00:00:00
so in my last video I talked about why I
00:00:02
think vibe coding actually sucks but I
00:00:04
do want to kind of give you another
00:00:06
honest perspective i think sometimes the
00:00:08
content I post is a little bit more
00:00:10
clickbaity i know it's going to get
00:00:11
views this one I just want to give you
00:00:13
my honest opinion on AI because honestly
00:00:15
I feel like I'm the original vibe coder
00:00:17
if you go back 6 months ago um I coded
00:00:19
an application only using voice to text
00:00:22
okay here's the video right here and I
00:00:24
also did a live stream on it so using AI
00:00:27
exclusively without like diving through
00:00:29
the code and having to actually touch
00:00:30
code uh itself I've been doing for a
00:00:33
while and I've been using Copilot like
00:00:35
since it came out off and on i've been
00:00:36
using cursor to help build out various
00:00:38
other projects i'm not sure where I
00:00:40
actually officially switched to cursor
00:00:42
on my uh my my videos it was probably
00:00:44
like 6 months ago but I do believe I was
00:00:47
using different AI tools along the way
00:00:48
like Claude and ChatgBT and stuff with
00:00:51
that being said I've been using AI for a
00:00:52
while and I've been in the industry
00:00:54
coding for 11 plus years now so I want
00:00:56
to give you my perspective on when I
00:00:58
like to use AI when I just ignore AI and
00:01:01
I go straight to coding and also should
00:01:03
you be learning AI as a junior when
00:01:05
should you like leverage it more when
00:01:07
should you not and honestly I'll
00:01:09
probably just ramble about things so I
00:01:10
think the best way to exemplify AI is
00:01:12
let's talk about real things that I've
00:01:13
tried building using AI so recently this
00:01:16
morning I wanted to refactor my homepage
00:01:18
for my course platform that I'm trying
00:01:19
to build out and I wanted to make the
00:01:21
homepage a little bit more interesting
00:01:22
so I added a library called 3.js which
00:01:25
builds out 3D nodes and it puts them
00:01:27
around a 3D space around this this sun
00:01:31
basically now I use AI exclusively for
00:01:34
this right i went to cursor and I
00:01:36
basically kept asking it to you know
00:01:39
build out this 3D system i probably went
00:01:41
through like 10 or 15 iterations of
00:01:43
prompting but finally I got to something
00:01:44
that I want overall I think it took me
00:01:47
maybe like 30 minutes to an hour to get
00:01:48
it working with AI it did run into a
00:01:51
bunch of bugs along the way i have to
00:01:53
basically help tweak it a little bit so
00:01:55
the real question is is why did I use AI
00:01:57
to build this out versus just doing it
00:02:00
myself and there's a variety of reasons
00:02:02
right the first one is this is just
00:02:04
throwaway code this isn't a core thing
00:02:06
in my codebase that I truly need to
00:02:08
understand this is like a marketing
00:02:10
thing where I'm like let's just get a 3D
00:02:12
thing going so it just looks cool when
00:02:14
someone lands on this page and from a
00:02:15
marketing standpoint maybe that'll drive
00:02:17
up um engagement or drive up sales
00:02:21
secondly this is not important to my
00:02:23
overall career path i would consider
00:02:25
myself more of a web developer full
00:02:26
stack developer or DevOps engineer
00:02:28
building out 3D scenes is not something
00:02:31
that's going to really elevate my
00:02:33
current niche in my professional career
00:02:36
with that being said this is like a
00:02:37
prototype i could throw it away and not
00:02:39
care and it's also not core knowledge to
00:02:41
my career now the third thing I want to
00:02:43
mention is that this is something I
00:02:45
could build myself right i just want to
00:02:48
give you some context back in college I
00:02:49
took a 3D course for a semester we had
00:02:51
to learn about 3D scenes we had to learn
00:02:53
about 3D objects we learned about vertex
00:02:55
shaders fragment shaders we learned
00:02:57
about camera perspective transforms the
00:03:00
linear algebra like they went in detail
