Latest / EGGS! The Podcast / Eggs 466: Less Vibe, More Code. The Future of Software Development with Tyler Wells

Transcript

Eggs 466: Less Vibe, More Code. The Future of Software Development with Tyler Wells · EGGS! The Podcast

0:12
Hey everybody, and welcome back to Eggs the podcast.
0:14
But before we get started, I want to talk to you about our
0:16
sponsor, Taylor Style. Most guys actually do want to
0:20
dress better. They want to look sharper, more
0:22
put together, more confident, better in business meetings,
0:24
date nights, conferences, traveling, wherever life is
0:27
happening. The problem is, most men
0:29
eventually abandon it because they don't want to think about
0:31
it. They don't want to shop.
0:32
They don't want to coordinate outfits.
0:33
They don't want to spend Saturday wandering through the
0:35
mall trying to figure out what smart casual means anymore.
0:38
So what's the easiest way to dress better, Taylor?
0:41
Style. Taylor is an AI powered menswear
0:44
styling and rental service backed up by real human
0:47
stylists. You fill out a quick profile, AI
0:49
learns your fit and style preferences, and then an actual
0:52
stylist fine tunes the looks and ships complete.
0:54
Outfits right to your door. And honestly that's the magic of
0:58
it. You still get to look like the
0:59
guy who has his act together without spending the mental
1:01
energy of getting there. I'm a subscriber right now and
1:05
every few weeks I get a box with hand picked outfits chosen by a
1:08
real stylist refined by their AI.
1:10
And the crazy part is they actually.
1:11
Nail it most of the time. Stuff I would never go by but I
1:14
end up wearing non-stop just because it works.
1:17
You wear the clothes, send them back when you're done and the
1:19
next box shows up. Easy.
1:21
No laundry, no shopping, no overthinking.
1:23
It. It's just one last thing pulling
1:24
on your attention every week and somehow you end up looking
1:27
better for it. Go to Taylor dot style.
1:29
That's TAELOR dot style and use promo code eggs to take 25% off
1:34
your first month. Once again, that's TAELOR dot
1:37
style. Use promo code eggs to take 25%
1:40
off your first month. They handle the clothes so you
1:43
can focus on everything else. And now back to the show.
1:46
Hey, everybody, welcome back to Eggs the podcast.
1:47
Today's special guest is Tyler Wells.
1:49
Tyler, how are you, man? Welcome to the show.
1:51
Thanks Ryan. It's a pleasure to be here.
1:53
Looking forward to the conversation.
1:55
Absolutely. Me too.
1:56
Thanks so much for making the time.
1:57
Let's start, I guess, kind of just at the top and talk a
2:00
little bit about who you are, what you do, and why we're
2:02
talking today. Sure.
2:04
So I'm Tyler Wells. I'm the Co founder and CTO of
2:07
bring grid dot AI. Been in this industry for Intel
2:11
by the Gray hair for probably north of 25 years.
2:14
I've worked in a whole bunch of different companies from small
2:18
to relatively large, spend a bunch of time at places like
2:22
Skype and Twilio and us now the third company that I've started
2:27
and we are kind of focused squarely in the in the world of
2:31
AI right now. I love that so much.
2:33
Well, let's talk a little bit just in the beginning to kind of
2:35
lay some groundwork about sort of your entrepreneurialism.
2:37
You said this is your third start up.
2:39
So obviously a lot of people who listen to this show go through
2:42
those things. Can you talk at all about just
2:43
sort of, I guess maybe if you want to focus on this project or
2:46
if you want to talk about other things in the past, but what was
2:49
sort of your first foray into entrepreneurship like what took
2:52
you there from working at like a Skype virtualia?
2:55
No, I think it's always just been AAI guess a dream that I've
2:59
had and and something that I want to do.
3:01
I've always wanted to run my own company.
3:04
You know, I watched my father, grandfather and uncles all run
3:06
their own company and I've seen others do it and, and decided to
3:11
take the crazy jump a number of years ago and, and say like,
3:14
let's just go for it and, and had an idea and let's try it
3:17
and, and see what happens. You know, the, the, the first
3:21
one was definitely a, a big learning experience.
3:25
And that was kind of in between my time at, at Skype and, and
3:28
Twilio obviously didn't go as, as planned.
3:31
And that's fine. But you know, from, from, from
3:34
all failures, you, you can learn a ton of things.
3:36
And, and I definitely did. And then I kind of spent the
3:38
next 7 1/2 years at, at Twilio, you know, which at that point in
3:42
time was, you know, under 200 people.
3:44
And when I left, it was probably over 5000.
3:47
So I got to see a, a different stage of, of growth and, and
3:51
sort of, you know, explosion of, of what that looks like going
3:56
from sub 200, you know, employees to over 5000 sub 100
4:00
million revenue to, you know, what it is today.
4:03
And, and then took all that experience and joined up with
4:06
another ex colleague at Twilio. And so let's do this again.
4:10
Let's jump back on this train and, and try to try to build
4:13
something. And, you know, for me, it's,
4:15
it's always been I, I like the idea of doing hard things.
4:18
I like the idea of, of being challenged.
4:21
And, you know, I think starting a company from zero, trying to
4:24
get it to 1 is, is sort of that, that ultimate challenge out
4:27
there. You know, how do you take
4:28
something that you see as a problem in the world or you see
4:32
as a, a, something that you want to, to build and, and how do you
4:35
turn that into something that can help more than just yourself
4:39
and, and put it out there and, and see what happens.
4:42
Yeah, I love that. Well, and one of the things you
4:44
talked about is sort of this like iterative growth, right?
4:47
You started out with your first project, you know, it went as
4:50
far as it went or whatever and then it fizzled and then you
4:53
moved back to regular employment or state, you know, regular W2
4:57
style employment. Then you jump back out again,
4:59
took another shot at it. And I imagine that second round
5:01
went a little further, maybe a little faster because you had a
5:04
little wind in your sails. Maybe you had a good founder or
5:06
Co founder with you. Can you talk about the
5:08
transition from that company to now, what you're doing and how
5:13
that sort of transition happened?
5:15
And, and the one thing that I'm really trying to understand
5:17
here, and we've seen this with a lot of guests and, and a lot of
5:19
people who are going through the entrepreneurial sphere.
5:22
There's this moment in a founder's life, this thing that
5:25
we call the lid. And the notion is that you run
5:27
into this kind of invisible barrier between what you know
5:29
how to do and what you don't know how to do.
5:31
And there's sort of a, you know, act that you have to go through
5:34
to transition this, you know, whatever we want to call it,
5:37
this invisible barrier, this energy or whatever that keeps
5:39
you from moving on to the next thing.
5:41
For some people, it can be, you know, through the help of a
5:44
mentor or something who can like hold your hand and drag you
5:46
through the lid. For other folks, it might be
5:49
more like, you know, you pick up the piece of knowledge in a
5:51
prior failed attempt and then you move forward.
5:54
And this time you can go further, faster, you know, that
5:56
kind of thing. So I wonder if you just talk
5:58
about, you know, that transition going from guy who knows nothing
6:01
and I'm assuming there was nothing in version 1, then
6:04
version 2, you know, whatever that was, and then and now to
6:07
version 3 of, of entrepreneurship, what that
6:09
looks like and just sort of maybe talk about failure on the
6:11
road too. Yeah, I mean, I think every time
6:14
you know, you, you come in with a little more experience.
6:18
I think you said it right. I mean, there's there's a whole
6:20
ton of things that you don't know, but you know, you have to
6:23
you just got to figure it out, right?
6:25
There's when there's only, you know, 2 of you as, as Co
6:28
founders, there's not a whole lot of, you know, there, there's
6:31
not a, a team of people that you can lean upon because, you know,
6:36
obviously you can't hire them. You don't have the type of funds
6:38
and, and really it's up to you to, to figure all of those
6:41
things out. And when we left Twilio to, to
6:43
start the the next company, which is called Propel, you
6:47
know, we had an idea that we felt that it been, you know,
6:49
more or less battle tested within Twilio.
6:53
You know, my team and I built built the platform that we sort
6:57
of pulled out of there from an idea perspective and said, hey,
7:01
look, this is applicable to not just Twilio could we kept seeing
7:05
it across a whole number of other customers inside of
7:08
Twilio, like, hey, how do we get a platform like this and how do
7:10
we do this, this, this, this thing that you guys do with the,
7:13
the analytics and everything else like that.
