WEBVTT

00:00:00.060 --> 00:00:03.700
Today,
I woke up to an interesting post by

00:00:03.780 --> 00:00:07.729
Rauch, the CEO of Vercel, about
which I wanna talk, and also

00:00:07.760 --> 00:00:11.620
about the implications
and my thoughts on the,

00:00:11.660 --> 00:00:15.620
how important the actual code is,
today and in

00:00:15.660 --> 00:00:18.950
the future. Now, in this post,
Guillermo says,

00:00:19.490 --> 00:00:21.410
"Code is an output. Nature is healing.

00:00:21.640 --> 00:00:23.560
For too long, we treated code as input.

00:00:24.020 --> 00:00:27.660
We glorified it, hand formatted it,
prettified it, obsessed over it.

00:00:28.140 --> 00:00:31.420
We built sophisticated GUIs to write it
in, IDEs.

00:00:31.900 --> 00:00:35.710
We syntax highlight, treeset,
minimap the code, keyboard

00:00:35.800 --> 00:00:39.660
triggers, inline autocompletes,
ghost text. What color scheme is that?

00:00:40.290 --> 00:00:43.900
We stayed up debating the ideal length of
APIs and function

00:00:44.000 --> 00:00:47.940
bodies.
Is this API going to look nice enough for

00:00:47.960 --> 00:00:51.460
read?
We're now turning our attention to the

00:00:51.680 --> 00:00:55.360
requirements, specs, feedback,
design inspiration.

00:00:55.800 --> 00:00:59.600
Crucially, production inputs.
Our coding agents

00:00:59.680 --> 00:01:03.480
need to understand how your users
are experiencing your application,

00:01:03.840 --> 00:01:07.480
what errors you're running into,
and turn that into code.

00:01:07.720 --> 00:01:11.620
We will inevitably glorify code less,
as well as coders.

00:01:12.060 --> 00:01:16.000
The best engineers I've worked with always
saw code as a means

00:01:16.040 --> 00:01:19.760
to an end anyway,
an output that's bound to soon be

00:01:19.820 --> 00:01:22.860
again." Wow, there's a lot in this post.

00:01:23.760 --> 00:01:23.960
Now,

00:01:25.900 --> 00:01:27.700
I have a lot of thoughts about this.

00:01:27.800 --> 00:01:31.210
Now obviously,
and I think we can all see this,

00:01:32.320 --> 00:01:35.960
the, the entire role
or the entire world of

00:01:36.000 --> 00:01:38.340
programming is changing rapidly.

00:01:38.380 --> 00:01:41.900
There's lots of stuff going on there and,

00:01:42.280 --> 00:01:46.120
especially, I,
and I guess many people in the industry

00:01:46.200 --> 00:01:50.020
feel that since December,
it accelerated again with

00:01:50.500 --> 00:01:53.840
Opus 4.5 and Cloud Code and now also

00:01:53.980 --> 00:01:57.940
Codex and we got new models
and new tools coming up almost every

00:01:58.040 --> 00:02:00.840
week. Things are changing quickly.

00:02:00.870 --> 00:02:04.630
And on a personal note, of course,
I'm trying to keep up with that,

00:02:04.840 --> 00:02:08.800
also regarding the content I create,
re-regarding the courses I create, and of

00:02:08.880 --> 00:02:11.460
course how I work, which
is the foundation of that all.

00:02:12.040 --> 00:02:15.880
And, uh,
that's why I released new courses on Cloud

00:02:15.940 --> 00:02:19.600
And I'm, by the way,
running a big promotion right now where

00:02:19.720 --> 00:02:23.640
my annual membership,
which gives you access to all the courses

00:02:23.700 --> 00:02:27.000
super low price. But it
is having an impact on all of us.

00:02:27.100 --> 00:02:31.020
That is my point here.
It's changing what we do, which

00:02:31.080 --> 00:02:34.540
content I create,
and of course how we build software.

00:02:35.720 --> 00:02:39.620
However, I'm not so sure
if we really move away

00:02:39.660 --> 00:02:43.300
from the code. And to me,
this post reads like code

00:02:43.540 --> 00:02:45.530
doesn't matter anymore, and

00:02:46.560 --> 00:02:49.820
I think I disagree with that quite a bit.