00:03:02
of using 3D and I think we also did it
00:03:05
in a browser as well so we kind of like
00:03:06
learned about
00:03:07
WebGL that knowledge that I learned
00:03:10
although it was 10 or 11 years ago it's
00:03:12
still somewhere up there right and that
00:03:14
was just one course in college i do
00:03:15
believe that I had to take like physics
00:03:17
3 in college or physics 2 where they
00:03:20
teach you how to take like magnetic
00:03:22
fluxes over 3D objects so you do have to
00:03:24
understand like mathematical formulas
00:03:26
that use like cosine and s and tangent
00:03:29
and how you can use them in equations to
00:03:31
calculate magnetic fields or flux around
00:03:34
objects and then on top of that I've
00:03:36
always been into gaming and 3D stuff so
00:03:38
like I've always played around with
00:03:39
Unity a little bit on the side i've
00:03:40
played around with these you know 3JS
00:03:42
libraries on the side i played around
00:03:44
with other 3D libraries and so 3D is not
00:03:47
something that's completely foreign to
00:03:48
me and so looking at this could I build
00:03:50
this myself i think I could the real
00:03:53
question is is how long would it take me
00:03:56
and how important is it for me to learn
00:03:58
it again like I said me knowing how to
00:04:01
build this by hand I don't think it's
00:04:03
valuable now if I were to switch careers
00:04:05
and become a game engineer designer then
00:04:08
yes I would probably should be able to
00:04:09
build this by hand and I should be able
00:04:10
to do it pretty fast it's probably not
00:04:12
too hard you probably just have to spawn
00:04:15
some random things around a 3D uh space
00:04:18
and then you just have to kind of draw
00:04:20
lines between these different nodes
00:04:22
randomly and if you look at the code I
00:04:23
mean it's not too long it's like 500
00:04:25
lines of code and I'm sure someone who
00:04:27
works in 3D space could easily build
00:04:29
this out themselves but using AI allowed
00:04:33
me to build this out much faster than I
00:04:34
could have done myself at some point in
00:04:36
your career you have to admit that using
00:04:39
a thirdparty library or using AI to just
00:04:42
get the job done is probably a much
00:04:44
better investment in your time honestly
00:04:47
I don't know what my original talking
00:04:48
point is but let's just keep on winging
00:04:50
it and see if we can talk about some
00:04:51
more stuff so if you're a beginner I
00:04:53
think there's a lot of confusion around
00:04:54
like should I be using AI when I'm
00:04:56
learning and when should I not use it
00:04:58
and going back to the story I just gave
00:05:01
I think if you understand how to build
00:05:03
something yourself or you feel confident
00:05:05
that you could build it yourself and you
00:05:06
know it's just going to take you a lot
00:05:08
of time to go back and refresh those
00:05:10
things could I have gone back and
00:05:12
refreshed my memory about all the things
00:05:13
I learned 10 years ago about 3D spaces
00:05:15
and you know try to really dive into 3JS
00:05:18
and learn it
00:05:19
absolutely do I want to spend 8 16 hours
00:05:22
doing that not really right i don't have
00:05:25
much time i want to focus on my YouTube
00:05:27
channel i want to focus on just shipping
00:05:28
this little feature and I don't want to
00:05:30
actually like improve my knowledge in 3D
00:05:33
space just yet maybe in the future I
00:05:35
will but right now I don't care that is
00:05:37
the mentality I would recommend you as a
00:05:39
junior if there's something that you're
00:05:41
doing and you don't think it's very
00:05:43
important to the overall goal of
00:05:45
learning then maybe you could just use
00:05:48
AI to do it for example let's say you
00:05:49
want to build out a website but you just
00:05:51
you you're like a backend engineer you
00:05:53
don't care about the UI then yeah use AI
00:05:56
to generate your landing page and people
00:05:58
argue that that's a bad take but I mean
00:05:59
honestly like if you're trying to build
00:06:01
out a landing page typically you go and
