Latest / Founder Mode / AI-First Product Design with Ritwik Pavan

Transcript

AI-First Product Design with Ritwik Pavan · Founder Mode

0:00
That's why Tesla was ahead of the curve when they were doing
0:03
FSD. And, you know, then you look at
0:05
Waymo or you look at, you know, in the past Cruise or Zooks
0:10
where they're having to basically send vehicles as test
0:13
vehicles to collect it. I mean, the reason Elon was
0:15
ahead of the curve with Tesla is because he was able to just have
0:18
drivers like us, You know, we were collecting.
0:22
Data for Elon for. A.
0:23
We're the Tesla training team right here at the Trio.
0:25
Exactly. Kevin, here we are founder mode.
0:36
Welcome back. Are you buddy?
0:37
So, so let's you know, let's talk about for you, given all of
0:41
the, you know, all of the stuff you're doing with AI right now,
0:44
like you, you're seeing a lot. Talk to me about AI first,
0:47
product design and, and the strategies that founders need to
0:51
kind of adapt, stay ahead. And just like not, you know, we
0:55
just talked about being too early.
0:56
Like it's very easy right now to kind of come upon it late.
0:59
Like what do you do? Yeah.
1:00
I think it's interesting. I think there's a couple things.
1:02
I think looking back, you know, historically, like there was
1:05
opportunities, you know, we had this kind of machine learning
1:07
sort of like Benton you, you'd have a data scientist, you'd
1:09
have like machine learning engineers and they'd build
1:11
features and then you'd train a model and it was incredibly slow
1:14
and incredibly manual. And I think the magic now is
1:17
these LLM's understand obviously large language models,
1:19
understand natural language. And so a lot of the way we
1:23
design and think about things today is like, how can you
1:25
collect enough context and how can you collect that in a
1:28
natural way so that you can let the LLM just do the work.
1:32
And what you're sort of amazed with is like, okay, if you give
1:35
it a reasonable response or a reasonable sort of request with
1:40
enough context, the actual responses are pretty good.
1:42
And so I think that's the first thing is how do you get enough
1:45
context and how do you just like let the LLM do its thing?
1:48
And then you start getting into the nuances of like, you know,
1:51
how do you make sure that it responds in the guardrails that
1:54
you have? Like one of our companies is
1:55
doing AI voice for medical. And so they're like the chance
1:58
of it getting off track and like confusing one very complicated
2:02
prescription regimen with another one is high.
2:05
So you have to be very careful of how you structure those sort
2:08
of kind of calls into the LLM or something.
2:10
That's a highly regulated space. You have to be really careful
2:13
that you don't. But I think, wow, the reason
2:16
though is that I think you know, and you've seen this like, you
2:19
know, things that are easy are now just not even, not even, you
2:22
know, that are just gone. And things that are sort of like
2:25
hard are now easy and things that were impossible are now
2:28
hard. And I think the part that is
2:30
important and what I think, you know, is we debate internally in
2:33
our companies a lot is like, are we doing stuff that used to be
2:36
impossible or is on the edge or fringe of hard to impossible?
2:40
Because if you're not, there's a good chance that whatever
2:43
momentum, whatever traction you have since it's brought to
2:46
market in public will be, you know, essentially an AI wrapper
2:50
company, it will get not. Adopted that's debate.
2:52
Is it is it interesting if it's just a wrapper?
2:55
I think I saw a post from the from YC being like, you know,
2:59
they're all rappers, like every single one on the list of the
3:02
lightest batch was like they're rappers.
3:04
So is it is it enough to build a wrapper or do you need your home
3:08
grown LLM and and data set and data model?
3:11
I, I love what you just said a moment ago about things that,
3:15
you know, things that were, that were hard are now easy.
3:18
Things that are impossible are now hard.
3:20
And I think that's, that's, and are you working on things that
3:22
used to be impossible that are now hard because like that is a
3:25
Moat. Let's get into it with our guest
3:26
today. We're thrilled to have Ritwick
3:27
Pavan on with us. He is a serial entrepreneur.
