The code still needs someone who cares

4:11:43All right, folks, welcome to Syntax.

4:11:45CJ and myself here.

4:11:47I'm Scott Talinsky.

4:11:47This is CJ Reynolds.

4:11:49We're going to be doing all kinds of fun stuff.

4:11:50We are going to be breaking down some of the latest AI news as well as getting into some fun games as to which AI model made this.

4:11:59So if you are out there, you're eating lunch or whatever, you can come hang out or you can just chill and just do whatever.

4:12:05But we'll be joining you online as well.

4:12:09So, CJ, what's up, man?

4:12:12How are you doing in San Francisco?

4:12:12I'm good.

4:12:13Let's introduce ourselves.

4:12:15And anybody that doesn't know who we are, we are Syntax.

4:12:17We're a podcast.

4:12:18We also have a YouTube channel all about web dev, programming, AI.

4:12:22We're both from Denver.

4:12:24We're both from Denver, yes.

4:12:27Yeah, I think we'll...

4:12:28Yeah, so Syntax is more than just a podcast these days.

4:12:30It's also a live stream, so check us out at Syntax on YouTube.

4:12:34We do all kinds of stuff, including a lot of this.

4:12:38So let's break it down here.

4:12:41Let's get into our first little bit, the first bit of news.

4:12:44They're calling it the greatest prompt ever.

4:12:47So this tweet has been going viral here because this is a good prompt.

4:12:53Find all implicit state machines in my app.

4:12:57CJ, why is this a good prompt?

4:13:00Have you used this prompt?

4:13:01I haven't used this prompt, but I've used similar prompts.

4:13:03And I think that there would be a follow-up to this prompt because an implicit state machine is essentially a state machine that you haven't defined.

4:13:12So you have a bunch of variables all over your code, like is loading or is running,

4:13:18and then you have code and if statements all over those.

4:13:21And so this prompt is essentially like, let's take that spaghetti code

4:13:25and turn it into something that is easier to reason about.

4:13:28Yes.

4:13:29Yes.

4:13:30So basically, if you don't know what an implicit state machine or an explicit state machine is,

4:13:35the easiest way to think about this is that an implicit state machine is stuff that you used to see

4:13:41talked about all over React code bases, stuff like is loading, has error, is success.

4:13:47These are often like a Boolean state, right, like this.

4:13:51And the big problem with that is that it can be bug-prone, it can be error-prone,

4:13:57but it can also put your UI into impossible states.

4:14:00So theoretically, if it's just two separate Booleans for these things, you could have

4:14:05a is loading and an is error, both true, and then what, you're showing both of them,

4:14:10or you're showing one of them or the other one.

4:14:12So the idea with this prompt,

4:14:15find all implicit state machines in my app,

4:14:18is that you are able to convert them

4:14:21to explicit state machines.

4:14:23Explicit state machines have a declared status.

4:14:27They have a value rather than a Boolean flag, right?

4:14:32And they have state transitions, right,

4:14:35with centralized rules.

4:14:36So if you've ever used something like Redux,

4:14:39which at a time was verbose and many people didn't love it.

4:14:44For those reasons, it ended up being a great idea for explicit state machines.

4:14:49There's a lot of libraries around explicit state machines like Redux, or not Redux, Xstate.

4:14:57Xstate, yeah.

4:14:57Xstate, yes.

4:14:59And it makes working with AI coding a little bit easier because these impossible states are no longer possible.

4:15:05and then you can also visualize around your code base a bit more.

4:15:11How useful do you find visualizing around your code base?

4:15:16Yeah, I mean, just in general, though, I find this useful,

4:15:20and it's actually an argument against a lot of people that are basically saying,

4:15:23don't worry about the code, let the AI make all the decisions.

4:15:27Yes.

4:15:28But I have found that a cleaner code base makes it easier for the AI to work in as well.

4:15:34And so if you give it a prompt like this, it's going to make it harder to have a weird UI state like you were mentioning.

4:15:40One prompt that I've been doing recently is find all duplication and opportunities for abstraction.

4:15:47Yes.

