Latest / Founder Mode / Will AI Replace Developers with Natalie Kaminski

Transcript

Will AI Replace Developers with Natalie Kaminski · Founder Mode

0:00
So first of all, again, I don't know what's happening behind the
0:02
closing doors of all these Silicon Valley, you know,
0:05
operations, right? Those cannot Privy to that.
0:07
But I'm just talking about the mainstream, the tools that are
0:09
available to us right now. None of it is AI.
0:12
There's no intelligence, there's no I in the AI.
0:15
I would argue that. You're listening to Founder
0:32
mode. The show where builders talk
0:33
honestly about what's changing and what still matters.
0:36
Today's episodes for every founder wondering if code
0:39
writing is about to become obsolete.
0:41
We're talking to somebody who's worked with hundreds of dev
0:43
teams and has lots of thoughts on where AI helps and where it
0:46
maybe doesn't. Jason there's so much hype about
0:49
AI writing full apps, but people forget software just isn't about
0:52
code, it's context. You've had hundreds of engineers
0:56
work for you at the various companies you've LED.
0:57
How many engineers are on the the stealth startup team now?
1:01
And I know you're running a few different businesses, but like,
1:03
you know, what's the average team size and what percentage of
1:05
their code is, is AI generated? Yeah.
1:08
So I think that that's it. There's no team more than 10.
1:10
But like, you know, startups do start small, but I think we're
1:13
doing and building at what would have typically been 1520 people.
1:18
So I mean, the number of humans is like, you know, 75% lower for
1:23
even a startup that's kind of, you know, getting early going.
1:26
And then I think the other piece is that I would say almost all
1:29
of the code is, you know, kind of touched by AI one way or the
1:33
other. And I mean, whether it's every
1:35
little lines written or gone. But I think like it is way more
1:38
prompt driven development than it is like typing code into a
1:42
browser or typing code into an editor.
1:44
And so I think for us like, oh, not to mention all the tests and
1:47
all of the like framework and infrastructure, and then, you
1:50
know, people are relying deeply on this to generate compliance
1:52
things. And, you know, we use, you know,
1:54
literally AI for every aspect of both building, delivering and
1:58
then sort of operating the business.
1:59
And so for us, it's, it's pretty interesting.
2:01
So I think it is like there's a really big shift happening here.
2:04
Yeah. My view is today, like I don't
2:06
want to work with or hire people that aren't like heavily using
2:09
AI, partly because I believe that that technology is a force
2:13
multiplier. And even for the most junior
2:15
folks, having AI is like, you know, sort of a superpower,
2:20
right? They can learn, they can go
2:21
quicker and they can also just get a lot deeper faster with
2:25
things that typically they would have had to like tag in another
2:27
senior person, right? So now a lot of that sort of
2:30
banter of like, you know, back and forth is working.
2:32
And, you know, for me, and even like, if you think of a lot of
2:35
the AI products out there, ours included, in terms of how
2:37
they're built, they're built on outcomes.
2:40
And so when you bring on an outsourced team, like, I want
2:42
that outcome to be the best, the fastest, and, you know, the most
2:45
efficient and potentially the cheapest is you say like for,
2:48
for what's best to breed right for a certain level, what are
2:50
you paying for? And so to me, I, I think AI is
2:53
imperative and the more junior you are, the guardrails you want
2:58
around the work they're doing. But you definitely want to
3:02
people to have the best and the latest tools.
3:04
Yeah, give them great tools. Yeah, you and I love to hit the
3:07
gym hard. Like we don't, we don't shy away
3:09
from the Smith machine and the and the squat rack and the the
3:13
bench press. So so should we pivot to our
3:16
interview today? Let's do it.
3:17
Yeah, Natalie's a founder, engineer and an operator who's
3:20
helped launch many products for the big names you've heard of.
3:22
Yeah, she works with technical and non-technical builders
3:25
alike, and she brings a different take on how AI fits
3:28
into the development stack. Let's get into it.