7:15
So we'd extracted it out of there with this idea of like,
7:19
look, we can build this because this has such a high degree of
7:22
applicability across not just a single vertical like Twilio, but
7:26
you know, all companies need analytics and they need it in a
7:29
certain way. Excuse me.
7:30
And so we, you know, went out there with the idea of like, you
7:34
know, focusing heavily on on the technology, focusing on building
7:37
the platform. But then where, you know, we
7:40
started bumping up against things that we didn't
7:43
necessarily know how to do well, which was that whole go to
7:45
market motion of, you know, how to take this platform that can
7:49
do all these things and sort of spread that gospel across, you
7:53
know, all the, the industry and, and, and turn them into
7:56
believers. And, you know, we had to bring
7:59
in, you know, marketing folks that were sort of specialists in
8:03
this. We had to, you know, we had to
8:04
augment every time where there were things that we couldn't
8:07
necessarily do, we started bumping up against, you know,
8:09
our knowledge. It was, you know, where can we
8:12
reach out to kind of find the, the, the best people in the
8:16
industry to, to come in and, and try to help us and, and what can
8:19
we actually afford? And so, you know, every single
8:23
time, like there's just there was just going to be things
8:26
that, you know, when you're AW2 employee and and you're kind of,
8:29
you know, a non founder, like all that structure is already in
8:32
place. So many things are just exist
8:34
for you that you kind of, you know, more or less take for
8:37
granted. Like, you know, how do I do
8:38
billing? How do I do my marketing?
8:40
How do I, you know, set up channel sales?
8:42
How do I all of that? I never had to think about that
8:44
from the engineering side. That was just already done.
8:46
But then you go start your own company.
8:48
I mean, you've got to do it. There's not really a, you know,
8:51
you can beg borrow playbooks out there and try them.
8:55
But, you know, at the end of the day, it's you've got to find
8:57
something that works and it's up to you to figure it out.
9:00
Yeah, I know that one. And I was going to say with that
9:02
one, I can't say that the lid was was ever removed.
9:05
You know, we bumped up, you know, we ended up getting
9:08
towards the end of 2024. And right around that same time
9:14
is is when you started seeing things like cursor and AI really
9:18
start to explode on the market. And I think that sort of shut
9:22
the lid on just so many things across the board because then
9:25
everything became like, well, aren't we just going to be able
9:28
to solve this with AI? Isn't AI just going to do this?
9:31
And I think it's stalled out. You know, a lot of the deals
9:33
that we're in discussions with a lot of our go to market,
9:37
everything was sort of met with, well, you know, AI is going to
9:40
solve this eventually for me. So I'm just going to wait to see
9:43
what happens. And, you know, yeah, ultimately
9:45
we ended up shutting the company down and then pivoting into
9:48
burning grit. Yeah, no, I love that.
9:50
And actually that little moment that you added there, I'm, I'm
9:52
sorry I interrupted you that, that little moment that you
9:55
described there is actually a great segue into what you're
9:57
doing now. I mean, for work, obviously, but
9:59
also, I think just sort of it in the, I don't know, you call it a
10:02
broader tech industry, but you're starting to see this
10:04
everywhere, which is companies are starting to lay off.
10:07
I assume some of this is head count that they probably had too
10:10
much of anyway and now they're just blaming AI.
10:12
But but you're definitely seeing people say AI is the reason
10:16
we're getting rid of people, things like that.
10:17
And I'm familiar enough with what's going on in AI right now
10:20
that I'm going, well, I don't think you're really replacing
10:23
those people with AI yet. I mean, maybe one day, but I
10:25
don't think we're there yet. And I guess maybe it depends on
10:28
the role, but but I think it's really interesting that, you
10:30
know, you started butting up against some of that stuff too,
10:33
which is just people making the excuse that, Oh, I don't need to
10:35
invest in this cause AI is going to solve it.
10:38
And so I wonder if we can help maybe level set for the rest of
10:41
this conversation. But I mean, clearly you took
10:44
that as AI don't know, like a jolt go get started digging deep
10:48
on AI. Assume obviously you probably
10:50
have been using it before, but I mean, you know, you this was a
10:53
maybe a reason to start digging in on on what it what can AI do?
10:56
What can't AI do? And where are the gaps?
10:59
And so. Oh, just to, to just finish that
11:02
thought. So the, the, I guess what I'd
11:05
like to try and do is like level set, like what should be the
11:09
expectations? Where are we actually in AI
11:11
world? And then we can talk about the
11:13
rest. Yeah.
11:15
And I, I, I mean it's changed a lot from obviously at the end of
11:19
2024 to now. I mean it's like insane amount
11:23
of change where we had to start using it was because we had to
11:28
do our own sort of headcount reduction as we saw Propel not
11:33
making the numbers we needed to make to make it a sustainable
11:36
business. We obviously start having to
11:38
start head, you know, cut headcount and but we still have
11:42
to ship product. You know, both myself and my Co
11:45
founder were like, we're not done.
11:46
We still think this hasn't a you know, has a chance.
11:48
So we'll have to, you know, cut our burn rate and sort of double
11:52
down. But that meant that, you know,
11:54
we have to start picking up a lot of the, you know, the
11:56
implementation and and building. And so in order for us to
11:59
continue to scale and build, you know, we kind of doubled down
12:02
into things like cursor because that was kind of the main ID and
12:05
the main sort of entry point into, you know, coding with AI
12:09
at the time, Like there wasn't even really a gentic coding yet.
12:13
And so, you know, we doubled down on that and started just
12:15
really learning a ton about, OK, how do we make these things
12:20
actually right production level code?
12:23
And it was through our own trial and error and, and, and research
12:26
and, and just constantly butting our heads up against it that,
12:30
you know, we kind of figured out this process.
12:32
We sort of figured out how important context was.
12:35
We started to, you know, put all of these pieces together and
12:38
then just realize like, OK, Propel's not going to make it,
12:41
but we feel like there's something here.
12:44
Let's go start, you know, brain grid.
12:46
And so as we started building brain grid, obviously we're
12:48
building all of that with AI. And, and, you know, we kind of
12:51
dove head first into, you know, how do we, how do we get these
12:55
agents to build everything for us?
12:57
And, you know, what is the, what is the state of their output?
13:02
And, and can we actually trust it?
13:05
And I would say like, you know, most of 25 was a lot of trust
13:10
and verify. You know, you obviously you've
13:14
got AI writing all of this code for you, producing a huge amount
13:19
of output, but all of that still has to be tested.
13:21
And so a lot of the same processes that we would have to
13:24
do that we put in place for managing human output are the
13:28
same types of things that we put in to manage the output from AI,
13:32
like the validation, the testing, the user acceptance,
13:35
everything like that. And then as we'd learn more and
13:38
more, we would start to fold that in.
13:39
And I would say the big shift came towards, you know, the end
13:43
of 25 into 26 when we started seeing, you know, like the, you
13:47
know, Opus 47 models start to to show up.
13:51
And I would say that was like a, a huge, you know, improvement in
13:57
terms of the output he now started getting from these
13:59
agents, whereas, you know, we could essentially, you know, one
14:03
shot, you know, product creation and everything else like that.
14:08
And the output we're getting was, was, was very good in terms
14:13
of layoffs and everything else like that.
14:15
I agree with you. I think there was a, you know,
14:18
organizations swelled tremendously with employees, you
14:21
know, especially during COVID, post COVID and things of that
14:24
nature. And, you know, they had to take
14:25
a step back and look at the bottom line of like, look, we're
14:27
not in, you know, the zero interest rate world anymore.
14:30
The world is changing. AI is changing everything.
14:33
It's a it's a pretty viable excuse to say like, oh, you were
14:35
just going to use AI to kind of, you know, trim the head count
14:38
and and everybody's going to kind of believe that because
14:41
there's so much hype around there about AI replacing jobs.
14:45
Is it replacing some? Yeah, But is it replacing all?
14:47
Is it making developers obsolete?
14:49
I think absolutely not. It's not there yet.
14:52
Is that day going to come? Possibly, but I don't think it's
14:56
going to the nature of what being a software developers I
14:59
think is changing more and evolving into this place of, you
15:03
know, sort of like stewardship, stewardship of the agents and
15:08
getting the agents to do what it is you need them to do and do it
15:11
correctly. Yeah, I know there's a couple
15:13
things in there I want to try and break down.
15:15
One of them I want to talk about it initially is you talked about
15:17
some of the framework that you guys are using in terms of or
15:20
had been using in terms of sort of establishing guard rails and
15:24
things like that for your systems.