00:06:02
you find pre-existing components right
00:06:04
and you just copy and paste them into
00:06:05
your application anyway and you don't go
00:06:08
and you ever learn about like okay how
00:06:09
does this drop down actually work under
00:06:11
the hood how does this type ahead
00:06:13
actually work under the hood because you
00:06:15
don't care you just want to build a
00:06:17
working application and you care more
00:06:19
about the backend engineering type of
00:06:21
thing so like you want to focus on using
00:06:23
AI less on the things that you're truly
00:06:26
passionate about or the things you truly
00:06:28
care about because you need to have a
00:06:29
core fundamental understanding of how
00:06:31
they all work now in a perfect world if
00:06:34
you have infinite time then you should
00:06:35
probably be learning everything and just
00:06:36
understand how all this stuff works and
00:06:38
how it all fits together and how you can
00:06:40
do it yourself but realistically there's
00:06:42
only so many hours in a day and once you
00:06:45
get a job you're asked to ship features
00:06:47
and deliver value right okay and so
00:06:49
there's like a trade-off of how much
00:06:51
time do I spend actually learning it and
00:06:53
understanding it versus how much time do
00:06:55
I just get it done now on that point if
00:06:58
you're on a team and you have a bunch of
00:07:00
engineers that are working with you it's
00:07:03
probably a good idea to understand how
00:07:05
it all works because what's going to
00:07:06
happen is you're going to ship a feature
00:07:09
and then someone's going to come back
00:07:10
and ask you okay there's a little bug
00:07:12
with it how did you implement this how
00:07:14
does this work and I don't think sitting
00:07:17
in a meeting in front of eight other
00:07:18
people saying "I don't know i use AI to
00:07:21
generate it." I don't think that's going
00:07:22
to really reflect good on your character
00:07:25
and who you are i think managers are
00:07:27
going to be kind of like scratching
00:07:28
their head of like "Okay does this guy
00:07:29
actually know how to code?" So in a
00:07:31
professional setting I think there's
00:07:33
more weight into understanding the code
00:07:35
that's being shipped because a lot of
00:07:37
the times like if you look at this code
00:07:39
here it just fills the code with magic
00:07:41
numbers it's not very configurable like
00:07:43
if you look at the calculations here of
00:07:45
the tilt angle and all this orbit speed
00:07:47
it just hard-coded random numbers
00:07:49
everywhere which is very unmaintainable
00:07:52
this is a very bad way to code like you
00:07:54
should not have magic numbers sprinkled
00:07:55
out throughout your codebase you should
00:07:57
at least have a centralized place for
00:07:59
configuration now granted I could ask AI
00:08:02
to refactor or remove all magic numbers
00:08:04
and just make it a little bit cleaner um
00:08:07
but you probably won't know to ask AI
00:08:09
that until you gain more experience so
00:08:12
some more things I would recommend with
00:08:13
AI is get good at asking it things such
00:08:15
as like can you refactor this code to
00:08:17
make it cleaner can you make it more
00:08:18
modular can you make it more abstract
00:08:20
can you make it more decoupled can you
00:08:22
make it more secure i think just asking
00:08:24
it to do things over your codebase along
00:08:26
the way is a very good approach um and
00:08:29
if you're a beginner I think you should
00:08:30
ask those questions and then review the
00:08:32
diffs right if you're using AI and if
00:08:34
you're a junior or trying to learn
00:08:36
review the diffs because again you don't
00:08:38
want to get caught up in a meeting and
00:08:39
they're asking you you know what is what
00:08:41
is this line doing how does this
00:08:42
actually work and you're like dude I
00:08:44
don't know i just I I don't actually
00:08:46
work i just you know let AI do my job i
00:08:48
just think it's going to look pretty bad
00:08:50
now let's talk about the quality of AI
00:08:52
because a lot of people say AI is