3:30
Natalie, welcome to founder mode.
3:31
You've spent your career sort of helping non-technical founders
3:34
turn ideas into scalable software.
3:36
And now, you know, building in a world where everyone's asking
3:38
like, is AI going to take your job?
3:39
And you know, today we're really kind of unpacking like do
3:42
developers still matter and and really see from your view what's
3:44
coming next? Hi.
3:45
Yes, well, thank you for having me.
3:47
And absolutely it's, it's a question I get to answer almost
3:50
on a daily basis as I meet people, as I, you know, I speak
3:53
to potential leads, workers, colleagues.
3:55
It's it's the $1,000,000 question I unequivocal answer
4:00
right now as of October 6th, 2025 because things change so
4:04
rapidly is that developers, engineers today matter more than
4:08
ever and that. You're safe, Kevin.
4:11
You're absolutely safe. So are we and jet rockets and
4:15
there's so much noise, right? There's so much hype happening
4:18
over what AI can and can I do. And I think a lot of these
4:22
conversation is borderline irresponsible because dealing
4:27
with non tech founders, you know, people often come to me
4:29
and they say, you know, I just spent the weekend vibe coding
4:32
this platform and now I'm going to put it on on the Internet.
4:36
I'm going to launch it to production and I'm going to
4:38
start taking people sensitive information in.
4:41
Let's let's dive in deeper. So you started as a developer.
4:44
Now you lead a team that helps others build.
4:46
How did that journey shape the way that you see tools like AI
4:50
fitting into everything well? At first, I have to admit when
4:53
AI first came out as hey, we can code now, you know, above
4:58
generative content and stuff, we can actually code and we can
5:02
seemingly put together apps. I was a little concerned.
5:05
I was like, are we done, you know, as a profession like that?
5:07
That was a very legitimate concern.
5:09
But as we to dive in and we actually, as a matter of fact,
5:11
this is perfect timing for this recording because we just
5:14
completed like a 5 month experiment at Jetrockets,
5:17
building a platform that, you know, using AI tools as much as
5:22
possible, like AI coding assistance just to understand
5:25
what it actually can and cannot do.
5:27
And so I can say without any doubt that right now, AI is not
5:32
going to replace, but AI is going augment us.
5:35
AI is going to make, you know, engineers a little bit more
5:38
efficient, you know, very deliberately, I say a little bit
5:41
more efficient. And, and I'm also going to say
5:44
that in order for you to gain the most benefits out of these
5:46
tools, you have to know what you're doing.
5:49
I, you know, my favorite analogy is saying, well, if you, if
5:53
somebody were to hand me a scientific calculator and said,
5:55
Hey, you can solve any mathematical fraction right now,
5:58
whatever, and I'd say, oh, this is amazing.
6:00
Let me do this. But like, what do I do with the
6:02
result? How do I apply it in real life?
6:04
What, like, what does this even mean?
6:06
Right? Having access to a scientific
6:08
calculator is not going to make me mathematician.
6:10
And so I claim that having access to cloud, cloud code or
6:15
any other tools, cursor, whatever your tool might be, is
6:18
not going to make you an engineer.
6:20
Now, if you know what you're doing, you can use those tools
6:23
to be faster, more productive and higher quality engineer.
6:27
And so when you think about workable AI, like the stuff that
6:31
is actually useful, you know, maybe engineers using AI to help
6:34
them be faster, be more efficient, right, cleaner code.
6:36
Can you explain what it really means in real product to have
6:40
workable AI? Yeah, absolutely.
6:41
So there are certain tasks in engineering in like daily
6:45
software engineers lives that nobody wants to do.
6:48
But you kind of have to do those tasks, right.
6:50
I mean, think of your data migrations, think of your
6:52
scaffolding like sort of boilerplate application, all the
6:55
crowd which is create, update, delete records, right.
