📺 Develpreneur YouTube Episode

Video + transcript

AI Project Failure: Why AI Projects Fail (And How to Fix It)

2026-03-27 •Youtube

Detailed Notes

AI project failure is more common than most teams realize—and it’s usually not because of the technology.

In this episode of Building Better Developers, we break down why AI projects fail, what companies get wrong about adoption, and how to actually use AI in a way that delivers real value.

Most organizations jump straight to tools like AI or CRMs without defining their goals, processes, or requirements. That leads to wasted time, frustrated teams, and expensive failures.

We cover: - Why starting with tools leads to failure - What AI readiness really means - How AI exposes broken processes - Why small implementations work better - How to use requirements-driven development

If you're a developer, tech leader, or founder trying to figure out how AI fits into your workflow—this episode is for you.

Chapters 00:00 Introduction 02:00 The Problem with AI Adoption 05:00 Using AI as a Team Member 08:30 Why AI Projects Fail 12:00 AI Readiness Explained 16:00 Adoption Challenges 20:00 Large Project Risks 24:00 Start Small Approach 28:00 AI and Developer Skills 32:00 Requirements Matter 36:00 30-Day Challenge 40:00 Wrap Up

Call to Action

👉 Subscribe for more developer and business insights 👉 Share this with your team before your next implementation 👉 Read more at https://develpreneur.com