00:08:53
garbage it generates slop and from my
00:08:56
perspective it's that's true it can
00:08:58
generate a lot of slop and you have to
00:09:00
have a proactive approach to accepting
00:09:03
the things that it generates which is
00:09:04
why I'm kind of against vibe coding
00:09:06
although vi coding is great if you just
00:09:08
want to build out something standard
00:09:09
like one thing I think AI is really
00:09:11
great for if you're doing web
00:09:12
development it's almost like these
00:09:14
models have been specifically trained on
00:09:15
JavaScript and TypeScript and web
00:09:17
development maybe that's because there's
00:09:19
just a bunch more code out there it can
00:09:20
train with but when you try to ask it to
00:09:22
do 3D stuff and like 3D games it seems
00:09:25
like it struggles a little bit
00:09:27
and I just think that there's just less
00:09:29
demo repos out there i think there's
00:09:31
less examples that it can be trained to
00:09:33
kind of give you code examples from but
00:09:36
in terms of what I've seen it done like
00:09:37
I have done vibe coding on web
00:09:39
applications and if you can learn how to
00:09:42
give it the right amount of context it's
00:09:45
going to generate a decent amount of
00:09:47
code that works right i would say 90 to
00:09:49
80% of the time it works pretty well if
00:09:51
you provide it the context if you just
00:09:53
go in blind and like make a new chat
00:09:54
window and tell it to achieve something
00:09:57
usually it's not going to work that
00:09:59
great so some ways that you can make it
00:10:01
better is there's a cursor file if
00:10:03
you're using cursor you typically want
00:10:05
to give it rules so like you should have
00:10:07
a MD file that explains the architecture
00:10:10
of your system how do you have your API
00:10:12
set up like what layer of your codebase
00:10:14
is responsible for talking to your
00:10:16
database what layer is responsible for
00:10:18
your components like where does stuff
00:10:20
live in your codebase the less you give
00:10:22
AI the more bad code it's going to
00:10:25
generate so I would think you should
00:10:27
leverage the rules in cursor or whatever
00:10:29
AI system that you're using leverage
00:10:31
giving as much context as possible i
00:10:33
think rules are automatically injected
00:10:35
when you ask prompts but I mean you can
00:10:37
do the at symbol and you can actually
00:10:38
click and drag like folders into here
00:10:40
you can click and drag files you can
00:10:42
reference various other things and the
00:10:45
more you give it the better it typically
00:10:47
performs and also the more guardrails
00:10:49
you put up like let's say you only
00:10:51
wanted to modify this page then I would
00:10:54
probably make sure that your chat prompt
00:10:56
only has this as context and nothing
00:10:58
else otherwise AI might start touching
00:11:00
other files so in order to give AI the
00:11:04
context that makes it work the best you
00:11:05
have to understand your codebase for
00:11:07
example if you opened up this really
00:11:09
small you know project and then you
00:11:12
started asking AI to start implementing
00:11:13
features it's probably not going to
00:11:15
perform as well for you than it would
00:11:17
for me because I know the different
00:11:20
files I need to modify to achieve the
00:11:23
best results for example if I wanted to
00:11:24
modify my course content page that has
00:11:26
like this video player well then I'd
00:11:29
probably reference the layout that does
00:11:32
that okay I already know off the bat
00:11:34
what file is responsible for this and I
00:11:37
can just ask it to modify this file and
00:11:40
that helps guide AI and give you a lot
00:11:43
better results let me hide my face real
00:11:45
quick so here's an example prompt I
00:11:46
would do if I wanted to add a new
00:11:49
feature to this codebase that basically
00:11:50
added a button here and it did something
00:11:53
maybe it inserted some data in the back
00:11:54
end so I'd say modify this page to show
00:11:57
a YOLO
00:11:59
button so that when a user clicks it