00:02:50.080 --> 00:02:53.760
Now I will of course admit,
and obviously it's true,

00:02:53.860 --> 00:02:57.780
that in the past, yes, we did obsess

00:02:57.920 --> 00:03:01.880
over API design, over naming

00:03:01.940 --> 00:03:05.080
of variables and functions, of the

00:03:05.160 --> 00:03:09.020
aesthetics of the code.
And I would actually argue

00:03:09.720 --> 00:03:13.580
parts of this are still important today,
and

00:03:13.660 --> 00:03:16.490
very likely also in the future

00:03:17.060 --> 00:03:20.960
because even if the role of humans shifts
away from

00:03:21.240 --> 00:03:23.980
writing the code towards reviewing it,

00:03:24.860 --> 00:03:28.820
what is easier to review? A,
a messy code base

00:03:28.900 --> 00:03:32.780
with ten thousands lines of codes that are

00:03:32.900 --> 00:03:36.360
partially redundant
or unnecessarily complex with

00:03:36.720 --> 00:03:40.440
weirdly named or shaped functions, or a

00:03:40.520 --> 00:03:43.900
clean code base?
What will be easier to review?

00:03:44.300 --> 00:03:48.049
And even if parts or even big parts of
that review

00:03:48.200 --> 00:03:51.940
process shift towards machines
and AI in the future, which of

00:03:52.000 --> 00:03:55.720
course is possible, even then I'm not

00:03:55.780 --> 00:03:59.400
convinced that the code quality does

00:03:59.540 --> 00:04:03.360
not matter at all anymore in that world,
and if it's

00:04:03.520 --> 00:04:07.180
just that you're paying for all those
tokens that are being

00:04:07.300 --> 00:04:11.180
generated and reviewed.
If something can be built

00:04:11.320 --> 00:04:14.960
cleaner, it will very likely consume less

00:04:14.970 --> 00:04:18.890
tokens for both generating
that code as well as

00:04:18.920 --> 00:04:22.430
for reviewing it.
And token cost looks like it will

00:04:23.080 --> 00:04:26.920
be very,
very important in the future because right

00:04:27.020 --> 00:04:30.520
subsidized plans by Anthropic, by OpenAI.

00:04:30.540 --> 00:04:33.020
We're not paying the real token cost.

00:04:33.370 --> 00:04:37.120
The $20 or $200 subscriptions w- you might

00:04:37.200 --> 00:04:37.960
have, and I have,

00:04:38.840 --> 00:04:42.600
these are not subscriptions where there
was vendors necessarily

00:04:42.920 --> 00:04:46.870
make any money off. The true token cost
is higher than that,

00:04:47.270 --> 00:04:51.140
and it's quite likely to go up in the
future, or

00:04:51.180 --> 00:04:54.820
what we pay to go up in the future
when those

00:04:54.840 --> 00:04:57.360
companies need to earn money at some
point.

00:04:57.480 --> 00:05:01.290
Obviously,
maybe due to all the technical progress,

00:05:01.440 --> 00:05:05.260
start at a point where the token cost as a
whole got down a bit, but

00:05:05.320 --> 00:05:07.640
we don't know that yet with certainty.

00:05:07.680 --> 00:05:11.540
We don't know how
that market will look like in the future,

00:05:11.620 --> 00:05:15.380
down theoretically, of course,
if companies can

00:05:15.480 --> 00:05:19.300
charge you a certain price,
which may be substantially above

00:05:19.340 --> 00:05:23.180
the true price, well,
they will happily take that margin.

00:05:23.500 --> 00:05:25.340
But yeah, th-
that's a totally different topic.

00:05:25.400 --> 00:05:29.140
But tokens and token efficiency matters
today and very

00:05:29.180 --> 00:05:33.140
likely also matters in the future,
and that of course is related

00:05:33.460 --> 00:05:37.160
to code quality because, in general, good

00:05:37.260 --> 00:05:39.960
code, clean code is

00:05:41.080 --> 00:05:44.580
code that is not allowed to grow endlessly
and

00:05:44.600 --> 00:05:48.060
become endlessly complex and complicated.

00:05:48.580 --> 00:05:49.460
Now of course,

00:05:50.360 --> 00:05:54.308
there is an argument to be made
that w-Programmers in the past

00:05:54.488 --> 00:05:58.428
also sometimes had the tendency to write

00:05:58.708 --> 00:06:02.528
a bit too much code maybe in certain
points to come up with a

00:06:02.708 --> 00:06:06.548
clean API and maybe some unnecessary
abstractions,

00:06:06.588 --> 00:06:09.848
which they didn't really need at the point
of time where they were implemented.

00:06:09.888 --> 00:06:13.588
We've probably all been there where we've
worked on a side project

00:06:13.648 --> 00:06:17.148
with zero users, and we decided
that we want

00:06:17.188 --> 00:06:21.158
to implement our database access such
that we could easily

00:06:21.308 --> 00:06:24.038
swap the database even though we never
intended to do that.

00:06:24.068 --> 00:06:27.708
And yeah, therefore,
we wrote some adapter that was, of

00:06:27.748 --> 00:06:30.408
course,
way more complex than it needed to be.

00:06:30.448 --> 00:06:32.388
We've all been there and wasn't great.

