Exploring Vibe Coding and the Future of Product Management with Gunnar Berger - JSJ 690 - JavaScript Jabber

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:05):
Hello everybody, and welcome to another exciting episode of JavaScript Jabber.
I am Steve Edwards, a host with the face for
radio and the voice for being your mind. Amine, excuse me,
but I'm the host at least I am. Today with
me on the panel, I have coming with us live
from Tel Aviv, Israel, mister Dan Shapire.

Speaker 2 (00:22):
How are you doing, Dan, I'm doing well.

Speaker 3 (00:24):
I'm perfectly fine with you being my mind. I could
use the mind.

Speaker 1 (00:29):
Yes, mine is a terrible thing to waste, as they
used to say. So, how's weather in Tel Aviv? We
always have to talk about this. It's still warm and hot.

Speaker 3 (00:38):
Yeah, well, I don't know about lovely, but it is hot.

Speaker 2 (00:44):
It's a it's a.

Speaker 3 (00:45):
It's humid because we're on the Mediterranean's humunity too. Okay, yeah,
but on the other hand, we've got the beach, so,
you know, some of the nicer things that we have
among the also lots of not so nice things we
recently have. But that's a topic for a different discuss.

Speaker 1 (01:00):
Yes, exactly. And coming as our special guest today is
mister gunner Berger. How you doing Gunner? All right, I'm
doing great. While these voice messages come on my phone,
so we are here to talk today about vibe coding,
that hot topic across the web and across the world

(01:22):
these days, it's at least in the tech space. Before
we get into the topic, Gunner, can you give us
a little backgrounded on yourself?

Speaker 2 (01:30):
Sure, full background or vibe coding background or I'm.

Speaker 1 (01:34):
Whatever you deem relevant to the topic at hand.

Speaker 4 (01:38):
All right, sure, I'm I'm a true nerd, been a
nerd for twenty plus years, been in tech industry for
twenty plus years. First half my career is centered it traditional.
It switched over about fifteen years ago. I was a
Gartner analyst for a while. I covered in computing. It's
worked Dan and I met.

Speaker 2 (02:00):
I went over to Citrics, was there for about six years.

Speaker 4 (02:03):
Have been currently now at Amazon. Been here about six
years and I've been leading product in some form another.
It constantly changes, but I've been doing some form of
products for a while now. At CIDRICSS, I would say
more new product initiatives and PI. I always kind of
choked at our CEO, Mark Templeton would always have like

(02:25):
and one more thing on stage at Synergy that's typically
where the CTO department would have something to do with
that and one more thing. So I did a lot
of MPI mergency acquisitions, that kind of stuff at Citric
and yeah, and I jumped over to product, which is
where I've been ever since. Yeah, and so I'm I'm

(02:47):
really interested in this topic because I think the definition
of product manager or at Amazon we hire product PMT
product manager technicals, and I lead a team of these professionals,
and I think that that definition is changing rapidly because
of what AI is empowering product managers to do. So Yeah,

(03:09):
I thought it's really fun to have a conversation with
technical people about this coming as kind of a I'm
a nerd technical, but I'm definitely not like a developer technical.

Speaker 3 (03:18):
So I would like to add to that, Soner. If
you recall, we kind of started talking about you coming
on the show after I posted on X kind of
a shout out inviting people who are involved with vibe
coding in some form or shape to come on our
show to talk about it. And about a month ago
we had Anthony Compolo talking about vibe coding from the

(03:42):
developer perspective. So I thought it would be great to
have somebody coming at it from a different perspective. One
of product management because I think again, looking at product
managers in the company that I work at, Size Sense
and other companies I talk to other people who are
in product management, I do see it making a whole

(04:04):
lot of impact. And I do think that one of
the most impacted fields are developers and product people, but
they're impacted in somewhat different ways, although there are a
lot of similarities. So I thought your your take on this,
your view, your point of view, would be very pertinent
to our listeners.

Speaker 4 (04:24):
I think it's interesting that even I'm assuming your listeners
are typical like in the development world, which is exactly
kind of what I'm seeing happen within the product management worlds,
Like those worlds are crossing more because it's almost like
we spoke two different languages in the past, right, I
speak the language the customer, and where I work we
are very big writing culture, so I'm used to like

(04:46):
great stories you go deep to the customer that world.
But because of what this new technology has created, it's
almost like having a translator with us at all times,
where now I'm speaking developer language, even though I'm not right,
I'm I'm vibing, and I think that has has massive

(05:08):
change to how we approach building products.

Speaker 2 (05:12):
Uh, throughout the industry, and I think everyone.

Speaker 4 (05:15):
A conversation I've had with a lot of my you
know buddies at work is if you had to start
a new company today, would it look the same as
if you did it five years ago? And I argued
that it would not, like you know we have. I'm
sure we'll talk about what is the role of junior devs?
How do you get senior devs in a world where
junior devs don't have a job anymore? Like, Uh, there's
so many different directions we can come to this, but

(05:38):
let me let me get one more dig in before
you jumps some work. I do want to have like
a legal note here.

Speaker 2 (05:45):
I'm Gunner. I'm just here talking about my experience.

Speaker 4 (05:49):
I in no way in speaking on behalf of my
company at all, So please don't edit that boasts. Just
make sure I'm required to have like a legal disclaimer
that I'm speaking on behalf of myself and experience.

Speaker 2 (06:01):
I haven't product well, I don't know.

Speaker 1 (06:03):
We just got a response to our sweet on job
script jabber from the head of AWS that said tell
Gunner to be careful.

Speaker 2 (06:09):
No, I'm kidding turn go ahead, turning my back.

Speaker 3 (06:12):
Yeah, I do have to say that it's it's essentially
the same for us. We're all here talking on behalf
of ourselves, not representing our employers. It's uh, you know,
take our opinions for what they're worth.

Speaker 2 (06:27):
Uh.

Speaker 3 (06:28):
Interestingly, I recently spoke with one of our senior developer
slash architects at size Sense, and he's one of the
people that has really bought into AI. He's the one
using up all the tokens in his development environment. And
the way he talked about it is how he feels

(06:49):
in a lot of ways like he is becoming kind
of product working with a team of junior devs that
he's instructing on what they should be doing doing and
then critiquing their work.

Speaker 2 (07:03):
So it's it. It is very.

Speaker 3 (07:06):
Interesting that we are coming at it from different perspectives,
but it seems that they are kind of converging on
on on this kind of a central place. But again,
I'll hand it over to you to get your opinion
on this.

Speaker 4 (07:21):
Yeah, I mean, it's what is the output that a
product manager should have these days? Like I still think
the input is very much the same, Like you need
to go deep with your customers, like be customer obsessed.
Understand what the problem is that you're trying to solve.
Like that's something that is still very firmly a product
manager role.

Speaker 2 (07:40):
But then it's what's the output.

Speaker 4 (07:42):
Well, previously the output is you know, writing a PRD,
you know, a business requirement doc, product requirement docs, product
feature request, whatever, it's it's in the world that I
live in, as always some form of a document that
then we have some paperwork, right, and we have a meeting,
and I have a meeting with you my head of
engineering or you know, some SDEs whoever it is with

(08:03):
someone on the development side, and we have conversations, they
have questions. So then I go back and I modify
my documented. Maybe I'm doing focus group testing. Maybe I
have a UX team coming in and uh sharing you know,
their perfection how this thing should look right, and just
this is just how it works.

Speaker 2 (08:18):
This is you know, in the industry, that's just this
is normal product manager one of one stuff.

Speaker 4 (08:23):
And so we spend weeks, if not months, in meetings
and doc reviews and back and forth, back and forth,
back and forth in what I like to call agile
full because it's neither actual nor waterfall. But you're doing
this over and over and over and over. Whereas that
iteration changes, everything changes. When I can just go into

(08:48):
a tool like Cursor, like Caro, like you know vs
Code with some MCP's running, Like there's so many tools
out there right now, I can go on a tool
and literally.

Speaker 2 (08:59):
Just like bullet point my thoughts.

Speaker 4 (09:01):
It's like this is this is this, and then have
that tool take my thoughts, stretch them out and either
write me a better BRD that makes me think of,
oh I didn't think about the yeah QA testing, I
should have done that. Yeah, And it thinks about that
for me just fantastic, because then I add more questions.
Now I'm having this back and forth from a product
manager perspective with the AI on what are the requirements? Like, fully,

(09:26):
what am I? What am I missing? Just ask the
prompt what am I missing? And it it's like, well,
did you think, well this is this? So you have
these better docs that are written in much much faster times.
But and I think, more importantly, I don't need to
generate a DOC.