00:12:03
it'll create a new record in my YOLO
00:12:07
table please create a YOLO table in
00:12:11
schema so I'm going to pass it the
00:12:13
schema file i'm also passing the
00:12:15
relevant file that should add the button
00:12:17
remember to use server functions
00:12:21
in data access layer and use case to
00:12:26
achieve it let's just try this out i'm
00:12:28
going to let it generate the code real
00:12:29
quick and we're going to kind of analyze
00:12:31
the code and what it's doing and see if
00:12:33
it's on the right path one thing I would
00:12:34
recommend doing is read through it as
00:12:38
it's coming through like just scan
00:12:39
through it there's been so many times
00:12:42
where like the first code example that
00:12:43
pops up is in a file that it should
00:12:45
never have touched and it's modifying
00:12:47
code in a way I would never have done
00:12:49
and when it starts doing that honestly
00:12:50
I'll just click stop here and I'll just
00:12:52
say you are doing the wrong thing and
00:12:55
I'll kind of correct it along the way
00:12:57
now if you have to do all this extra
00:12:58
handholding to get AI to be useful like
00:13:01
why would I still use it because
00:13:04
honestly having to type all this by hand
00:13:06
or tab completing it is still going to
00:13:09
be slower than just letting AI generate
00:13:11
hundreds if not thousands of lines of
00:13:13
code in a single prompt and that's the
00:13:17
main selling point is that now we don't
00:13:18
have to worry about being very efficient
00:13:20
at our idees and editing code and typing
00:13:24
you could be a terrible typist but you
00:13:26
could still build some amazing
00:13:28
applications so we're not limited by our
00:13:30
physical capabilities of typing on a
00:13:32
keyboard anymore we are limited by our
00:13:35
abilities to describe and communicate
00:13:38
what our needs are to the software AI
00:13:41
which again this is a skill that's going
00:13:43
to translate to your job right if you
00:13:44
can become good at communicating your
00:13:46
needs to not only your co-workers and
00:13:48
your managers or maybe even your you
00:13:50
your wife or your spouse like being good
00:13:53
at communicating and really describing
00:13:55
what you want is something that you're
00:13:57
going to learn the more you use AI and I
00:14:00
think you're going to shoot yourself in
00:14:01
the foot if you're not trying now to
00:14:03
understand how you need to talk to the
00:14:06
LLM because I found that there's a very
00:14:08
particular way that you need to talk to
00:14:10
the LLMs to make it more efficient and I
00:14:12
just don't think people actually
00:14:13
understand how to prompt very well with
00:14:15
so we got a new schema that was
00:14:17
generated we got a YOLO table okay and
00:14:19
it set up relations so again I'll kind
00:14:20
of skim through the code like it made a
00:14:22
new table um you should know how your OM
00:14:24
works you should know how your table
00:14:26
should be generated what columns should
00:14:28
be like what what are these serial
00:14:30
values versus like strings and
00:14:31
timestamps these are things that you as
00:14:33
an engineer need to understand okay
00:14:35
right here this actually put the DB YOLO
00:14:37
file in the wrong place so at this point
00:14:40
you might want to ask yourself is my
00:14:42
cursor rules are they set up wrong am I
00:14:45
not explaining enough in these rules of
00:14:48
where my data access layer needs to be
00:14:51
because right now it just put this in a
00:14:53
complete wrong directory this DB
00:14:55
directory is not for my actual like um
00:14:58
methods so at this point I'd probably
00:15:00
switch this to data access and there we
00:15:02
go i fixed that issue but what I'm
00:15:04
trying to show you is that the AI is
00:15:05
going to get stuff wrong but it's your
00:15:06
responsibility to go and remember okay I
00:15:09
need to tweak my rules and tweak my
00:15:11
context so this will not happen again in
00:15:14
the codebase that I'm looking at we look
00:15:16
at this file i would say that this put