6:58
Think about, I don't know, thinking even even simple stuff
7:02
like thinking of a good naming convention for certain variables
7:06
within your application can't be tedious and the boring.
7:10
And AI excels at those things, right?
7:13
AI can automate a lot of this work for you.
7:16
Another thing is test coverage for your sort of like code test
7:20
coverage. A young is great at producing
7:22
quality, thorough test coverage. And I am yet to meet a developer
7:28
who actually enjoys writing tests for their own code.
7:31
So having those tools and knowing how to implement and
7:34
integrate those tools into your daily life is that is the skill
7:37
that if you are an engineer today, you must develop, you
7:41
must learn, and you must bring into your daily operation and
7:44
then that this is what gives you this efficiency.
7:46
And my best engineers are jet rockets.
7:48
They probably gain about 25 to 30% efficiency, although it's
7:52
hard to tell because like how do you measure something like this?
7:54
But I like to say 30% sounds like a good number.
7:58
I like it. I mean like a simple question
8:00
like how many of your is there any of your devs or anybody
8:02
who's not using AI today or 100% are using something?
8:05
So we have one rule at our company is that we do not allow
8:09
interns and junior engineers to use AI because then they don't
8:13
know what they're doing. And AI as you know, AI, if you
8:17
don't know what good code looks like, you're going to accept
8:20
whatever AI gives you and AI gives you.
8:23
So you know, AI provides the false sense.
8:26
That means they can't even use like ChatGPT to look stuff up.
8:28
So you're still making them use Google to go to slat, stack
8:30
Overflow or like? No, they can use checks.
8:33
You could see, of course, but you know, I'm, I'm mostly
8:35
talking about, you know, code generators and and stuff like
8:38
that. Got it.
8:39
Copilot and stuff, stuff integrated into your ID, your
8:42
development environment. Yes, because I want them to
8:44
learn. I want them to understand what
8:46
AI does and I want them to recognize when AI makes mistakes
8:49
because it does make mistake every.
8:51
Every do you use like AI to do code?
8:53
Like are the senior people then are using AI to generate are
8:56
using AI to do code reviews or like where, where is it like
8:58
sort of like click in? So I'm a junior engineer.
9:00
I just RIP a bunch of code. Line of code at Jet Rockets
9:03
reviewed by a human engineer. Not every line of code.
9:06
I mean, that would be in saying it's, it's impossible.
9:08
You know, like you have a pull request that you submit and then
9:12
you're, you know, an engineer has to review like another
9:14
engineer that you're paired, but must review kind of the constant
9:18
context of that code. And some issues are being caught
9:21
by AI before it even gets to a human.
9:23
But ultimately, the human is going to make this determination
9:26
whether or not this code passes the litmus test of OK, is this
9:29
good enough to go to production? Yeah, makes sense.
9:31
And then do you guys like how much faster are you today at
9:35
writing code and like how much cheaper are the things that
9:38
you're building versus say a year ago?
9:40
So this is another kind of like argument I'm going to challenge,
9:43
right? And I'm going to say, why should
9:44
it be cheaper? Like if we've, if we've learned
9:47
a skill, which is if we've learned to integrate AI into our
9:51
daily operation, we took the time we learned it, we
9:55
implemented it in such a way that it makes us more effective
9:58
and more productive. Why should it be cheaper to my?
10:00
Yeah. So you're staying, right, If
10:02
there's improvements in technology, you guys incur the
10:04
value, clients still pay full price kind of or as a mental
10:06
model? Well, it's sort of the stuff
10:09
that it yes, definitely certainly thinks that in the
10:12
past took longer that take less time now for sure.
10:15
But ultimately it's not because we don't we don't charge for
10:19
time, right. We don't charge for hours or
10:22
lines of code produced. We charge for value.