15:25
You know what you're doing, you guys have the system has
15:29
generated code, but now it's going through a review and it's
15:31
doing some of these other steps, things that you would do with a
15:33
normal software team. What do you think the difference
15:36
is? Or is there any difference in
15:38
sort of thinking like an AI versus thinking like a human?
15:41
And so, and, and the question is that AI was built by humans.
15:45
And so theoretically, things that work on humans feel like
15:48
the things that should work on AI.
15:49
Like we should be able to talk to it sort of in a human way,
15:52
tell it to run these audits and do the kinds of things, you
15:55
know, like we would a human engineer.
15:57
But is there a difference or or is AI somehow different in its
16:01
software development modes where you actually need to talk to it
16:05
using different language or different way of thinking all
16:08
together versus just like this is another human engineer.
16:10
Does that make sense? It's kind.
16:11
Of yeah, it does. I mean, I see what?
16:13
I'm saying. Yeah, I don't think you can
16:15
treat it just like another human.
16:19
And I think where I see that is if you say to another human,
16:23
like, Hey, you need to test ABC and and D and you tell the AI
16:30
that it's going to test exactly ABC and D.
16:34
If you give that to a human and they're going through and
16:36
testing A&B and suddenly they hit something during B, they may
16:40
not move on to C. They may think like, Hey, I
16:43
actually need another step after B.
16:45
Like there's, there's sort of this, this afterthought, right?
16:47
And they're sort of like they tend to have more memory and
16:52
understanding of a system that they're working in as opposed
16:57
to, you know, I start a new session and clawed.
17:00
It's got, you know, it. It's basically that's that
17:03
memory is starting all over again.
17:06
And it's got to go look at the system.
17:08
It's got to read code. It's got a sort of Orient
17:10
itself. And if you're giving it
17:12
something like a very like, hey, pointed command of go to AB and
17:15
CI mean largely what I've seen is it's designed as it's going
17:18
to go to AB and C, you know, regardless of C may be totally
17:23
wrong. B may be this.
17:25
And so if you're not, you know, if you're providing sort of a
17:28
very terse description of what you're asking it to do as
17:30
opposed to, hey, I want you to do A.
17:33
If you find problems in A, you need to go back and figure out
17:35
what those problems are before ever moving on to B.
17:38
And then when you get to B, if you, you know, and sort of
17:40
setting all of that stuff up, I feel like because it's not
17:44
really thinking beyond, right, It's not necessarily designed,
17:47
it's more prediction, right? It's kind of like, hey, you need
17:49
to do AB C&D, I'm going to do ABC and D.
17:51
That's what I'm going to do. It's why a lot in the early
17:54
days, you would give it something like, Hey, I want you
17:56
to go work on, you know, implement this function over
17:59
here. And we ended up finding it not
18:01
only to implement that function over here, but it saw something
18:03
else and decided to go refactor that as well.
18:05
And so you end up with sort of like all this collateral
18:09
probably, you know, at that time in the early days was, was, was
18:11
damaged. That would then have to, you
18:13
know, be resolved by the human going like, well, I asked you to
18:16
do one thing and you ended up in like 3 different spots and, and,
18:19
you know, kind of screwed everything up.
18:21
And you know what? That's where we had to spend a
18:24
lot of time, especially. And, and one of the big reasons
18:26
we built Brain Grid was to to sort of create those guard rails
18:29
to keep it on track, to keep it focused, to keep it working in
18:33
the right places as opposed to that kind of wandering all over
18:36
your code base and making unwanted changes.
18:39
Yeah, I love that one, especially as like, you know,
18:42
novice engineer myself who's pretty familiar with these
18:44
tools. It's very easy if you don't have
18:47
foundational knowledge to ask the wrong thing or use the wrong
18:50
language or something like that. And I think to your point
18:52
earlier about the, you know, the software engineers not being
18:56
obsolete just yet, I think that's really where they start
19:00
shine, right? And and there's parallels to
19:02
this, right? I remember like when I was a
19:03
kid, I'm just old enough to have remembered a little time pre
19:06
Internet. And so, but I know that there
19:08
was an era where like if I wanted to be online, I had to
19:11
learn to code. So I had to do, I had to be able
19:12
to do a little HTML if I wanted to put up my stupid little
19:15
landing page so that I could get out of my little, you know, Palm
19:18
Pilot or whatever I had. And so, oh, BlackBerry Pearl,
19:21
that's what it was. And and so anyway, so I had to
19:24
learn a little bit of code to get there.
19:25
And so, but learning that little bit of code was then you this
19:28
the differentiation between me and like a pure novice or a pure
19:31
amateur, right. Then there was this shift that
19:33
happened where you started to see things like WordPress and
19:36
they'd be got got more sophisticated over time, but you
19:39
know, sort of the WYSIWYG style editors where people who didn't
19:42
really know anything could now start delving into stuff that
19:44
used to require team. And so there was sort of that
19:46
transition where it was like, OK, well, now my HTML knowledge
19:49
is kind of irrelevant. Like I'm not such a developer
19:52
that my code blank, you know, was strong, but I'm actually
19:55
stronger now in WordPress or using whatever WYSIWYG type
19:59
styled editor because I have some foundational background in
20:03
HTML. And then in my own experience, I
20:06
took a stint as or took like a year as a CMO at a tech startup,
20:11
software startup. And so I learned a bunch about,
20:14
you know, APIs and building out schemas and doing all this
20:16
stuff, you know, actual software development work.
20:19
And now I've gone to this transition where again, the the
20:22
world has shifted a little bit. And now amateurs can dally in
20:24
the land of the experts. You know, we can use these tools
20:27
to be able to do some of the things that it used to require a
20:30
high level engineer. But because I've got this extra
20:33
foundational knowledge, I can push that tool further than, you
20:36
know, your average amateur. But so to your point about the,
20:40
the need of of serious software engineers like you guys, it
20:44
seems like could push this stuff so much further.
20:46
Even still that, you know, you mentioned earlier being able to
20:49
kind of one shot your, your approach, you know, and, and
20:52
land something that was reasonably good in one shot.
20:55
I think that is where the line is different.
20:57
So I think a lot of people are looking at this is like this
20:59
really giant change. And I mean, AI, don't get me
21:02
wrong, is a wild technology, but at the same time, it's like
21:06
almost stage 4 of this web development cycle that's been
21:09
going on for years, you know? And so even though it is a push,
21:12
it's made the amateurs better, but it will make the pros even
21:15
more, you know, more better. It's probably not the way to say
21:18
that they're going to make sense.
21:19
Stronger development. Well, I mean, it's, it's, I
21:23
don't think it necessarily makes the pros better.
21:26
It makes them more effective and more efficient.
21:32
It increases their ability to produce output, right.
21:35
So at the end of the day, as an engineer, you know, you, you
21:39
need to produce output and then, and the output of that is, is
21:41
typically software, well designed software and, and code
21:44
that is providing, you know, performing some level of, of
21:47
functionality or feature for the product.
21:50
If you take a principal level engineer and they, you know,
21:57
dive into the tooling, especially something around like
22:00
Claude code, and they get really good at sort of orchestrating
22:04
that, you know, you're going to see their productivity
22:07
essentially go through the roof. They are going to be able to
22:11
replicate essentially what's in their head multiple times across
22:15
multiple agents and orchestrate the outcomes that they expect as
22:19
if they were doing themselves. And they're probably also going
22:22
to be able to cover more because they're going to allow the
22:24
agents to sort of deal with some of the things that maybe they
22:28
don't necessarily. A lot of times they're like, I
22:30
don't want to do this logging or, you know, this sort of stuff
22:32
over here. I'm not super interested in or
22:34
the testing, but they're going to be able to cover all of that
22:37
now because they're going to essentially, you know, manage
22:39
and orchestrate the swarm of agents to increase their
22:42
efficiency and output. And, and literally take what's
22:44
in their head, their craft, what they've been doing.
22:47
And sort of translate that into, you know, that the prompts and
22:52
the the instructions for those agents to more or less like, you
22:56
know, replicate them in, in a much more meaningful and sort of
22:59
larger manner. And then if you look at the
23:02
amateurs, what's happening is all of that knowledge of what
23:08
we've been doing for, you know, decades or the last 10 years,
23:11
depending on where you're at, however long, we're sort of
23:13
bringing that knowledge into these harnesses and into these
23:17
agents. So that way somebody that's
23:19
never written a line of software that in their life can
23:22
essentially express themselves and their idea in, you know, in
23:27
in natural language and an agents or you know, team of
23:31
agents can take that and actually build that into
23:35
correctly built software. Yeah, no, I love that.
23:38
Well, and I think that's a good delineation too.