3:30
He started a company called Vaid, then Krava, and now he has
3:33
his third venture called Fragile.
3:35
We'll talk about some of that work in a moment.
3:37
And he's just tackling these hard problems, especially in
3:40
hardware, software, smart cities, modular construction,
3:43
Exciting, exciting guests on the podcast today.
3:45
And and I'm just stoked to have me.
3:51
Ray Wake, welcome to founder mode.
3:52
It's great to have you here. Great to be here.
3:54
Thanks, Jason. Awesome, man.
3:55
Well, yeah, it's cool because we met what, a while back now.
3:58
We met through Hampton and actually met Jason through
4:00
Hampton, which is kind of cool. Little Hampton Pod.
4:02
We were at the a retreat and so, yeah, little Hampton, shout out,
4:05
shout out. But it was good.
4:07
And yeah, so I think jumping into it, I mean, what is
4:10
building AI first product mean in practice?
4:12
And like, you know, I don't know, how does that change the
4:14
user experience? I know you've got it done a ton
4:16
of stuff with talking to companies and he just came back
4:18
from an international trip. And I'm, I'm curious kind of
4:21
what you're seeing with, you know, AI and how that compares
4:24
to the sort of traditional approach.
4:26
It's a it's a good question. So I think that the most
4:29
interesting part about AI first when it comes to let me get into
4:32
like consumer products specifically is just there is no
4:35
one-size-fits-all. I'm seeing so many different
4:38
broad applications when we're looking at AI first.
4:40
I mean, today I just featured a consumer product where it's
4:43
basically a tennis, it's AAI tennis player.
4:46
So basically this, it's able to do all sorts of different swings
4:49
based on like what you're trying to practice for.
4:52
And it's AAI device that basically gives you a tennis
4:55
player and then eventually they're evolving to pickleball.
4:58
There's another AI device I've featured before where it's a
5:00
chess, it's AAI chess player. So you know, I don't know if
5:03
you've ever played chess.com, but you know, you play on your
5:06
phone and you're moving the pieces.
5:07
Now imagine actual chess board where the pieces move on its own
5:12
and it's like it's magnetic and it moves on its own.
5:15
And then, you know, it's able to teach you instead.
5:17
So basically you have a trainer across the world, AAI trainer
5:21
too, where it's like training you on how to play.
5:24
So it's got the voice and everything else and it and it
5:26
shows you how to play so tennis. Pickleball one also physical
5:30
world like on the other side of the court with you.
5:32
Yeah, yeah. So it's like it's on other side.
5:35
Yeah, it's it's robotics. Ball machine because I was like,
5:39
yeah, absolutely. We had like this little turtle
5:40
ball machine back when I was playing tennis and like it would
5:42
just spit him out at you at like, you know, some relatively
5:44
just one. Thing we do so exactly and I
5:47
used to play tennis and that was the standard right?
5:49
Like you could go to a tennis center, you could rent one of
5:51
those tennis machines and it just like throws balls at you.
5:54
But now taking that to the next level, it's basically an AI
5:57
tennis machine that reacts to the shots you're taking.
6:00
Then they've got an entire app. They've got an app that
6:03
basically understands all of your shots and then gives you a
6:06
write up at the end of your tennis session is like, this is
6:08
what you need to improve on. This is where like you're
6:11
lacking. And so to answer your question,
6:13
Kevin, I think it's like AI first to me is like just
6:17
ultimately a better customer experience.
6:18
It's taking everything that before was like, in my opinion,
6:23
like not the best customer experience and now it knows you
6:26
as an individual and is basically able to improve the
6:30
customer experience per, you know, whatever it is you're
6:33
looking for. A few months ago I had the
6:36
chance to meet with a company called Matic Matic robots and
6:38
and I think they're a really exciting AI first product
6:41
because they're effectively they built this robot vacuum, right.
6:44
So at first glance it looks like a Roomba.