Speaker 2 (09:39):
I still do.

Speaker 4 (09:40):
That's still kind of the expectation, but more often than not,
I'm actually more interested in building a prototype, and you know,
working in sandbox environments and saying, you know what, I
have this idea, let's.

Speaker 2 (09:52):
Just run with it.

Speaker 4 (09:53):
And I don't have I don't need a code review,
I don't need security all this kind of stuff because
I'm not a developer, right, But now I can develop
and I can actually show a working system to my
development team, where now the handoff is not necessarily just
a document, it is a working experience prototype that I vibed,

(10:16):
and I think that's where things change. And then we
have this whole question of like what does that mean
for developers? How many developers you need? If you're able
to do quick prototyping like it used to be, I
would maybe write a one pager up, send it to
a junior dev. The dev like whips up a prototype
in a sprint. We look at the prototype like that's
we're talking weeks of time down to an afternoon. For me,

(10:39):
we're talking less communication errors because now I'm actually showing you,
not just telling you. I've said this multiple times with
never on a podcast. It's you know, there's this phrase
that a picture it's worth a thousand words, but a
prototype is worth a thousand meetings. Just showing me the
thing that you're trying to build, and it shouldn't matter
that the code is not that good, Like it doesn't matter.

(11:02):
What matters from a product standpoint. It's because I'm not
I'm not trying to say build it like this as
far as like the code goes. I'm saying, build it
like this because this is how I fundamentally want this
thing to work.

Speaker 1 (11:14):
So are you saying then that it's a it's from
a Okay, you're saying so the underlying code may not
be what they want, but the UI, the appearance, the flow,
this is what you're trying to give them. The idea
is gives me. The ideas is the is the says,
is it very close to what you want? As an
end product ors? You'r I would assume this is just

(11:36):
sort of okay, this is a general idea of what
I'm looking for. It's not going to be you know,
fit and finish isn't going to be there. It's just
sort of a general, yeah, this is what I want
it to do, so you can see it instead of
me'b describe it on paper.

Speaker 4 (11:48):
That's just it. It depends. I would agree with what
you said. The fit and finish doesn't necessarily have to
be there. But at the same time, it depends how
deep we go into using these AI agents.

Speaker 2 (11:58):
Okay, I gotta be someone careful.

Speaker 4 (12:00):
Let's just say there's an MCP that exists that has
my UI framework built into it. Well, now my fit
and finish is my fit and finish.

Speaker 2 (12:09):
Now is the code fit? I don't know.

Speaker 4 (12:13):
I'm not intelligent enough to tell you if the code
is right, but I can tell you that the fit
and finish like the when I pull it up on.

Speaker 2 (12:19):
You know, a browser.

Speaker 4 (12:21):
Yeah, that's actually what the end result should look like
because I can use an AI that takes like a
figma and turns it into working code that's public information,
right doing taking advantage of these tools out there, Like now,
all of a sudden, like my UI is my UI,
even though it maybe only took me an afternoon to
do it, I actually have a rich UI using the components,

(12:44):
using everything that I wanted to use. But again, I
do want to draw a really firm line that I'm
not saying that the actual code makes any sense.

Speaker 1 (12:54):
That right, for sure?

Speaker 2 (12:55):
Yeah, we get that, but the fit and finish might
depending on the tools you're using.

Speaker 3 (13:00):
A few questions about that. So, first of all, I
totally agree with you and understand the concept that the
code is not production ready. For example, it might not
be secure, it might have you know, it might not
require our log in or anything. It just it just
looks like, you know, there might not be issues like
permissions and all this stuff, although you might actually vibe

(13:22):
code some of that in order to show how that works,
if that's part of what you're designing. I totally get that,
and I totally agree with that. I am curious about
some other aspects though. So first of all, how likely
are you actually going to be to try to actually
connect it to back end data versus just mocking some data?

(13:48):
So that would be one question. Another question, you mentioned
that you can take Figma designs, but will you actually
even be waiting for the UX people to you have
the designs ready or will you also kind of quote
unquote jump the gun on them and say this is
kind of I'm not like, I'm not a developer. I'm

(14:10):
also not a UX expert, but this is kind of
like how I envision it working. I don't know if
this is the proper color scheme, if this is like
it has accessibility and stuff like that. I just want
to show you how I feel it should kind of look.
So those are the two questions that I have.

Speaker 4 (14:32):
I come from a position of privilege here. I just
want to be really clear about this. I have a
UI framework, and in fact, I can pick from a
set of UI frameworks that have a wonderful UX team
behind them that are built on it. So your question
of like as a product manager, when I just like, hey,

(14:53):
this is how it works, honestly, like I would point
it to an MCP of like, hey, using this framework,
right to talk to this MVP about that framework, I
can within a few minutes, I can have something like
I don't know if you've messed with.

Speaker 2 (15:06):
Like lovable AI or these other ways.

Speaker 4 (15:08):
There's a lot of these guys out there where they
can actually give you the fit and finish that is
aligned to your standards. So even accessibility, this is actually
one of that I'm glad you brought that up, Like
that's a common one.

Speaker 2 (15:21):
It's like, oh, well, you know, a product manager.

Speaker 4 (15:23):
Doesn't necessarily think about that because they're just thinking like
the end customer problem is, Dan needs to talk to Steve,
and this is how we got to make it work right,
and they're not thinking about light mid dark mode and
whatever else is going on within the Accessibility to I framework.
But that's where I point to, like that may be
true and how a product manager thinks today. It's like, hey,
I'm just going to solve this problem, but these tools

(15:46):
make it so even though I'm not necessary to think
about it, they're embedded into it. Even your comment Steve
about security, if you're using the right tool, if you're
using the right I don't want to call it prompt
because they're like the prompts before the prompts, I forgot
what they're called the rules.

Speaker 2 (16:00):
I think that's what they're rules.

Speaker 4 (16:02):
Like if you have the right rules in there is
like and when you respond, like think about securities, think
about this, think about that, like when you have those
embedded in the background, even though I might not me
thinking about it, which is what I love about it.
By the way, it's like, I'm just thinking, I want
to get Dan to talk to Steve using my new
in semessenger app.

Speaker 2 (16:18):
I'm making it up.

Speaker 4 (16:20):
It's like cool, Okay, well I didn't think about this,
that and the other, but it did. And that's where
I was saying earlier is like where I used to
write a BRD and I have to think about every
single edge case scenario. Getting a meeting is like, Gunnard,
did you think about this edge case?

Speaker 2 (16:35):
And then I get mad.

Speaker 4 (16:36):
It's like, my gosh, we're going to be naval gazing
for the next four months about every edge case. Well,
now I can throw it in here and it's it's
got doing a very good job. Like these tools six
months ago to today are just completely different as far
as I'm concerned. Like I remember trying to do this
about six months ago and I'm like copying and pasting
stuff into like chat, GBT or Rock two five or

(16:59):
whatever it was at the time.

Speaker 2 (17:00):
Now you know it's fully embedded.

Speaker 4 (17:02):
It's thinking the latest one that I'm very impressed with
this amount of QA testing that it built. I've never
once asked it to build a QA test for this
feature that I'm biting. But these tools are as they're
getting more mature, they're thinking about scale, They're doing all
this stuff where it is thinking about accessibility, is thinking
about security, is thinking about how do you test this

(17:23):
stuff at scale so you automatically alert when things are
working properly or latency is getting too high or you
know you're.

Speaker 3 (17:31):
Yeah, but even when you do have a design system,
UX stood as a job. I mean, you know how
to organize stuff on the page, how to break things
between pages.

Speaker 4 (17:46):
That's where I say it came from a position of
privilege in that, yes, you exit the job. So I
have a rich library of components that I can talk
to an AI and have it use those commonents. Now
the question is what happens when you come up with
an idea that doesn't it within that library component? My
position privilege is, well, then I that would be kind
of traditional product management of like, hey, go talk to

(18:07):
the UX team, have a meeting.

Speaker 2 (18:09):
This is what I'm thinking. They build a component and
so yeah, that would be maybe I'm moving back into
the traditional world.

Speaker 4 (18:16):
But you only have so much of that because once
you built this library up, I can in that library
is understood by AI.

Speaker 2 (18:25):
It just doesn't.

Speaker 3 (18:26):
So in effect, you've not only cut kind of the
developer out of building the MVP, you've also cut the
designer out of building the MVP. You could, let's let's
put it bluntly.