23:39
It's sort of, you know, it's pouring gas on the people who
23:42
are great, but it's not necessarily that they're
23:44
becoming themselves a better engineer, for example.
23:47
So I want him to just touch on and and we can pivot off of this
23:51
here in just a second. But I wanted to ask you about
23:53
what do you think this means then?
23:54
Or like what's the gap between current engineers and those that
23:58
might be coming through the pipeline now?
23:59
I always joke about this. I have a Co host here, but he's
24:01
got some work that conflicts sometimes.
24:03
So we can't always get together, but he went through this big
24:07
software accelerator to become a software engineer maybe two
24:11
years ago, right? So maybe the worst time to get
24:13
to junior software engineering. But he nonetheless, he did it
24:17
and he learned a ton. And you know, to your point or
24:19
the point that you just made about, you know, he's a much
24:22
better AI pilot than your average bear, right, because he
24:26
at least has some foundational knowledge.
24:28
But the job market, I have to imagine is going to get a little
24:32
funky for these people as they kind of transition from the old
24:36
school, the guys that have years of experience in engineering to
24:40
sort of whatever this new version is.
24:42
And I don't know if it's the same where you just come
24:44
through, you learn your code, you practice, you do your code
24:46
challenges, etcetera. Or do you actually just train to
24:49
become like an AI cut? You know, I always use the word
24:51
pilot, you know, So do you learn to pilot AI?
24:54
And then, you know, that's the gig now.
24:57
Like the software engineer is the Super AI pilot that knows
25:00
that platform and knows that tool inside.
25:02
Now. What do you think the the future
25:03
looks like for those guys? I mean, yeah, I mean, I, I think
25:06
that's largely where it's going, right.
25:09
So, I mean, you start to take, you know, the, the, the folks at
25:13
my level have been doing this for a long time.
25:15
Obviously, we're going to start to age out and, you know,
25:19
juniors are going to have to come in and, and start to
25:21
replace us as, you know, we're retiring or going off to do
25:25
different things. And, you know, those juniors are
25:28
coming into an entirely different world than, you know,
25:31
what I came into. You know, they're coming into a
25:34
world where, you know, there is a high probability, if not 100%
25:39
certainty, that that agent is going to be able to write, you
25:43
know, tremendously better software than they are because
25:46
of the they just don't have the experience and, and, and really
25:49
it's just been trained on so much more.
25:51
And so like you said, they're going to have to become sort of
25:54
the, the pilots of, you know, what gets fed into those agents
25:58
and how do they keep those agents, you know, performant
26:01
and, and doing what they're supposed to be doing?
26:04
You know, I think a lot of people talk about the software
26:08
factory and, and you can probably see it all over X and,
26:11
and things happening out there and I don't think it's there
26:14
yet. And when I sort of describe the
26:17
factory, it's literally like you have inputs in the form of like
26:20
requirements and feature requests and different things
26:24
like that. And the factory spits out, you
26:26
know, fully tested, you know, operational code that can then
26:29
just be deployed right to production.
26:31
And that's probably an end state that shows up at some point.
26:35
But it's all these, I don't think they're yet.
26:39
And so you're still requiring, you know, sort of humans on the
26:42
keyboard and in a sense, or you know, at least humans speaking
26:45
into a microphone essentially in natural language, you know,
26:49
describing what it is they need to do.
26:51
And then augmenting that with context from specifications,
26:56
requirements, you know, what have you and, and getting these
26:59
things to, to build it and build it the right way.
27:02
And so that still needs sort of that shepherding across the line
27:05
to get it all the way through. All right, yeah, no, I think
27:08
that's right. Let's kind of pivot back into
27:12
more of what you're doing that bring bring grid today and some
27:15
of the common problems. So for people who maybe aren't
27:17
super familiar with this kind of work, there's a number of these
27:20
platforms out here. And I I will probably say some
27:22
stuff that's wrong about your business, Tyler, so feel free to
27:24
correct me, but there's a number of people using these different
27:27
platforms out there. Like Lovable is 1, but there's,
27:30
you know, lots of these things where people, I think Replit is
27:33
another one. People are using these things to
27:35
build apps, They're using it to build websites, They're using
27:37
all these things and basically you can prompt a natural
27:39
language kind of what you want it to do.
27:41
So for, you know, our conversation earlier where we
27:43
were talking about the novice being able to just kind of speak
27:46
their mind and then have an AI build something out of it.
27:49
That's that's possibility through these platforms.
27:51
It's not always great. Sometimes it yields bad output,
27:54
you know, whatever, but it's getting better all the time.
27:56
So I wanted to talk to you a little bit because I think that
27:58
where you guys start to step in is in getting past the sort of
28:02
amateur level where people are toying with the stuff and trying
28:05
to develop real software using magentic tools or AI tools, you
28:08
know, whatever your platforms are.
28:11
And like we've talked about, you know, AI is definitely made
28:13
building software faster, but what has it not made easier in
28:17
your view? I think two things.
28:22
First and foremost is what to build.
28:25
I don't think it's made that necessarily easier yet.
28:28
I think from what I've seen, you know, the humans still have to
28:33
curate through their own ingenuity and understanding of,
28:40
you know, their customers, their marketplace of what needs to be
28:43
built. I think there's still a lot of
28:46
work there that that is largely human driven.
28:49
So they're providing that input in terms of like, hey, here's
28:52
here's what's going to unlock our next level of growth.
28:54
Here are the things that we feel like our customers need.
28:57
Here's what we've heard from our customers and some of that that
29:00
has to be curated and then turned into software that has to
29:04
be built and whether that's turning into requirements.
29:06
But there's some sort of level, there's some input that's coming
29:08
out of all of that that humans are doing that's in being fed
29:11
into AI to actually go in and build.
29:16
And then I think the, you know, the, the next phase is, is just,
29:22
you know, the, the ingenuity, right?
29:24
So, you know, the humans are still the ones that that have
29:28
that level of creativity of something new that needs to be
29:31
built or something new that doesn't exist or a novel way of
29:35
of, you know, approaching a problem through the lens of
29:39
software. You know, they're the ones that
29:41
are coming up with those ideas. And then they're the ones that
29:44
are having to, you know, to work with the agents in order to get
29:47
that built. You know, we're not seeing yet
29:50
where, you know, all of a sudden, you know, an agent has
29:53
launched a, a brand new, you know, software application that
29:58
it came up with all on its own, that it decided there was a
30:01
market for it. It decided there was, you know,
30:04
customers and, and just all of a sudden, you know, automatically
30:06
built everything. I mean, humans are still the
30:08
ones that are sort of like the the drivers of all of that
30:10
ingenuity and and the new things that we're seeing.
30:13
Yeah, I think that's right. And I think it this maybe lands
30:16
into a little bit, but I mean, I think people spend maybe too
30:19
much time focusing on the intelligence part of AI and not
30:22
enough time focusing on the artificial part, which is to say
30:25
that it's not really thinking. You know, you're talking to it
30:28
like it's a human and you're treating it like it is, but it
30:30
isn't. And it's trained on a bunch of
30:32
knowledge. And it's sort of, you know,
30:34
without excessive prompting will sort of give you kind of an
30:37
average of things or give you kind of a middle, you know,
30:39
middle of the ground kind of answer on a lot of stuff.
30:41
And, you know, so, but I think people think they're talking to
30:44
something that's truly smart, You know, the same way that you
30:46
might talk to, I don't know, some super genius name your guy.
30:49
And you know, like where they're conjuring this out of, you know,
30:53
just their sheer brain power. But the, the reality is it's
30:56
only as good as the input really in the sense that, you know,
30:59
it's, it's just, it's not smart. It's not thinking about things
31:02
in the same way that humans still think.
31:04
And so, and maybe we'll get there one day or we'll be able
31:06
to fake it well enough, but right now I don't think we're
31:08
there. And, but I think that's where
31:10
brain grid really starts to step in, which is around the product
31:12
planning piece of this. And, you know, so to your point,
31:15
I mean, yes, the ingenuity and all those things, but I think
31:17
actually documenting this, you know, not to, you know, beat it
31:21
too hard. But the, the thing you said too,
31:22
about 1 shotting, the way that you're capable of 1 shotting a
31:25
great output from the very beginning is through great
31:29
planning upfront. Like, I mean, you didn't just
31:31
say, Hey, I want an app that does this.
31:32
Like that'll get you something. But you know, if you actually
31:36
have a full blown set of documentation or whatever, and
31:38
you know, maybe you're using AI to help you write the
31:40
documentation, but you know, you're definitely creating a, a
31:44
product plan or something around this.