Transcript Text
Understand?
Good morning.
>> Set up.
>> How is that? How's that audio level?
>> Now I can hear you.
You see that was good.
>> Yeah.
>> Okay. Good.
See, grab that.
Grab that.
That the right one. There's the right
one. It's here.
Bait.
All right. We're hitting record. Should
I go here? Um, I'm going to move my So,
hello everybody. We are doing the
standard
studio arrangement here. Just a second
while I get off camera and hopefully get
in on camera where it's a little better
lighting.
Oh, that was looks much better now. So,
I'm not looking down at everybody.
>> Yeah, now you're better.
There we go.
Okay. So,
>> yeah, that's a good point. Let me fix
that.
Get these straightened out a little bit.
Bam.
There. Yep.
Pardon us while we just sort of like do
our basic setup for you guys.
There we go. How's that?
>> Looks good. So, excellent. So, I don't
feel like I'm getting I'm bright but not
being blinded. Um,
>> I still have to work on that. I don't
have good lighting in this room, but
this is what I got to work with today.
>> Yeah, I get different. It's different
based on my uh what time of the day
obviously is I get a little bit
different light depending on whether
because I got a window right here. So,
it very much can change what I end up
with. Um, all right. Well, welcome
everybody. Welcome back to our weekly
challenge. Uh this week we're going to
talk about Dustin do uh doice.
I forget how to pronounce his last name,
but that's okay. Dusty. Um this is
really good conversation that we had.
We're talking about really like right
sizing how much you bite off uh when
you're actually building out a project.
So this is great. Um welcome to our
weekly challenge for developer, building
better developers. I am Rob Broadhead,
one of the founders of developure. If
you don't know already, check all of our
back catalog. Um, also the founder of RB
Consulting where we help you straighten
out your uh we help you prepare for big
projects and make sure that you are
ready that you are properly going in
with the eyes wide open uh as you jump
into one of these big investments of an
IT project. Uh good thing and bad thing.
I'll talk a little bit about more of
this after our introductions, but good
thing is I've had some really fun uh AI
chats this week. Uh numerous of them
with actually with real people and uh a
couple with AI itself. uh made some
adjustments and things like that that
were uh fun, but also I'm going to talk
about them a little bit because in the
conversations with chat uh with the chat
group, the chat bots of varying sorts, I
got some interesting feedback and
thought it just some stuff that'd be
cool to share. Uh bad thing is
my sun just like moved. I just had
clouds go over. So now my lighting is
going to change a little bit as we go
through this. But better yet, let's just
pass this over to Michael while I try to
find my glass of water. Hey everyone, my
name is Michael Malash. I'm one of the
co-founders of Building Better
Developers, also known as Developer. I'm
also the founder of Envision QA, where
we create reliable, tailored software
that helps you work smarter, scale
faster, and stay in control. Good thing,
bad thing. Uh good thing, uh like Rob,
uh we've had some interesting
conversations with some different people
lately. Uh been getting out, networking
a little bit. Um kind of meeting some
new people and learning some cool new
things about what's going on in the
world. uh bad thing uh what's going on
in the world. I'm trying to I'm on a
news fast and uh basically social media
fast trying to stay away from all the
negativity going on in the world right
now.
Wow, that's quite the fast because the
negativity is uh pretty much everywhere
except for here. We are positivity all
over the place. Let me tell you about my
AI thing. This was actually really uh
fun, amusing and and educational. uh one
of the things that we have done at RB
consulting and we have we have embraced
AI substantially. So we are
as an AI consultant I'm also using it uh
drinking your own you know drinking from
your own Kool-Aid as it were something
like that scratching your own inch it uh
built our old LM LLM we have the RB
operating system which is basically our
uh and our document library and things
of that nature several other things. Uh
so we we've actually nicknamed as we've
mentioned before AI uh typically chat
GPT we call him chip uh which him is I
know misgendering possibly uh which is
where this story goes uh but we also
utilize uh we use Grock we use
perplexity we use Gemini
we use Microsofts a little bit uh and
mostly just because that's just not in
our primary tool set uh but we do have
it available because these things are
becoming sort of uh you know just you're
going to see them everywhere.
So in doing so actually we've had a lot
of conversations uh we being primarily
myself and Natalie our operations
manager and talked about the team and
development and how do we what are some
ways that we may want to integrate or
uh advertise essentially or tell people
that like hey we use AI and it came to
me at some point um that I was like well
maybe because we've got a our team page
maybe we should add AI as a team member.
And so I thought about it as more it's
it's a little bit of a gimmicky thing
and stuff like that. But then I actually
went down that rabbit hole with AI
itself and said, well, how would this
how should we do this? What would this
look like? There's a lot of interesting
stuff that came out of this. Uh the
short answer is that we will have it's
not out there as I speak, but probably
by the time well by the time this shows
up by the time you guys are watching
this, we will have uh Chip will be one
of our member of our staff. And I had
ship build its own uh page and
everything else. Uh because we have a
staff page. We have and it it really was
there's a conversation that went with
that to make sure that we are clear that
chip is a tool that we use. Chip is
something that we're using. Uh we're not
letting it run. I'm I'm not a fan of
letting it just run wild and do a bunch
of bots and do a bunch of stuff. It's
it's really a virtual assistant for lack
of a better term. It really is u just
something that we use to be like hey I
need you to create this report blah blah
blah. It does help with coding and other
things. As we got into this uh it did a
really good job. Surprising to me it did
a really good job of we'll call it
humanizing
uh the chip personality on that page.
Now granted I gave it really good um
templates. I said like here's what our
others look like and it did a really
good job about talking about like here's
what chip does and here's some of the
tasks it does because that it knows very
well. Uh but then it also said like what
chip does when chip is you know when
chip's not working which was sort of
funny because it's like well basically
chip doesn't it addressed it in a
serious way. It says well chip isn't but
if chip isn't working then it's probably
hiding behind a spreadsheet or something
like that. So it gave it like a little
bit of a human touch which I thought was
uh was very interesting. More
interesting was when I said give me an
image because
we have two images. We have a business
and a casual image for all of our our
team. And I said I need a business and a
casual image. Now initially it didn't
give me anything. And I said well okay I
need a business and a casual image. And
it gave me initially um I'm not going to
show it here but I may throw that
somewhere in the links at some point. It
gave me this uh cartoonized very young
female.
Think of like manga type you know those
kinds of of uh cartoons very much the
big eyes and wide eyes all that kind of
stuff. So it felt it was cartoonish but
not overly cartoonish.
Uh but the interesting thing was like
and we actually had the conversation why
did you choose female
and because it was just a it was just
one of those it's like we had never
gendered it and so the whole thing was
to put it to to AI to say like how do
you see yourself and it addressed it and
said well I really don't have a gender
however when we asked why did you pick
female it said that um AI is generally
considered female because it's an
assistant, because it helps you out,
because it's it is trying to be helpful
and a lot like it gave a nice list of
these traits that essentially it said
these are female traits. These are
typically attributed to females. We also
asked a question like how is that look
if we have a female with a mostly male
staff? Uh would it be better to be uh
male or not? And it definitely addressed
that and talked about balance and things
like that. And so it's really
interesting that we we went through this
and it actually sees itself and it
actually gave the example of Alexa,
Siri, all of these and I hadn't thought
about it, but all of those are are
female voices. They're all, you know,
you can change it to whatever you want,
but essentially it's treated out of the
box as a as a female. So, it's a really
interesting thing about how we'll say AI
sees itself. uh having watched Tron
Aries just weeks ago, it actually is
like an interesting little like, you
know, that stuff's all in your head as
well.
But it's uh it's it's one of these
things that I think
it'll be I'm really interested as we go
into this is how we recognize
what AI can do for us and what it
assumes because there's a lot of those
things that goes back to what we're
going to talk about today. what we
talked about this week is that there are
assumptions that go into some of these
things and if those assumptions are
wrong then the whole project's going to
fail. Um wrapping that story up is we
did go back and we said we would like to
have something. It actually suggested
what it called human adjacent uh picture
and it turned out that actually the
human adjacent is just a a generic
looking mid probably 30 40ish probably
more 40 to 50 year old female more or
less. I mean it had a short haircut some
stuff like that but you would think this
would be like a woman of some sort or
someone who identifies as a woman. Uh
initially it was very young. uh looked a
little more like 20some, early 30some.
Um somebody I will not mention her name
uh mentioned that like maybe we can make
them a little older. And so we asked to
make could you be a little older and it
jumped probably 20 years in I would
think in the image that it used. Uh we
also asked could you be a little more
robotic and it just slapped metal all
over like a really bad setting of mine.
So
um it was a really fun This is one of
those things. It's a it's a really fun
um exercise, but it's also I think very
educational as to what
>> what we give AI and and how it thinks
and what it saves and and what it
doesn't. And I think if you haven't done
if you haven't gone down this road a
little bit and actually pursued AI
beyond like beyond just like a stateful
AI of some sort, not just like I ask it
a question or I do a search and then I'm
done, but actually utilizing the we'll
call it the memory of it. Um you will
find I think that there's a lot to a lot
to learn there and there's a lot of apps
you can build related to this. Um this
is something that we will Michael and
actually will be talking about that we
may have something down the road that we
we offer some of these kinds of things.
Uh because it is it's something that's
like it's here. There are practical uses
to it today. You just have to be very
careful as you as you get into it. That
was a very long story. I know. So and
you it looks like you may have even had
some thoughts on that, Michael. So I'm
going to throw that out to you so I can
take a breath. So, it's funny because we
talked to Dustin about, you know,
software systems and software as a
service, things that talking about like
the business problem and it's like
almost everything you went through
through this process kind of touched on
different points in the conversation we
had with Dustin, you know, talking about
adoption, you know, adopting is adoption
is a real problem. you know, buying
tools too early, using them, and you
just walk through like, okay, we're kind
of past the adoption phase of AI. It's
like, we need to start embracing it. We
need to start using it, you know, but
the problem is companies don't know how
to use it, right? They don't understand
what this tool really can do for them.
So, it's just funny and I love how you
went through the process of, you know,
starting with Chip. you know, you've
been talking about chip for weeks and
now you've like tried to humanize it,
trying to incorporate into your business
that I find that extremely funny. Uh,
unfortunately,
when when uh you did the whole manga
thing, all I could think of was Twitch.
Um, so for those of you that may not
know what Twitch TV is, it's the video
gaming streaming service where you can
go watch people play video games. And a
lot of people have funky uh animated uh
icons that kind of simulate them. Uh,
which I think we can even do in Zoom,
like we can put on like little cow faces
or whatever, and they'll kind of emulate
us um our speech and our hands. It it's
just
I really find it interesting how you
took an idea and kind of this problem
you had and it's like well how do you
embrace this? How do you how do you
adopt it? How do you apply it? And
it's just really cool because I've been
dabbling with AI but from a different
perspective. Uh I'm working on getting
it more into the business side of
things, but I've been working more on
the automation side.
Similarly, trying to take that code uh
that test generator tool of mine and
make it even more automated using AI to
get AI to
basically real time uh test software uh
instead of just doing it uh you know
build all this test. I kind of want a
rules engine that would basically test
my uh test for me. And basically, you
don't have to keep the test sort
anymore. You basically have these rules
and then it just goes through and tests
the site for you. So, it can kind of be
um it can work with your system. It can
grow with your system instead of being
stagnant. So, if your code changes, it
would be like, "Oh, hey, you've got new
code changes. Let me look at our test
and see if there's anything new that you
haven't covered yet." and it looks to
grow the the test that way. So, these
are just kind of things that
I've been working on similar to what
you're doing. But I love the idea though
of the virtual assistant. That's really
cool. It's just funny though. It I keep
going back to the conversation with us.
It It's like a lot of the things
he talked about for adoption, for
growing your business, for you know,
getting things uh going,
this is where we need to begin. But the
problem is a lot of businesses don't
follow that. They go buy these tools.
They go buy like earpiece, CRM or some
AI package, hand it to their employees
and say here go with no real guidance,
no real instructions and
they want to see return on the
investment immediately. It's like, well,
what are you doing with this? How are
you doing it? Well, I don't know. you
know, you've I've got like this pile of
work to do with these do deadlines and
now you're giving me something else to
do on top of this and expect immediate
return on investment. You're expecting
me to immediately becoming more
productive, more focused. And that's the
opposite effect because when you
introduce something new in an already
stacked work space, you're going to see
an increase in work and uh and a
decrease in performance in the short
term because you've now overwhelmed your
staff. You've overwhelmed your employees
and now you have to get to that point
where they have to have time to play
with the tools. They have to have time
to figure out what how it works. But the
problem is if you don't give them that
time and you don't give them that
direction,
it's not going to work. You're going to
fail or you're going to break things
that are already broken. Like it this
could extremely highlight bad processes,
uh, procedures, and I'm going way off on
a soap box here, so I'm going to pause
and head this back to you for a minute.
>> Um, it is easy to go there. And this is
it actually does
circles back to a lot of what we talked
about with with Dusty this week is that
uh and it's actually interesting. I had
a a conversation on another podcast. Um
and one of the things we talked I was
actually asked what is because this is
what RB consulting does. It's like what
is AI readiness? If you're providing an
AI readiness assessments, what is AI
readiness? And it's interesting because
I said well you know the funny thing is
is like we're taking everything and just
slapping the name AI on top of like what
are you an architect? No. now I'm an AI
architect, but it's like no, you're not.
You're just an architect. What's an AI
project manager? It's a project manager.
Is we're all putting this cart before
the horse. And I said, really, like AI
readiness is actually just business
readiness. It's the exact same stuff
really that you would need to do that we
talked about with Dustin. If you're
going to do whether it's like a CRM or
if you're going to go all the way to a
full-blown ERP or something like that,
you need your proverbial ducks in a row
in order to do that and be successful.
If you're just going to take a big
honking tool and say we're going to use
this and we're going to shove our
business into it, and you don't know
what your business is, if you don't have
enough awareness of how your business
works, it's going to fail. You're going
to spend a lot of time on that. And
that's what we're seeing now is
Gartner's got stats out there that it's
going to like I think there it changes a
little bit each time I see it, but it's
roughly about 75% or 70% of the AI
projects this year will fail. And the
average cost is going to be I think $7
million per failed project for you know
mid to largeiz businesses. That is not a
small amount of of change. And it goes
back to and they even mentioned that
companies aren't ready. they're diving
into this. And so to sort of add a
little caveat to my earlier story there
is like we as technologists should be
playing around with AI. We should be
finding ways that it's and picking at it
a little bit and saying like does it do
this does it solve this problem or this
really it's this style of problem. For
example, um emails like how does it
really help you write an email? How does
it really write help you write a status
report? How does it really help you
track KPIs?
Because part of it is going to be
understanding where its limitations are,
which you know, jumping to the end, it's
it is a pattern recognizing engine. So,
it has to have the patterns first. If
you don't have the data, it's not going
to it honestly, it will magically create
data sometimes, but it's not going to
create the right data. it's just going
to fill in the blanks just like a, you
know, a a very junior staff member
that's like, "Oh, I've got to just put
data here, so I'm just going to put
whatever. It doesn't matter." AI will do
that kind of stuff. Even if you tell it
not to. So, we need to be, if you're on
the cutting edge of this, uh, if your
business wants to be, you need to
actually be there and be able to say,
"Well, these are some things it does and
these are some things it doesn't." And
in most of cases, the whole like jump to
AI is no. If you aren't in a point where
you can solidly automate your processes,