10:25
Hypothetically say that there's other dev shops out there or AI
10:28
dev shops or some, you know, kids come out of like school and
10:31
be like, hey, I could just ChatGPT this stuff and kind of
10:34
vibe code slash whatever. Like do you feel there'll be
10:37
price pressure in the market? If you're like, hey, our price
10:39
is still $10 and somebody else is able to build it and wants to
10:42
sell it for 2. And then you're like, well, like
10:44
do you think that'll compress like margins or sort of like
10:47
more of like, you know what, what do you see this going?
10:49
Again, I think it depends on the, on the agency, right?
10:52
So I, I can only speak for jet rockets, of course, you know,
10:55
right. So and jet rockets, we don't, we
10:58
don't sell lines of code. We sell value.
11:00
And the value is in our knowledge and our expertise And
11:03
the fact that if you come to us and you hire us, we're going to
11:05
build you a platform that's going to be secure and scalable
11:08
and you know, all those great things.
11:09
And we're going to do it collectively.
11:11
I don't charge. Oh, you're going to get 2
11:12
engineers, 1 is yeah, yeah, none of that exists.
11:16
Yeah. I don't compete with two
11:19
teenagers who are vibe coding, so let them charge you 10 bucks
11:22
an hour and then enjoy the results of their work.
11:25
Yeah, yeah, codecs did this in like few minutes and it took you
11:27
10 and you had bugs. Look, I I think that's right and
11:30
like I've heard of the derogatory slur about AI device.
11:33
We can't say it on the show, it's just too derogatory.
11:35
But like, you know, it, all jokes aside, like I think what
11:38
we're getting at here, Natalie, is the future is very weird
11:41
right now. Like the idea that all you know,
11:44
the media loves to tell scary stories, right, because they it
11:47
bleeds, it leads. So it's like AI is coming for
11:49
everybody's jobs, right? And I'm talking to two technical
11:51
people right now and I'm the non-technical guy and I'm in
11:54
this world where I can go into whether it's Kodak's with from
11:57
open AI or it's it's clawed from anthropic or it's my lousy
12:02
lovable web app builder. And it's not lousy.
12:03
I like lovable, but I can use any of those things and I can
12:05
take my idea and turn into the thing.
12:07
I think the point you're making though is like there is there is
12:10
inherent risk in what I'm going to get in terms of a something
12:13
that's sales and I can take to market and I could actually
12:15
commercialize. And then if I'm a bigger
12:18
enterprise that wants enterprise grade security and compliance
12:21
requirements, that's basically all out the window.
12:23
But at some point to Kevin's questions for you, doesn't it
12:25
catch up? Don't we get there where like AI
12:28
can do a lot of that or maybe all that stuff And then it's
12:30
like, what is really the human value that these human engineers
12:35
with human experience? Because AI at some point is
12:36
going to hit a GIASI and be able to do anything better than a
12:40
human can do it at some point, presumably.
12:43
So let's let's dig into that because I think that's the
12:45
philosophical conversation that's underlying.
12:48
This is like, what do you believe of the future of super
12:50
intelligence? Where are we going to get to and
12:52
what's your timeline for it? So.
12:54
Yes, thank you. So first of all, again, I don't
12:56
know what's happening behind the closed doors of all these
12:58
Silicon Valley, you know, operations, right?
13:01
I'm not Privy to that. But I'm just talking about the
13:04
mainstream, the tools that are available to us right now.
13:06
None of it is AI. There's no intelligence, there's
13:09
no I in the AII would argue that there's pattern recognition,
13:13
there's repatriveness, There is, you know, projection models.
13:19
Absolutely, yes. Think about it this way, Like
13:22
the code that it generates, how does it know what code to
13:25
generate? Because it learned on the code
13:27
available on the Internet, right?
13:29
The more it's going to generate garbage, which it has been
13:32
generating garbage, the worse the code quality is going to get
13:35
over time. I mean, look at the content that
13:38
we're now seeing a LinkedIn everywhere else, right?