Speaker 4 (18:43):
I would say compress. We've compressed the amount of designers
we've needed. We've compressed the amount of developers we've needed
within this process of creation. It just allows me to
iterate much quicker now. I think you actually kind of
hit an interesting point. And I think AI in general,
AI is only backwards facing, right, it can only tell

(19:04):
you what it's been taught. I think it struggles more
with like, hey, I want to do something that's new
and different. I have a much harder time and I'm like, hey,
I want a UI component that does this. Those prompts
have done that, Like those are painful, whereas it's lot
easier when if it's something that's more common, like hey,
this thing don't exist, it can do that no time.

Speaker 2 (19:25):
But thinking I don't know. At the same time, you see.

Speaker 4 (19:28):
These freaking crock AI videos that do interesting stuff. But
in my experience that the creation side of it that's
like new and different, I find that to be a
little bit.

Speaker 3 (19:41):
I had that amusing situation recently. I had I was
trying to get and I was playing around with a
new library that does a certain thing, an alternative to
an existing library that we're using, and I asked it
to build a mall application for me using that does

(20:01):
a certain thing using this new library, And I was going.
Cursor was chugging along, and a certain point it says,
I failed, I'm going to revert to that old library,
which I found to be really amusing. It's kind of
almost a human type response in a sense. So I
stopped it because that the whole point was building the

(20:22):
thing using the new library, not just doing it again using.

Speaker 2 (20:25):
The old library first.

Speaker 4 (20:27):
Or deleted my entire front end because it got tired
of my type script airs the entire front and said
compiled successfully.

Speaker 2 (20:33):
Your airs are gone.

Speaker 4 (20:34):
And I literally tweeted at the time, I said, a
I will kill us all. Don't ever tell it to
fix the human condition.

Speaker 3 (20:41):
Yeah, sometimes sometimes it, you know, there was this kind
of funny story, it's it's a bit old about back
in the COVID days, they were looking to use AI
to analyze X rays chest x rays to try to
figure out to see I kind of a doctor assistant
trying to figure out based on the X ray if

(21:02):
that person has COVID or not. And they came out
with some amusing results. So, for example, it turned out
that a lot of the healthy people were the picture
was taken while they were standing up, while people were sick.
Pictures were taking while they were laying down, and the
chest is kind of positioned differently when you were laying down,

(21:23):
So basically they I based its decision on whether they
were standing up or laying down.

Speaker 2 (21:29):
Such for machine learning use case, I'm like, generally they either.

Speaker 3 (21:32):
Yeah, I know, but it's kind of like, of course,
I was just giving it as an example with that.
With these systems, you all don't always get what you
expect to get. Sometimes you get funny shortcuts like you mentioned.
But fixing all the errors by deleting all.

Speaker 1 (21:48):
The farms, well, basically your computers do what you tell them,
do you know what we want them to do? Is
the adage I've always Yeah, I.

Speaker 3 (21:56):
Had the problem with my kids as well.

Speaker 1 (21:58):
By the way, that's amen to that, brother, Amen Gunner.
I wanted to ask you a lot of this has
been pretty high level, and I'm sort of a nuts
and bolts kinds of guy, so I'm curious if you
could walk through this, maybe the steps and the tools
as much as you can to. When you're vibe coding,

(22:18):
you know how you're creating, what are you opening up?
What do you talk to me? What are you typing into?
How is it giving you your output? Sort of hard
to do audio, you know, without the visual I'm sure,
and that would be very nice. But I'm just curious
to see what the actual mechanics are of vibe coding,
because everything I've heard had been is a high level.

(22:40):
Oh hey, it's as great. I can tell to do
this and it gives me this, Okay, great, how do
you do it?

Speaker 4 (22:45):
I I have quite a few friends that are engineers, developers,
and they all laugh at me on my text chains
these days because I'm learning things that developers learned ages ago. Right,
it's not in my world as a perfect example, and
the thing is, like I spent ten years in it,
I should know better. Is the answer to some of

(23:06):
this stuff, But just something as simple as using Git.
I didn't bother using Git until it deleted my entire
front end and I had no backup.

Speaker 2 (23:14):
Damn it.

Speaker 4 (23:15):
I spent a couple of days on the front end
and literally I couldn't get it back.

Speaker 2 (23:19):
It was just John.

Speaker 3 (23:20):
I have to interject that I've come to a similar
conclusion that Git is one of the is like the
thing to use together with five coding slash AI coding,
because it usually asks you do you want to keep
something or reject it, but I found that it's easier
to just mostly accept it, use see that it works,

(23:44):
and do comparisons using git, and if it's bad, then
just revert using Git.

Speaker 4 (23:50):
Yes, because there you're talking about curser.

Speaker 2 (23:53):
I did a lot with Kurser.

Speaker 4 (23:55):
I have more recently moved to clock code because cursor
just I don't know what they're ll and they're using
as band scene. I don't think they're using clots on it,
but it's just not great. Like I just I'm sure
you guys have had the experience where like I'm gonna
I got promise a okay to fixate, I got to
go change B B broke, C C broke, and then

(24:16):
you're just watching cursor go in circles for an hour,
even though you can read it literally like even though
I'm not a developer to answer your questions, see like.

Speaker 2 (24:23):
How do I'm kind of get in the middle of it,
but how do I actually do it?

Speaker 4 (24:28):
Is I'm reading the AI, I'm reading NonStop of what
it's doing. Again, I'm not a developer, so what am
I doing is like it's code code, code, code code,
But then I start to see things like tenant ID
and it starts to show up often enough, and I
start having airs often enough where I have actually caught
it doing the wrong thing, like from a developed code standpoint,

(24:51):
where it's like I did one five thing where I'm like,
I was doing a single database using rols and then
I switch it to like one database per tenant to
make the permissions easier by the way, things I couldn't
say a month ago, and in that I could say,
it's like I did this migration between these two database
search for multi tenancy, and I could see it like

(25:13):
doing it wrong, and so to interject and like no,
like that variable doesn't exist anymore, like you're doing it wrong,
and they're like oh, and then it changed it and
I literally helped it troubleshoot its own stupid air cursor
drives me crazy. Cloud Code have had far, far fewer
actual coding errors as I'm using the software. But anyways,
so how do I use this stuff? Well, I've tested

(25:36):
a lot of different AI systems. The one that's currently
on my screen is visual Studio, and I'm just using
cloud Code via the CLI, which is how clock code works.
It's got a great in it function where we'll actually
look at your entire project, and in my opinion, it's
the best at actually understanding the project. And yes, I've

(25:57):
done curser from from zero to one in cursor. Cursor
is very good at this and understanding your project as well.
But I've just got too mad at the Cursor circular
chasing of errors.

Speaker 2 (26:11):
That just drove me crazy. I have. I've found that
Claude is much much or cloud code is much better
at that. Unless is no plug for anyone, I don't
really care.

Speaker 1 (26:19):
So cloud code is a terminal application basically, if I'm
looking at it right, I haven't used it yet. Yeah, Like,
it's not like an I d ee. You open up
and here's your files and here's your code. Net is
basically command Linehere you're typing in text.

Speaker 2 (26:30):
Hey do this for me?

Speaker 4 (26:32):
Yes, I mean I could screen share it with you,
but I'm literally doing a terminal in the side of
my ide.

Speaker 2 (26:36):
Right.

Speaker 4 (26:37):
So I'm in there at typing Claude. Boom, it launches,
I do slash a knit. It then goes and reads
the entire Fulder structure that I'm that I'm in uh,
and then it creates a empty file.

Speaker 2 (26:49):
And it's just very similar to Cursor, right, Uh.

Speaker 4 (26:52):
Curser's just looking at that fuller UH structure, and it's
creating its own documentation of what it sees.

Speaker 2 (27:00):
A Cursor and Claude are doing very small things.

Speaker 4 (27:03):
I wish Claude actually had that more id rich integration.
I like the UI of Cursor.

Speaker 2 (27:10):
Better than I like the UI of Claude, but I will.

Speaker 4 (27:13):
Say I like the results of Claude better, so I'll
deal with the small amount of changes there.

Speaker 1 (27:18):
So you said it reads your fulder structure. Do you
have to have a Fulder structure in place before you
start or are you starting with basically nothing that says here,
I'm going to create this project.

Speaker 2 (27:27):
Yeah, yes, you can always start playing.