4:15:48And that's another thing that for a long time people were saying, oh, copy-paste is better than abstraction.

4:15:52But when you have a massive sprawling code base and AI is having to work in all of that,

4:15:57having a single source of truth or one way to do things in your code base also makes it a lot easier for the AI.

4:16:03Yeah. No, the biggest problem with that, not that that's a problem, the biggest problem with that is getting the AI to adhere to your abstractions, right?

4:16:12Because otherwise you'll end up with copies of abstractions or, yes.

4:16:15Exactly. And personally, I've found like lint rules to be really good for that.

4:16:19So like one example is a CLI tool that I'm building that does a lot of file read and writes.

4:16:24So it's writing files all over your system. It's saving them.

4:16:27And before I refactored, it had like 10 different places with 10 different ways of writing the file.

4:16:31but then I brought it all into a single file manager,

4:16:35and any part of my code that wants to write a file has to use the file manager.

4:16:38So there's a lint rule that says no bare FS read, no bare FS write,

4:16:43and that basically forces the AI to use that abstraction.

4:16:46Totally, yeah.

4:16:47So I like this prompt, and I like this type of thing.

4:16:51So finding your implicit state machines and changing them to be explicit state machines

4:16:56and really getting a handle on the state of your application,

4:16:59in my mind is something that does decrease the slop level of a code base.

4:17:04Definitely.

4:17:04And this is the AI engineer conference.

4:17:08So, like, we still need engineering.

4:17:10It's not just prompts.

4:17:11Yes, yes, yes.

4:17:12It is all prompts, CJ.

4:17:14All right.

4:17:14The next thing that we brought here is unfair play.

4:17:18So Fair Play is basically how an iOS app can be encrypted so that way it cannot be decompiled.

4:17:32So Unfair Play decrypts an IPA and it allows you to recover the actual application that was built for a macOS app.

4:17:45And so this project's really neat because what it's doing is it's kind of like the ROM hacking scene of video games where you aren't supplying a counterfeit app to somebody or an app that is – you're not hosting the YouTube app.

4:18:03But potentially what you could be doing is you could take your own YouTube app that you've downloaded yourself.

4:18:10You can unfair play it.

4:18:12You can get the IPA.

4:18:14You can modify the application to your heart's content and then reinstall it on your device.

4:18:21How does that sound to you, CJ?

4:18:23Sounds amazing.

4:18:24So, I mean, I'm an Android person.

4:18:25If you listen to the Syntax podcast, you know that.

4:18:27Yeah.

4:18:28And we've been able to do a similar thing.

4:18:29So there is this project called Revance in the world of Android.

4:18:34that is very similar.

4:18:36So, I mean, it's very common to take a Spotify app,

4:18:38take a YouTube app, take an Instagram app,

4:18:40and remove the parts that show you ads or something like

4:18:43that. We're not condoning that necessarily, but that's something.

4:18:45No, remove those ads, yeah.

4:18:47No, and I think what I saw with UnfairPlay

4:18:49is that people were taking YouTube

4:18:51and then removing all YouTube shorts from it,

4:18:54which honestly, that's not a bad idea, yeah.

4:18:57I mean, imagine modifying Instagram to give you a timeout.

4:19:00There's probably some built-in rules that are like,

4:19:02don't scroll for too long.

4:19:03But they don't always work.

4:19:04You could enforce it.

4:19:06You could make it so that it only shows you feeds for a certain amount of time,

4:19:10and then you have to put the app down or something like that.

4:19:12Or just hiding stuff you don't need.

4:19:14There's so many functionalities in apps, especially like YouTube now,

4:19:18that they did put shorts in the main bar,

4:19:20and they removed subscriptions from the main navigation.

4:19:24So, hey, with UnfairPlay on iOS, you could just swap that back.

4:19:29Take back your apps.

4:19:30Take back your apps.

4:19:31We're taking back our hardware.

4:19:32We're taking back our apps.

4:19:34Yeah, and this is super relevant with the reverse engineer everything repo that came up.

4:19:40This is the good part of AI.

4:19:43Yes.

4:19:44Stuff like this.