31:46
That's that's making it or that's drawing the line between
31:49
sort of amateur software and professional software, I think.
31:52
Yeah, I mean 100%. I mean, if you come in and and
31:55
you describe like pay, I want to build a, you know, a Kanban
31:59
style task board to manage my my household.
32:03
And you know, you just drop that in somewhere and you let AI sort
32:07
of reason about what it is that that means.
32:11
I mean, you're going, it's, it's going to come up with something,
32:14
but is that something necessarily what actually exists
32:18
in your head, your, your idea? You know, you gave it a, a few
32:22
words, maybe a couple of sentences, and you left a whole
32:25
bunch of stuff open to interpretation, which is largely
32:28
going to be interpreted by its training models and everything
32:30
else. And it, it, and it's, it's sort
32:32
of designed to go now build that.
32:35
Whereas what we try to do is take a step back.
32:37
So you take those, those same couple of sentences and words
32:40
about what it is you can describe that's in your head.
32:43
And we want to tease that out into an entire plan.
32:45
And we want to actually interact with you to ensure that we that
32:49
we're extracting what does, what does this actually mean?
32:52
Like, what do you want that Kanban board to look like?
32:55
What do you want to have happen when, you know, I drag one from
32:58
the other? And we want to have that sort of
33:00
conversational interaction with the user to really ensure that,
33:05
you know, those couple of sentences that they wrote are
33:08
going to be turned into what they're really thinking about.
33:11
And I think where that becomes really important is for the, the
33:14
non software folks, so that there's plenty of folks out
33:17
there that have, you know, deep domain expertise in, in areas,
33:23
but they don't know, I mean, they're not engineers.
33:25
And, and, you know, in the past, somebody with, you know, deep
33:27
expertise would have to go raise money or have some friends that
33:31
were engineers. They would have to come together
33:33
and, and the engineers would try to distill everything out of
33:36
their heads and, and try to build it.
33:38
And I think now that that's not necessary anymore.
33:41
Now that person can sit down, could sit down with Brinker and
33:45
say, OK, here's my idea and how do I build this?
33:49
And, you know, we're going to look at that idea.
33:52
Our agents are sorry. Our agents are going to look at
33:53
that idea. They're going to break it down.
33:56
They're going to turn that into a project plan.
33:59
It's going to start with an epic and a bunch of requirements, you
34:02
know, depending on what is and the complexity of what you're
34:04
trying to build, it's going to understand, hey, does this need
34:07
a database? Hey, what kind of security does
34:09
this need? You know, what are the, what are
34:11
the, the components that are going to make up the system?
34:14
And it's going to think about that.
34:15
And then it's going to start to present you with, with
34:17
questions. It's going to ask you things
34:19
like what did you really mean by the statement?
34:21
And, and what do you want it to do in this, in this case?
34:23
And it's going to help you work through those edge cases that
34:26
you would, you know, in the past would be a product manager and,
34:29
and group of engineers sitting down and working through all
34:31
that stuff. Now you're able to do that
34:32
yourself, literally having conversations with our agents
34:36
inside a brain grid. And then as you work through
34:38
that and you get that requirement specified and, and
34:41
and built out, you're going to be able to take it and actually
34:44
build it with brain grid. Brain Good is going to go
34:46
through that. Take that requirement and then
34:47
start building it on your behalf.
34:49
Yeah. There's two points that I just
34:51
kind of want to reaffirm. I think you're totally right.
34:54
One is, you know, sort of for the non engineer type,
34:57
especially, you know, if you're using one of these vibe coding
34:59
platforms, it's really easy to ask it for something.
35:02
And like you said, the system will start to do its thing and
35:05
it'll start filling gaps and and make assumptions on your behalf.
35:08
It'll generate something and a lot of times that something
35:10
looks like what you think you wanted.
35:11
So like to use your example of a Kanban system or something, you
35:14
know, I'm like, oh, cool, I can drag my cards and blah, blah,
35:16
blah. But the part that's missing is I
35:19
don't know, the database management of these cards as
35:21
they move from system, you know, spot to spot and how are we
35:23
tracking that? And do they persist and blah,
35:25
blah, blah. Like there's all these different
35:26
things that, you know, if you're not engineering focused or if
35:29
you don't have any experience in that, you know, maybe AI picked
35:31
up the slack and they knew what you meant, but maybe it didn't.
35:35
And, you know, without rigorous testing and other things like
35:37
this, you just don't know. And so I think a lot of times,
35:39
and I've seen examples of this, I'm sure you have as well,
35:41
things that people actually ship because they have this level of
35:44
confidence that, Oh yeah, I did this thing.
35:45
Look, it's awesome, That works great, but it works only in a
35:48
singular environment with a singular use case.
35:51
As soon as you hit refresh, you lose your data, you know,
35:53
whatever it is. And, and so I think there's a
35:56
lot of that. So I think, you know, to your
35:57
point and, you know, having a good plan and building something
35:59
from the beginning and having a tool like brain grid that allows
36:02
people to sort out those edge cases because really you don't
36:05
think about them unless you're thinking sort of engineering
36:07
minded anyway. So I think that that's really
36:10
powerful. The other thing I wanted to just
36:11
kind of add to that, I guess it's sort of related is just how
36:15
important the language is. And I will say sort of as a
36:18
novice and, you know, ask my wife, she thinks the robots are
36:20
going to come out to get me as soon as they're fully sentient.
36:22
Because as soon as it starts slipping, I'm like, dude, what
36:25
are you thinking? Like, come on, like, why would
36:27
you make such a stupid decision? You know, like I talked to it
36:29
like it's bad, you know, a bad friend or something.
36:32
And, and so, you know, you know, at some point my professionalism
36:37
slips and I start getting really loose with my language.
36:39
But I find that that's when I start getting the worst output.
36:42
You know, if I treat, if I speak a little more mechanically and I
36:45
try and use the technical terms, although I will say my 1 little
36:48
hack as I always lead conversations or any big feature
36:51
addition or feature change to say, I say I'm not a web
36:55
developer. So just understand, you know, so
36:57
as I'm prompting the system, I'll be like, I'm not a web
36:59
developer. I don't know what I'm talking
37:01
about. I'm going to try and explain
37:02
this, but I might use the wrong language, you know, help me
37:05
through that. And, and it usually does a
37:07
pretty charitable job of, of working with me.
37:09
But so, but I think just to back up your point that I think that
37:12
that's really where products back and understanding, you
37:14
know, first of all, the capability of these tools, but
37:17
understanding the you, you just don't want to be leaving it up
37:21
to assumption to the system. And the parallel that I draw is
37:25
in the olden days, I guess even as recently as a few years ago,
37:29
we were still using like development teams out of India
37:32
and Pakistan and places like this to to help us with web
37:35
projects. And you know, and we still do on
37:38
bigger builds because I still think it's good to have a human.
37:40
But the the point I wanted to make is that, and you can
37:44
probably relate to this too if you've done some of the same, is
37:47
if you don't have your product spec perfect.
37:49
If you don't have a PRD that uses the right language and
37:51
explains what you want it to do in, in the way that's the proper
37:54
way of saying it and really articulate your points.
37:57
If you don't do a good job planning, even with human agents
38:00
on the other side of the world, you know, as they're working on
38:02
these things, they will start to make assumptions or they'll
38:04
guess or they'll do whatever. And then you end up with bad
38:07
output and you end up with extra cycles and you end up with more
38:10
billing. And that's where these web
38:11
projects get the the reputation of just, you know, becoming
38:14
wildly expensive and flying off the rails and getting way out of
38:16
scope is that a lot of times they just didn't take the time
38:19
to plan upfront and, or, or said something and left something
38:22
open to interpretation. I mean, 100%, I mean, I've seen
38:27
that in, in real time and we always had that.
38:30
The, the saying is like, you know, you're throwing it over
38:32
the wall, right? So like, basically, you know, if
38:35
you write this, you know, you write this requirement in
38:38
isolation and then you throw it over the wall to the development
38:41
team and expect to get perfection back or expect to
38:44
get, you know, what was in your head that you didn't clearly
38:47
document or you didn't write down.
38:49
But you, you know, you wrote something in that spec, you
38:52
know, you're going to get wildly disappointed.
38:54
And I think with AI, it's, it's exactly the same.