which means if you can't solidly define
and document your processes, AI is going
to be a failure. And this goes back to
what we talked about with Dusty is
there's all these big projects and these
big budgets and all this stuff comes in
and sometimes it's a it's a rush to get
something done. And it's like, okay,
we've got 6 months to put Netswuite into
all of our business because our current
whatever the product is, our license is
about to run out. And so, we want to
make sure that we can get off of this
license in time before we get whacked
with whatever our our bills are. That is
all well and good and very good reasons.
You know, those are solid reasons to get
this done. However, if you go off
halfcocked, you're not gonna you're
going to end up spending more time
fixing stuff and chasing stuff down
because now you've already started the
process and now you're having to rewrite
uh documentation and migrations and
integrations and all of those shuns that
we talk about in the many times. Then
you have to rebuild those. You have to
redo those. So you need to really sit
down and and plot out more I think more
importantly than it has been in the
past. You really need to be specific on
requirements and specific on your design
because if you're going to utilize AI or
utilizing a resource, a person if you
want to call it that, that needs
requires very detailed instructions. You
can give a human something and say go do
this. Um,
I had something from the editorial
thing. Why? This says, "My son doesn't
like AI." How should I mention that?
Should talk that story.
>> I just also add in that some people
don't like it.
>> Ah, okay.
>> And Tim went into details with you about
why he does it.
>> So,
okay, good. Um, I had something my my
editor just like threw some ideas at me
to talk about here. So, uh, I will talk
about this in just a second. Um, because
I wanted to close in. Oh,
the key is small. If we jump into
something that's really big, then you
have to have all of your ducks. You have
to know all of the pieces of your
business in order to do it. It is going
to be much better. It is much easier for
us to take this in a peacemeal approach.
take something that is a small
well-defined process
and then let's make sure that we can
detail it out, we can automate it and
then we can maybe use AI to get there.
And honestly, I think you're going to
find that by the time you get to the
using AI portion of it. A lot of times
AI actually doesn't add anything to it
or if it does, it's just very minute
pieces because all it's doing is pattern
recognition, which guess what? We've
been writing code to do pattern
recognition for decades. And so
sometimes we can build a better, tighter
solution because we can very tightly
define what the pattern is, what we're
looking for, and how to do it as opposed
to having AI do it and its costs. Simple
example of that would be if you have a
CSV file that is just first name, last
name, address, city, state, zip,
you know what that pattern is. you know
like if it's uh you know pipe separated
or or comma delimited or whatever it is
you know what that pattern is it is very
easy to do that yes you could tell AI to
do it and it could bring it all back but
what was the cost for AI to do it and
even the time to process for AI to do it
versus if you just had a script that did
it as well and I would be aiss if I
didn't address my editor and say while I
have very strong opinions towards this
and we are obviously diving right into
AI
Um there's nothing wrong if you're
waiting if you're sitting there and my
son for example uh who is in it and has
been doing it for several years now uh
is not a fan of AI and he's brought up
some really good places where it is not
he is in a I don't want to name his
product uh that he works on regularly
but it is one of many products that is
out there that is just saying we are
going to add AI and that's really this
goes to Michael's conversation what he
said is like they're just trying to add
AI. Well, it's one of these that I don't
think anybody ever asked the people that
used the tool where whether they needed
AI, whether they wanted AI, or where it
would even work. Because my son actually
spent a a decent amount of time laying
out a solid list of all the reasons AI
is not needed. It's not going to help.
And it really is going to probably cause
more harm than good when all is said and
done. We had a conversation that you
guys will hear about as well with it's
like this also poisons the the
employment pool to some extent because
now you're seeing junior developers
their jobs are disappearing because
people are saying well I can just use AI
for that but that means then the m the
mid-level developers don't have junior
de developers to train so that they
don't grow up to become mid-level
developers and then eventually your
whole you know your whole ecosystem is
is poisoned and you end up in trouble.
So, I will now get off of that side.
Apologize to all you guys, but I'm
sorry, but not sorry. But I think this
is something it's worth uh spending a
little time on a Friday talking about
because it is a a very common theme. And
obviously, Michael and I both have, you
know, strong opinions and experiences
with it. So, I will step down so you can
step back up onto the the soap box.
>> Yeah. So, it's interesting. Um Ian's
point is actually very interesting
because I've seen some
applications
just like he said that have introduced
AI and I see no useful purpose for it at
this time. It's just hey here's AI in
your tool. It's like why is it just so
we can uh say we're using AI?
It it was interesting because
I read an article or we were in a
conversation with someone the other day,
but one of the automation comp or auto
uh companies, their chatbot is literally
an API in the chat GPT. So, if you can
figure out which one that is, you don't
have to waste your tokens or your money.
You literally just go to their chatbot
and you have a full access to the
professional level of chat GPT. It's
like these are
examples of why
adapting too early, not understanding
why you're you why you're getting this
tool. Just hey, we're getting it can be
a problem
really. And Dustin kind of laid this
out. It's like start with the goal. You
know, we we mention this all the time.
Understanding your why. Define your
outcome. Start with the outcome. So one
of the examples he gave was company
needs a CRM. He's like that no what the
company needs we want to increase
customer retention. Two different
things. One is the process one is a
tool.
Figure out your processes figure out
your procedures.
Write those down. AI might be able to
help you with that. But you need to
really think through what it is that you
need. What your business needs. What are
you trying to accomplish? Then walk
backwards. It It's funny in software we
call this test-driven development. We
start with the end and work our way back
to building the code. So you start from
the solution and then um or you start
with the end product and then you work
your way back and write the code. So for
instance, if you need a JSON parser, so
you're going to end up with JSON. So you
start with a little method that just
spits out JSON and the test just reads
the output. Okay, yes, I got JSON. And
then you work your way back and you
build the parser. You build all the
pieces. But at the end, the test never
changes. The test should always get this
response. No matter what you do
upstream, you should always end with
this uh
JSON. You should end with this process.
Same thing. So you start with the goal.
What is the end product? So your end
product is never going to change. What
you do upstream should fall in line to
get that end product. If it doesn't,
you're doing something wrong. So that's
one of the things to figure out.
Interestingly enough though, just going
out and buying tools, throwing AI at
doesn't mean you're going to get there.
Those tools and things you buy may
highlight that whoops, we don't have the
policies and procedures or the process
in place to get to this end product. So
just throwing this tool in just
highlights that we're way up here when
we need to be down here. So
take the time and revisit. This is not
something that you do once and walk
away. You need to constantly be
reviewing your policies and procedures,
your processes, and making sure that
you're on track to hit your goal, that
you're always focused on your goal.
you're focused on what it is that your
company wants to produce, what your
solutions are, and that you're hitting
that, that you're not going off on all
these tangents or rabbit holes. Now,
adoption can still be a problem. You
know, time matters. Like Ian, he's
waiting, so he is not one of those early
adapters. Me and Rob,
we're jumping in. You know, it to us
it's a new toy, but we're taking this
toy and we are figuring out how to turn
this toy into a BMW. We're trying to
take this toy car and turn it into a
real car, making it productive, making
it sleek, making it useful.
However,
if Ian waits too long
and AI becomes too embedded in companies
and companies really only want
developers with AI skills,
then he might miss the boat. He might
lose jobs or be unable to get contracts
because he is not versed enough in AI to
keep up with the current status quo. Now
is that right now? Probably not. But it
could be coming. So these are just
things just in any tech stack any
development. Over the years we've seen
this you know back in the 80s it was
Cobalt for train 90s it became C uh C or
uh C C++ Java and then over time we've
seen Python and you know Groovy, Ruby,
all these other Rust uh languages come
up. Those that jumped on those early
were kind of the forerunners, but they
had a hard time finding jobs. They had a
hard time figuring out what to do with
these. That's kind of where we're at
with AI right now. It's like it's
everywhere. It's there. People can jump
in. They can play with it. We're still
figuring out what to do with it. But the
problem is once people start really
figuring out what to do with this, that
is the time to start jumping on, start
learning it. Get in there, figure it
out. Because if you're not, you'll be on
that late side and then you're going to
be playing catch-up and it's going to
take you a little bit longer to get
there. And if AI goes the way everyone
thinks it's going to go, if you're too
late in adapting to AI, you may be too
late to stay on this trajectory. You
might be playing catch-up for a long
time.
A lot to unpack there. I've got a couple
things to show out. First one, it's not
Ian, so I don't I want to protect the
innocent. Ian is not the son. is
anti-AI. Um Ian has got a different
approach. Uh Ian happens to work for me
for those of you guys that don't know.
And and our approach for the we'll call
the the junior developers I guess uh
which are actually all mid-level
developers at this point is to figure
out where to best utilize AI. They have
found that is useful as a tool. Um they
don't use it in the same way I do
because they're using it as developers.
I'm using it as running a business
operating system as well. like I'm using
using it fully as a a virtual assistant.
Um, if you're just starting out, if
you're a if you are a junior developer,
I would just say you need to understand
how to talk to AI to utilize it, but do
not make it your primary way of getting
your work done because you need to
understand the underpinnings of what
you're building. And if you're not doing
that on a regular basis, then you're
going to end up using AI is going to be
your crutch instead of being a tool. and
then you end up in a a very bad
situation. And I think that's what we're
running into with businesses is they're
just like, "Okay, I'm gonna I've got
something bad. I don't have a process
that I I understand, so I'm going to
slap AI on it and I'm going to fix it."
Um, I would say, uh, Michael referred to
it as being a new toy and stuff. It is,
yes, you can see it that way, but I
would say it is a toy like a double
barrel shotgun is a toy is that you can
actually do a lot of damage and you need
to be very intentional with what you're
doing. There are already patterns that
are out there for uh AI use, for AI
architecture. There are definitely
patterns of solutions that AI work very
well with that that are proven that are
documented that have um background
behind it. Uh if you want some fun
research is look into uh Albania's AI
cabinet member u called Dileia and what
they've done uh with that. there are a
lot of there's a lot of places to go to
realize find out where AI uh mistakes
have been made but also where it can be
very valuable and where power very
powerful. So um I would which sort of
gets us into the challenge for the week.
Um Michael also mentioned you know this
is all test driven development sort of
follows this but it's actually bottom
line it's like at the end of the day
it's requirements driven development.
It's like you have requirements and this
is why I'm I'm going to harp on that
just a bit is you have requirements
and then that leads to your solution.
Now if you do it one of the ways you can
do it is you do requirements and then
you have tests that actually validate
those requirements and so now you say
what I build after that my test should
all should always succeed because they
have to output essentially what the
requirements are expecting.
This is what you should do is actually
let's practice requirements. I think I
want to go to a 30-day challenge instead
of a 7-day challenge. And I'll see what
your your thoughts are on this, Michael,
because we have not talked about this
before because we almost never do. Um,
I think is take a 30-day challenge,
build a tool, build an app, and start
with scratching your own itch. So, maybe
your problem is um like as a developer,
maybe your problem is I want a I need to
have a better way to find my old code.
is find code that is like, you know,
find that JavaScript function I wrote
two years ago that I know is out there
and I just don't want to rewrite it
again. Basically, building your own, you
know, AI code repository that you can go
refer to. That could be an example. But
the key to this is write out first
before you write any code or before you
have AI write any code or anything like
that is write out your requirements. Get
yourself and it doesn't have to be word
and all that kind of stuff. use like
markdown or something just very
specific. It has to do this and this and
this and this and this and go into all
of those and ask yourself questions
about all of that is like how can I
provide more information about each of
those requirements and then
start having a conversation with I pick
your tool and basically walk through it
and say maybe give it from the start say
I want to build this. Give it all your
requirements and start with what do you
need to know? What have I missed? What
can I, you know, what am I
needing to add that I have not? Where
are the gaps? Uh, is this is this
valuable? Maybe you ask, is this
something that already exists somewhere?
What would it look like in the real
world? What would the market look like
for this? How would I talk about it?
What would I call it? You can have these
conversations, but the goal is to move
you from requirements in that day one to
a working application by day 30. Now,
doing so that means that you're going to
like give yourself this sprint. You're
going to have to put some milestones in
there. So, maybe by the end of the first
week, you need to have something that is
a solid uh data model and flow diagram
or something of those n of that nature.
Since you can use AI to generate code,
you may have like maybe it's a clickable
demo of some sort by week two. Then
maybe it's uh I need to know what I have
to have all of the functions fleshed
out. So, I need to know essentially this
is the feature set that I'm going to be
generating. By the end of week three,
maybe you need to be like, "All right, I
need to have something that I can now
play with, that I can work on, that I
can test. I need to have the uh the
test, the unit test generated for it. I
need to have documentation generated for
it." Those kinds of thing, and then get
yourself to day 30. What are your
thoughts on that as a a bigger than
normal challenge?
I think that works because it kind of
flows with the conversation we had with
Dustin and it gets us back to focusing
on the requirements, focusing on what it
is that we're trying to accomplish. What
is our goal? Because a lot of times
that's where we fail. We think we have
the requirements, but they're so loosey
goosey that you don't have the
requirements. What you have is you have
an idea that it's not fully fleshed out.
And when you just try to implement that
idea, you end up with a whole bunch of
if then, what ifs. And if you have what,
well, what if this, what if this, then
you're not there. You don't have the
requirements. Requirements should be
able to be read as a statement. If you
read the statement and there is no if
then or well, what about this? Then you
have a good requirement. If you have a
statement that prompts, well, okay,
well, what about this? What about this?
What then you don't have a requirement.
you have a just an idea. So get your
statements, get your requirements
flushed out so that they are as
definitive as possible and then work
through build the software. I love it.
Um,
one little tidbit I kind of want to
throw in, uh, when we were talking
about, you know, train, you know, the
mid-level developers training the, uh,
lower level developers to be and the
mid-level developers to become senior.
I'm already seeing this trend in a
negative way. I'm seeing mid-level
developers or team leads say, "I'm too
busy. Go talk to AI. Go have AI solve
your problem. I don't have time to deal
with you. I'm too busy working this
problem." That is very dangerous. If you
are not helping your team problem solve
your problems and you're pushing them
off on AI,
one of two things is going to happen.
they're going to leave.
Or two, your software is going to become
a dumpster fire if it's not already. Or
three, you're never going to move up.
You're going to be stuck putting out
fires. You're going to be stuck in that
busy mode. You're not going to learn the
management skills necessary to become
that senior, to continue analyzing, and
continue to grow.
>> I think it's a good parting thought
there. uh a good way to grow, a good way
to uh help your career is like yeah you
can use uh and you can use AI to get
used to like ordering something around
and and sending commands and see how
that works. But then you definitely are
going to have a you need to have that
personal uh experience because that is
what's going to help you grow as a
developer and uh not only the soft
skills but also the technical skills
because it is always better to help you
uh solidify something in your head if
you have to explain it to somebody else.
And the more you have to explain it, the
more that becomes just something that
now it's just embedded in your psyche
essentially.
Hopefully our ugly faces are not
embedded in your psyche or the back of
your uh eyeballs. And we thank you so
much for hanging out with us. Uh thank
you for spending your time and uh just
enjoy your challenge, enjoy your
weekend. And as always, go out there and
have yourself a great day, a great week,
and we will talk to you next week.
Transcript Segments
68.799