00:06:32.748 --> 00:06:36.658
No, it was not,
and it will not be today or in the future,

00:06:36.728 --> 00:06:38.888
if that's code written by you or by AI.

00:06:38.917 --> 00:06:42.488
And right now, and
that can of course change, but right now,

00:06:42.948 --> 00:06:46.868
in my experience, these AI models
and tools like

00:06:46.968 --> 00:06:50.448
Codex, Cloud Code, they have a tendency

00:06:50.508 --> 00:06:53.548
to overcomplicate things, to

00:06:54.148 --> 00:06:57.588
introduce unnecessary complexity,
to suggest

00:06:57.888 --> 00:07:01.688
changes that, yeah,
really don't make a lot

00:07:01.748 --> 00:07:04.668
of sense that are unnecessarily complex.

00:07:05.088 --> 00:07:08.968
Codex, especially right now,
has a tendency to

00:07:09.188 --> 00:07:12.928
never remove any code, just add more
and more and more code and

00:07:13.068 --> 00:07:16.328
add fallback code and legacy handling,
and you have to

00:07:16.368 --> 00:07:19.848
explicitly almost force it to delete code
and to let

00:07:19.908 --> 00:07:23.788
go, uh, of old APIs. So that, of course,
can

00:07:23.848 --> 00:07:27.508
all change, but these
are things you have to fight

00:07:27.608 --> 00:07:31.508
about, or, uh, for today. And that is, of

00:07:31.608 --> 00:07:35.488
course,
kind of what this post says doesn't matter

00:07:35.528 --> 00:07:37.588
anymore, but I say, yes, it does matter.

00:07:37.848 --> 00:07:41.648
You, as a developer,
you have to take those fights and you

00:07:41.748 --> 00:07:45.268
have to care about the actual code
and the code quality

00:07:45.908 --> 00:07:46.368
today,

00:07:47.188 --> 00:07:48.708
and very likely also in the future.

00:07:49.088 --> 00:07:52.888
Obviously,
nobody knows what the future holds,

00:07:52.968 --> 00:07:56.768
always be important
that you have a code base

00:07:56.828 --> 00:08:00.528
that doesn't grow unnecessarily complex
for many

00:08:00.648 --> 00:08:02.448
reasons.
Token efficiency being one of them.

00:08:02.698 --> 00:08:06.528
Another reason, of course, uh,
and also the review part, by the

00:08:06.568 --> 00:08:10.058
way. A- another reason, of course, is that

00:08:11.708 --> 00:08:12.168
code,

00:08:13.268 --> 00:08:16.668
of course, has an impact on the

00:08:16.708 --> 00:08:20.638
performance of an application. Now,
I'm not sharing any

00:08:20.688 --> 00:08:24.428
groundbreaking truth here or news, but, of

00:08:24.588 --> 00:08:27.768
course,
this is something that's easy to overlook.

00:08:27.988 --> 00:08:31.808
You could say that with AI generating
code, the

00:08:31.948 --> 00:08:35.628
actual code doesn't matter too much
anymore because it's easy

00:08:35.688 --> 00:08:39.368
to refactor, regenerate, replace, and that

00:08:39.468 --> 00:08:42.688
is all true if you ignore the token cost
part

00:08:42.728 --> 00:08:46.408
again. But, of course, the

00:08:46.888 --> 00:08:50.868
code that's being generated has an impact
or could have an

00:08:50.928 --> 00:08:53.008
impact on the performance of your
application.

00:08:53.408 --> 00:08:56.368
There are multiple ways of querying a
database.

00:08:56.668 --> 00:09:00.328
You could run nested queries
or unnecessarily

00:09:00.488 --> 00:09:03.548
many queries because, hey, the result
is the same.

00:09:03.628 --> 00:09:07.148
You got the data you need. Yeah,
but a poorly optimized

00:09:07.188 --> 00:09:10.628
database query, of course,
can show its true

00:09:10.708 --> 00:09:14.528
cost as soon as you do have a significant
amount of users.

00:09:15.008 --> 00:09:18.828
The same, of course,
is true for nested loops

00:09:19.168 --> 00:09:22.648
that can creep into your code base
that leads to worse

00:09:22.688 --> 00:09:26.248
performance, and that is why, of course,
still the code

00:09:26.508 --> 00:09:29.808
quality matters and optimizing the code
matters.

00:09:30.848 --> 00:09:34.678
You can definitely say
that future AI models and

00:09:34.728 --> 00:09:38.528
tools will be very good at that
and will be able to do

00:09:38.628 --> 00:09:42.488
that.
That future models will be able to produce

00:09:42.628 --> 00:09:46.008
right from the start and also evaluate
and then

00:09:46.048 --> 00:09:49.228
improve code, and that
is absolutely possible.