6:46
It doesn't look like one, but it sounds like a Roomba or AU fee
6:50
or any one of these robot vacuums.
6:51
But when you actually look into it, it's learning all of the
6:54
behaviors around your home and is then to know that, OK, your
6:58
kids are coming home at 4:00 PM and they're going to make a mess
7:00
around the dining table. So let me go out the.
7:03
Floor, it's like, get ready. Exactly.
7:05
And so just, you know, ultimately I think AI first is
7:09
going to allow for humans to have a better customer
7:11
experience and every product is going to be focused on the
7:15
individual themselves rather than this kind of standardized
7:18
product that we used to see in. The past personalization first
7:21
is like we could be another way to say it's like now this AI is
7:24
now enabled personalization at this like new and incredible
7:27
sort of rate that we wasn't possible before.
7:29
Yeah. Absolutely.
7:30
And I think it also allows for better outcomes where, you know,
7:34
before we were OK with mediocrity, but now it's like,
7:37
how much can we actually push these products to their limit to
7:41
deliver the best experience? I love it.
7:43
I think what's interesting when you juxtapose that against AI
7:48
later, right? So AI first, amazing.
7:51
We talked, we're talking about hyper personalization of product
7:54
experiences, but for founders that are kind of falling into
7:58
the trap of adding AI later and even companies have been around
8:02
a long time and like they just weren't AI first.
8:04
How do you think about that mindset of building AI 1st And,
8:09
and maybe a good example of this would be, you know, your days at
8:12
starting vade and then also like another player in the space, a
8:15
very big one sidewalk like smart cities in the pre IAI explosion.
8:21
And now how how you'd think about that?
8:22
I'm just super interested in this space because it there's so
8:25
much that we don't even think about when we walk around in the
8:27
world in our in our urban or suburban environments.
8:30
And like this is stuff that I can imagine going off of what we
8:34
just talked about that that hyper personalization and AI can
8:36
make our world around us so much better.
8:38
And I know you have a strong opinions about this, so I'm
8:41
eager to hear them. Yeah.
8:42
So you had mentioned my first startup, Vaid, which was a Gov
8:45
tech startup where we built cameras that used computer
8:48
vision to collect real time parking data and then
8:51
effectively improve traffic congestion, help citizens, you
8:54
know, with finding parking spots.
8:56
And if I were to build that company today, it would look a
8:58
lot different than it was back in 2017.
9:01
And that's just because of the tools that are available today
9:04
to, you know, everything from being able to use AI to predict,
9:10
you know, when a parking spot would be available.
9:13
Also on that note down the road, I mean, I think there's it's
9:17
it's going to be inevitable, you know, bike lane safety, getting
9:20
traffic lights to actually react to all of this behavior that is
9:24
being learned around like, what are the prime times when, you
9:27
know, people are looking for parking spots?
9:29
When is like that? What are the busiest times when,
9:32
you know, these curbs are taken? And then being able to, yeah, in
9:35
real time, and then being able to connect to this hardware
9:38
where it can now be AI first as well.
9:40
Because today there's effectively no real time data to
9:44
be able to know, you know what I consider parking spaces as a
9:47
city's inventory, right? There's no way to balance the
9:51
supply and demand, even when like some level of data is
9:54
available to the parking meters. So if we can actually use all of
9:57
that data and then connect it across through like pedestrian,
10:00
for example, sample with like the pedestrian science, right?
10:03
And then the traffic lights today, it's kind of just it's
10:05
kind of all just like a guess. But if we can use AI, I mean,
10:08
the company would look a lot different today than it was even
10:11
six years ago. So I think AI first is, yeah,
10:13
sorry, go ahead. No, no.
10:14
Have you seen approaches to this strategy in terms of like
10:17
training data feedback loops? Because I think part of what
10:20
you're getting into is like, you know, we didn't have the models,
10:22
but then it's also like, did we have like the product road map
10:25
to make that strategy work where it's like, hey, you need, you
10:28
need smarts on both sides, right?