13:41
Like most people use AI to generate their LinkedIn post and
13:44
you can sell. You can like the first 3 seconds
13:46
you can sell, oh, this is garbage put out by some sort of
13:49
AI, right? And this, the same applies to,
13:52
to, to the code that it generates.
13:53
Now again, and I'm going to repeat this, AI has its place
13:57
even in today's engineering and coding practices.
14:00
If you know what you're doing, you can use it as sort of like,
14:04
like an assistant, right? Like a junior engineer who's
14:07
sitting next to you and who's willing to do boring, intidious
14:10
tasks that you do not want to do.
14:12
And if you know how to prompt it properly and if you can check
14:15
its work and if you can fix its arrows, then you're fine.
14:19
Then you can use it and you can become more, you know,
14:21
efficient, more productive and so on and so forth.
14:24
Now, where is AI going to be on October 6th, 2027 or 2028?
14:29
Who the hell knows, right? I mean, we may have this
14:31
conversation a year from now and I'm going to say I take it all
14:33
back, but I'm talking about today, right?
14:36
I first of all, I don't believe that if you've built any
14:39
products for anyone in the past or in your life, you know, so
14:42
that the biggest issue is understanding what the client
14:45
wants, because often times the clients don't know themselves
14:48
what they want, right? So being able to clearly
14:51
understand the problem and clearly articulate the solution
14:54
is the biggest value, for instance, that Jet rockets
14:56
brings to his clients, right? So I think that yes, absolutely
15:01
we're going to use Lovable and droplet and like other tools to
15:05
generate commodity apps that, you know, sort of like input in
15:09
output, you know, data in data out, something very basic.
15:12
Maybe it'd be like, I don't know, like a wish list registry
15:16
for your wedding, you know, that you want to send to your friends
15:18
and family instead of using something online and stuff like
15:21
that. Something very basic, simple
15:23
that does not require a whole lot of thinking, processing, you
15:26
know, what if scenarios, etcetera.
15:28
The moment things get a little bit more complex, AI is so far
15:32
away from even like getting anything done right now.
15:36
I guess I made it wrong. Well, we're.
15:38
Going to find out together exactly.
15:40
We'll have to do a follow up conversation.
15:42
I completely agree with you. I think the truth is somewhere
15:45
in the middle, right? So it's as often in life.
15:47
But I'm going to use my my CTO. He is by far the strongest
15:52
engineer and he is the hands on and.
15:53
Your CTO still writes a lot of code because you guys are so
15:56
like. I said we're very small and my
15:58
CTO is active on several projects.
16:00
He overseas the code, he makes all of our architectural
16:02
decisions, etcetera and so on and so forth.
16:04
So like he's, he's very much, you know, very much hands on.
16:07
As a matter of fact, he enjoys the hands on.
16:09
Got it. Yeah, exactly.
16:11
So he's been using a assistance since I mean since they became
16:15
widely available. I think at first he started
16:17
using copilot built into his Idi think that was over a year ago,
16:21
maybe because he was one of the he's a contributor to many open
16:24
source projects. So he got it for free whatever.
16:27
So we used that. We tried Devon then cursor was a
16:30
tool and we still have a few people who prefer cursor.
16:33
But my city on again talking about him.
16:36
He loves Claude Claude code. That's what he says in the best,
16:40
by far the most advanced and can, you know, produce the most
16:44
quality output, right. So that's that's the tool that
16:47
he prefers now and of course uses ChatGPT for like other
16:50
things, but like Claude code that that that is the tool that
16:54
is embedded now into his daily operation.
16:57
I can tell you what he kind of finds, what are the shortfalls
17:01
of of the code, and then I'll tell you how he actually uses.
17:03
What would you say like percentage wise if you know?
17:05
I'd say about 30. Percent 30% of the code
17:07
generated. Yeah, about, well, it's not
17:10
about 30% of generated code. It's about I was going to answer
17:13
a different question. I should have listened to you.