Speaker 4 (27:30):
It's just I'm often code or vibing on something that
you know, I didn't vibeuse. This isn't my day drop like,
this is just something I do from about six pm
to midnight most nights. So that's the other thing I
find really interesting is that as a non I'm a creator.
This gets into a different sidebar. But like I do
in my spare time, I am a woodworker. I like

(27:52):
to I like the active creation of things. That's actually
why I choose product management is because I like the
act of creation, but it's often been too difficult for
me to understand.

Speaker 2 (28:03):
The languages are changing all the time.

Speaker 4 (28:04):
I did computer science twenty years ago to college, like
the language of college is changing.

Speaker 2 (28:08):
I don't think my C plus p bus knowledge helps
me anymore.

Speaker 4 (28:12):
And so it's nicest to be able to truly create
and feel like I can see an output of an
idea in very short amount of time.

Speaker 2 (28:22):
But now I told you I went on a tangent there.
What was your question.

Speaker 1 (28:25):
Well, I was just going to say, we can call
you gunn or the creator, but no, I was just
I'm just trying to get my head around what it
is what you're actually doing when you're vibing, you know,
how are you doing it?

Speaker 4 (28:39):
So I start like, right now I've got ID open
a blank I just start open new folder, blank folder.

Speaker 2 (28:45):
Right now.

Speaker 4 (28:47):
The product I'm currently looking at, I got two of
them open. One of them is Caro, which is ki
r dot dev. It's on a wait list right now,
but it is public and it's very similar to Cursor
powered by clouds. On it you can switch it to three,
seven or four of course fron on four. And the
thing that I found really interesting about this one to

(29:08):
get to your question, is it has the vibe It
literally is called vibe as a function, and the other
one is spec as a function. And you can think
of this and other ones like ask versus agent. I
think is what curser calls it, like am I going
to write you code? Or am I just going to
respond to your questions?

Speaker 2 (29:24):
Similar?

Speaker 4 (29:26):
But the SPEC in this thing is really interesting because
that's kind of like as if I'm talking to a
really good product manager or program manager depending on.

Speaker 2 (29:35):
Your miro suft.

Speaker 4 (29:36):
But this this thing is going to start taking my
idea is, hey, I want to create a new instant
messaging app.

Speaker 2 (29:42):
Okay, cool? Uh.

Speaker 4 (29:44):
The spec is going to start asking me questions, what
what do you want to do with the instment? Is
some interesting app? Blah blah blah, And we kind of
go through this whole thing together. It's going to then
create my PRD. From there, it will we're talking about
compressing a different use cases. Product manager is compressed, right
because now I'm able to write docs much faster. You

(30:04):
talked about UX, We talked about that. The other thing
is impresses the program manager's job. Everyone is effected by
this thing inside this industry. So it goes back to
my question, if you were to start a company, what
would it look like. So the next step of this
spec is to design it. Right, So here's all the
normal stuff you get from a product manager. I needed
to do this, you know, as a user. I want

(30:26):
to do this so that I can do the standard
user story, very product manager centric user story. So now
I can do that by just talking like a human
and saying I want this messaging app that allows me
to connect to these people. I wanted to only be
for government employees, you know, so let's bring bring in
some you know, tough security stuff into it. It's like,
do all this kind of stuff.

Speaker 2 (30:46):
Okay. Now after you do that, now it's going to
go to a design phase.

Speaker 4 (30:49):
So now I'm in a principal engineer level at least
you know where I work, be a principal engineer.

Speaker 2 (30:54):
Okay.

Speaker 4 (30:55):
So now it's going to look at this and I
can say, you might ask me the question or follow
up and to say, like where do you want to
design songs? I want it all built on a bus.
Every service is a bus for obvious reasons for me.
But I could say is your I could say, GCP,
I could say, go use some startup, doesn't matter, so
then it will do a full design spec. And that
one is really impressive. How well, because that's actually something

(31:17):
I'm smarter act. If you're outside of the development world,
you're like, you know, my world of actually understanding how
different services work together.

Speaker 1 (31:25):
Okay, so real quick design, you're talking about designing the
structure of the app. When I think design, I'm thinking
you I design.

Speaker 2 (31:31):
And then.

Speaker 4 (31:34):
So just a clarify are you going to use for
this type of workload the system, market system market that role?

Speaker 1 (31:41):
Okay, I just want to clarify what design we're talking
about here.

Speaker 4 (31:44):
So we're designing the entire system, So what is the
front end going to use? We're going to use React whatever.

Speaker 2 (31:49):
Uh. So we go through all the things they will
do it.

Speaker 4 (31:51):
I have some back and forth with it on that,
and then it gives me a design spec.

Speaker 2 (31:56):
Now this is me vibing. You asked me a vibe question,
not what I'm doing it like for my day job.
Just this is me vibing.

Speaker 4 (32:01):
So give me a full design spec, which I enjoy
because now I can, as a product manager turn back
around and I start looking at.

Speaker 2 (32:09):
It wants me to use Eacy two containers.

Speaker 4 (32:11):
It doesn't just as an example, or wants me to
use a far Gate container and make sure you use
a doctor container. Well, now I actually do my little
product management thing of like how is this going to
cost me?

Speaker 2 (32:21):
Right?

Speaker 4 (32:21):
So then I started looking at these different design specs
and I say, you know what, that one's pretty expensive.
Can we use serverless here, don't use serverless here, use
already s here, use my squel here, right, And to
have these different conversations, and then just like you with
chat GBT, it's going to start telling you is like, well,
here's the advantages of this type of databas system, resist
type system. These are the different you know, running it

(32:41):
as a service, and these cost tructures for these things.
So all these things kind of again get compressed in
you in you you know, my job to build a
five year p and l of these things. So as
I'm looking at the design, I'm also looking at the
cost of the design that there's ways that we can
adjust that design to make it more cost effective without
necessarily to directly impacting customer experience. I had one the

(33:03):
other day with use puss grass versus Aurora, and you know.

Speaker 2 (33:08):
The gives and takes there.

Speaker 4 (33:09):
So it's doing design doc which I would actually in
my day job send it over to a principal engineer,
some senior level architect type person to just to design that.
From that, this is where the program manager comes in.
Once you have the agreed like product specs, you've now
got the design of the system done that it will

(33:31):
literally break down. I did it. I did it by
accident the other day. I didn't realize I was going
to do it. It will break it down by sprint.
I will get an entire burndown chart using this AI.
Now I haven't tested any of that, but it was like, dang,
like talk about yet another role in this industry where
you look at this and it's like this will take

(33:52):
I think the one idea is like seven months to complete.

Speaker 2 (33:55):
I'm like, no, you just do it. It's a good
about maur. That's scary.

Speaker 3 (34:02):
That is scary because the obvious question that that this
brings up is where do, aside from your you, yourself
kind of sitting in the director's chair as it were,
where do other humans fit into this process.

Speaker 4 (34:21):
I'm biased your developers, I'm not someone has to sit
in director shares. You put it like what are the
right problems to go solve? Like, I don't know about you,
but I've got family members that love to call me
up with their million dollar idea, sure, like I can

(34:42):
get you.

Speaker 3 (34:42):
Remind me there was this old Dilbot strip which now
gets a whole different meaning thanks to vibecoding, where the
boss comes to Dilbert and says, I have an idea
for a product. How can I get it off the ground?
And the but ask them, well, do you have the

(35:02):
technical know how? It says obviously not, So do you
have the money?

Speaker 2 (35:07):
Says no?

Speaker 3 (35:08):
Then basically what you have is nothing. Well, now, thanks
to vibe coding, you might have actually have something even
if you don't have the technical know how or money.

Speaker 4 (35:18):
So that gets into another interesting conversation that I think
one of the advantages I have is because of my background.
I'm not like I've hired NBA grads with no technical background.

Speaker 2 (35:31):
I think that's a different struggle.

Speaker 4 (35:34):
But for my background, like I started in it, like
I was a systems engineer at some point in time,
Like I've had a lot of different roles in my life.
So a lot of times when it's giving me a design,
like I understand like even a simple term like database.

Speaker 2 (35:49):
I know that's asbout as high level as we can go.

Speaker 4 (35:52):
Some some pms fresh out of school may not really
understand what that means, right, And that's being kind of
but you get into these lower levels, like you do
have to have a fundamental understanding of how how building
applications work. And if you don't, you can ask the
AI to teach you. But you do need to have that.
So when I say to Steve earlier, like I'm reading it,

(36:15):
I'm reading it because I do have a fundamental knowledge
of what it's doing. Right, Oh, it's doing this thing
to do this, and that's going to go plug in
over here. It's like, okay, I have some fundamental knowledge
over twenty years of history here where I kind of
get it now. I don't necessarily on the code, but
as they say in the Matrix, I don't even look

(36:35):
at the code anymore.