4:19:45The ownership and the control, all that good stuff.

4:19:48I'm here for it.

4:19:49Yes, me too.

4:19:50Okay.

4:19:52Let's keep going on some news.

4:19:54How do you feel about that?

4:19:55Let's do it, yeah.

4:19:56We have some coding games lined up for you too, but I think people are eating right

4:20:00now, so maybe not the best time to introduce those.

4:20:02Okay, so let's do the rise of coding languages for agents.

4:20:08So very recently, Kit from the Anomaly, which is the Open Code team, has been talking about his new language, Blood.

4:20:20and he hasn't shown it really beyond just a couple of glimpses of even what he had described as outdated syntax.

4:20:29But blood, as well as many of these other AI languages,

4:20:33have kind of turned into this conversation of our AI languages, AI first languages,

4:20:40really something that is worth exploring, worth investigating, or whatever.

4:20:46And so we've talked about a few of these on syntax before.

4:20:49But now with Blood, there's also BendLang at bend-lang.com, as well as ZeroLang from the Vercel team.

4:20:59So I will say that those are just three that I got off the top of my head.

4:21:05I've probably seen like 15 of these bad boys.

4:21:07So before it was everybody is creating a Safari fork or a browser fork or a Chrome fork.

4:21:14Then everybody is creating a front-end language or whatever.

4:21:17Now we got vibe-coded languages, and you better believe, CJ, all of them are faster than Rust.

4:21:24They're all fast.

4:21:25They're all very fast.

4:21:26That's their benchmarks.

4:21:27Yeah, I mean, I think it's super interesting.

4:21:29I mean, my initial reaction to this is AI was trained on existing programming languages,

4:21:35so why would it be better at a new programming language?

4:21:38But if you look at some of their tenants and, like, how they're actually making it useful for AI,

4:21:42it really comes down to giving it proofs and preventing it from making mistakes.

4:21:49Almost similar to Elm Lang.

4:21:51There's various programming languages that make it really hard to get into error states

4:21:54or have memory leaks and stuff like that.

4:21:58Elm is low-key getting a lot of attention right now because of all kinds of things.

4:22:05And in fact, effect and fold kit, which is the UI framework in effect, is an Elm style framework.

4:22:14And so I see Elm as one of these things that seems to be talked about quite a bit more in this space now.

4:22:22Well, it's interesting that you bring up effect because I don't think of effect as a programming language.

4:22:26It's more of a runtime.

4:22:27It is.

4:22:27But it gives you some of these same guarantees that these languages do, right?

4:22:30It makes it so that you have explicit error states, and errors have to pass through in a certain way.

4:22:38And again, it just makes AI more predictable when it's writing in that code base.

4:22:41Yeah, yeah.

4:22:42And along with effect, there is a front-end framework called FoldKit that is built on effect.

4:22:48And the developer behind FoldKit just announced that he's going to be working on this full-time now.

4:22:54This thing absolutely rips for a lot of reasons.

4:22:57You have one state tree.

4:22:59It's all effect-based.

4:23:00Everything's a schema.

4:23:01Agents are good at this framework.

4:23:03And their premise isn't, you know, with a lot of front-end frameworks, JavaScript frameworks, it was always this is faster, this is faster than this or whatever.

4:23:12This is correct, and the AI is good at writing it.

4:23:17And I actually saw some alternative renderers for this.

4:23:20Someone was getting GPUI output from a FoldKit app.

4:23:24Wow.

4:23:24So going straight native on it.

4:23:26I think that's huge.

4:23:28It's so funny.

4:23:28I mean, times are changing because I would see something like FoldKit and think about it, okay, would I use this?

4:23:34And I mean, that's one of the reasons I haven't picked up effect because it's a little obtuse.

4:23:38It takes a while for me to figure out what is this thing actually doing.

4:23:41But if it's for the agent, then it doesn't matter as much as long as it makes the app correct and better.

4:23:49Give the agent a little treat.

4:23:50Yeah.

4:23:51Yeah, just a little treat for the agent.

4:23:53Let them hack a little computer competitor.

4:23:55Who knows, right?