10:29
Like especially in a city case where like there's this legacy
10:32
tech that you're trying to like kind of like light up with AI or
10:35
personalization. Yeah.
10:36
I mean, I'm seeing it on the road map for a lot of companies.
10:39
I think it's a good and bad because I think there's some
10:41
companies that are just trying to insert this AI first when it
10:44
doesn't make makes sense to go close a a fundraising check or
10:47
investor check. And then there's the.
10:48
Company. No, nobody's doing that.
10:50
Just by the way, anybody, just. Legitimately it's like golden
10:53
20X right? Exactly right.
10:55
Throw some AI use case in there and then you've got a yeah,
10:58
check. So I think there's a good and
11:00
bad to it. I think there's like real
11:01
applications where it makes sense.
11:03
But then there's also a stretch where like, I mean, ultimately I
11:06
do see a lot of these companies predominantly having a positive
11:10
use case. But you know, it's of course a
11:12
bubble right now where there's a lot of companies and he's kind
11:14
of just throwing it out as a buzzword.
11:16
So let's, I mean, let's double click there, right?
11:18
Like when, when you think about that building of that initial
11:23
high quality data set for a truly AI first product, not a
11:28
pretend AI first product like we just talked about, how do you
11:31
solve that that cold start problem and, and really get
11:34
there without the scale, right? Yeah, it's a good question.
11:37
And I think like the first use case or the first AI or the
11:41
first product that comes to mind is humanoid robots.
11:43
Because when you know, you've got all these companies kind of
11:47
fighting, you've got optimists, you've got, you know, unitary,
11:51
you've got a handful of these humanoid companies that are out
11:54
there where people think that the bottleneck is actually the
11:57
hardware. When in reality it's not the
11:59
hardware, It's the real like time data and the software side
12:04
where there's just simply not enough models to kind of train
12:07
it on, right? Like for example, like when
12:09
you're talking about a humanoid robot in every single persons
12:12
home, every single home is like very different layouts.
12:15
You've got like different washing machines.
12:17
So so it is an ongoing challenge that I you know, that I mean,
12:22
that comes to mind when we're looking at training these
12:24
models. I think it's far more complex to
12:27
be able to like just add this layer.
12:28
I mean, even with like the robot companies, right?
12:31
There's so many challenges around, like every home has a
12:33
different layout. And so I think it's just, you
12:36
know, it's going to come with time.
12:38
I mean, AI is moving at lightning speed right now, but I
12:41
think when some use cases, there's just not enough out
12:44
there to train these models on. I mean, with software, it's a
12:47
lot different because you can just train it off of X or Reddit
12:50
or like you know, with Grok, for example, just pulling from all
12:53
the X posts. But I think to get real life
12:55
footage is, you know, that's why Tesla was ahead of the curve
12:59
when they were doing FSD. And you know, then you look at
13:02
Waymo or you look at, you know, in the past Cruise or Zooks
13:07
where they're having to basically send vehicles as test
13:10
vehicles to collect it. I mean, the reason Elon was
13:12
ahead of the curve with Tesla because he was able to just have
13:15
drivers like us, we. Were collecting data for Elon.
13:20
For the Tesla training team right here, the trio.
13:22
Exactly right. And so I think that was a
13:25
brilliant move on his part because everyone else is is
13:27
behind, right? Like so it is an ongoing
13:30
challenge, but I think like as different applications come up,
13:33
it'll be interesting to see, you know, how the models improve.
13:36
It is good because I think, you know, we're working on a lot of
13:38
software stuff in voice and, and one of the things for us is like
13:41
building simulators for that. When it's software or things, if
13:45
you, if you can mock, you know, the other side of that
13:47
transaction or you can have synthetic transactions that
13:50
you're training against where I think to your point, like with
13:52
physical world stuff and humanoids, like it's hard to
13:55
mock, like how many houses can you mock or what?
13:57
You know, you can at some level, but it's trickier to kind of put
14:01
that in software. But pivoting a little bit to
14:03
like the teams behind this, because I think a lot of this is
14:06
the teams that are going to be required or the talent that's
14:08
required, sort of take some of these AI first challenges head
14:11
on. I mean, you've talked to a bunch
14:12
of founders across across these consumer hardware space.