Speaker 2 (36:36):
I just see blonde girl, blue dress. Whatever he says
to the matrix.

Speaker 4 (36:39):
If you understand my reference there, Yeah, that's kind of
how I think about it, Like it's doing a bunch
of stuff, but I'm starting to just kind of see
what it's doing, right.

Speaker 2 (36:49):
Yeah.

Speaker 3 (36:49):
The other aspect is something that came up with our
discussion with Anthony when he was talking about vibe coding
from the developer's perspective. He talked about the there's still
the need to be able to take complex problems and
break them down into parts. So obviously, you know, as

(37:10):
AI gets more sophisticated, it can also start doing that itself,
but it can go veer very far off track if
you just give it a very general problem and tell
it to solve it. So it's like, you know, if
you tell it, like, I don't know, bake me a cake,
you might not get you get the cake, but you

(37:33):
might get a cupcake, or you might get the cake
in a savory cake, or you might get the cake
in a totally different flavor than what you want, or
you might get the cake that's not edible at all
or whatever.

Speaker 1 (37:47):
Yeah, you can get yellow cake uranium.

Speaker 4 (37:48):
I mean, that's kind of the point I'm saying, Like
I would agree that was true about six months ago,
I don't think that's as true now. These tools have
gotten significantly better. I literally just put in my I
want to create an instant messaging app for the government
use case. When I used earlier and it is now
giving me exceptance criteria across the board of my requirements

(38:10):
that I go through, and it is thinking about scale,
it is thinking about how do you actually break this
thing down into components that are more easily manageable. It's
doing all of that for me. And that was kind
of like the mind blown thing I had a couple
of months ago because I used see you, I'd build
a I'd go into chat TBT and have.

Speaker 2 (38:32):
It build a lambd, Python and function.

Speaker 4 (38:34):
Of some sort for me, like it was good at
kind of a single page, single use.

Speaker 2 (38:38):
I've been doing that for a couple of years. It's
only more recently where when you were wanting to do
some things at greater scale that I feel like these
tools have caught up to that. I'm not saying it's
there yet.

Speaker 4 (38:52):
I'm not saying I can vibe and have a production
ready anything, but I can vibe and have it think
around the big stuff that you don't normally do. But again,
that requires a certain level of background and expertise to
understand why that's important.

Speaker 2 (39:09):
My family member calling me up.

Speaker 4 (39:10):
With their million dollar idea, they don't have that background,
they don't have that understanding of like why it might
be important to understand permissions or I don't know, that's
too easy, but you get my point. So these tools
have gotten significantly smarter. And I'm saying this is in
the last this year, last six months where I've seen

(39:31):
a significant shift and the tools ability to just write
me a quick script that's been going on for a
couple of years now to thinking bigger, thinking, scalable, thinking, modular.

Speaker 2 (39:44):
It's they're much much better.

Speaker 4 (39:46):
And I would actually point to kiro Is I think
one of the better ones of doing exactly that. Even
but even like Claude Code and others, I've noticed that
they are looking at how I structure out a folder
from zero to one?

Speaker 2 (40:00):
Right, how do I structure this thing out?

Speaker 4 (40:02):
And if you ask, if you put in the right prompts,
which that's where the user experience starts to fall apart,
because you have to know to ask the right questions.
But if you put the right prompts in, it is
going to give you the type of experience I would expect,
you know, working at a large, high scale company. And

(40:23):
I think that that's an interesting thing is when it
is smart enough to do this with a bigger picture
in mind.

Speaker 3 (40:29):
So if we think about that, and let's not even
talk about today, let's talk about I don't know, two
or three years down the road, which seems like a
whole lot, but it's essentially nothing. Are we going to
have like a convergence of the developer role in the
product role where in both cases you're basically describing to

(40:52):
the AI what you want and you're technical enough to
use the correct terms and proper and proper explanations and
be specific enough in what you request to get to
actual the actual desired results. Uh. And that's essentially going
to be programming, Like being a programmer is going to

(41:14):
be being is going to be what used to be
known as being a product person.

Speaker 4 (41:18):
The product persons their number one job is still to
understand the customer problem that we're trying to solve. That
doesn't change, Ever, the output of that does change, Like
does that mean I'm writing a PRFIC q br D,
Like what.

Speaker 2 (41:30):
Is the output of that? I still think there is
plenty of room for us.

Speaker 4 (41:35):
You know, perificu is like an Amazon way of writing
any kind of business business strategy doc. I still think
there's a lot of room for As you know, Jeff
Bezos would say, working backwards from the customer problem, you
never go wrong there. So I think product managers still
have a very important role in that. And I still
as I surrounded by developers, I know them well enough

(41:57):
that they have no desire to do that, like they
want to build shit.

Speaker 2 (42:01):
Sorry, I don't know the rules on language on this podcast,
but yeah, so like that that role is still there.

Speaker 4 (42:08):
But to your point, what is the output of me
to the input of developer. That is where it gets
more gray area. And I'm you know, I'm literally like
meeting with my team and saying, you need to have
an ID running as your primary these days, like instead
of Microsoft word right, it is work with your ide

(42:30):
Don't give me a doc that you didn't write with
an AI because we have a very high bar for
writing where it work. And you know, I don't have
to deal with grammatical issues anymore. I don't have to
deal with tense issues or hyperbole issues because of how
we work as a company. It's like so grammarly, yeah, yeah,
grammatically like all this kind of stuff.

Speaker 2 (42:51):
It's well, grammarly is different.

Speaker 4 (42:53):
I'm just saying, like you have someone in your office
that's you know, just you have an expert in any
field that you want for the most part right here,
put that on your screen, write your document, and collaboration
with this thing. And as you get better at this,
and as these these tools get better, you're gonna get

(43:15):
better docts. You're going to start doing a better job
of not just understanding the customer problem that's still a
human endeavor, but putting that in terms that and making
it easier for developers to then build the next thing.

Speaker 2 (43:30):
There's another conversation. We don't necessarily talk about it.

Speaker 4 (43:32):
I do think the idea of sprint changes once this
stuff really starts to steamroll, because I don't necessarily think
we need to be working at two weeks prints anymore.
I know some people do one three, But I do
think there's a fundamental shift in program management.

Speaker 2 (43:51):
That's another topic.

Speaker 3 (43:52):
We might touch on that. But I have a question
when you so you said that Amazon is is very
kind of strict in terms of, you know, what the
doc what the product, role is the product, the documents
that are created, the standards that they adhere to.

Speaker 2 (44:13):
Et cetera. A high bar. It's gonna be a high
bar positive.

Speaker 3 (44:21):
If you were working at the company that was not Amazon,
would you would you still have these documents or would
you just.

Speaker 2 (44:31):
Use the vibe code result, I'm fully in on the
kool aid over.

Speaker 4 (44:35):
Here of our leadership principles as a company. And I'm
not saying that because I'm recorded. I really do believe.
I think we have great leadership principles here, and I
love the perific format and you can google it.

Speaker 2 (44:48):
This is publicly stuff.

Speaker 4 (44:50):
I don't know how much talk about it, but I
know there's a lot of stuff out there, so I
want to somewhat avoid talking about that because again, I'm
just here as Gunner. I'm not here as any Amazon appointee.
But yeah, I would use them because you still need
to have you need to understand the problem that you're
trying to solve.

Speaker 2 (45:09):
Period.

Speaker 4 (45:10):
Now, how I build a doc, I would build it
with AI, and I have built any doc I've built
of that scale in the last two three years has
had some type of AI assistance involved in it. So
I build a doc because I can go I can
be more rapid in my in building of that document.

Speaker 2 (45:28):
And you know the pr press release epic.

Speaker 4 (45:31):
You frequent last questions, the questions what questions should I
be asking, Hey, you've you're I'm working on this thing.
What questions am I not asking here that I should
be asking? I might ask the AI. It's like, well,
you know someone might ask about this is this is
It allows you to look around corners that you may
not naturally look around.

Speaker 2 (45:47):
And that's why i'm you.

Speaker 4 (45:48):
Know, you're you're You've got an expert of the room
with you as you're, you know, a human, and you're
not always going to think about all of the questions
that I should be asking.

Speaker 2 (45:57):
So having this this.