00:15:18
this in the wrong location and then also
00:15:20
the layout this this is not the file I
00:15:23
wanted it to change so at this point you
00:15:25
didn't prompt correctly i'm going to
00:15:26
reject everything because I just don't
00:15:28
think I prompted it very well and I'm
00:15:29
going to try to start fresh okay we
00:15:33
instead are going to give
00:15:36
it the specific file that I want and
00:15:40
then I'm going to give it my data access
00:15:43
layer and my use cases so that it knows
00:15:46
what folders I'm talking
00:15:48
about okay so let's try prompting this
00:15:51
one more time i want you to add a YOLO
00:15:54
button to this index page at the top
00:15:58
near the other edit admin buttons okay
00:16:01
again I'm being specific i wanted to add
00:16:03
it here i want it to be to the left of
00:16:04
the edit segment so I should probably be
00:16:06
very specific so I'm going to say at the
00:16:09
top to the left of the edit
00:16:12
segment button i know my head's hiding
00:16:15
this let me change
00:16:18
this when a user clicks the
00:16:21
button invoke a server
00:16:24
function define it in
00:16:27
the the same layout file so I'll just do
00:16:31
this which should call a use case to
00:16:36
create a YOLO entry inside the
00:16:41
database that use case
00:16:43
should invoke a data access
00:16:48
method also update
00:16:52
schema to provide a new yellow table now
00:16:57
honestly I'd probably use voice to text
00:16:58
to get this working but let's just try
00:17:00
this one more time so let's see how much
00:17:02
better it gets when I gave it a little
00:17:05
bit more context about where I wanted my
00:17:07
stuff to live how I wanted it to create
00:17:09
it right i'm getting a little technical
00:17:10
here i'm telling it where to put the
00:17:12
server function i wanted to put it
00:17:14
inside this file okay so now let's look
00:17:16
through this so we have a YOLO table
00:17:18
that I got created it looks like it's
00:17:20
pretty good we have a YOLO data access
00:17:23
method that creates it this looks fine
00:17:26
uh it does have a little database error
00:17:27
so I'm not sure what's going on there
00:17:28
but like fixed typescript there we have
00:17:30
a use case okay so it created a use case
00:17:33
function in the correct location it used
00:17:35
the proper suffix on my files if we look
00:17:38
at this we now have a button here
00:17:42
um that looks like it's creating a YOLO
00:17:44
function which is calling my use case so
00:17:46
all this code looks perfectly fine this
00:17:48
code looks like it did a good job and if
00:17:50
I click it um it might crash because I
00:17:52
don't actually have like a my schema set
00:17:54
up um but at least the alert is popping
00:17:57
up for us okay okay so this worked a lot
00:17:59
better and I think the reason is because
00:18:01
I prompted it a lot better with a lot
00:18:04
more context again this all boils down
00:18:06
to the question of could I have done
00:18:07
this faster myself and honestly I would
00:18:09
probably still say no i still think that
00:18:12
modifying these four files putting all
00:18:15
the imports and making sure that the
00:18:17
code is in the right place and trying to
00:18:18
scroll through the code and figure out
00:18:20
the right place to put these things and
00:18:22
then double-checking that I didn't maybe
00:18:24
forget something or I did something
00:18:25
wrong and if I were to mistype something
00:18:28
that could add another step of me having
00:18:30
to debug my codebase because I typed in
00:18:33
the wrong character somewhere and now I
00:18:34
broke my codebase i would say that AI
00:18:37
for the most part if it does it correct
00:18:40
then you will save a lot more time and
00:18:43
the goal is to make it generate correct
00:18:45
output at a higher percentage most of
00:18:48
the time now I will say leave a comment
00:18:50
if there's certain strategies that you
00:18:52
have found when doing prompting to make
00:18:54
it be more accurate um like what what do
00:18:58
you do do you just use cursor rule files
00:19:00
do you just provide it a bunch of