14:15
Like what are you seeing in terms of the makeup of these
14:18
teams? And like what does it take to
14:20
win at an early sort of AI first hardware startup versus say like
14:24
a SAS company that we will all worked in, you know, 5-10 years
14:26
ago? Now with tariffs that you know,
14:29
I, I think it's a lot harder to vibe develop a product than it
14:32
is to vibe code. Yeah, it's an interesting one.
14:35
One of my close friends ASA, he's building kind of
14:37
construction robots and you know, when he built his first
14:40
robot startup, which was drones, they like custom built boards
14:44
and custom cameras and like super highly custom firmware.
14:48
And now it's like literally off the shelf robots and like using,
14:52
like I said, more simple techniques, right, to sort of
14:55
put this stuff together partly I think fuzzes scale.
14:57
So I think it, it's interesting that you're seeing the same
14:59
thing when you and the thing that for him, it's about how
15:02
quickly can they ship. And so I think you mentioned
15:04
this like, hey, people buying stuff on Alibaba and kind of
15:06
like, you know, rebranding it and then just shipping it.
15:09
I'm curious what you're seeing in terms of the speed of getting
15:12
stuff to market. And then how does that impact
15:15
like the ethical concern of like, putting a vacuum in my
15:18
house that maybe, you know, chasing my dog down or my kid
15:20
around versus like, you know, spending a little more time
15:23
making sure that that thing's got the safety controls to not,
15:26
you know, Elon's had some challenges with this as well.
15:28
Yeah. I think the biggest thing that I
15:31
have seen is the recent is that people are rushing to get to
15:36
market. And because of that, often times
15:38
the product that you see advertised, you know, people are
15:41
are pushing towards like marketing first companies rather
15:45
than product first, which I see is like a big problem because,
15:48
you know, and of course, like Elon is Elon has done this in
15:51
the past. But I think with Elon like that,
15:53
you know that the product is going to get delivered.
15:55
It might be like 8 years late, but it's going to get delivered,
15:57
right. I think it's that's
15:59
unfortunately not going to be the case with a lot of these
16:01
other companies where, you know, they set the price even before
16:04
the product is out of manufacturing and then they've
16:07
got this entire video where it shows the product at its V10.
16:12
I don't understand when my cyber truck wasn't.
16:14
Was it 50 grand? What was the number it was
16:16
supposed to? Be maybe that's the vibe coding
16:17
of hardware is you release a YouTube video on X.
16:20
Yeah, which like I get it from, I get it from like getting
16:23
customer feedback as soon as possible before investing, you
16:25
know, millions of dollars to building the product.
16:28
But at the same time, I think like showing the V10 and then
16:31
customer getting the V1 version often times leads to like, like
16:35
lack of trust as well. I mean, this is something you
16:37
even see with the FSD, right? People are like, oh, why is the
16:40
car not driving itself and picking me up today?
16:42
You know, why do I pay ten $10,000 for this?
16:45
And so except there's some level of trust that has been built up
16:48
within Tesla that has not been built up with a lot of these
16:51
newer AI wearable companies, for example, right?
16:54
So they launched the product on X, it goes viral, gets a million
16:58
impressions, and then it's going to take a year to get the
17:01
product right. So at that point, who knows
17:03
where we're going to be in AI at that point, You know, Meadow,
17:05
you might as well get a Meadow Orion, so.
17:08
How do you think about the housing crisis and how
17:10
technology and AI can help create more affordable housing
17:14
at scale? And I'm just curious your
17:16
thoughts on other tech startups in this space, since this is a,
17:19
you know, a couple companies ago.
17:20
Like talk to me about cover, if you're familiar with that.
17:23
I'm, I live outside of LA and I, I know Alexis and that team.