4:23:56Right. The full kit, I think, is an interesting thing, because one of the things that we've talked about in the past, and this was like in the past, like what, like two months ago.

4:24:06So it was ancient history was that, oh, I don't know if these AI languages will ever catch on because what are the models going to be able to write them?

4:24:14But the thing that you get with FoldKit in effect is that you dump an example repo into your code base, and you just tell the agent to consult the examples and consult the code base, and the agent's got no problem with it.

4:24:31It's got no problem.

4:24:32So in my experience, I don't think that matters anymore.

4:24:37At one point in time, you had to really worry about, was it trained on this specific framework?

4:24:44And as somebody who wrote SvelteKit for a long time, the big problem with that was that it wasn't in the training data.

4:24:50So therefore, you had to then incorporate all kinds of deterministic checks and skills and MCP servers to validate your code base.

4:24:59And you still do to an extent.

4:25:00Right.

4:25:01Yeah.

4:25:01But now you can just dump all this stuff in context, and it can write it.

4:25:06The context is big enough, yeah.

4:25:08I mean, one thing I checked is foldkit.dev slash llms.txt.

4:25:11It exists.

4:25:12So you pop that into an agent, and it finds all the pages that give examples.

4:25:17And like you were saying, the context is big enough now that it can have those docs in context and do the right thing instead of, like, hallucinating tag names or using Svelte 3 instead of Svelte 5 or something like that.

4:25:31Yeah.

4:25:31And so with FoldKit, they even have an MCP to connect directly to their dev tools, which is sick because the agent can then use the dev tools in an agent-focused way, right?

4:25:42Yeah.

4:25:43And I think that's a really great thing that has changed my mind on agentic-first programming languages.

4:25:50Because at first I would have said, yeah, I don't know if they'll hang in there.

4:25:56But seeing how easy it was to develop something on this, which is clearly not any of the training data or whatever, it has really changed my mind on this.

4:26:05That said, just like with everything else vibe-coded these days, there is a battle for attention.

4:26:12Right.

4:26:12Right?

4:26:12It's all about who's using it, how many people are using it, rather than – I mean, maybe someday we'll all just have our own programming languages that we like.

4:26:21Yeah.

4:26:22because like Kit said in regards to blood-laying,

4:26:25that the syntax can be whatever you want it to be,

4:26:30which he hasn't released too much.

4:26:31What does that mean?

4:26:32Okay, so he's just talking about it in his tweets

4:26:33because I haven't seen this, so yeah.

4:26:35So the blood-laying thing,

4:26:37he hasn't really announced too much of it.

4:26:39But again, I think with any of the things

4:26:41that we're getting into,

4:26:42agentic-first dev tools are going to be really helpful,

4:26:45but I also think it's very possible

4:26:47that the best language for agents

4:26:50may not be TypeScript Rust or whatever.

4:26:53Right.

4:26:53Possibly.

4:26:56Which isn't that bad for me.

4:26:58No.

4:26:58I mean, I also am lamenting the death of having a favorite language or framework.

4:27:05Yes.

4:27:06Languages are like little mascots.

4:27:08We like them.

4:27:08Yeah.

4:27:08I mean, it was our identity as developers.

4:27:12So, yeah, now it's the agent.

4:27:15The agent should have a favorite.

4:27:16And it's not even a favorite.

4:27:17If the agent works well in that language or framework, that's probably the one that you should use.

4:27:21I mean, it's one of the reasons I started using React again is because the agents are really good at React.

4:27:26Yes.

4:27:27Yes.

4:27:28And you know what?

4:27:28I will say I did start a project with FoldKit, and I kind of really liked its output.

4:27:34So we will see.

4:27:36Yeah, for sure.

4:27:37You heard it here first.

4:27:39No more Svelte for Scott.

4:27:41Yes.

4:27:41Well, no.

4:27:42That's not necessarily true.

4:27:44That's the Svelte.

4:27:44But I was having problems specifically around state in a large application.

4:27:49And FoldKit basically, like we were talking about with implicit state machines,

4:27:56FoldKit makes a lot of that less possible.

4:27:59Right.