Speaker 4 (45:59):
Tool at your disposal that's been trained on basically the
human intelligence collective human intelligence like you, you'd be remiss
not to use it. So I'm really interest stive you
said you're the you'd be the anti AI, because I've
been speaking down for thirty minutes of very pro AI,
So I'd love to hear kind of like the anti
version of this.

Speaker 2 (46:19):
But from my perspective, like I can move faster, I can.

Speaker 4 (46:22):
Have better documents, I have better questions, I have everything
about It improves in.

Speaker 2 (46:28):
My ability to make good product.

Speaker 4 (46:31):
The negative, if you're going to have one is I
do think it's the doom for all of us that
we are having a job. But that's a whole other issue
that I'm going to just sweep down the road into
the future.

Speaker 1 (46:40):
Well, so you want to say you want to hear
my vay point, and then is that what you're saying Gunner.
I'll start out by saying that there is not a
tool that exists that, when not used in its proper
context or as proper purpose, can be abused and is
not going to work. Well, that is true of anything

(47:03):
from I'll give a weird example carbon fourteen dating. In
terms of dating, you know, geology and stuff like that.
I see people uh twist and use that, you know,
to in various contexts and stuff. In my experience, you know,
AI has can be a great tool. It can do

(47:23):
a lot of things, but there are other things that
it really sucks at. We had a discussion in my
workplace the other day about somebody said, I am so
waiting for the AI bubble to burst. And there's an
article that I was going to do for picks that
and I had to tweeted it out the other day
and I got to find it again. But the gist

(47:44):
of the title is basically, I'm going to e think
pile drive. The next person who talks to me about
AI because people are so you know, sick of it,
especially in the development where about everything everything's AI AI AI.
And as a person who uh prior to you know,
Chat dpat Chat GPT's release that sort of made everything

(48:05):
explode a couple of years ago, I lived a lot
in the search world, Lucine based search, you know, Google
type stuff, views and Lucine PATCHI, solar elastic search, that
kind of stuff, and knowing one that AI is a
model and models, whether it's climate models that I see
have used all the time, whether it's economical models, whether whatever,

(48:25):
garbage in, garbage out. And also the developers have a
very very large impact on the AI and what it generates.
You can see models with bias, you know, from a
political standpoint, from any other nandpoint. You see it all
the time in terms of oh, I'll comment on this guy,
but no, I can't say anything good about this guy,

(48:47):
even though they're both political figures in the same positions
that have been You know, there's stories that I've seen,
horror stories that I've seen from a legal standpoint with AI,
where I basically literally made up cases that did not
exist and attached them to some guy and he had
to clear his own reputation, you know, an attorney. There

(49:08):
was a recent legal case in California that I've mentioned
before where a judge censored attorney because she said, this
is obviously AI. He quotes these cases and it goes
to look, they don't even exist. The AI made them
up to try to fulfill, you know, its request instead
of just saying say, I don't know it now. From
a code standpoint, you know, when IT first came out,

(49:30):
a lot of it is, oh, you can tell a
I to write your code for you, and great, here
writes are your code for you, And then you start
seeing surveys that eighty percent of code that was generated
AI eventually gets reverted.

Speaker 3 (49:42):
I have to interject with a funny thing I saw
next a while back where somebody wrote the dumbest person
you know is currently being told by AI that that's
the very good point.

Speaker 1 (49:55):
Yes, right, exactly, you know, and so you know AI,
it's it's like I said, garbage in garbage. I used
Geminia to use chat GTP for troubleshooting code a lot
of times, like shoot, why am I getting this error?
You know, help me figure this out? Help me make
PHP stand happy that kind of stuff, and even then
I got corrected a lot of times because it gives
me bad data. You know, last night I was messing

(50:15):
around with the JavaScript inertia thing and we use this function.
I go look it like, no, that doesn't exist, and
it's like, oh, yeah, sorry, I was wrong.

Speaker 2 (50:23):
You know.

Speaker 1 (50:24):
So there's so much that it gets wrong. There's so
many dependencies, there's so many things that can influence it.
And you know, one of the things that I think
is obviously having to be work on is that if
you give it a task and it can't do it,
instead of saying I can't do it, it makes up
a bunch of crap to try to fulfill what it's
told to do. And so to me, we had a

(50:47):
guy on and I forget his name, and we were
talking about AI models and he was going into detail
about how AI models work and what they do, and
I came out with that thinking that's basically just search
on steroids, is what it is. And so so from
a generative standpoint, you mentioned this earlier when you wanted
to create something from scratch, it really can't do that.

(51:07):
Generative generative AI is not something I would trust at all.
You know, if you're asking it to work with something
that's already there and maybe helping you figure out in
your case, is generating stuff, it's generating products and stuff
like that. But it shouldn't be everything. In all you know,
in all that it's being used in abuse for and

(51:27):
I think, so back to my original point, I think
as a tool, when it's used properly, it's very efficient.
You've mentioned yourself vibe coding, how it saved lots of
time because it can do stuff so quickly and generate
a demo for you, where in years past it would
have taken documents written in meetings. And I know because
I went through all those meetings and writing documents and
that kind of stuff. But it's not the solution to everything.

(51:50):
And you know, I've seen it takeover jobs. My daughter's
fiance lost his job because it got you know, farmed
out to AI. And we see stories about Junia developers
getting hired now because AI can do it for you.
So there's pros and cons to you know that part
of the argument too. But my point is, yes, it's
a great tool, but let's not use it for everything

(52:11):
and just saturate our entire lives with AI and it
can solve all their problems because.

Speaker 2 (52:15):
It can't well that it can't generate from zero.

Speaker 4 (52:18):
I would fundamentally disagree with that, but let me caveat
the point because you were you were jumping on what
I said earlier. Using djingo, it will build a great thing.
You have a framework, and it will build from that framework.
My point was that we were talking about UI at
the time. So if I already have a framework with
a UI framework with a bunch of bones in it

(52:40):
or assets in it, it can use those assets into
a fantastic job. It's when I want to create something
that's fundamentally different from that where I would start to struggle.
But if you think about that, that's that's not necessarily
a bad thing. Like it's not creating type script JavaScript
like it's it's not creating.

Speaker 2 (52:58):
A new script, right. It's using that which it has
been trained on and does a very good job of writing.

Speaker 4 (53:05):
Code because it has billions of lines of code that
it has been trained on to do. But I don't
think it's going to come up with gunner script anytime
soon that's going to completely change it. Maybe because I
have read some of the non generative AI, but some
of these AI tests where like all of a sudden,
the AI are talking different AI in a completely different

(53:26):
language and the researchers can figure out what it was doing. Maybe,
but that's like way outside the realm of like generative
AI and what we're talking about. So my point is
just simply like it's really good at taking something that
exists like JavaScript, like typescript, like I don't know C
plus plus whatever, and it's going to write that better
than human I think, like to your point, you're arguing that,
oh no, it's not. I would take the other argument.

(53:48):
I think it will do it better. And we're talking
less than twelve months time. I've seen in the last
six months how much better it is.

Speaker 2 (53:56):
Just switching from Kurser to Claude and saying how much
better that is.

Speaker 4 (54:00):
It's like significantly better in very short periods of time.
So I do think the reason it can be better
is because it has billions of data points to look from.
The problem is like, if I want to create Gunner script,
is the new better than Java script, Well, it has
zero lines of code to learn from that, so it
will be zero helpful in that endeavor.

Speaker 2 (54:22):
Right. That's where I say, like it can't generate from zero.

Speaker 4 (54:25):
From that standpoint, it can, but it can do a
fantastic job if it has billions of points of reference
to draw from to write code. So almost everything I
do starts from a blank slate. I just want to
be really clear, and I've gotten very far with prototypes
from a blank slate, So I do believe it can
do the zero to one thing very well.

Speaker 2 (54:45):
But you also are having another comment. You said someone
lost their job, and start to hear.

Speaker 4 (54:48):
That I talk to me in a year, I who knows,
Like I'm literally like I've talked about us the other day.

Speaker 2 (54:56):
It's like, look, I just.

Speaker 4 (54:57):
Can't worry about that anymore. If it comes from me,
it comes from me. Do I think it can and
should come for me? Yeah, I do. I think this
stuff is getting closer, uh that less and less jobs
are needed. And this is as far as like I
said earlier, like I think AI is going to be

(55:17):
a downfall, Like ultimately in the end, it's it's it's
going to have a bigger negative outcome than a positive outcome.
But unfortunately, I'm going to sweep that to the future.
As I've been telling my kids. I used to tell
you go to computer science. Now, I say, flee from it.
I don't want any of.