38:57
I mean, context is keying. And, and that's why, you know,
39:00
that the first iteration of our product was very focused on
39:04
bringing, you know, some rigor and structure to the sagentic
39:08
process. And that rigor and structure was
39:10
in the form of, you know, well specified requirements, epics
39:15
and, and, you know, breaking that down into tasks and, and,
39:18
and, you know, really giving it something that had been well
39:21
thought out instead of you just going like, hey, here's this one
39:24
line. And I expect magic because as
39:27
soon as you expect magic from that thing, you're just, you're
39:30
going to be wildly disappointed and you're not going to get it.
39:33
I think some of that has been reduced with the newer models.
39:38
They've gotten a hell of a lot better and, and you know, the,
39:41
the harnesses and cloud code and things like that, they've,
39:44
they've done better training and they've, they've thought through
39:46
some of those things, probably because they kept seeing people
39:49
get wildly disappointed when they were giving him one line,
39:51
you know, requirements. But you know, we saw that very
39:54
early on because, you know, we were still trying to keep our,
39:57
our business running and, and, and ship product and, and we
40:01
kept getting things that were like, man, this is garbage.
40:02
We can't use this. How do we get this?
40:04
We know it can do it. So what do we have to do?
40:07
And you know, it's, it's became very important to get that thing
40:10
specified and get it broken down.
40:12
But what's hard is when you've got users now.
40:15
And our user base, I think a lot large for a large part are are
40:18
definitely non-technical folks. How do you get them to have that
40:23
conversation? And so, you know, we have to
40:25
sort of ask the questions like, Hey, is this, you know, again,
40:27
like, is this what you're really thinking?
40:29
And here's the here's the recommendation.
40:32
You know, the agent would would say you should take, but here's
40:34
some other ones. And if you know, or do you want
40:36
to do something completely different?
40:38
And so it's, you know, again, it's, it's trying to take those
40:41
non-technical folks and sort of like walk, you know, hand hold
40:44
them through the path of, of building software.
40:47
Well, and I think that that, you know, not to blow too much smoke
40:50
up your butt, but I think this is a really good place for a
40:53
business like this because it's, it's clear that people will try
40:56
and do more themselves than maybe they would have in the
40:59
past as it pertains to web development because of tools
41:02
like this. And so having tools like yours
41:05
or having organizations like yours that will kind of shepherd
41:07
people through this, that will help them get the training or
41:09
understand the questions or even just know the questions to ask,
41:12
right? Because for most people, it's
41:13
like, maybe to your point earlier, you're talking about
41:15
people having a clear vision of what they want, but they just
41:17
don't know how to say it. And so, you know, so in those
41:20
situations, I, I think a lot of times just because of their lack
41:23
of engineering training, they're asking the wrong questions or
41:25
not providing the right data. And you know, just by doing that
41:29
planning in advance, you can greatly increase your, your
41:32
output. I wanted to talk to you a little
41:34
bit about just to sort of, and we've talked about it a little
41:36
bit throughout the show, but I, I kind of hate the term, but the
41:39
vibe coding thing I wanted to talk to you about, you know,
41:42
these would be the platforms like Lovable and Ruplet and
41:44
these other things that I mentioned earlier.
41:45
But like, what do you think that the vibe coding is getting
41:48
right? So for people who are trying to
41:50
use it or maybe striking out on their own and they want to give
41:53
it a shot, they don't realize they have a problem that they
41:55
need brain grid for yet. They will, but they haven't got
41:58
there yet. So, but what, what do you think?
42:00
Vibe coding or the the platforms and we can just stick to the
42:03
major players or whoever you have the most experience with
42:05
are getting right. And then where do you see that
42:07
stuff start to break down? I mean, largely what we've seen
42:10
is what they're getting right is that you know that idea to the
42:14
aha moment. And so again, if you're, you
42:17
know, non-technical, non developer, you've got this idea,
42:21
you know, you can go on these platforms and you know, to the
42:24
best of your ability, articulate that idea and you know, you're
42:29
going to see something and you're going to be like, holy
42:31
crap, like that, that is actually on the web.
42:34
Like that is that appears to you as as being real.
42:37
And so it kind of like, you know, you get that dopamine hit
42:41
of like, I took an idea, now I've got software, but you know,
42:46
is it, is it really, and, and what level is that software
42:49
actually at? Is it something that like, Hey,
42:52
I had an idea and all of a sudden magically I've got a, you
42:55
know, you know, highly scalable production level system that I
42:59
can ship and serve a million people.
43:00
No, it's not going to happen. Like that's, that's, that's a
43:03
pipe dream right now. But again, what they're getting
43:06
right is, you know, they're taking people that, you know,
43:10
largely have had ideas and could never do anything in software
43:13
because they, again, they weren't developers and sort of I
43:16
would say, you know, giving them hope in in terms of like, hey, I
43:20
can take this and maybe I can do something with it.
43:22
And, you know, I can get it to a certain stage of maybe now I
43:27
need to go get a technical Co founder.
43:30
And the reason I say that is we had a lot of early customers
43:32
coming to us and we're like, hey, I've built this thing on
43:35
lovable and it's getting to a, you know, a level of
43:38
sophistication that anytime I ask it to do this, it breaks the
43:41
rest of the system. And I just can't get beyond
43:43
this. And then, you know, I, I went
43:46
over to you guys and I basically told it the same thing.
43:49
You built a requirement, you gave me a plan and I let you
43:53
know, execute it with Claude and holy shit, it just works now.
43:57
And you know, that was kind of like our, you know, our aha
44:00
moment is like, look, you if you do the planning and you get the
44:03
context right, these agents are going to build it, build what
44:06
you need, and they're going to do it the right way.
44:08
And so, you know, largely where we would see people leaving
44:11
those platforms is, you know, they were like, Oh my God, I
44:14
just spent you know, weeks putting this thing together and
44:18
it's just not quite right or it keeps breaking or it, you know,
44:21
completely just, you know, I'll ask it to do one thing and it
44:23
destroys everything. I'm so frustrated like save us
44:27
basically and you know, and that's, you know, that's how we
44:31
end up getting most of our, a lot of our customers coming that
44:33
way. Yeah, well, and I'll say it's
44:35
interesting too, because with a lot of these platforms they're
44:37
sort of token based or you get so many fries at something.
44:40
And so in having similar experiences myself in, in
44:44
lovable, like if I didn't have a high enough token limit, for
44:46
example, you can spend a lot of your time, your tokens, your
44:49
bandwidth, whatever the, the metric they're, you know, your
44:52
supplier is using. You can spend a big chunk of
44:54
that just fixing the problem that you just made through 1 bad
44:57
prompt and then you've burned your month's usage or you, you
45:00
know, whatever. So now your project has kicked
45:02
another month down the road. And so efficiency in that sense
45:06
is really important. You know, I mean, unless you
45:07
have some kind of, I don't know, multi gajillion dollar
45:09
grandmaster, you know, level version of of whatever platform,
45:14
but like, I mean, I use Claude Max and you know, I'm constantly
45:17
hitting usage limits and stuff like that.
45:19
And and that's the big plan. You know, I mean, well, the
45:22
bigger plan, the bigger consumer level plan.
45:24
And so so anyway, so I think that, you know, to your point,
45:28
like getting it right or having efficiency around those things
45:30
is great because if you're constantly going back and fixing
45:32
what you did, you end up burning a lot of that stuff.
45:36
So that illusion of completion idea is.
45:38
Is, yeah, you burn a lot of it and you're just, you're, you're
45:41
frustrated and you're like, now what do I do?
45:44
Like I'm kind of screwed. I've got to go find something
45:46
else because, you know, I'm out of credits.
45:49
And now I've got a now I've got like, oh shit, I've got to pull
45:53
that code down. Do I have to go sign up for a
45:55
Claude Max? Do I have to get a Claude Pro?
45:58
Do I need to go to codecs? Now I've got to try to set up a
46:00
development environment. And, you know, as as we sort of
46:04
evolved our platform, you know, we kept seeing people, you know,
46:08
writing in the messages like they built this plan out.
46:10
It was a beautiful plan. They're like, OK, now implement
46:12
it. And our agents like, I'm sorry,
46:14
I can't do that. You know, I'm a planning agent.
46:17
I don't I don't do that. You need to use plot or cursor,
46:20
whatever else it was. And we found that to be a like a
46:24
drop off point because, you know, think of the non tech
46:29
folks trying to set up a local development environment, you
46:33
know, on Windows or you know, their MacBook like that's, oh,
46:37
you got to get in the terminal. You know, how many?
46:39
What's a terminal, you know, and it's like they don't necessarily
46:43
want to spend the time or the investment to kind of walk
46:47
through getting all of that stuff set up.