4:27:59So because of that, you end up getting a lot of those benefits for free just based out of the framework.

4:28:05Okay.

4:28:05Let's talk about another programming language, which is Rust.

4:28:12Oh.

4:28:12Okay, we mentioned Rust.

4:28:14Rust being great to write is awesome.

4:28:16And Discord just announced that they were moving Discord themselves to Rust and using GPUI to do it.

4:28:26Incredible.

4:28:27Yeah, so there's a tweet from Stanislav.

4:28:30He works at Discord.

4:28:31Yes.

4:28:31Yeah.

4:28:32And he has a screenshot of his, like, Windows task manager.

4:28:37And Discord Phoenix is the codename, only using 290 megabytes of RAM.

4:28:43And so, I mean, this is, we talked about it on the podcast before, like, this is the age of everything getting faster and being rewritten with Rust.

4:28:50So I'm here for it.

4:28:52I mean, I know you've used GPUI quite a bit, but I think, like, as long as the app works, all of a sudden I've got a gig extra RAM to use, I'm here for it.

4:29:01Yes, folks.

4:29:02So let this be a thing.

4:29:04You just heard that Discord is moving to GPUI.

4:29:07I've been talking about GPUI for months, for many months now.

4:29:12And you heard me talking about FoldKit earlier in about six months when other people start to catch on.

4:29:18I think FoldKit's going to be getting some major press here.

4:29:21Just listen to the pod.

4:29:22Listen to Scott.

4:29:23You'll know what's coming next, basically.

4:29:25Syntax.fm.

4:29:26We're live every single Monday, folks.

4:29:28Check us out.

4:29:29So how do you feel about it?

4:29:31This is clearly an agentic decision, right?

4:29:33Right. Discord is one of those apps that opens every single or like re reinstalls itself every single time you open it.

4:29:40Right. There's always an update coming. So it's also an app that I've never thought was non-performant.

4:29:46I subscribe to maybe like 40, 50 Discord channels. Yeah.

4:29:52And I've always been surprised at how well it handles that.

4:29:56That said, the fact that they were

4:30:00able to rewrite the entire application, Discord, no small application, in GPUI, means that, like, man, we can optimize now easily for performance.

4:30:11Definitely. And I mean, you think about, like, why did it take so long to do this? Like, one of the reasons they were using Electron is because it's cross-platform.

4:30:19Yes.

4:30:19Right? They write one app. It works on Windows, Mac, and Linux. They have a massive user base.

4:30:23Distributed anywhere.

4:30:24Yeah, and so Electron is great for that.

4:30:27And, I mean, you've talked about some of the woes of Tari.

4:30:30Yes.

4:30:30Like, it's similar to Electron, but it's going to be a little bit faster, and it ships with the system WebView.

4:30:34But you run into a lot of potential compatibility issues with, like, WebKit on Mac OS versus, like, Chrome on Linux.

4:30:42And with Electron, you don't get that.

4:30:44So I understand why it was Electron for so long.

4:30:46Oh, yeah.

4:30:46And if you were talking Electron versus anything else, Electron still wins for a number.

4:30:50I mean, you can choose your exact Chrome version with flags in Electron.

4:30:55Yes, and then every user gets that.

4:30:57But that's what you also get with GPUI, right?

4:30:59You have one code base with GPUI, and then you get a cross-platform app that is native.

4:31:04It uses native UI elements on every platform.

4:31:08So now's the time to do it.

4:31:10Like we saw Bun was rewritten in Rust.

4:31:12Yes.

4:31:13And we talked about it on the pod a few months back.

4:31:16but Jared had an entire agentic flow of like take the existing code base and tests

4:31:22and then gradually port them, gradually get them passing.

4:31:26I'm not sure if they could do a similar thing with an Electron app in its test

4:31:30because it's very like browser focused.

4:31:32But at the same time, agents have gotten really good at like taking a large code base

4:31:37and making sense of it in some other language or porting it to another.

4:31:41Yeah, I do wonder what that – how that portability worked with testing

4:31:45and how they're testing infrastructure.

4:31:47Like, did they convert the test?

