51):
And it's kind of funny, you know, pilot analogy there of
like, you know, who's flying theplane on this one?
Is the code flying or is it, youknow, is it on autopilot or is
it on copilot mode? And then you know, of course,
fast software, you know, that whole mindset of like, how do
you exploit AI everywhere? And then you really got to like
be careful and sift through the agent hype and make sure that
you apply it, you know, sort of when it's ready.
Yeah, if this sparks some ideas,fire it off to a founder who is
(26:14):
hustling with a tiny crew. Yeah, Smash.
Subscribe, drop a quick review and help more builders find the
show. Links to J, JS, LinkedIn and X
will be in every tool that we mentioned will be in the the
show notes in the description. And next time, we're jumping
from building lean teams to thinking about how commerce and
direct consumer has changed across the logistics side of the
(26:37):
business and D to C businesses. Yeah, we'll breakdown guerrilla
growth tactics and kind of why they still matter in 2025.
Remember, when you're short on headcount, the boldest move is
to let AI take the night shift so your team can build smarter.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy, Jess Hilarious, And Charlamagne Tha God!
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.