17:15
I'm sorry, about 30% improvement in his efficiency.
17:19
That's how he would say like, I'm about 30% more efficient,
17:23
more. But all jokes aside, like, you
17:25
know, when we talk about these 10X engineers or 100 X engineer
17:29
and giving them, you know, I'm going to use Kevin's analogy
17:32
going into the gym and like, OK, you know, you can lift weights.
17:35
But like, here's a machine over here that if you use this
17:37
machine, like your ability to lift weights and how quickly you
17:42
will progress in the amount of weight you can put up is going
17:45
to, is just going to skyrocket like a 30% improvement.
17:47
It's like, OK, like 30% more efficient.
17:49
That's not bad. I just, I talked to folks in the
17:51
Bay all the time that are like, oh, I was, you know, I was doing
17:55
this much work and now I can do like a 10X engineer is now doing
17:58
two or three or 4X the amount of output.
18:02
I would love to meet a person who's actually.
18:04
I mean, you're looking at one of them like.
18:06
Hey, I want to be able to automate a process.
18:09
Yeah, yeah, by all means, I mean absolutely, you can automate any
18:13
process very quickly and very rapidly nowadays.
18:16
Now take an application for instance, right.
18:18
That and I'm going to use an actual example.
18:20
I'm going to use an application of one of our clients.
18:23
The client name is Safari Portal.
18:25
You can check it out on Safari Portal dot app.
18:27
It is an enterprise level end to end cure operator managing
18:32
platform. It does everything like from
18:34
building beautiful custom itineraries for, you know, for,
18:38
for customers to providing sort of booking options and
18:42
commissions and say like everything, it's like a CRI,
18:45
everything is built into it, right?
18:46
And we continuously build it and develop it.
18:49
And it's been developed in development for the past 5-6
18:52
years. Have you been working on it from
18:53
the start or did you join kind of mid?
18:55
Yes. No, no, no, that's we build it
18:57
from scratch. Yeah, it's an end, right.
18:59
We build it from start. It's an award-winning tool.
19:01
It's probably the best on the market right now in this
19:03
particular industry. It's it's, it's a great product
19:06
and our client is making a lot of money.
19:07
Everything is great now and the way we're building it right the
19:10
way it's been architected, it allows us to build new features
19:13
without really affecting a lot of like old legacy code.
19:16
So it's not like every time we go in, we have to like go
19:19
through some legacy and you know five years of code.
19:22
It's it's pretty legacy right now, but it has a lot of logic
19:25
that it has to consider right when when building something
19:28
like this. And here's what happens.
19:30
What we see is AI very often ignores existing components and
19:34
duplicates code instead of using it, right?
19:36
So now my engineer is going to say, oh great, you've produced
19:39
all these garbage, you've produced 10,000 lines of code.
19:42
But I didn't need it because I have a code base that you could
19:45
have referred to, right? So now he has to go clean it all
19:48
up, etcetera. It would have been faster for
19:50
him to do it without AI, right? It often forgets like earlier
19:55
instructions, like mid session context.
19:57
And it forces right? It forces the engineer to
20:00
constantly repeat themselves over and over and over and over.
20:02
Engineers don't like doing that. That's a time kill, but that's a
20:05
time kill. Forget liking versus not versus
20:08
not. Like, yeah, it's going to kill
20:10
the time, right? No, I get it.
20:11
And you're paying for this time. Look, I think, I think Natalie,
20:15
the point is well made, which is, you know, people matter is
20:18
not the solution and AGIASI is like is not here.
20:22
And so I think I, I like, I come back to the quote you said,
20:25
which is like the AI we're missing the I in it.
20:27
And like, I think Kevin and I are very bullish on the on the I
20:31
part, but I I agree, like the truth is somewhere in the middle
20:33
and that endurance mindset has to be helpful here.
20:36
Like how do you maybe you're on a 26.2 mile run, You've got to
20:40
be thinking about other things. That's a lot of time.