Speaker 2 (55:35):
My be aware which job is not good to good
luck with that one. There.

Speaker 1 (55:42):
You know, there's a in the US. There's UH to
answer your question about where to go, dan Uh. There's
a guy named mic Ro who goes around talking about
trades and the dearth of people that there are for them.
That for the people that are as compared to the
people that are needed for jobs in the trades where
they can make good mind me without going to college
and sinking hundreds of thousands of dollars into debt and

(56:04):
so and the joke, and he made a great point
a little while ago, he said, And the Democrats used
to be really bad about this is they would say, oh,
you lost all your job because EPA shut down. Your
come mines learned to code. Well, now you can't do
that anymore because there's no junior debs being hired. Because
people are using AI. So there are places to go.
It's just from the developer standpoint, people like you and

(56:26):
I who have been around a long time and already
know how to do some of this stuff more at
a senior level, we're fine. It's the newer people who
don't know this and want to get a job, they're
in the cast twenty two. Okay, I need more experience
if I want to get hired, because they're only hiring
FEUs a lot of experience. But where do I get
the experience?

Speaker 2 (56:42):
I think?

Speaker 1 (56:43):
And to me telling an AI to do something, I'm
a you know, I'm a ground level, nuts and bolts
kind of guy. I want to know how something works
at it's a base level. And if I'm just telling
AI to do it, what am I learning?

Speaker 2 (56:53):
Nothing?

Speaker 4 (56:53):
But is it a bad thing for those junior debs
to start in my position in the product world? Is
it a bad thing for developers to really start with
a firm understanding and with the customer problem is that
we're trying to solve with this firm understanding of the
cost model of solving it, how different database structures can
cost you significantly different expenses on your P and L.

(57:14):
I don't know why the things a bad thing. You
just might gravitate more towards the development side. So you
start with this AI prompting PM type role that does
prototypes in your traditional junior dev role. But now you're
maybe just learning a little bit differently than how it's
been taught today that you can't be separated from the
customer problem, that the only way into this field might
be through the customer problem. But then you're like, hey,

(57:36):
I'm going to gravitate towards the actual code that it's building,
and you build a whole new chain of Like how
you get to senior dev is you start from a
different place like I didn't.

Speaker 2 (57:47):
I didn't started as product manager.

Speaker 4 (57:48):
I started it having problems with technology and I wanted
to go to the vendor side because I incorrectly assumed
I could make a huge difference over these technologies.

Speaker 1 (58:00):
Okay, but where are you getting the actual coding experience?

Speaker 2 (58:02):
Then? Well, I'm just saying like if would it be
a blended thing?

Speaker 4 (58:07):
Would it be as I'm not just going to learn
CS in a vacuum of CS because I'm going to
go sit in the cubicle and code all day. Maybe
it's I'm learning CS within the context of AI is
out there helping you along the way. But what are
we talking about about AI? It is the inputs that
we're giving it, right, the prompts that we're giving it.

(58:27):
So are you learning the type of prompts that I
am giving it. I would argue your customer centric prompts
are we learning to start with the customer problem, which
I would very firmly believe in. As a product manager,
you should understand your customer base, understand your customer problem
before you write a line of code.

Speaker 1 (58:46):
Well, that's always been true, A bad thing, that's always
been true. I mean, what's the classic line about building.
You know, if you build, they will come, or you
build something for a problem doesn't exist. You see companies
fail all the time because they built, Hey, this would
be a great product, will nobody wants? So who cared
about it?

Speaker 2 (59:01):
Right?

Speaker 1 (59:01):
That hasn't changed. That's always been that way with software, right,
So I don't think that's anything new. What I the
way I see it.

Speaker 2 (59:10):
That is what I'm saying. I'm not saying it's new.

Speaker 4 (59:12):
I'm saying the responsibility could shift where that junior role
is a mix between the responsibility of that and understanding
the code behind it.

Speaker 1 (59:21):
Yeah, but I see as more as writing to the
tool instead of underneath. And I think one of the
best examples, you know, you hear people talk about prompt
engineering as a whole new field. So what are you
doing you're learning how to tell AI to do something.
To me, that's like a grade school And this is
a common thing you'll hear in education. My daughter is
a teacher, and I remember hearing this when I was
going through school, where.

Speaker 3 (59:42):
You have.

Speaker 1 (59:44):
State tests that you know your kids have to pass
in order to do well at the end of the year.
You want they make your school look better. In teacher
if they do better on the state test. So what
do you end up doing. You teach for the test
instead of teaching for the material. To me, it's the
same thing. Now you're not learning how to do the task,
you're learning how to tell something else to do the
task for you.

Speaker 3 (01:00:06):
Before we cut to picks, I have to mention something
amusing in this vein that recently happened to me. And
this might not be indicative of anything, or maybe it is,
I don't know. So I was reviewing code that one
of our mid level engineers had written and she used
AI quite a bit, and AI implemented a certain is

(01:00:28):
it needed to iterate through database entries, and she had
to generate the code for it, and it generated working code.
But when I reviewed the code, I noticed that it
used recursion to do the looping, which would mean in
the case of JavaScript, that if it did enough iterations,
the stack would blow up. So, just out of curiosity

(01:00:53):
I had, I asked the AI myself, hey, look, changes
code not to and she didn't fix it. She she
did not review that part of the code, so she
was not aware of that problem. She just ran some
unit tests and the unit tests passed, and therefore it
looked because none of them used enough data to cause

(01:01:15):
the problem. And so I asked the AI, please fix
the code by writing this loop non recursively, and it
said fine, and it did it, but it implemented it
as an as a loop within a loop, so it
made an O of N problem into an.

Speaker 2 (01:01:32):
O of N squared problem.

Speaker 3 (01:01:34):
And then I basically got fed up with it and
I just rewrote those ten lines of code lines of
code myself.

Speaker 4 (01:01:40):
But that's where I think an MCP steps then to soay,
you know, like, that's my point I made earlier. It's like,
you need to have an AI that's the smartest AI
on JavaScript, and as you would build something like that,
you would teach it. These are the things you look
out for, right, So then your code review would be, oh,
this isn't jobs or whatever? Did you send that through

(01:02:02):
the MCP for the for jobscript review and literally an
AI you have taught it. Look for this, look for that,
look for the other type of thing, and you can
have the output if you want it.

Speaker 2 (01:02:11):
Don't output new code that creates an N squared problem, Like,
just tell me when you see it right and give
you a proper code review, not new code.

Speaker 4 (01:02:19):
Just capture the things that you see that are wrong.
That That would be my kind of reverse to that.

Speaker 3 (01:02:25):
It's like, yeah, I agree, and I think we will
get there, and we'll get there pretty quickly. I think
that eventually AI is coming from all for all of us,
except for certain service jobs. Like I guess we would
still like to interact with actual humans. I don't know
in restaurants, go to McDonald's reacting out with the computer. Yeah,
but that's not a fancy risk. I anyway, I think

(01:02:49):
it's time. So before we go to pics, is there
anything else gonna that you would like to say that
you didn't get around to say about this thing?

Speaker 2 (01:02:57):
Let me see my notes. No, I think we're all right.

Speaker 4 (01:03:01):
I I haven't didn't get a chance to talk about
the downfall of humanity and how.

Speaker 2 (01:03:06):
It all started with social media. But outside of that not, Yeah,
I think we're all called end of that, Gunnar.

Speaker 3 (01:03:14):
If people want to get in touch with you, In
case you actually want people to get in touch with you,
how should they go about it?

Speaker 4 (01:03:20):
You have LinkedIn probably i EX or LinkedIn Gunner w
B either either one of those.

Speaker 2 (01:03:28):
If you want to chat, I'm happy to debate.

Speaker 4 (01:03:32):
My points on AI and product management with anybody it's
out there.

Speaker 2 (01:03:35):
Just don't ask me questions about code. Not I think
we should go to pics.

Speaker 1 (01:03:41):
Then, right, Steve, Right, let's do it. So I will
start today and get the high point of the of
the broadcasts right here, and then we'll just sort of
slide down hill from there, the high point being the
dad jokes of the week. So when I interviewed my
current place of employment, my interview asked me what makes

(01:04:04):
you a good fit for this position? And I said,
I broke into your system and scheduled this interview myself.

Speaker 2 (01:04:10):
Right.

Speaker 1 (01:04:13):
I decided that, you know, having that latent desire for
fame and fortune, that if I ever start a band,
I'm going to call it day job. So when people say,
don't quit your day job, I can reply thanks. We
practice a lot, and then finally my son is now
at the age where he's curious about the human body.
I guess I'll have to hide it somewhere else.

