Code quality still matters - Why you should care about & glorify your code!

00:00:00: Today, I woke up to an interesting post by

00:00:03: Rauch, the CEO of Vercel, about which I wanna talk, and also

00:00:07: about the implications and my thoughts on the,

00:00:11: how important the actual code is, today and in

00:00:15: the future. Now, in this post, Guillermo says,

00:00:19: "Code is an output. Nature is healing.

00:00:21: For too long, we treated code as input.

00:00:24: We glorified it, hand formatted it, prettified it, obsessed over it.

00:00:28: We built sophisticated GUIs to write it in, IDEs.

00:00:31: We syntax highlight, treeset, minimap the code, keyboard

00:00:35: triggers, inline autocompletes, ghost text. What color scheme is that?

00:00:40: We stayed up debating the ideal length of APIs and function

00:00:44: bodies. Is this API going to look nice enough for

00:00:47: read? We're now turning our attention to the

00:00:51: requirements, specs, feedback, design inspiration.

00:00:55: Crucially, production inputs. Our coding agents

00:00:59: need to understand how your users are experiencing your application,

00:01:03: what errors you're running into, and turn that into code.

00:01:07: We will inevitably glorify code less, as well as coders.

00:01:12: The best engineers I've worked with always saw code as a means

00:01:16: to an end anyway, an output that's bound to soon be

00:01:19: again." Wow, there's a lot in this post.

00:01:23: Now,

00:01:25: I have a lot of thoughts about this.

00:01:27: Now obviously, and I think we can all see this,

00:01:32: the, the entire role or the entire world of

00:01:36: programming is changing rapidly.

00:01:38: There's lots of stuff going on there and,

00:01:42: especially, I, and I guess many people in the industry

00:01:46: feel that since December, it accelerated again with

00:01:50: Opus 4.5 and Cloud Code and now also

00:01:53: Codex and we got new models and new tools coming up almost every

00:01:58: week. Things are changing quickly.

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

00:02:04: also regarding the content I create, re-regarding the courses I create, and of

00:02:08: course how I work, which is the foundation of that all.

00:02:12: And, uh, that's why I released new courses on Cloud

00:02:15: And I'm, by the way, running a big promotion right now where

00:02:19: my annual membership, which gives you access to all the courses

00:02:23: super low price. But it is having an impact on all of us.

00:02:27: That is my point here. It's changing what we do, which

00:02:31: content I create, and of course how we build software.

00:02:35: However, I'm not so sure if we really move away

00:02:39: from the code. And to me, this post reads like code

00:02:43: doesn't matter anymore, and

00:02:46: I think I disagree with that quite a bit.

00:02:50: Now I will of course admit, and obviously it's true,

00:02:53: that in the past, yes, we did obsess

00:02:57: over API design, over naming

00:03:01: of variables and functions, of the

00:03:05: aesthetics of the code. And I would actually argue

00:03:09: parts of this are still important today, and

00:03:13: very likely also in the future

00:03:17: because even if the role of humans shifts away from

00:03:21: writing the code towards reviewing it,

00:03:24: what is easier to review? A, a messy code base

00:03:28: with ten thousands lines of codes that are

00:03:32: partially redundant or unnecessarily complex with

00:03:36: weirdly named or shaped functions, or a

00:03:40: clean code base? What will be easier to review?

00:03:44: And even if parts or even big parts of that review

00:03:48: process shift towards machines and AI in the future, which of

00:03:52: course is possible, even then I'm not

00:03:55: convinced that the code quality does

00:03:59: not matter at all anymore in that world, and if it's

00:04:03: just that you're paying for all those tokens that are being

00:04:07: generated and reviewed. If something can be built

00:04:11: cleaner, it will very likely consume less

00:04:14: tokens for both generating that code as well as

00:04:18: for reviewing it. And token cost looks like it will

00:04:23: be very, very important in the future because right

00:04:27: subsidized plans by Anthropic, by OpenAI.

00:04:30: We're not paying the real token cost.

00:04:33: The $20 or $200 subscriptions w- you might

00:04:37: have, and I have,

00:04:38: these are not subscriptions where there was vendors necessarily

00:04:42: make any money off. The true token cost is higher than that,

00:04:47: and it's quite likely to go up in the future, or

00:04:51: what we pay to go up in the future when those

00:04:54: companies need to earn money at some point.

00:04:57: Obviously, maybe due to all the technical progress,

00:05:01: start at a point where the token cost as a whole got down a bit, but

00:05:05: we don't know that yet with certainty.

00:05:07: We don't know how that market will look like in the future,

00:05:11: down theoretically, of course, if companies can

00:05:15: charge you a certain price, which may be substantially above

00:05:19: the true price, well, they will happily take that margin.

00:05:23: But yeah, th- that's a totally different topic.

00:05:25: But tokens and token efficiency matters today and very

00:05:29: likely also matters in the future, and that of course is related

00:05:33: to code quality because, in general, good

00:05:37: code, clean code is

00:05:41: code that is not allowed to grow endlessly and

00:05:44: become endlessly complex and complicated.

00:05:48: Now of course,

00:05:50: there is an argument to be made that w-Programmers in the past

00:05:54: also sometimes had the tendency to write

00:05:58: a bit too much code maybe in certain points to come up with a

00:06:02: clean API and maybe some unnecessary abstractions,

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

00:06:09: We've probably all been there where we've worked on a side project

00:06:13: with zero users, and we decided that we want

00:06:17: to implement our database access such that we could easily

00:06:21: swap the database even though we never intended to do that.

00:06:24: And yeah, therefore, we wrote some adapter that was, of

00:06:27: course, way more complex than it needed to be.

00:06:30: We've all been there and wasn't great.