Understand?

91.52

Good morning.

93.759

>> Set up.

112.079

>> How is that? How's that audio level?

115.119

>> Now I can hear you.

117.52

You see that was good.

118.96

>> Yeah.

119.92

>> Okay. Good.

125.04

See, grab that.

127.84

Grab that.

130.319

That the right one. There's the right

131.76

one. It's here.

134.64

Bait.

140.08

All right. We're hitting record. Should

142.16

I go here? Um, I'm going to move my So,

146

hello everybody. We are doing the

148.08

standard

149.76

studio arrangement here. Just a second

154.319

while I get off camera and hopefully get

156.239

in on camera where it's a little better

158.239

lighting.

168.48

Oh, that was looks much better now. So,

173.12

I'm not looking down at everybody.

176.8

>> Yeah, now you're better.

181.76

There we go.

184.56

Okay. So,

185.2

>> yeah, that's a good point. Let me fix

186.879

that.

193.28

Get these straightened out a little bit.

196.56

Bam.

198.08

There. Yep.

204.239

Pardon us while we just sort of like do

206

our basic setup for you guys.

209.36

There we go. How's that?

211.68

>> Looks good. So, excellent. So, I don't

213.92

feel like I'm getting I'm bright but not

216.319

being blinded. Um,

219.84

>> I still have to work on that. I don't

221.519

have good lighting in this room, but

224

this is what I got to work with today.

226.56

>> Yeah, I get different. It's different

228.159

based on my uh what time of the day

231.519

obviously is I get a little bit

232.72

different light depending on whether

233.92

because I got a window right here. So,

235.28

it very much can change what I end up

237.599

with. Um, all right. Well, welcome

241.2

everybody. Welcome back to our weekly

243.68

challenge. Uh this week we're going to

245.04

talk about Dustin do uh doice.

249.76

I forget how to pronounce his last name,

251.28

but that's okay. Dusty. Um this is

253.84

really good conversation that we had.

255.599

We're talking about really like right

258.479

sizing how much you bite off uh when

260.639

you're actually building out a project.

262

So this is great. Um welcome to our

264.96

weekly challenge for developer, building

266.8

better developers. I am Rob Broadhead,

268.56

one of the founders of developure. If

270.4

you don't know already, check all of our

272.72

back catalog. Um, also the founder of RB

276.24

Consulting where we help you straighten

278

out your uh we help you prepare for big

281.199

projects and make sure that you are

282.56

ready that you are properly going in

284.8

with the eyes wide open uh as you jump

286.8

into one of these big investments of an

288.96

IT project. Uh good thing and bad thing.

292.96

I'll talk a little bit about more of

294.16

this after our introductions, but good

296.4

thing is I've had some really fun uh AI

300.4

chats this week. Uh numerous of them

302.479

with actually with real people and uh a

305.36

couple with AI itself. uh made some

307.84

adjustments and things like that that

309.28

were uh fun, but also I'm going to talk

311.6

about them a little bit because in the

313.44

conversations with chat uh with the chat

316.16

group, the chat bots of varying sorts, I

318.56

got some interesting feedback and

320.639

thought it just some stuff that'd be

322

cool to share. Uh bad thing is

326.479

my sun just like moved. I just had

328.16

clouds go over. So now my lighting is

329.6

going to change a little bit as we go

330.8

through this. But better yet, let's just

333.12

pass this over to Michael while I try to

335.44

find my glass of water. Hey everyone, my

337.759

name is Michael Malash. I'm one of the

338.88

co-founders of Building Better

340.08

Developers, also known as Developer. I'm

342.16

also the founder of Envision QA, where

343.84

we create reliable, tailored software

346.32

that helps you work smarter, scale

348.16

faster, and stay in control. Good thing,

351.199

bad thing. Uh good thing, uh like Rob,

354.24

uh we've had some interesting

355.28

conversations with some different people

356.8

lately. Uh been getting out, networking

359.759

a little bit. Um kind of meeting some

362.24

new people and learning some cool new

364.16

things about what's going on in the

365.52

world. uh bad thing uh what's going on

369.039

in the world. I'm trying to I'm on a

371.12

news fast and uh basically social media

374.319

fast trying to stay away from all the

376

negativity going on in the world right

378

now.

379.68

Wow, that's quite the fast because the

381.36

negativity is uh pretty much everywhere

384

except for here. We are positivity all

387.039

over the place. Let me tell you about my

388.88

AI thing. This was actually really uh

391.039

fun, amusing and and educational. uh one

394.4

of the things that we have done at RB

395.84

consulting and we have we have embraced

398.08

AI substantially. So we are

402

as an AI consultant I'm also using it uh

404.72

drinking your own you know drinking from

406.319

your own Kool-Aid as it were something

407.84

like that scratching your own inch it uh

410.4

built our old LM LLM we have the RB

413.28

operating system which is basically our

416.24

uh and our document library and things

417.919

of that nature several other things. Uh

420.88

so we we've actually nicknamed as we've

423.36

mentioned before AI uh typically chat

425.919

GPT we call him chip uh which him is I

429.12

know misgendering possibly uh which is

431.52

where this story goes uh but we also

433.68

utilize uh we use Grock we use

436

perplexity we use Gemini

439.36

we use Microsofts a little bit uh and

441.599

mostly just because that's just not in

443.12

our primary tool set uh but we do have

445.039

it available because these things are

447.039

becoming sort of uh you know just you're

449.36

going to see them everywhere.