4:31:49Yeah, I really am curious about that.

4:31:51That's really interesting.

4:31:51And on top of this, Anomaly also released Rune.

4:31:57Have you seen Rune?

4:31:58I haven't seen Rune.

4:31:58What is this?

4:31:59R-U-N-E, as you typically spell Rune.

4:32:02Rune is their GPUI competitor.

4:32:05Oh, yeah.

4:32:06It is a Rust-based GPU framework.

4:32:09And the idea here is, again, that it's going to look nice.

4:32:13and I haven't dove into it beyond just seeing that it was released

4:32:17and that it was a different Rust-based framework for GPU UI work.

4:32:23Okay, so they're going to rebuild, like, OpenCode Desktop,

4:32:27or I guess they have built OpenCode Desktop with this,

4:32:29and now they're releasing it as open source for other people to use.

4:32:32Yes.

4:32:33That app that we talked about last week, oh, man, Hex.

4:32:37Hex, yeah.

4:32:37I got Dex, I got Rex, I got Hex, I got all these different apps on my thing,

4:32:41But that was built in Rune, and they were touting some of the abilities to have it look nice and fit in and have good UI across all platforms.

4:32:52Right.

4:32:53And make it easy to do so.

4:32:55So this is another one to keep your eye on if you are wondering about making cross-platform apps performant with Rust.

4:33:03But obviously, you get into the web space with some of these, and then you're what?

4:33:08but you're rendering on a canvas and you're not using DOM elements at that point.

4:33:14So it's not, you know, there's all these different write-once, deploy-anywhere solutions,

4:33:20whether it's like React Native translating it and then React Native Web translating it back to web

4:33:25or any of that, or taking web and rendering it as a web view in a phone or whatever.

4:33:29And there's Flutter is another one.

4:33:30There's Flutter.

4:33:31The Flutter folks are very, they will let us know if we forget Flutter again

4:33:36because I think there's too many times we've forgotten Flutter.

4:33:38But what there hasn't necessarily been is truly the right once deploy anywhere and have that actually be native everywhere in a way that's effortless.

4:33:50I think the closest that we've gotten with this besides Flutter is like the React Native, React Native web world.

4:33:56But React Native is a tough platform, and we're seeing people move off of it all the time, right?

4:34:02Yeah.

4:34:03I mean, again, I never thought I'd see the day, but it's like, again, the reason Electron was so popular is like web devs already knew it and also made it super simple to be cross-platform.

4:34:14But now if the agent's writing the code and it's performant, why not?

4:34:19Why not native?

4:34:20Why not?

4:34:20Yeah.

4:34:21Why not?

4:34:21Yeah, I know.

4:34:22I totally agree with that.

4:34:23I think there was actually there was another one that I just saw somebody talk about that was a right once to play everywhere.

4:34:29I'm trying to pull it up, but I cannot.

4:34:30It just was announced like 12 hours ago.

4:34:34Is it Rust-based?

4:34:35Oh, CJ, that's a good question.

4:34:37I don't know.

4:34:38There's too many things.

4:34:39There's too many dang things released all the time.

4:34:41Let it sit there.

4:34:42Maybe it'll come to you.

4:34:42Yes, yeah, absolutely.

4:34:44Okay, cool.

4:34:46Let's get into some questions.

4:34:49We only get a ton of questions over here at Syntax.

4:34:51So I want to dive into some questions that are coming in across the web development and the development ecosystem overall.

4:34:58So we had a number of these questions come in.

4:35:00First one that we had come in was, how are y'all feeling about brain rot in big tech?

4:35:06Do we think software engineers at big companies are losing their skills without knowing it?

4:35:11And I kind of talked about this in my talk today.

4:35:14I did a talk today.

4:35:15We're at AI Engineer.

4:35:18Obviously, people are losing their skills.

4:35:20In fact, in my survey that I did, I think it was like 67% of people listed that they feel like their skills are diminishing since using AI.

4:35:29But I think the bigger question isn't are our skills in coding diminishing?

4:35:35The bigger question is does that matter?

4:35:39Yes.

4:35:40Right.