46:48
And it's, it's a pretty big leap for, you know, a number of folks
46:51
to be able to operate at that level.
46:55
You know, maybe for you and I that's like, OK, whatever second
46:57
nature. We've done it a million times.
46:59
But you know, the, the person from logistics that, you know,
47:03
has this amazing idea. They're probably not going to
47:06
want to, to do that sort of stuff.
47:08
And then they're going to hop into a level bowl and they're
47:10
going to get something to a certain point.
47:12
And then when it gets to the level of complexity, you know,
47:15
they make it frustrating. Now where do they go?
47:18
Yeah. So I wonder if we could just
47:19
take a second and just talk about maybe the difference
47:21
between prototype and product. Because I wonder if this isn't
47:24
maybe a line where we could draw, you know, something
47:26
definitive in the sand and say, hey, basically this is kind of
47:29
where you're kind of good to operate on your own, which is
47:32
maybe you're building sort of a a rough draft or you're just
47:35
trying to get a concept out or maybe get a proof of theory or
47:37
something. And then there's like this other
47:41
thing that's the, the production level, you know, ready to go.
47:44
And I wonder if you just talk about sort of the difference
47:46
between making a prototype versus making something real.
47:50
And you know, I put real on quotes because, I mean, I guess
47:52
if you can see it, it's real. But you know, but like you made
47:55
the point earlier, you know, could you actually run this with
47:57
a million users? And, and the answer's no.
47:59
So what's the difference between sort of a prototype versus a
48:01
real product? Yeah.
48:03
I mean the, the prototype to me allows you to, you know, prove a
48:09
part of a hypothesis. And you know, a lot of times
48:13
being able to see and interact with that prototype is going to
48:18
allow you to form a, a more educated opinion about, hey, is
48:23
this, is this the right layout? Is this the right sort of
48:26
workflow that we want? And obviously with AI now you
48:29
can do that very cheaply and quickly.
48:33
And I think, you know, most product managers today should be
48:37
doing this, you know, probably at the same time they're either
48:41
writing that spec or working on that spec, or they're taking
48:44
that spec and saying, all right, let's turn this into a
48:46
prototype. And like, you know, walking
48:48
through the the workflow, walking through the buttons, is
48:51
that in the right place? Do I like this?
48:53
Does this make a lot of sense before sort of giving it off to
48:57
maybe it's the engineering team or, or taking it to the next
48:59
level? The, the prototype should allow
49:02
you to answer the questions that you have of the feature that you
49:07
want to build and allow you to sort of visualize what it is,
49:10
you know, possibly you know, or at least prove your, your
49:14
hypothesis that this makes sense.
49:16
The production level is can I take that and operate that in a
49:23
repeatable, reliable manner where I can observe it, where I
49:27
can scale it, where I can understand it, where it's, it's
49:30
going to stand up to continual usage.
49:34
It's going to stand up to people doing things to it that I didn't
49:38
necessarily anticipate and not blow up in extraordinary ways.
49:43
Or at least when it does blow up, it it blows up in a way
49:46
that, you know, the users not left in a completely crashed out
49:50
mode or something else like that.
49:51
It's, it's taking it from something that's helping you,
49:54
you know, answer questions, prove an idea to something that
49:57
can be used in a more general purpose at a scalable.
50:01
And whether it has to scale to, you know, 1050 to 1,000,000,
50:05
that's going to be the next level of it.
50:08
And there's a whole lot more engineering aside from just
50:11
like, hey, I built it to, you know, to scaling it to operating
50:15
it, to observing it, to being able to ensure that you know,
50:19
you're reacting to errors properly and, and, and people
50:22
are having the right type of experience with it.
50:25
Yeah, No, I think that's right. Well into, you know, just kind
50:28
of use again, it's personal example, but like with the
50:30
recent website I built for our company, I had built it using
50:34
Figma make as a way to generate the UI basically for it.
50:38
And what it does is it builds sort of a React JavaScript kind
50:42
of platform that, you know, it looks great, it's really
50:45
clickable, it loads really fast. It does all this stuff.
50:48
Well, it wasn't until much later that I realized as a non
50:50
engineer, non developer myself, that those kinds of websites are
50:53
like not crawlable. Basically they just, you know,
50:56
it's in five lines of text. It's you know, when you, when
50:58
you look at the source for those things or whatever.
51:00
So all the data I was putting on there that I was hoping I could
51:02
rank in Google or whatever for none of that stuff was visible.
51:06
So after a long heart to heart with Claude, we had a
51:08
conversation. We put together an idea for how
51:10
I can take this website that I really like, but how how can I
51:13
make it so that it works a little better?
51:15
And we ended up moving to a platform where back end
51:17
infrastructure called Astro and, and basically rebuilt the
51:21
website or refactored it on top of this new platform so that it
51:24
would render full HTML pages and all of our pages could then be
51:27
crawlable. And so, but those I think are
51:30
the kinds of things that are the big delineator, one of the big
51:34
boundaries between sort of the amateur and and the
51:36
professional, which is it didn't even occur to me until I
51:40
discovered the problem. And then luckily I had enough
51:42
wherewithal to be able to fix the problem.
51:44
But, but if I just shipped that or had gone to market with it or
51:48
something like that, like I'd be like, oh God, how come nobody
51:50
can find my website? Like, why am I not coming up on
51:52
Google? What's going on?
51:54
You know, and, and so it would end up being a bigger problem.
51:57
And if I didn't have the language or the right questions
51:59
to ask, I wouldn't even know how to solve it, you know, or where
52:01
to go. No, you would, you'd basically
52:03
be stuck like everyone else. It's push something out there
52:06
that hey, I've I've got a domain and I've got something behind
52:10
it, but no one can discover it. Nobody knows about it.
52:14
Nobody, you know, unless I somehow got very lucky because,
52:18
oh, I, you know, I never published my site map.
52:21
Oh my, you know, my SEOI open up.
52:23
I don't even know what Lighthouse is, you know, you
52:26
know, all of these, these other things that come with sort of
52:29
that level of professional software development, you're
52:32
just not, you're just not going to know about them.
52:34
And, you know, that's another thing that we, when we think
52:37
about our platform, because again, we're sort of targeted.
52:40
It's, you know, it's sort of like those, those consumers that
52:42
are not necessarily technical. Those are all things that we we
52:46
have to think about, like we have to think about for you on
52:48
your behalf, because we can't have the expectation that you're
52:52
going to, to even know about these things.
52:55
So how do we allow our builders, our agents to sort of like jam
52:59
all of that stuff in there? So it's there and available and
53:02
is actually just behind the scenes, already there and
53:05
working. Yeah, Yep.
53:07
No, but I think that's right. So in these last few minutes, I
53:10
want to spend a few minutes just talking about the brain grid
53:12
structure. And if I have this wrong, feel
53:14
free to correct me. But as I understand, you guys
53:16
use sort of this build, verify, trust kind of framework where
53:18
you start with capturing, structuring, building and then
53:21
verifying. And, you know, obviously we've
53:23
been talking about the reasons why verification and, and these
53:26
things are such a big deal. But I wonder if you'd sort of
53:28
just take us through the process of running a project with you
53:30
guys. You know, what does that mean to
53:32
capture it, develop the structure you know, to build.
53:34
You know what it what does this look like in terms of an
53:36
engagement with brain grit? Yeah, so I mean it.
53:39
It all starts with the idea. So everything you know we do
53:42
starts with an idea. So you provide that idea to us
53:47
in natural language, we will then our agents will then I keep
53:51
saying we, but it's our agents will essentially, you know, look
53:55
at that idea. And the, the first thing that
53:58
it'll do is we find most, most people that come to us don't
54:02
even have like a, a GitHub repo, right?
54:04
And so you don't need any of that stuff.
54:06
We'll take care of that for you. So we will create a brain grid,
54:09
brain grid manage repo for you. And then the first question we
54:12
look at is based upon that idea that they've given us, does this
54:18
need data persistence? Does it need a database?
54:20
If it does, we'll provision one for you.
54:22
So we'll then provision a brain grid managed database for you.
54:25
Then we'll go into actually right, turning that idea into
54:29
sort of that first specification.
54:32
And again, our agents are going to look at it and say like, hey,
54:36
can I do this in just a single requirement?
54:41
If it's sufficiently easy enough that the agent feels it can do
54:46
it in a single requirement, it's going to write that single
54:50
requirement and send you straight into the build process.
54:53
And then the builders going to take it over and start doing all
54:55
the implementation. If it's not, it's going to look
54:59
at it and say, OK, let's actually break this down more
55:02
because I don't quite have enough information.