451.759

So in doing so actually we've had a lot

454.24

of conversations uh we being primarily

456.72

myself and Natalie our operations

458.8

manager and talked about the team and

461.199

development and how do we what are some

463.759

ways that we may want to integrate or

468.08

uh advertise essentially or tell people

470.16

that like hey we use AI and it came to

473.52

me at some point um that I was like well

476.08

maybe because we've got a our team page

479.12

maybe we should add AI as a team member.

481.919

And so I thought about it as more it's

483.52

it's a little bit of a gimmicky thing

484.8

and stuff like that. But then I actually

486.319

went down that rabbit hole with AI

488.319

itself and said, well, how would this

490.319

how should we do this? What would this

491.84

look like? There's a lot of interesting

494.479

stuff that came out of this. Uh the

496.4

short answer is that we will have it's

498.24

not out there as I speak, but probably

499.84

by the time well by the time this shows

501.84

up by the time you guys are watching

503.68

this, we will have uh Chip will be one

506.319

of our member of our staff. And I had

509.44

ship build its own uh page and

512.08

everything else. Uh because we have a

513.919

staff page. We have and it it really was

516.56

there's a conversation that went with

518

that to make sure that we are clear that

520.399

chip is a tool that we use. Chip is

522.479

something that we're using. Uh we're not

524.48

letting it run. I'm I'm not a fan of

526.399

letting it just run wild and do a bunch

528

of bots and do a bunch of stuff. It's

529.68

it's really a virtual assistant for lack

531.839

of a better term. It really is u just

534.72

something that we use to be like hey I

536.24

need you to create this report blah blah

538.8

blah. It does help with coding and other

540.64

things. As we got into this uh it did a

544.64

really good job. Surprising to me it did

547.36

a really good job of we'll call it

549.44

humanizing

551.279

uh the chip personality on that page.

554.56

Now granted I gave it really good um

557.36

templates. I said like here's what our

559.36

others look like and it did a really

561.44

good job about talking about like here's

563.2

what chip does and here's some of the

564.64

tasks it does because that it knows very

566.56

well. Uh but then it also said like what

569.12

chip does when chip is you know when

571.2

chip's not working which was sort of

573.12

funny because it's like well basically

574.88

chip doesn't it addressed it in a

577.2

serious way. It says well chip isn't but

579.36

if chip isn't working then it's probably

581.04

hiding behind a spreadsheet or something

582.88

like that. So it gave it like a little

584.32

bit of a human touch which I thought was

586.72

uh was very interesting. More

588.959

interesting was when I said give me an

591.12

image because

594.24

we have two images. We have a business

596.64

and a casual image for all of our our

599.2

team. And I said I need a business and a

601.6

casual image. Now initially it didn't

603.279

give me anything. And I said well okay I

605.04

need a business and a casual image. And

607.839

it gave me initially um I'm not going to

610.88

show it here but I may throw that

612.72

somewhere in the links at some point. It

614.399

gave me this uh cartoonized very young

619.76

female.

621.519

Think of like manga type you know those

624.48

kinds of of uh cartoons very much the

627.279

big eyes and wide eyes all that kind of

629.12

stuff. So it felt it was cartoonish but

631.68

not overly cartoonish.

634

Uh but the interesting thing was like

635.44

and we actually had the conversation why

637.839

did you choose female

640

and because it was just a it was just

641.92

one of those it's like we had never

643.2

gendered it and so the whole thing was

645.2

to put it to to AI to say like how do

648.8

you see yourself and it addressed it and

651.279

said well I really don't have a gender

653.279

however when we asked why did you pick

656.64

female it said that um AI is generally

662.32

considered female because it's an

664.16

assistant, because it helps you out,

665.92

because it's it is trying to be helpful

668.399

and a lot like it gave a nice list of

670.48

these traits that essentially it said

672.16

these are female traits. These are

674

typically attributed to females. We also

677.04

asked a question like how is that look

679.279

if we have a female with a mostly male

682

staff? Uh would it be better to be uh

685.519

male or not? And it definitely addressed

688.72

that and talked about balance and things

690.56

like that. And so it's really

692.88

interesting that we we went through this

694.64

and it actually sees itself and it

696.399

actually gave the example of Alexa,

698.32

Siri, all of these and I hadn't thought

700.48

about it, but all of those are are

702.079

female voices. They're all, you know,

705.04

you can change it to whatever you want,

707.04

but essentially it's treated out of the

709.36

box as a as a female. So, it's a really

711.92

interesting thing about how we'll say AI

714.32

sees itself. uh having watched Tron

717.04

Aries just weeks ago, it actually is

719.36

like an interesting little like, you

720.8

know, that stuff's all in your head as

722.24

well.

724.24

But it's uh it's it's one of these

725.92

things that I think

728.24

it'll be I'm really interested as we go

729.92

into this is how we recognize

733.279

what AI can do for us and what it

736.16

assumes because there's a lot of those

738.24

things that goes back to what we're

739.68

going to talk about today. what we

740.88

talked about this week is that there are

742.32

assumptions that go into some of these

744

things and if those assumptions are

745.519

wrong then the whole project's going to

747.92

fail. Um wrapping that story up is we

751.44

did go back and we said we would like to

752.959

have something. It actually suggested

754.32

what it called human adjacent uh picture

757.6

and it turned out that actually the

759.519

human adjacent is just a a generic

763.36

looking mid probably 30 40ish probably

767.2

more 40 to 50 year old female more or

769.76

less. I mean it had a short haircut some

771.36

stuff like that but you would think this

772.56

would be like a woman of some sort or

774.88

someone who identifies as a woman. Uh

777.2

initially it was very young. uh looked a

779.04

little more like 20some, early 30some.

781.76

Um somebody I will not mention her name

784.399

uh mentioned that like maybe we can make

785.839

them a little older. And so we asked to

787.2

make could you be a little older and it

790.24

jumped probably 20 years in I would

792.88

think in the image that it used. Uh we

795.04

also asked could you be a little more

796.399

robotic and it just slapped metal all

798.24

over like a really bad setting of mine.

800.72

So

802.72

um it was a really fun This is one of

805.2

those things. It's a it's a really fun

807.12

um exercise, but it's also I think very

809.44

educational as to what

811.68

>> what we give AI and and how it thinks

814.16

and what it saves and and what it

816.16

doesn't. And I think if you haven't done

818.88

if you haven't gone down this road a

820.8

little bit and actually pursued AI

823.44

beyond like beyond just like a stateful

826.32

AI of some sort, not just like I ask it

828.32

a question or I do a search and then I'm

829.76

done, but actually utilizing the we'll

832.24

call it the memory of it. Um you will

835.12

find I think that there's a lot to a lot

836.88

to learn there and there's a lot of apps

838

you can build related to this. Um this

840.16

is something that we will Michael and

842

actually will be talking about that we

843.36

may have something down the road that we

845.04

we offer some of these kinds of things.

847.36

Uh because it is it's something that's

849.04

like it's here. There are practical uses

852.079

to it today. You just have to be very

854.079

careful as you as you get into it. That

856.88

was a very long story. I know. So and

858.56

you it looks like you may have even had

859.92

some thoughts on that, Michael. So I'm

861.12

going to throw that out to you so I can

862.72

take a breath. So, it's funny because we

865.199

talked to Dustin about, you know,

867.76

software systems and software as a

871.199

service, things that talking about like

872.72

the business problem and it's like

874.16

almost everything you went through

875.92

through this process kind of touched on

878.639

different points in the conversation we

880.56

had with Dustin, you know, talking about

882.72

adoption, you know, adopting is adoption

885.92

is a real problem. you know, buying

887.6

tools too early, using them, and you

889.519

just walk through like, okay, we're kind

891.76

of past the adoption phase of AI. It's

894.56

like, we need to start embracing it. We

896.399

need to start using it, you know, but

898.399

the problem is companies don't know how

900.8

to use it, right? They don't understand

904.24

what this tool really can do for them.

906.24

So, it's just funny and I love how you

908.959

went through the process of, you know,

912.32

starting with Chip. you know, you've

913.839

been talking about chip for weeks and

915.6

now you've like tried to humanize it,

917.6

trying to incorporate into your business

919.6

that I find that extremely funny. Uh,

922.48

unfortunately,

924.16

when when uh you did the whole manga

926.079

thing, all I could think of was Twitch.

928.639

Um, so for those of you that may not

931.12

know what Twitch TV is, it's the video

933.279

gaming streaming service where you can

935.199

go watch people play video games. And a

936.8

lot of people have funky uh animated uh

940.72

icons that kind of simulate them. Uh,

943.44

which I think we can even do in Zoom,

944.88

like we can put on like little cow faces

946.639

or whatever, and they'll kind of emulate

948.399

us um our speech and our hands. It it's

952.24

just

954.16

I really find it interesting how you

957.92

took an idea and kind of this problem

961.759

you had and it's like well how do you

963.199

embrace this? How do you how do you

965.839

adopt it? How do you apply it? And

969.199

it's just really cool because I've been

971.839

dabbling with AI but from a different

973.68

perspective. Uh I'm working on getting

976.32

it more into the business side of

977.92

things, but I've been working more on

979.279

the automation side.