17:26
Like, I think it's very interesting this space and you
17:28
have a ton of experience in it. Yeah, I think the I think
17:32
modular construction, you know as a whole is necessary to be
17:36
able to accelerate the speed at which construction is going to
17:39
be done over as an excellent example.
17:42
I in terms of AI in itself as an application here, I actually
17:46
think the number one use case will be in permitting and
17:48
regulations because just to sift through all of the regulations
17:54
and permitting that goes in housing per county, not even per
17:57
like city, right, like perk zip code is just such a tedious
18:01
process and the you know, and I think this goes in with like
18:06
dodge and. Everything.
18:07
I knew we were going to go there.
18:09
Everything to do, you know, that can be done to cut regulations
18:12
today we're not there. And, and a large reason why an
18:15
SF and these other cities you're not seeing more housing is
18:18
because of the fact that it's so complex to even go get a permit.
18:21
And so I think the biggest application initially is going
18:24
to be being able to understand, you know, what are the
18:28
regulations that exist within the zip code and then being able
18:31
to, you know, quickly get permits out.
18:34
And then also know whether you're in the confines of being
18:37
able to like, you know, get exhaler X design built.
18:40
And then to kind of take that to the next level.
18:43
What would be really cool is, you know, the AI, basically you
18:47
use AI to understand all of the regulations and then it can put
18:51
together a floor plan for your house and say, OK, this is the
18:54
plot of land where you want to build a house, you know, build
18:57
me a house or build me a, you know, whatever in office based
19:01
on the restrictions that you understand in the zip code.
19:04
Like, you know, that's just an idea that came to mind, right?
19:07
It works like backs into the hard space versus the like
19:10
setbacks and all that sort of pre calculated.
19:12
I mean, it's pretty awesome because I mean, I think when
19:13
we've done some of our permitting up here in the
19:15
barriers, like it wild, right? Like how long, I mean this many
19:18
feet, Oh, the fence can't be here.
19:19
The pool's got to be offset here.
19:20
Oh, but it's grandfathered in so that if it's there, if you do
19:23
it, but if you rebuild it, it doesn't move and you're just
19:25
like, man, it's crazy. So it was very straightforward.
19:28
You've talked to a ton kind of companies and I think like, you
19:31
know, we went down the kind of like the permitting stuff and
19:33
pieces. But where from like the hardware
19:35
perspective, consumer hardware, like where do you think are like
19:38
sort of the untapped opportunities?
19:39
Where do you think there's just not enough innovation or there
19:42
there hasn't sort of been a breakthrough?
19:44
And what categories or sort of even products or companies do
19:47
you think like, you know are kind of like under underrated or
19:50
sort of just haven't had their moment in time yet?
19:52
Yeah. I mean, I think it's like very
19:54
clear, the ones that are a little bit overrated or hyped up
19:57
right now. Yeah, the high ones are good.
19:58
It's like, what are the under, what are the ones that like
20:00
where we the product categories that haven't had the investment
20:02
or the excitement? Yeah.
20:04
I think one of the exciting underrated ones is like sports.
20:07
I'm seeing like a lot of early stage companies where I'm like,
20:10
why is this company not been able to or like, why is this
20:13
company not been seeked out by, you know, some of the bigger
20:16
investors is like there is a very real application in AI to,
20:22
you know, help in lifestyle, right?
20:24
Like there's so many like as everybody kind of shifts and,
20:27
you know, eventually AI takes over, you know, jobs and, and,
20:30
you know, helps you become, you know, easier, older or anything
20:34
else. It's like, how can we use AI to
20:36
actually improve people's lives, whether it be sports, whether it
20:39
be, you know, any activity, right?
20:41
I think there's like, I think one of the ones where it has
20:44
been used and is going to continue evolve is like health
20:47
tech, right? There's a lot of health tech
20:48
companies being funded like crazy right now and you know,
20:52
health wearables, but also like health platforms.