20:43
So, so how do you stay sharp and how do you use that endurance
20:46
mindset when it when it comes to the work that you do and, and,
20:49
and kind of this ever changing, very fast moving landscape?
20:52
Well, I think this is a perfect analogy because like in
20:55
marathon, you have to ignore, ignore the hype, ignore the
20:58
case. You know, you go out way too
21:00
fast. You're not going to be able to
21:01
hold it for 26.2 miles. So being able to stay
21:04
consistent, being able to ignore the hype and being, you know, it
21:08
will allow you to outlast all these spring chasers that you
21:11
see everywhere around, you know, social media now, right?
21:14
I mean, we've seen it happen in the past.
21:16
Yeah, sort of, you know, I mean, there was this whole like been
21:19
long enough, you know, in this in the streets.
21:21
You have gone through several of these hype cycles.
21:25
And yes, this one is probably more significant than others.
21:28
100% I would not argue. And again, I just want to
21:31
clarify, I'm not a skept like I'm a healthy skeptic.
21:34
We embrace AI, we use AI, but we're not going to lose our head
21:37
over it and we're not going to shut our doors because Oh my
21:40
God, everything is agentic now. No, it's not.
21:43
This is this has been awesome. Natalie, where can people find
21:46
you online or connect with you or your companies?
21:48
Yes. So you can find me online at on
21:50
LinkedIn. Actually, I'm quite active
21:52
there. My name is Natalie Kaminski.
21:54
You can search for me. You can e-mail me
21:56
natalie@jetrockets.com and come and visit
21:58
ourwebsite@jetrockets.com. Thank you.
22:00
Awesome, Natalie, thank you so much for coming.
22:03
This was, you know, very clear on a fun kind of dynamic
22:06
conversation and and lots of interesting insight from you.
22:08
So just appreciate you coming on the show.
22:10
Thank you for the opportunity. Yeah, enjoy that.
22:13
So that was that was cool. I mean, Natalie definitely comes
22:16
into this with a unique perspective, a conversation that
22:19
you and I don't usually have on this show, frankly.
22:21
And, you know, I'd love to hear you try to repeat back and
22:24
summarize Kevin, the the takeaways from from Natalie's
22:27
perspective. Yeah.
22:28
And so I think like I said, we probably don't agree on
22:30
everything, which is cool, I think to hear sort of an
22:32
alternate perspective of like how much AI and how deep you
22:35
should go. But I think, you know, some of
22:37
the things we hit on that kind of came from her side were like,
22:40
you know, developers matter and that, you know, sort of AI is an
22:42
augment and so it doesn't really replace and then guardrails, you
22:45
know, junior engineers shouldn't use AI at all.
22:48
And that, you know, humans need to do, you know, the majority of
22:50
the review or at least the final sign off and that real.
22:52
Contradiction to your to your strong position.
22:54
Give everybody the tools. But I think an interesting point
22:56
she made. Yeah, I think a little different
22:58
there, I think, but I think the the real gain that she's seeing
23:01
is saying like, hey, the top kind of best engineers are
23:03
getting a 30% efficiency. Yeah.
23:05
And then ignore the hype really, right, That endurance beats turn
23:08
chasing. And I think, you know, I think
23:09
it would be fun to sort of revisit or think back, you know,
23:11
where does this like put us in a year from now and really try to
23:14
like, you know, see how that sort of how some of these
23:16
predictions and some of these, you know, our opinions may
23:18
change either way. Yeah, nobody has a crystal ball
23:21
for sure. If this shifted how you think
23:23
about AI and product development, send it to someone
23:25
trying to build smarter, not just faster and.
23:28
If you're enjoying these conversations, tap the subscribe
23:30
button. A review share goes a long way.
23:32
And that's it for this one. Keep building, but don't forget
23:35
shortcuts only take you as far as your fundamentals.