982.24

Similarly, trying to take that code uh

985.12

that test generator tool of mine and

988.72

make it even more automated using AI to

991.759

get AI to

995.12

basically real time uh test software uh

1000.8

instead of just doing it uh you know

1003.12

build all this test. I kind of want a

1005.759

rules engine that would basically test

1007.44

my uh test for me. And basically, you

1010.8

don't have to keep the test sort

1011.92

anymore. You basically have these rules

1013.12

and then it just goes through and tests

1014.639

the site for you. So, it can kind of be

1018.48

um it can work with your system. It can

1022

grow with your system instead of being

1023.759

stagnant. So, if your code changes, it

1026.16

would be like, "Oh, hey, you've got new

1028

code changes. Let me look at our test

1030.559

and see if there's anything new that you

1033.28

haven't covered yet." and it looks to

1035.199

grow the the test that way. So, these

1037.839

are just kind of things that

1040.88

I've been working on similar to what

1042.48

you're doing. But I love the idea though

1044.24

of the virtual assistant. That's really

1046.88

cool. It's just funny though. It I keep

1050.24

going back to the conversation with us.

1052.16

It It's like a lot of the things

1055.76

he talked about for adoption, for

1057.44

growing your business, for you know,

1058.88

getting things uh going,

1062.16

this is where we need to begin. But the

1064.32

problem is a lot of businesses don't

1068.08

follow that. They go buy these tools.

1070.08

They go buy like earpiece, CRM or some

1074.32

AI package, hand it to their employees

1077.039

and say here go with no real guidance,

1080

no real instructions and

1083.6

they want to see return on the

1085.6

investment immediately. It's like, well,

1087.28

what are you doing with this? How are

1088.32

you doing it? Well, I don't know. you

1090

know, you've I've got like this pile of

1092.4

work to do with these do deadlines and

1095.28

now you're giving me something else to

1096.88

do on top of this and expect immediate

1100.16

return on investment. You're expecting

1101.6

me to immediately becoming more

1103.679

productive, more focused. And that's the

1105.919

opposite effect because when you

1107.6

introduce something new in an already

1110.48

stacked work space, you're going to see

1113.52

an increase in work and uh and a

1118

decrease in performance in the short

1120.559

term because you've now overwhelmed your

1123.36

staff. You've overwhelmed your employees

1125.36

and now you have to get to that point

1127.2

where they have to have time to play

1129.6

with the tools. They have to have time

1130.799

to figure out what how it works. But the

1133.44

problem is if you don't give them that

1134.72

time and you don't give them that

1136.32

direction,

1138.24

it's not going to work. You're going to

1140.24

fail or you're going to break things

1144.08

that are already broken. Like it this

1145.76

could extremely highlight bad processes,

1148.24

uh, procedures, and I'm going way off on

1151.039

a soap box here, so I'm going to pause

1153.36

and head this back to you for a minute.

1156.32

>> Um, it is easy to go there. And this is

1158.16

it actually does

1160.559

circles back to a lot of what we talked

1162.32

about with with Dusty this week is that

1165.36

uh and it's actually interesting. I had

1166.64

a a conversation on another podcast. Um

1169.44

and one of the things we talked I was

1170.88

actually asked what is because this is

1172.559

what RB consulting does. It's like what

1174.48

is AI readiness? If you're providing an

1176.72

AI readiness assessments, what is AI

1179.039

readiness? And it's interesting because

1180.559

I said well you know the funny thing is

1182.16

is like we're taking everything and just

1184

slapping the name AI on top of like what

1186

are you an architect? No. now I'm an AI

1187.919

architect, but it's like no, you're not.

1191.28

You're just an architect. What's an AI

1192.88

project manager? It's a project manager.

1195.44

Is we're all putting this cart before

1198.32

the horse. And I said, really, like AI

1201.12

readiness is actually just business

1202.96

readiness. It's the exact same stuff

1204.88

really that you would need to do that we

1206.4

talked about with Dustin. If you're

1207.44

going to do whether it's like a CRM or

1210.08

if you're going to go all the way to a

1211.28

full-blown ERP or something like that,

1212.96

you need your proverbial ducks in a row

1216.32

in order to do that and be successful.

1218.799

If you're just going to take a big

1221.039

honking tool and say we're going to use

1222.96

this and we're going to shove our

1224.24

business into it, and you don't know

1226.32

what your business is, if you don't have

1228.159

enough awareness of how your business

1229.919

works, it's going to fail. You're going

1232.08

to spend a lot of time on that. And

1233.84

that's what we're seeing now is

1235.039

Gartner's got stats out there that it's

1237.12

going to like I think there it changes a

1239.039

little bit each time I see it, but it's

1240.159

roughly about 75% or 70% of the AI

1243.28

projects this year will fail. And the

1245.6

average cost is going to be I think $7

1248.08

million per failed project for you know

1250.64

mid to largeiz businesses. That is not a

1253.28

small amount of of change. And it goes

1257.039

back to and they even mentioned that

1259.52

companies aren't ready. they're diving

1261.12

into this. And so to sort of add a

1263.6

little caveat to my earlier story there

1266.24

is like we as technologists should be

1269.52

playing around with AI. We should be

1271.6

finding ways that it's and picking at it

1274.32

a little bit and saying like does it do

1276.159

this does it solve this problem or this

1278.72

really it's this style of problem. For

1280.64

example, um emails like how does it

1285.12

really help you write an email? How does

1287.76

it really write help you write a status

1290.24

report? How does it really help you

1292.08

track KPIs?

1294.08

Because part of it is going to be

1295.679

understanding where its limitations are,

1297.679

which you know, jumping to the end, it's

1300.24

it is a pattern recognizing engine. So,

1303.6

it has to have the patterns first. If

1306.08

you don't have the data, it's not going

1307.84

to it honestly, it will magically create

1310.32

data sometimes, but it's not going to

1311.52

create the right data. it's just going

1312.64

to fill in the blanks just like a, you

1314.96

know, a a very junior staff member

1318.4

that's like, "Oh, I've got to just put

1319.76

data here, so I'm just going to put

1321.039

whatever. It doesn't matter." AI will do

1322.96

that kind of stuff. Even if you tell it

1324.48

not to. So, we need to be, if you're on

1328.559

the cutting edge of this, uh, if your

1330.799

business wants to be, you need to

1332.4

actually be there and be able to say,

1333.679

"Well, these are some things it does and

1335.28

these are some things it doesn't." And

1337.12

in most of cases, the whole like jump to

1340.159

AI is no. If you aren't in a point where

1343.52

you can solidly automate your processes,

1345.84

which means if you can't solidly define

1347.84

and document your processes, AI is going

1350.88

to be a failure. And this goes back to

1352.88

what we talked about with Dusty is

1354.32

there's all these big projects and these

1355.919

big budgets and all this stuff comes in

1357.84

and sometimes it's a it's a rush to get

1360.559

something done. And it's like, okay,

1361.679

we've got 6 months to put Netswuite into

1364.24

all of our business because our current

1366.96

whatever the product is, our license is

1369.36

about to run out. And so, we want to

1370.96

make sure that we can get off of this

1372.4

license in time before we get whacked

1374

with whatever our our bills are. That is

1376.48

all well and good and very good reasons.

1379.76

You know, those are solid reasons to get

1382.24

this done. However, if you go off

1385.84

halfcocked, you're not gonna you're

1387.52

going to end up spending more time

1388.96

fixing stuff and chasing stuff down

1391.12

because now you've already started the

1392.64

process and now you're having to rewrite

1395.84

uh documentation and migrations and

1398.159

integrations and all of those shuns that

1400.559

we talk about in the many times. Then

1402.96

you have to rebuild those. You have to

1404.559

redo those. So you need to really sit

1407.6

down and and plot out more I think more

1411.2

importantly than it has been in the

1412.72

past. You really need to be specific on

1415.44

requirements and specific on your design

1418.159

because if you're going to utilize AI or

1420.72

utilizing a resource, a person if you

1424

want to call it that, that needs

1426.48

requires very detailed instructions. You

1429.76

can give a human something and say go do

1431.76

this. Um,

1436

I had something from the editorial

1438

thing. Why? This says, "My son doesn't

1440.72

like AI." How should I mention that?

1442.799

Should talk that story.

1445.039

>> I just also add in that some people

1448

don't like it.

1449.2

>> Ah, okay.

1449.919

>> And Tim went into details with you about

1451.679

why he does it.

1454.159

>> So,

1455.76

okay, good. Um, I had something my my

1458.48

editor just like threw some ideas at me

1460.559

to talk about here. So, uh, I will talk

1463.039

about this in just a second. Um, because

1465.2

I wanted to close in. Oh,

1468.559

the key is small. If we jump into

1472.24

something that's really big, then you

1474.799

have to have all of your ducks. You have

1476.72

to know all of the pieces of your

1478.559

business in order to do it. It is going

1480.799

to be much better. It is much easier for

1482.799

us to take this in a peacemeal approach.

1485.84

take something that is a small

1488.159

well-defined process

1490.48

and then let's make sure that we can

1492.48

detail it out, we can automate it and

1494.24

then we can maybe use AI to get there.