Speaker 4 (01:04:32):
Now I got one for you, Steve. What's that my
family's favorite dad joke? What did Pardikus do when the
lion ate his wife? I know this one, tell me nothing.
He was Gladiator?

Speaker 1 (01:04:45):
Gladiator? I knew it, dang it. I've used that one
before Gladiator. And what's the song is that? My son
listens song called Gladiator and we always make cannibal jokes
about that one.

Speaker 3 (01:04:57):
I still the dad joke that you told that I
still like. The best is the one about the half brother?

Speaker 2 (01:05:03):
All right?

Speaker 1 (01:05:04):
Oh, yes, my dad was a magician. So I just
have half brothers now or something?

Speaker 2 (01:05:09):
No, it was.

Speaker 3 (01:05:12):
I have a half brother. Oh is does he have
a different mother? No?

Speaker 2 (01:05:16):
He met a shark. Yeah that's right. That is so good.

Speaker 1 (01:05:22):
And I'll throw out a couple of titles to some
I guess you want to call them anti ai a
blog posts that somebody else had brought up work lately.
One is called I will I think pile drive you
if you mentioned AI again, it's from about a year ago.
Pretty funny and one thing that we didn't talk about
AI with a little bit, and maybe this is included

(01:05:45):
in your part about the dawnfall of humanity later, Gunner,
there's a document called the Hater's Guide to the AI
bubble is the money involved, the losses that a lot
of these companies are taking simply because of the computing
resources that are required in terms of energy, in terms
of servers power. It's just insane.

Speaker 3 (01:06:08):
If there's an energy crisis in the world. A alternatively,
if China invades Taiwan, then AI will be delayed significantly.

Speaker 4 (01:06:17):
In either case, bitcoin bitcoin probably burns through more than
anybody else, and that thing is.

Speaker 1 (01:06:23):
What it does too. You're right, but sure they suck
power like nothing.

Speaker 3 (01:06:29):
Yeah, but with bitcoin, when the power cost goes up,
then bitcoin money stops. And the point is exactly that
that if the if the cost goes up or the
chip availability goes down, that would put a significant break
on on AI.

Speaker 1 (01:06:48):
So anyway, that's enough of my ranting, Dan, What do
you got for picks anything?

Speaker 3 (01:06:53):
Okay, I'm going to So I was I surprisingly enjoyed
the show that I guess I wasn't supposed to enjoy.
It's on Netflix. It's called WWE Unreal. It's it's a
behind the scene look at the WWE. Now, obviously this
is not a sport. This is a show. This is
kind of like a soap opera with people throwing each

(01:07:17):
other around, and it's literally interesting. It's it's kind of
interesting to see how the sausage is made behind the scene,
how they you know, beat each other up and then
hug each other behind because they're close friends or stuff
like that.

Speaker 2 (01:07:32):
And some of the most like.

Speaker 3 (01:07:35):
Supposedly violent people are actually very you know, nice homely
type people behind the scene. Obviously they're all kind of
fed in the head to be doing that, but it's
still interesting. The other kind of interesting pick that I
have so it's it's a show worth watching even if
you're not into wrestling, in my opinion. The other thing

(01:08:00):
is is, as I was searching for it, Google, instead
of just providing me with the link to either the
Netflix show or to the IMDb section about it, just
gave me an overview of it from its ai, which
is the fact that you know, we are a podcast

(01:08:23):
talking about you know, using JavaScript ostensibly to build websites,
and people are visiting websites a lot less than they
used to because they're getting their results from Google itself directly.
And I'm you know, thinking about the point in time again,
unless energy costs go up through the roof, where instead

(01:08:48):
of actually being served some existing website, Google might generate
custom website dynamically on the fly for you based on
the query that you with it. I think we're on
the way there again, unless it becomes too expensive to
do so, I guess those would be my pick for today.

Speaker 2 (01:09:12):
Yeah, I don't even It's very rare that I Google anymore.
I most always use aii chat et.

Speaker 3 (01:09:18):
Yeah, well, Googling these days is AI because nine times
out of ten you get the answer.

Speaker 4 (01:09:23):
F I'm just saying like themes where I go to
chat GBT and I ask my question, so I don't
get a bunch of garbage result, I just get the
answer I want. I have a Google thing in my kitchen.
I ask a question, it always gives me like a
paragraph and the answer with like forty two. It's like,
I don't need the paragraph, I just need forty two.

Speaker 1 (01:09:42):
The paragraph for the answer to everything.

Speaker 4 (01:09:44):
Then yeah, yeah, it's way too much, all right, So
I'm supposed to picks here I did give you. I
put a link in our chat. So white people are
called cognitive empathy. And it's actually talking about the issue
that you guys mentioned earlier, which is that AI will
always agree with even.

Speaker 2 (01:09:59):
When you're wrong.

Speaker 4 (01:10:00):
And it's an interesting documents on GitHub to read about
this problem and how AI will I think, adapt change
over time to stop always a green and actually have
a sense its own stance. Not necessarily political, but when
it comes to something that's scientifically code that it would

(01:10:22):
actually say the best way.

Speaker 3 (01:10:23):
Hitler was actually a good person.

Speaker 2 (01:10:26):
My gosh, I'm glad that's coming from someone who lives
in Israel.

Speaker 4 (01:10:29):
Anyway, I did one as a joke with my kid.
I'm like one plus I proved to my kid that
one plus one is three, and I just wanted to
see how it respond to that, and the fact that
it tried to like give me every option it could
be like why you might say that, well, because it's
synergy and businesses like to use this term like and

(01:10:51):
I said, no, you should just tell me one plus
one is two like I'm I'm wrong. And so this
cognitive paper is really interesting one. But it's called cognitive empathy.
It's a short read. It's like a ten minute reading
to look at it. What other things?

Speaker 2 (01:11:06):
I don't know. You said a show? I'm up. I
just got done watching season two of Netflix's Tires Shane
Jane Gillis.

Speaker 4 (01:11:13):
Yeah, that guy is like he's got a direct line
to my funny bone.

Speaker 2 (01:11:17):
I love that guy.

Speaker 1 (01:11:18):
He does the best Trump impression I've ever seen. Good,
whether you like him or hate him, just the hand,
the voice, the mannerisms, the hand movements, even with the
wig on it.

Speaker 2 (01:11:30):
That's so funny.

Speaker 1 (01:11:31):
He's so good at it.

Speaker 2 (01:11:32):
Yeah, he did kill Tony and Madison Square. Yes, was
dying watching that thing. Uh, something of that. I'll end
with kind of what we talked about this whole time.

Speaker 4 (01:11:40):
I encourage anyone listening to check out the cure ok
I r dot dev id. It's an interesting one. And
then you know, you guys, check out cloud code if
you're using cursor. If I was still stuck using curser,
I'd probably be as mad at ai as some of
the comments of her today. Uh, clod code is what
actually brought me to the next page. Or I'm like, hey,
look a I can write half decent code. It's a

(01:12:03):
pretty big jump. So yeah, give back you can get
you know, seventy trial or whatever. It's like twenty bucks.
I do them.

Speaker 2 (01:12:09):
I'm actually using at work.

Speaker 3 (01:12:13):
I'm using a cursor with Claude four Sonnet okay, and
I'm getting fun probably the more results because yeah, it's
all cloud for it so and for Sonnet thinking whatever
that means.

Speaker 2 (01:12:28):
Cool. Well, thanks for having me on.

Speaker 1 (01:12:29):
So just for a reference for Kiro is the new
aws UI ide that uses AI and we will be
having Eric Canshit from ABS or the DevRel been on
here multiple times before to talk exactly about that topic
about four weeks from this recording. Not sure when this

(01:12:50):
will have come out, but we will be discussing that
in detail.

Speaker 3 (01:12:53):
Okay, thank you very much Gunner for jumping on. I
know that you have to drop off about right now,
so it was very informative and I very much was
happy to catch up with you and talk about all
this stuff.

Speaker 1 (01:13:05):
Yes, thanks everybody for listening to job scrip drabber and
we'll talk at you next time.

All Episodes

Exploring Vibe Coding and the Future of Product Management with Gunnar Berger - JSJ 690

Episode Transcript

Popular Podcasts

Dateline NBC

Stuff You Should Know

The Breakfast Club

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}Exploring Vibe Coding and the Future of Product Management with Gunnar Berger - JSJ 690