20:54
One I'm not seeing as much as like more the lifestyle side of
20:57
it. You know, as I mentioned the
20:59
earlier, like I think there's going to be so many different
21:01
products that can be applicable, whether it's sports or
21:03
activities or hobbies, which which gets very interesting
21:06
because you obviously have like companies like 8 sleep or aura
21:08
ring or woop right where where they will have this like they
21:12
will eventually become AI first if they if they already aren't
21:15
offering it. Another use case I think that's
21:19
very interesting is outside of the like kind of lifestyle side
21:23
is I'm trying to figure out the best way to to there's public
21:27
safety is a very interesting 1. So I think there's a company
21:30
that you guys might be familiar with called Flock Safety, which
21:33
is Flock is absolutely crushing it.
21:36
And I think there's going to be a very real use case in security
21:40
systems, even for homes. I'm not talking about like the
21:43
cameras for HOA and neighborhoods, but for your home
21:46
there. I forget that it's a very
21:47
prolific founder. And I think in Miami, I'm
21:49
forgetting who it is. I think is the the a star
21:52
founder, somebody that's doing like this deterrence of, you
21:56
know, crime in your home. But like there there could be we
21:59
could take AI to the next level for like break insurance, right?
22:03
Like what if we're now able to use that to actively like deter
22:07
criminals? So I think there's going to be a
22:09
real application in crime, public safety, but also like
22:11
just, you know, home access control, like what if we could
22:14
take ring or, you know, some of these other secure ADT or some
22:17
of these other security systems to the next level with AII love
22:21
the. Focus on lifestyle and public
22:23
safety. I think these are interesting
22:25
categories that I agree. I'm not seeing nearly as much
22:28
interest and excitement. So as we think about maybe just
22:31
shouting out to the founders out there that are building this
22:33
right, like you've been very successful at raising capital.
22:36
So when we talk about how early stage founders in these AI first
22:39
companies that are doing super innovative, cool stuff, how
22:42
should they communicate that vision to their investors and
22:45
ultimately customers without overhyping and to land that big,
22:49
you know, first check or you know, maybe the A round, the B
22:52
round, right, the thing that gets them to escape velocity.
22:55
Yeah, I think it's so. Also you just reminded me one of
22:58
the other underrated ones is AI with manufacturing or just self
23:03
building. One of my favorite companies is
23:05
Bamboo Lab, if you guys are familiar.
23:07
It's like the top 3D printer company, but one of the exciting
23:10
parts about it is, you know, what if you could use AI but
23:15
also like this whole manufacturing site to basically
23:17
build products on your own without and like having to go
23:21
reach out to factory, you know, and get parts and components,
23:25
you know, done. And then it takes like 3 weeks
23:27
to deliver The reason I mention. The CNC business, basically.
23:30
Yeah. So the reason I'm mentioning
23:32
that point Jason is because to answer your question with
23:34
fundraising, I think by being able to have founders build
23:38
these ad first companies that allows them to potentially have
23:41
a recurring revenue channel. Often times hardware was looked
23:44
down upon because it's like one time cost very difficult to get
23:47
out to market. And then once you get out to
23:49
market like now how are you actually making money getting
23:52
like recurring revenue. But I think like when you have a
23:55
real use case where you're understanding customer feedback
23:58
or you're, you know, using all this data to basically improve
24:01
their life or you know, build a better experience for them, They
24:05
are going to be more willing to continue investing in the
24:07
product as well. Whether that be, you know, like
24:10
a monthly subscription, whether it be like, you know, a
24:12
different interactions. But I do think AI unlocks this
24:16
new use case or like this kind of feature set where people are
24:21
now going to be using those products on a more recurring
24:24
basis, which is something that investors look for, like when
24:26
they're looking at apps, for example, not only are they
24:29
looking for. Retention.
24:31
Yeah, and with apps, they're not only looking at like retention,
24:33
but they're also looking at like how frequently is this app being
24:35
used, right, like activity wise. And I think that just improves
24:39
radically when you're able to have this AI bot or like this AI
24:42
system, for example, with the tennis, like just showing, you
24:45
know, all all the like where all you hit it.
24:48
But then like what can be improved in in real time?