1496.4

And honestly, I think you're going to

1498

find that by the time you get to the

1500.799

using AI portion of it. A lot of times

1503.2

AI actually doesn't add anything to it

1506

or if it does, it's just very minute

1507.76

pieces because all it's doing is pattern

1509.84

recognition, which guess what? We've

1511.279

been writing code to do pattern

1512.48

recognition for decades. And so

1515.679

sometimes we can build a better, tighter

1518.48

solution because we can very tightly

1520.64

define what the pattern is, what we're

1522.799

looking for, and how to do it as opposed

1524.96

to having AI do it and its costs. Simple

1528.159

example of that would be if you have a

1530.64

CSV file that is just first name, last

1533.12

name, address, city, state, zip,

1536.559

you know what that pattern is. you know

1538.96

like if it's uh you know pipe separated

1541.039

or or comma delimited or whatever it is

1543.44

you know what that pattern is it is very

1545.039

easy to do that yes you could tell AI to

1547.36

do it and it could bring it all back but

1549.36

what was the cost for AI to do it and

1551.52

even the time to process for AI to do it

1554.72

versus if you just had a script that did

1556.559

it as well and I would be aiss if I

1559.039

didn't address my editor and say while I

1562.88

have very strong opinions towards this

1565.279

and we are obviously diving right into

1567.36

AI

1568.32

Um there's nothing wrong if you're

1570.48

waiting if you're sitting there and my

1572.96

son for example uh who is in it and has

1576.48

been doing it for several years now uh

1578.48

is not a fan of AI and he's brought up

1580.88

some really good places where it is not

1583.76

he is in a I don't want to name his

1585.6

product uh that he works on regularly

1587.52

but it is one of many products that is

1590.08

out there that is just saying we are

1591.84

going to add AI and that's really this

1594.4

goes to Michael's conversation what he

1596

said is like they're just trying to add

1597.919

AI. Well, it's one of these that I don't

1600.559

think anybody ever asked the people that

1602.559

used the tool where whether they needed

1605.679

AI, whether they wanted AI, or where it

1608

would even work. Because my son actually

1610

spent a a decent amount of time laying

1612.159

out a solid list of all the reasons AI

1615.52

is not needed. It's not going to help.

1618.24

And it really is going to probably cause

1620.64

more harm than good when all is said and

1622.64

done. We had a conversation that you

1624.48

guys will hear about as well with it's

1626.159

like this also poisons the the

1629.52

employment pool to some extent because

1631.44

now you're seeing junior developers

1634.32

their jobs are disappearing because

1636.4

people are saying well I can just use AI

1638.24

for that but that means then the m the

1640.48

mid-level developers don't have junior

1642

de developers to train so that they

1643.76

don't grow up to become mid-level

1645.12

developers and then eventually your

1647.12

whole you know your whole ecosystem is

1649.919

is poisoned and you end up in trouble.

1652.559

So, I will now get off of that side.

1654.88

Apologize to all you guys, but I'm

1656.72

sorry, but not sorry. But I think this

1658.24

is something it's worth uh spending a

1660

little time on a Friday talking about

1662.32

because it is a a very common theme. And

1665.84

obviously, Michael and I both have, you

1667.52

know, strong opinions and experiences

1669.12

with it. So, I will step down so you can

1671.12

step back up onto the the soap box.

1673.52

>> Yeah. So, it's interesting. Um Ian's

1677.279

point is actually very interesting

1678.96

because I've seen some

1681.679

applications

1683.36

just like he said that have introduced

1685.279

AI and I see no useful purpose for it at

1688.96

this time. It's just hey here's AI in

1691.12

your tool. It's like why is it just so

1693.84

we can uh say we're using AI?

1698.48

It it was interesting because

1701.36

I read an article or we were in a

1703.039

conversation with someone the other day,

1704.64

but one of the automation comp or auto

1707.919

uh companies, their chatbot is literally

1710.72

an API in the chat GPT. So, if you can

1713.76

figure out which one that is, you don't

1715.36

have to waste your tokens or your money.

1716.96

You literally just go to their chatbot

1718.799

and you have a full access to the

1721.12

professional level of chat GPT. It's

1724.08

like these are

1727.12

examples of why

1730.32

adapting too early, not understanding

1732.799

why you're you why you're getting this

1734.88

tool. Just hey, we're getting it can be

1737.76

a problem

1740.08

really. And Dustin kind of laid this

1742.32

out. It's like start with the goal. You

1744.559

know, we we mention this all the time.

1746.72

Understanding your why. Define your

1749.679

outcome. Start with the outcome. So one

1752.64

of the examples he gave was company

1755.36

needs a CRM. He's like that no what the

1758.72

company needs we want to increase

1760.88

customer retention. Two different

1763.12

things. One is the process one is a

1765.2

tool.

1766.88

Figure out your processes figure out

1769.2

your procedures.

1771.52

Write those down. AI might be able to

1773.52

help you with that. But you need to

1775.84

really think through what it is that you

1779.44

need. What your business needs. What are

1781.52

you trying to accomplish? Then walk

1785.12

backwards. It It's funny in software we

1788.159

call this test-driven development. We

1790

start with the end and work our way back

1794.399

to building the code. So you start from

1796.32

the solution and then um or you start

1800.08

with the end product and then you work

1802.159

your way back and write the code. So for

1804.32

instance, if you need a JSON parser, so

1807.36

you're going to end up with JSON. So you

1808.799

start with a little method that just

1810.72

spits out JSON and the test just reads

1813.12

the output. Okay, yes, I got JSON. And

1815.36

then you work your way back and you

1816.64

build the parser. You build all the

1818

pieces. But at the end, the test never

1820.24

changes. The test should always get this

1822.799

response. No matter what you do

1825.12

upstream, you should always end with

1827.76

this uh

1830.72

JSON. You should end with this process.

1832.88

Same thing. So you start with the goal.

1835.2

What is the end product? So your end

1837.2

product is never going to change. What

1839.44

you do upstream should fall in line to

1842.96

get that end product. If it doesn't,

1845.12

you're doing something wrong. So that's

1846.88

one of the things to figure out.

1850.24

Interestingly enough though, just going

1851.84

out and buying tools, throwing AI at

1854.399

doesn't mean you're going to get there.

1856.48

Those tools and things you buy may

1858.799

highlight that whoops, we don't have the

1861.52

policies and procedures or the process

1863.279

in place to get to this end product. So

1865.2

just throwing this tool in just

1866.799

highlights that we're way up here when

1869.36

we need to be down here. So

1873.2

take the time and revisit. This is not

1876.159

something that you do once and walk

1877.76

away. You need to constantly be

1880.799

reviewing your policies and procedures,

1882.64

your processes, and making sure that

1884.64

you're on track to hit your goal, that

1888.08

you're always focused on your goal.

1890.399

you're focused on what it is that your

1892.88

company wants to produce, what your

1896.159

solutions are, and that you're hitting

1898.32

that, that you're not going off on all

1900.08

these tangents or rabbit holes. Now,

1904.399

adoption can still be a problem. You

1907.039

know, time matters. Like Ian, he's

1909.919

waiting, so he is not one of those early

1912.559

adapters. Me and Rob,

1915.6

we're jumping in. You know, it to us

1917.919

it's a new toy, but we're taking this

1920.88

toy and we are figuring out how to turn

1923.919

this toy into a BMW. We're trying to

1927.2

take this toy car and turn it into a

1929.12

real car, making it productive, making

1931.12

it sleek, making it useful.

1934.559

However,

1936.559

if Ian waits too long

1939.519

and AI becomes too embedded in companies

1943.2

and companies really only want

1946.96

developers with AI skills,

1950.72

then he might miss the boat. He might

1952.399

lose jobs or be unable to get contracts

1956.399

because he is not versed enough in AI to

1959.679

keep up with the current status quo. Now

1963.039

is that right now? Probably not. But it

1966.799

could be coming. So these are just

1968.32

things just in any tech stack any

1971.039

development. Over the years we've seen

1972.48

this you know back in the 80s it was

1975.2

Cobalt for train 90s it became C uh C or

1980.399

uh C C++ Java and then over time we've

1983.6

seen Python and you know Groovy, Ruby,

1986.96

all these other Rust uh languages come

1989.36

up. Those that jumped on those early

1993.679

were kind of the forerunners, but they

1995.6

had a hard time finding jobs. They had a

1997.519

hard time figuring out what to do with

1999.36

these. That's kind of where we're at

2000.88

with AI right now. It's like it's

2002.96

everywhere. It's there. People can jump

2005.44

in. They can play with it. We're still

2007.36

figuring out what to do with it. But the

2009.279

problem is once people start really

2010.96

figuring out what to do with this, that

2012.64

is the time to start jumping on, start

2015.76

learning it. Get in there, figure it

2018.08

out. Because if you're not, you'll be on

2020.08

that late side and then you're going to

2021.44

be playing catch-up and it's going to

2023.679

take you a little bit longer to get

2025.12

there. And if AI goes the way everyone

2027.6

thinks it's going to go, if you're too

2029.76

late in adapting to AI, you may be too

2032.24

late to stay on this trajectory. You

2035.2

might be playing catch-up for a long

2036.72

time.

2039.279

A lot to unpack there. I've got a couple

2040.96

things to show out. First one, it's not

2042.64

Ian, so I don't I want to protect the

2044.48

innocent. Ian is not the son. is

2046.88

anti-AI. Um Ian has got a different

2050.079

approach. Uh Ian happens to work for me

2051.76

for those of you guys that don't know.

2052.8

And and our approach for the we'll call

2055.2

the the junior developers I guess uh

2057.28

which are actually all mid-level

2059.04

developers at this point is to figure

2061.119

out where to best utilize AI. They have

2064.48

found that is useful as a tool. Um they

2067.52

don't use it in the same way I do

2068.879

because they're using it as developers.