00:09:50.008 --> 00:09:53.908
That still doesn't defeat my point
that code quality matters,

00:09:53.988 --> 00:09:57.788
though,
no matter if a future model generates it

00:09:57.908 --> 00:10:01.657
you as a developer. Now, of course,
what is the big difference is that

00:10:01.768 --> 00:10:05.358
in the, in the world where the AI
is able to do all of that,

00:10:05.748 --> 00:10:09.368
we humans are taken out of the loop,
and I guess that is

00:10:09.508 --> 00:10:13.407
kind of the point Guillermo
is making here, but, of

00:10:13.468 --> 00:10:16.888
course, we don't know how exactly
that future will look like.

00:10:16.948 --> 00:10:18.797
Now, what I will say about this post is,

00:10:19.888 --> 00:10:23.688
and I get it, that
is just how the internet is these days.

00:10:23.728 --> 00:10:27.498
You have to be provocative. You,
you have to kind of be edgy so that people

00:10:28.308 --> 00:10:32.008
like me create a video about it.
I totally get it, but, of course,

00:10:32.268 --> 00:10:35.928
in this post it,
it sounds like all these things we did in

00:10:35.988 --> 00:10:39.908
past were pretty stupid, right?
Building dedicated

00:10:40.068 --> 00:10:43.228
tools,
GUIs that make it easier to write code,

00:10:43.488 --> 00:10:46.648
discuss API design. Oh,
that's pretty stupid.

00:10:47.368 --> 00:10:51.048
Well, no. That is... That shows that you

00:10:51.388 --> 00:10:55.328
care,
and as a human you wanna care about your

00:10:55.368 --> 00:10:58.808
work. That's like saying, oh,
pretty stupid that

00:10:58.848 --> 00:11:02.828
designers used to first sketch on a piece
of paper

00:11:02.848 --> 00:11:06.628
and then use dedicated software like
Photoshop to turn this into

00:11:06.748 --> 00:11:10.008
actual digital art. No, that
is not stupid.

00:11:10.128 --> 00:11:13.928
That is (laughs) exactly what sets a human
that

00:11:14.108 --> 00:11:17.678
cares about their craft
and their job apart from

00:11:17.728 --> 00:11:21.668
someone who doesn't. And framing this as

00:11:21.728 --> 00:11:25.588
kind of stupid or weird or something that

00:11:25.688 --> 00:11:27.088
is gone for good

00:11:27.988 --> 00:11:31.908
is just a bad take. It's just not cool

00:11:31.968 --> 00:11:35.808
in any way. And, of course,
now we're building new software,

00:11:35.908 --> 00:11:39.848
by the way.
We're building new software for the, the,

00:11:39.908 --> 00:11:43.708
agentic engineering world,
something like the Codex app and

00:11:43.808 --> 00:11:47.748
all these AI agents and h- agent apps and

00:11:47.808 --> 00:11:51.678
chewies we have these days.
That's the new kind of software we're

00:11:51.728 --> 00:11:55.348
and we'll keep on doing stuff like
that and we'll, of course,

00:11:55.628 --> 00:11:59.528
invent new ways to discuss about like how
a properly

00:11:59.848 --> 00:12:03.808
written skill for agents should look like
or how

00:12:03.848 --> 00:12:07.248
to set up a good agent's MD file
or whatever.

00:12:07.268 --> 00:12:10.718
We'll keep on discussing until we're

00:12:10.768 --> 00:12:14.608
totally out of the loop and we, as humans,
don't need

00:12:14.648 --> 00:12:18.468
to do anything anymore, which
is a point I don't think

00:12:18.968 --> 00:12:22.088
will arrive and I certainly don't hope
will arrive.

00:12:22.708 --> 00:12:26.648
But until that point,
it's good to have opinions

00:12:26.668 --> 00:12:30.368
about the stuff you work on
and to care about the stuff you work on

00:12:31.328 --> 00:12:33.188
and I think that is really important.

00:12:33.708 --> 00:12:34.568
So yeah,

00:12:35.688 --> 00:12:39.398
I think code, the, the,
the structure of the code,

00:12:39.688 --> 00:12:43.538
that will stay important.
The only question is if

00:12:43.568 --> 00:12:47.508
the AI can do it all on its own
and produce perfect code, and I

00:12:47.548 --> 00:12:50.428
don't know that,
and nobody can know right now.

00:12:50.528 --> 00:12:51.788
You can't look into the future.

00:12:53.088 --> 00:12:56.758
But in general, as a human,
you should care about what you're doing.

00:12:56.888 --> 00:13:00.608
That's, that is what sets the people
that can be super

00:13:00.648 --> 00:13:04.628
successful and have fun at their work
apart from the people

00:13:04.668 --> 00:13:05.908
that don't, I think.