24:50
So in terms of like how I would pitch it to investors, I would
24:53
just, you know, try to tap into the the long term use case of
24:57
the hardware products rather than just like, hey, we're
25:00
building this product and it's going to help people do XYZ.
25:02
Transactional. We're building a relationship
25:04
with the user. Yeah.
25:06
Now and I think the the cool thing about that is I mean it,
25:08
it kind of takes hardware from this one time thing to now be a
25:10
subscription of the there is a subscription that makes sense
25:13
there, right. And so the hardware can just be
25:15
like amortized over that and then personalization and all
25:17
that. So I think I mean, right, look,
25:18
this has been an incredibly insightful, you know, time with
25:22
you. And I mean, I really appreciate
25:23
you sharing all the different pieces here.
25:25
And I think sort of breath of knowledge of this consumer
25:28
hardware space in particular is pretty amazing.
25:30
And you know, seeing all the the videos you do and and your
25:33
newsletter is obviously pretty impactful.
25:34
So where can people find you out and follow more about your work?
25:37
It's been great chatting with you guys.
25:39
I'm on Twitter, it's just at Rithwik Pavan.
25:41
So RITWAKPAVANI also write a a newsletter weekly sharing
25:47
exciting consumer product launches.
25:49
It's called Hardware Herald, so you can just go to
25:51
hardwareherald.com. It's been incredible to see that
25:53
grow over the last year or so. Thanks.
25:55
For thank you. Glad to have you on.
25:57
Thank you guys. That conversation with Ridwick
26:03
was like true master class and building, I mean, intelligence
26:05
on the ground up. And I think for me, if you're
26:07
just treating AI like a feature and you bolt it on later, like
26:11
hopefully that shifted your perspective, right?
26:12
And now you realize that like, you really have to think about
26:15
if you're going to do AI, have it lead with AI.
26:17
And then how do you have that integrated in a way that makes
26:20
the most sense? And so let's recap a few of the
26:22
things that he went over. I think first was this notion of
26:24
AI first hardware is really this hyper personalization, right?
26:28
Like every piece of hardware now feels like it's very much for
26:31
you. And it's that real time user
26:33
experience that just gets better and better as you use it.
26:36
But to build that, we need really good training data.
26:38
And so he talked a lot about the quality of the training data and
26:41
not necessarily the hardware, but it's like that becomes the
26:43
chief growth bottleneck, right? What is that ability for you to
26:46
get accurate training data to then be able to convert that
26:50
back and the winning devices are going to really be integrated,
26:52
right? It's that hardware, software,
26:54
ML, it's like and the talent that you need to go put those
26:58
together, you're no longer going to have just one type of
27:00
engineer. The full stack is now really
27:02
starting from like mechanical engineer, you know, electrical
27:04
engineer and all of that. And then I think that they
27:07
highlighted this notion of like, where's the white space
27:09
opportunities, right? So this idea of lifestyle,
27:11
sports and just lifestyle in general.
27:13
And then how does that bring AI driven stuff into the home
27:16
public safety, maybe self manufacturing tech.
27:18
And then finally, like I just converts these one off gadgets
27:21
that, you know, used to buy something off the shelf
27:23
hardware. Now it's a subscription business
27:25
and investors are going to love that, right?
27:27
And it's just the size of these businesses and the value of
27:29
these businesses is going to continue to advance.
27:31
Nailed it It's a it's a lot of wisdom packed into a short
27:34
amount of time. So if you found this
27:36
conversation valuable, do us a favor share it with another
27:39
founder or builder who's navigating the world of AI and
27:42
and trying to build AI 1st and make sure you're subscribed
27:44
because the next time on founder mode we're we're also tackling a
27:47
pretty interesting trend hacking direct to consumer growth with a
27:51
surprising strategy. So you can find links to
27:53
everything that we talked about, including how to connect with
27:55
quick in the show notes. And just thanks again for tuning
27:58
into Founder mode. Remember, don't just add
27:59
intelligence building around it. Build smart and build now.