00:19:02
context or do you just think that the
00:19:04
benefits that we're getting from this
00:19:05
just truly isn't as good as people think
00:19:07
it is so again like let's look at this
00:19:09
application and let's try to break down
00:19:11
like where I used AI to generate stuff
00:19:14
the layout of this I used AI to generate
00:19:17
most of the stuff you're seeing here I
00:19:19
used AI to generate again could I have
00:19:20
done it myself absolutely could I have
00:19:22
gone and found a third party React
00:19:25
component library that does this type of
00:19:27
layout for me sure could I have done
00:19:30
something where when I go to mobile like
00:19:32
it still creates a mobile navigation
00:19:34
menu that's the exact same probably but
00:19:38
would doing it myself have saved me
00:19:41
time probably not i think AI has made
00:19:44
this a lot faster like even all these
00:19:45
components that I added in here's
00:19:47
another thing i asked AI to add this
00:19:49
little hover effect this is something
00:19:50
I've seen done on other websites is this
00:19:52
something I could have done
00:19:54
myself probably but I would probably
00:19:56
still go to Stack Overflow and I go and
00:19:58
look up examples or I go on GitHub and
00:20:00
find examples and then I copy and paste
00:20:02
those examples into my codebase and then
00:20:03
I tweak them myself so at the end of the
00:20:05
day am I truly doing it myself without
00:20:07
assistance from you know the internet no
00:20:11
i'd probably still be looking for
00:20:13
example things now some people are good
00:20:15
enough where they understand all the
00:20:16
nuances of CSS and animations and
00:20:19
opacity and radials and stuff like that
00:20:21
that they could have easily done this
00:20:23
themselves but when it takes me 10
00:20:25
seconds to prompt AI and 5 seconds to
00:20:28
wait for it to add this hover style on
00:20:30
cards you really have to acknowledge the
00:20:33
fact that you're going to move faster
00:20:35
just using AI in certain situations now
00:20:38
I've been rambling for a while and I'm
00:20:39
going to give my opinions on it from an
00:20:40
engineer who's been doing this for a
00:20:42
while my
00:20:43
ultimate strategy right now is I
00:20:46
typically prompt AI first then I review
00:20:49
the code the prompts I give I will make
00:20:51
them as specific as possible to the
00:20:54
knowledge that I have of my codebase so
00:20:56
that I get good results and then I
00:20:58
review the code and accept it and then
00:20:59
I'll reprompt it to a certain amount of
00:21:02
times if I'm prompting this thing five
00:21:03
or six times and it keeps giving me bad
00:21:07
results to me that's a signal that
00:21:09
you're just not prompting it correctly
00:21:11
or you've encountered something that the
00:21:12
AI just cannot figure out the LLM just
00:21:15
cannot solve it and it might be more
00:21:17
beneficial for you to jump into the
00:21:19
files yourself and start doing tab
00:21:21
completions or start understanding so
00:21:23
you can get more context but even like I
00:21:25
said like I prompted AI maybe you know
00:21:28
10 or 15 times maybe even 20 times to
00:21:30
get this little 3D thing working and it
00:21:33
still is probably a significantly less
00:21:36
amount of time than if I were to try to
00:21:38
do this myself and learn the 3D spaces
00:21:40
again and learn the correct animations
00:21:43
to rotate the camera and stuff a lot of
00:21:45
work probably could go into this not
00:21:47
necessarily hard work it's just stuff
00:21:48
that I have to go back and relearn and
00:21:51
those are the strategies that I've been
00:21:53
taking but I do want to reiterate that
00:21:54
when people say AI generates slop I mean
00:21:56
this code it's pretty sloppy right so
00:21:59
you still have to come back with like a
00:22:00
fine tooth comb and tell it to refactor
00:22:02
and put things and get rid of magic
00:22:04
numbers and make it better i think you
00:22:06
can just keep on adding a lot of bad
00:22:07
code to your codebase if you don't have
00:22:10