00:06:32: No, it was not, and it will not be today or in the future,

00:06:36: if that's code written by you or by AI.

00:06:38: And right now, and that can of course change, but right now,

00:06:42: in my experience, these AI models and tools like

00:06:46: Codex, Cloud Code, they have a tendency

00:06:50: to overcomplicate things, to

00:06:54: introduce unnecessary complexity, to suggest

00:06:57: changes that, yeah, really don't make a lot

00:07:01: of sense that are unnecessarily complex.

00:07:05: Codex, especially right now, has a tendency to

00:07:09: never remove any code, just add more and more and more code and

00:07:13: add fallback code and legacy handling, and you have to

00:07:16: explicitly almost force it to delete code and to let

00:07:19: go, uh, of old APIs. So that, of course, can

00:07:23: all change, but these are things you have to fight

00:07:27: about, or, uh, for today. And that is, of

00:07:31: course, kind of what this post says doesn't matter

00:07:35: anymore, but I say, yes, it does matter.

00:07:37: You, as a developer, you have to take those fights and you

00:07:41: have to care about the actual code and the code quality

00:07:45: today,

00:07:47: and very likely also in the future.

00:07:49: Obviously, nobody knows what the future holds,

00:07:52: always be important that you have a code base

00:07:56: that doesn't grow unnecessarily complex for many

00:08:00: reasons. Token efficiency being one of them.

00:08:02: Another reason, of course, uh, and also the review part, by the

00:08:06: way. A- another reason, of course, is that

00:08:11: code,

00:08:13: of course, has an impact on the

00:08:16: performance of an application. Now, I'm not sharing any

00:08:20: groundbreaking truth here or news, but, of

00:08:24: course, this is something that's easy to overlook.

00:08:27: You could say that with AI generating code, the

00:08:31: actual code doesn't matter too much anymore because it's easy

00:08:35: to refactor, regenerate, replace, and that

00:08:39: is all true if you ignore the token cost part

00:08:42: again. But, of course, the

00:08:46: code that's being generated has an impact or could have an

00:08:50: impact on the performance of your application.

00:08:53: There are multiple ways of querying a database.

00:08:56: You could run nested queries or unnecessarily

00:09:00: many queries because, hey, the result is the same.

00:09:03: You got the data you need. Yeah, but a poorly optimized

00:09:07: database query, of course, can show its true

00:09:10: cost as soon as you do have a significant amount of users.

00:09:15: The same, of course, is true for nested loops

00:09:19: that can creep into your code base that leads to worse

00:09:22: performance, and that is why, of course, still the code

00:09:26: quality matters and optimizing the code matters.

00:09:30: You can definitely say that future AI models and

00:09:34: tools will be very good at that and will be able to do

00:09:38: that. That future models will be able to produce

00:09:42: right from the start and also evaluate and then

00:09:46: improve code, and that is absolutely possible.

00:09:50: That still doesn't defeat my point that code quality matters,

00:09:53: though, no matter if a future model generates it

00:09:57: you as a developer. Now, of course, what is the big difference is that

00:10:01: in the, in the world where the AI is able to do all of that,

00:10:05: we humans are taken out of the loop, and I guess that is

00:10:09: kind of the point Guillermo is making here, but, of

00:10:13: course, we don't know how exactly that future will look like.

00:10:16: Now, what I will say about this post is,

00:10:19: and I get it, that is just how the internet is these days.

00:10:23: You have to be provocative. You, you have to kind of be edgy so that people

00:10:28: like me create a video about it. I totally get it, but, of course,

00:10:32: in this post it, it sounds like all these things we did in

00:10:35: past were pretty stupid, right? Building dedicated

00:10:40: tools, GUIs that make it easier to write code,

00:10:43: discuss API design. Oh, that's pretty stupid.

00:10:47: Well, no. That is... That shows that you

00:10:51: care, and as a human you wanna care about your

00:10:55: work. That's like saying, oh, pretty stupid that

00:10:58: designers used to first sketch on a piece of paper

00:11:02: and then use dedicated software like Photoshop to turn this into

00:11:06: actual digital art. No, that is not stupid.

00:11:10: That is (laughs) exactly what sets a human that

00:11:14: cares about their craft and their job apart from

00:11:17: someone who doesn't. And framing this as

00:11:21: kind of stupid or weird or something that

00:11:25: is gone for good

00:11:27: is just a bad take. It's just not cool

00:11:31: in any way. And, of course, now we're building new software,

00:11:35: by the way. We're building new software for the, the,

00:11:39: agentic engineering world, something like the Codex app and

00:11:43: all these AI agents and h- agent apps and

00:11:47: chewies we have these days. That's the new kind of software we're

00:11:51: and we'll keep on doing stuff like that and we'll, of course,

00:11:55: invent new ways to discuss about like how a properly

00:11:59: written skill for agents should look like or how

00:12:03: to set up a good agent's MD file or whatever.

00:12:07: We'll keep on discussing until we're

00:12:10: totally out of the loop and we, as humans, don't need

00:12:14: to do anything anymore, which is a point I don't think

00:12:18: will arrive and I certainly don't hope will arrive.

00:12:22: But until that point, it's good to have opinions

00:12:26: about the stuff you work on and to care about the stuff you work on

00:12:31: and I think that is really important.

00:12:33: So yeah,

00:12:35: I think code, the, the, the structure of the code,

00:12:39: that will stay important. The only question is if

00:12:43: the AI can do it all on its own and produce perfect code, and I

00:12:47: don't know that, and nobody can know right now.

00:12:50: You can't look into the future.

00:12:53: But in general, as a human, you should care about what you're doing.

00:12:56: That's, that is what sets the people that can be super

00:13:00: successful and have fun at their work apart from the people

00:13:04: that don't, I think.