55:05
It's probably got questions. Let's turn this into an epic.
55:09
And then a certain number, you know, in requirements after that
55:12
in order to satisfy that, that epic and what it is that you're
55:16
trying to build. And so you've come in and let's
55:19
just say you take the, the more complex path.
55:22
The first thing you're going to get is a project brief.
55:24
And that project brief is going to be our interpretation or the
55:28
agent's interpretation from your idea of what it is you're trying
55:31
to build. And that's going to be the, your
55:33
first opportunity to have to start that engagement with the
55:35
agent. And it's going to ask you, hey,
55:37
here's what I think you're trying to do.
55:39
Does this make sense or am I missing something?
55:42
And so you can then go back to and say, like, that's close, but
55:46
I wish you would do this, this and this, and then we'll refine
55:48
it. And then it's going to break
55:50
that down into the requirements. And so let's imagine whatever
55:53
you're trying to build, you get 5 requirements out of it.
55:56
And so then you're going to, it's going to sort of put that
55:58
skeleton together. And then what you have to do is,
56:01
is now specify each one of those requirements.
56:04
So you go into the first requirement and you'll click a
56:07
button that says specify. That agent will then start to go
56:10
through everything and say, OK, I see you want to build this.
56:13
I can see the epic in the project plan.
56:16
I have a few questions. Let me go through those
56:18
questions. Great, you've answered them all.
56:21
Here's requirement 1. And now you could actually just
56:23
go straight into building it. And so essentially the pattern
56:27
is, you know, you've got this project plan, you specify those
56:31
requirements, you build the requirement, and then you could
56:34
publish that requirement, then you go back, specify the next
56:37
requirement. So it's essentially synchronous
56:39
of, you know, not necessarily parallel, but you know, one step
56:43
at a time of taking you through that development process
56:47
requirement by requirement until you've got the entire thing
56:50
built. The builder was going to take
56:52
care of like all the testing and everything else like that.
56:55
It's going to build all of that framework into, you know, the
56:59
application that it's building and then you're going to be able
57:02
to deploy it and you're going to get a URL that's essentially
57:04
like, OK, there's that thing that I just, you know, spent my
57:08
time on that was in my head a, you know, an hour or so ago.
57:11
And now I've actually got a a URL out there the some sort of
57:15
working software. Yeah, no, I love that.
57:18
Well and I love that it's just, I don't know that structured
57:21
framework I think really helps people, especially non-technical
57:23
founders to to figure things out.
57:26
When you are doing this verification step though and
57:28
you're trying to work against AI built software, are you
57:31
verifying against the original spec?
57:33
Are you verifying against needs like how are you or user needs
57:36
or business goals like what is your verification process look
57:40
like? Yeah.
57:40
So each requirement that we generate not only has functional
57:44
requirements, you know, technical requirements at the
57:48
end, it's got a big set of acceptance criteria.
57:51
And that acceptance criteria is going to be dependent upon what
57:54
it is that's being built in that requirement.
57:56
And then that acceptance criteria is then used by the
57:59
builder. So when that builder pulls that
58:02
requirement down and starts to implement it, it's then going to
58:06
obviously go through all the implementation.
58:08
It's going to make sure that, you know, TypeScript validation
58:11
passes, lending passes, tests are passing, all of that good
58:15
stuff. And then when it feels like, OK,
58:17
my implementation is done, it's going to go pull all of that
58:21
acceptance criteria, Go back and look at the at the code and then
58:25
ensure that it meets the acceptance criteria either
58:28
through testing or code and provide evidence back that, hey,
58:32
this acceptance criteria is met by this year.
58:35
And it's going to do that through every single one of
58:37
those. And if to find something that
58:38
it's not meeting the acceptance criteria, it's going to loop
58:41
again and come back and and keep working through it till it gets
58:44
it right. OK.
58:45
And then on the other side of the equation and maybe you guys
58:47
help with this, I just wanted to to get clarity around it.
58:50
Like what should the founders or the people who come to you with
58:52
the project, what should they be doing on their end in terms of
58:55
auditing that first draft of AI developed software?
58:59
I mean, if they don't know everything, but like what should
59:00
they be looking at user flows, data models, security, like what
59:04
are the things that you would authoritize with them and how
59:06
much of the onus is on them to do that vetting versus, you
59:10
know, you guys providing them? Well, I mean, the vetting of the
59:12
output today is going to largely be on them, right?
59:15
So I would say first and foremost, they want to pay
59:19
attention to the input. So as they've come with that
59:24
idea and that requirement has been generated or requirements
59:27
have been generated, you know, they want to read through, they
59:31
want to make sure that it's capturing what their ideas are
59:35
and what it is they're trying to build.
59:37
And if it's not or they want to change, it is have that
59:40
conversation with the agent in order to get it to what it is
59:43
that they want. And then after the post build
59:46
and they've they've shipped something or they're they're
59:48
interacting with in the preview window.
59:50
You know, it's going to be up to them to sort of go through that
59:53
and ensure that the agent didn't necessarily miss something
59:59
which, you know, obviously as we know they they could or maybe it
1:00:02
misinterpreted something. But you know, it's kind of up to
1:00:05
them to go through and, and then play that user role and, and
1:00:08
click through and, and actually use the software and make sure
1:00:11
it's doing what they want. And if it's not, provide the
1:00:13
feedback to the agent so we can go and fix it and get it the way
1:00:16
they want. Or maybe they've gone through
1:00:18
and it's doing exactly what they want, but now they want to
1:00:21
change it and they're sort of like this doesn't quite know.
1:00:24
I thought it was this way, but I don't really like it like this.
1:00:27
I want to move the button over here.
1:00:28
I want this drop down here. Well, they got to take that
1:00:30
information, you know, and, and give it back to the agent just
1:00:33
as you would a human and say, like, all right, let's change
1:00:35
this, here's what I want done. And then they sit back and watch
1:00:38
and it's going to go through everything and and make the
1:00:40
change. Awesome.
1:00:42
OK, Well, cool. So, yeah.
1:00:43
So, I mean, I don't know, maybe I I went too far too fast saying
1:00:46
that we shouldn't be treating AI like humans.
1:00:48
I mean, they're maybe just people just like us.
1:00:50
Well, I mean, I, I think, you know, it's, it's, it's there,
1:00:53
there are humans that never sleep, right?
1:00:54
So like they're, they're available.
1:00:56
I mean, they're just there and they never get tired.
1:00:57
They never complain. You don't have to give them a
1:00:59
performance report. You know, you're not paying them
1:01:01
necessarily salary. So, but still at the same time,
1:01:04
like, you know, I try to treat mine as like, you know, kind of
1:01:09
not necessarily, you know, maybe almost like a peer, right?
1:01:11
It's like, all right, we've got to solve this problem.
1:01:13
We're going to solve it together.
1:01:16
I'm not going to write the software for it, but I am going
1:01:18
to validate what it is you've done.
1:01:20
And here's a way that we want to do it.
1:01:21
I want you to do some of the validation.
1:01:23
I want you to create some of the the testing framework, but we
1:01:26
need to solve this problem ultimately so my customer you
1:01:29
know doesn't run into this issue or or gets this new feature.
1:01:32
Yeah, no, I love that. Tyler, we're coming up on the
1:01:34
end of the hour. Do you want to take a minute and
1:01:37
just let people know where they can track you down online, if
1:01:39
they want to learn more about Brain Grid, where they can go,
1:01:41
and if they wanted to try and connect with you individually,
1:01:44
how that might happen? Yeah, I mean, I think it's
1:01:46
pretty, pretty easy to find us brain grid dot AI.
1:01:49
We're also LinkedIn, Twitter, Instagram as as well.
1:01:53
So hopefully people are seeing some of the videos our marketing
1:01:56
people are are putting out there.
1:01:59
Saw a really cool one the other day they threw out there.
1:02:00
They they basically built like a Nike clone site that was like,
1:02:04
it was amazing, looked beautiful.
1:02:07
And then yeah, I mean, I'm on Twitter at Tyler Wells and and
1:02:10
LinkedIn, you know, Tyler S Wells and easy to find me.
1:02:14
Well, cool. Well, thank you so much, Tyler.
1:02:16
I really appreciate you taking the time to visit with us today.
1:02:18
It's been awesome. That my pleasure.
1:02:19
Enjoy the conversation. Thanks, Ryan.
1:02:21
Absolutely. And thanks so much to everybody
1:02:22
who tunes into the show this week and every week.
1:02:25
And we'll see you guys next time.