someone with experience of maintaining
00:22:13
large code bases and making them
00:22:14
maintainable coming back and asking AI
00:22:17
to fix things in a certain way so my
00:22:19
ultimate opinion is yes you should be
00:22:20
using AI you should be learning how to
00:22:23
work with it you should be leveraging it
00:22:25
when you think it's going to make you
00:22:26
move faster but at the same time take
00:22:29
some time to understand what it's doing
00:22:31
in the code that it's generating it's
00:22:33
just going to make you a better engineer
00:22:34
in the long run and I don't think we're
00:22:37
at the point yet that AI is going to
00:22:39
take our jobs i think right now it's a
00:22:41
tool that can expedite your progress and
00:22:44
the people who know how to correctly
00:22:45
prompt and use AI to get results done I
00:22:48
think are just going to be put above
00:22:51
others who don't um and maybe that's
00:22:53
ultimately going to take jobs right
00:22:54
maybe that's ultimately going to cut
00:22:56
jobs away because now you have one
00:22:57
person who's able to ship multiple apps
00:23:00
and functionality at the same time
00:23:01
compared to other people trying to do it
00:23:03
manually they're going to fall behind if
00:23:05
you're watching my content and you trust
00:23:07
my advice again I would say especially
00:23:11
if you're in the web development space
00:23:12
you should probably use AI uh to help
00:23:15
you generate some of this stuff again
00:23:17
that's why I use cursor in all my videos
00:23:18
if I didn't think AI was this useful I
00:23:21
would not have two different cursor
00:23:23
subscriptions i have a subscription for
00:23:25
my personal laptop i have a subscription
00:23:26
for work that's how useful I feel cursor
00:23:29
has made me and maybe that's a
00:23:31
subjective feeling it's hard to
00:23:33
quantitatively measure how much faster
00:23:35
AI is making us like some people say
00:23:37
that it makes you a thousand times
00:23:38
faster or 100 times faster which in some
00:23:41
scenarios yes I have spent days i
00:23:44
remember being on a project we had eight
00:23:46
developers all trying to solve the same
00:23:47
bug and we spent two weeks together
00:23:51
looking through the codebase trying to
00:23:52
figure out what was causing this bug and
00:23:55
it boiled down to a single boolean that
00:23:56
we had this flip from false to true
00:23:59
somewhere in our codebase okay now could
00:24:02
AI have helped us find that bug faster
00:24:05
potentially and so there's scenarios
00:24:07
like that that I don't think we measure
00:24:09
where it takes a developer like a couple
00:24:11
of days or hours to debug something and
00:24:14
if AI can just throw you that bone and
00:24:16
help you like solve it faster these are
00:24:18
the things we're not measuring and I'll
00:24:20
say I have seen this multiple times you
00:24:21
using AI where I'm trying to do
00:24:23
something myself the you know the
00:24:25
classical way of programming i run into
00:24:27
a bug I cannot figure it out and I'll
00:24:29
sit there for 10 or 20 minutes just
00:24:31
trying to figure it out and then I
00:24:33
remember oh I have cursor let me just go
00:24:34
ahead and have cursor try to figure out
00:24:36
for me it does it in like 10 seconds and
00:24:38
I'm like "Okay I should have actually
00:24:40
leveraged cursor first to help me get
00:24:42
through this instead of wasting 20
00:24:43
minutes because those are 20 minutes
00:24:44
I'll never get back." Okay and I know
00:24:47
for a fact the bug I just solved is
00:24:49
something dumb it's some stupid typing
00:24:51
error a misspelling of something me
00:24:53
using some type of API incorrectly and
00:24:56
whether or not I solve the bug myself or
00:24:58
I let AI do it for me like I'm going to
00:25:00
read through the code and understand so
00:25:02
I don't do it again in the future
00:25:03
hopefully anyway I know I've been
00:25:04
rambling a while around this video um
00:25:06
and leave a comment if you feel like I
00:25:08
just ramble for too long but uh that's
00:25:10
about it have a good day