2070.639

I'm using it as running a business

2072.96

operating system as well. like I'm using

2074.879

using it fully as a a virtual assistant.

2077.52

Um, if you're just starting out, if

2079.599

you're a if you are a junior developer,

2081.599

I would just say you need to understand

2084.8

how to talk to AI to utilize it, but do

2088.72

not make it your primary way of getting

2091.359

your work done because you need to

2093.679

understand the underpinnings of what

2096.8

you're building. And if you're not doing

2099.04

that on a regular basis, then you're

2100.32

going to end up using AI is going to be

2101.76

your crutch instead of being a tool. and

2103.839

then you end up in a a very bad

2105.359

situation. And I think that's what we're

2107.119

running into with businesses is they're

2108.56

just like, "Okay, I'm gonna I've got

2110.32

something bad. I don't have a process

2112.24

that I I understand, so I'm going to

2113.76

slap AI on it and I'm going to fix it."

2115.839

Um, I would say, uh, Michael referred to

2118.72

it as being a new toy and stuff. It is,

2121.28

yes, you can see it that way, but I

2123.2

would say it is a toy like a double

2125.119

barrel shotgun is a toy is that you can

2127.28

actually do a lot of damage and you need

2130.32

to be very intentional with what you're

2132.4

doing. There are already patterns that

2134.96

are out there for uh AI use, for AI

2137.599

architecture. There are definitely

2139.68

patterns of solutions that AI work very

2141.76

well with that that are proven that are

2144.32

documented that have um background

2147.04

behind it. Uh if you want some fun

2149.76

research is look into uh Albania's AI

2154.88

cabinet member u called Dileia and what

2158.48

they've done uh with that. there are a

2161.44

lot of there's a lot of places to go to

2163.68

realize find out where AI uh mistakes

2166.64

have been made but also where it can be

2168.72

very valuable and where power very

2170.56

powerful. So um I would which sort of

2174.32

gets us into the challenge for the week.

2177.839

Um Michael also mentioned you know this

2179.92

is all test driven development sort of

2181.68

follows this but it's actually bottom

2183.76

line it's like at the end of the day

2184.88

it's requirements driven development.

2186.64

It's like you have requirements and this

2188.16

is why I'm I'm going to harp on that

2189.839

just a bit is you have requirements

2193.119

and then that leads to your solution.

2195.359

Now if you do it one of the ways you can

2197.359

do it is you do requirements and then

2198.56

you have tests that actually validate

2200.56

those requirements and so now you say

2202.16

what I build after that my test should

2204.48

all should always succeed because they

2207.2

have to output essentially what the

2210

requirements are expecting.

2212.32

This is what you should do is actually

2214.16

let's practice requirements. I think I

2216.56

want to go to a 30-day challenge instead

2218.32

of a 7-day challenge. And I'll see what

2219.92

your your thoughts are on this, Michael,

2221.359

because we have not talked about this

2222.56

before because we almost never do. Um,

2226.8

I think is take a 30-day challenge,

2228.96

build a tool, build an app, and start

2233.52

with scratching your own itch. So, maybe

2236.24

your problem is um like as a developer,

2239.44

maybe your problem is I want a I need to

2241.76

have a better way to find my old code.

2244.4

is find code that is like, you know,

2246.24

find that JavaScript function I wrote

2248.64

two years ago that I know is out there

2250.24

and I just don't want to rewrite it

2251.599

again. Basically, building your own, you

2254.16

know, AI code repository that you can go

2257.119

refer to. That could be an example. But

2259.92

the key to this is write out first

2262.8

before you write any code or before you

2265.2

have AI write any code or anything like

2267.04

that is write out your requirements. Get

2269.52

yourself and it doesn't have to be word

2271.92

and all that kind of stuff. use like

2273.119

markdown or something just very

2274.72

specific. It has to do this and this and

2276.4

this and this and this and go into all

2278.24

of those and ask yourself questions

2279.599

about all of that is like how can I

2281.359

provide more information about each of

2282.8

those requirements and then

2285.839

start having a conversation with I pick

2287.68

your tool and basically walk through it

2290.24

and say maybe give it from the start say

2292.72

I want to build this. Give it all your

2294.96

requirements and start with what do you

2297.68

need to know? What have I missed? What

2299.44

can I, you know, what am I

2303.44

needing to add that I have not? Where

2305.52

are the gaps? Uh, is this is this

2308.48

valuable? Maybe you ask, is this

2310.48

something that already exists somewhere?

2312.24

What would it look like in the real

2313.52

world? What would the market look like

2315.04

for this? How would I talk about it?

2316.88

What would I call it? You can have these

2319.119

conversations, but the goal is to move

2321.28

you from requirements in that day one to

2324.72

a working application by day 30. Now,

2327.839

doing so that means that you're going to

2329.28

like give yourself this sprint. You're

2331.2

going to have to put some milestones in

2332.64

there. So, maybe by the end of the first

2334.079

week, you need to have something that is

2336.079

a solid uh data model and flow diagram

2339.68

or something of those n of that nature.

2342

Since you can use AI to generate code,

2343.92

you may have like maybe it's a clickable

2345.359

demo of some sort by week two. Then

2348.16

maybe it's uh I need to know what I have

2350.56

to have all of the functions fleshed

2352.24

out. So, I need to know essentially this

2353.68

is the feature set that I'm going to be

2355.2

generating. By the end of week three,

2356.88

maybe you need to be like, "All right, I

2358.56

need to have something that I can now

2360

play with, that I can work on, that I

2361.599

can test. I need to have the uh the

2364.56

test, the unit test generated for it. I

2366.72

need to have documentation generated for

2368.64

it." Those kinds of thing, and then get

2370.64

yourself to day 30. What are your

2372.64

thoughts on that as a a bigger than

2374.8

normal challenge?

2377.28

I think that works because it kind of

2379.2

flows with the conversation we had with

2380.96

Dustin and it gets us back to focusing

2383.68

on the requirements, focusing on what it

2386.56

is that we're trying to accomplish. What

2388.48

is our goal? Because a lot of times

2391.76

that's where we fail. We think we have

2394.079

the requirements, but they're so loosey

2396.16

goosey that you don't have the

2398

requirements. What you have is you have

2400.32

an idea that it's not fully fleshed out.

2403.2

And when you just try to implement that

2404.96

idea, you end up with a whole bunch of

2407.04

if then, what ifs. And if you have what,

2409.92

well, what if this, what if this, then

2411.52

you're not there. You don't have the

2412.88

requirements. Requirements should be

2415.599

able to be read as a statement. If you

2418.88

read the statement and there is no if

2421.359

then or well, what about this? Then you

2424.48

have a good requirement. If you have a

2426.48

statement that prompts, well, okay,

2428.48

well, what about this? What about this?

2429.92

What then you don't have a requirement.

2431.599

you have a just an idea. So get your

2436.32

statements, get your requirements

2438

flushed out so that they are as

2441.52

definitive as possible and then work

2444.079

through build the software. I love it.

2446.32

Um,

2448

one little tidbit I kind of want to

2449.839

throw in, uh, when we were talking

2451.92

about, you know, train, you know, the

2454.56

mid-level developers training the, uh,

2456.56

lower level developers to be and the

2458.48

mid-level developers to become senior.

2460.319

I'm already seeing this trend in a

2462.319

negative way. I'm seeing mid-level

2465.599

developers or team leads say, "I'm too

2469.52

busy. Go talk to AI. Go have AI solve

2473.359

your problem. I don't have time to deal

2474.72

with you. I'm too busy working this

2476.48

problem." That is very dangerous. If you

2479.92

are not helping your team problem solve

2483.28

your problems and you're pushing them

2485.28

off on AI,

2487.28

one of two things is going to happen.

2488.8

they're going to leave.

2490.96

Or two, your software is going to become

2493.28

a dumpster fire if it's not already. Or

2496.4

three, you're never going to move up.

2499.119

You're going to be stuck putting out

2501.44

fires. You're going to be stuck in that

2502.72

busy mode. You're not going to learn the

2505.68

management skills necessary to become

2507.92

that senior, to continue analyzing, and

2510.48

continue to grow.

2513.28

>> I think it's a good parting thought

2514.72

there. uh a good way to grow, a good way

2516.96

to uh help your career is like yeah you

2519.76

can use uh and you can use AI to get

2522.24

used to like ordering something around

2523.92

and and sending commands and see how

2525.599

that works. But then you definitely are

2527.04

going to have a you need to have that

2528.72

personal uh experience because that is

2531.2

what's going to help you grow as a

2532.64

developer and uh not only the soft

2535.28

skills but also the technical skills

2537.119

because it is always better to help you

2540.16

uh solidify something in your head if

2542.64

you have to explain it to somebody else.

2544.16

And the more you have to explain it, the

2545.44

more that becomes just something that

2546.88

now it's just embedded in your psyche

2549.68

essentially.

2551.2

Hopefully our ugly faces are not

2553.119

embedded in your psyche or the back of

2554.8

your uh eyeballs. And we thank you so

2558.24

much for hanging out with us. Uh thank

2559.839

you for spending your time and uh just

2562.72

enjoy your challenge, enjoy your

2564.8

weekend. And as always, go out there and

2566.8

have yourself a great day, a great week,

2569.119

and we will talk to you next week.