4:35:40I mean, again, I'm like biting my tongue on stuff I've said in the past.

4:35:44Like our skill sets are shifting.

4:35:47Yes, they're shifting.

4:35:47They really are.

4:35:48And you talked about lamenting – or you maybe didn't even think about it that much, but if you've been in this space long enough, things change and things move fast, especially if you're in the web dev space.

4:35:58New APIs, new browsers, browsers get deprecated.

4:36:01And so if you ever had to implement something for Internet Explorer 6, there is no reason for you to know how to do that anymore.

4:36:08It is no longer a thing.

4:36:10And so it's gradually becoming, well, what should our expertise be now?

4:36:15Yes.

4:36:15What, like, if it's not programming, then are we now orchestrators?

4:36:22Are we now, like, tech leads?

4:36:24And so I do see us moving more and more towards that.

4:36:28At the same time, like, in order to be a really good architect or tech lead,

4:36:33you kind of need to know how things work under the hood to be able to come up with good decisions

4:36:37and have good guidance.

4:36:38Yeah, the agents will still do whatever the hell they want.

4:36:41Right.

4:36:42I mean, and I found that with the tool that I'm working on right now,

4:36:44like it does whatever it wants, but then I, I reign it back in. I'm like, Hey,

4:36:48we need to be a little bit more performant over here,

4:36:50or we need to share code a little bit more.

4:36:52And I know that because I myself have worked in large code bases.

4:36:57So like those kinds of skills that definitely translate, um,

4:37:00in terms of where the brain rot is happening.

4:37:02Like I think it's at multiple levels though,

4:37:03because a lot of it comes from management.

4:37:05Like management is like use AI move fast.

4:37:08And then developers are having to stay caught up and then they're burning out.

4:37:12And so now they can't even worry about do they still know how to code.

4:37:18So it's happening, but it's all, I think, part of this big shift of, like, where are we going?

4:37:23Where are we headed?

4:37:24Yes.

4:37:25Yeah.

4:37:25Yeah.

4:37:26Okay.

4:37:26So a thing I will say, and I said this during my talk, so I will say there are things that I used to do in web development.

4:37:34I used to know every browser hack for Internet Explorer 5, Internet Explorer 6, and I haven't had to use it.

4:37:44I used to be able to write Angular components by hand.

4:37:48Yeah, me too.

4:37:49I haven't had to do this in so long that my brain has atrophied in those skills.

4:37:56So, like, does that matter anymore?

4:37:58It doesn't matter.

4:37:59I don't have to produce a website for Internet Explorer 5 ever again.

4:38:03Right.

4:38:03So those skills are gone.

4:38:05And some of those things you can lament and you can mourn the loss of those skills.

4:38:10But at the end of the day, things have shifted.

4:38:12Right.

4:38:13And so, yeah, they've shifted.

4:38:15And I think it becomes what should be important.

4:38:18Like, I absolutely don't think it should be we are just prompters, just prompt, prompt, prompt.

4:38:23There has to be some thought behind that prompt.

4:38:25Yes.

4:38:25And that's where the skills are shifting to.

4:38:27And so I do agree that there is actually brain rot in that gray area.

4:38:31Oh, yeah.

4:38:31Like, people are like, all I do is prompt now.

4:38:33I don't actually have to worry about.

4:38:34I think you still need – if you want to have a good product, if you want to have a good team, you should still care about those things.

4:38:41Well, yeah, yeah, because at the end of the day, it is looking for a simple solution at the end of the day.

4:38:46Because what we saw, especially with the rise of Vibe coding, was that people were just hucking all kinds of stuff, God knows what, into their browser.

4:38:56And you say, oh, who cares?

4:38:57The app works fine.

4:38:58It gets a decent lighthouse scores.

4:38:59and then you look at it, and the entire DOM is duplicated twice,

4:39:02once for desktop and once for mobile, and you're just like,

4:39:05what are we doing here?

4:39:06You know, that stuff still matters for sure.

4:39:07It still matters.

4:39:08Yeah.

4:39:08Yeah.

4:39:10Cool.

4:39:10All right, next question here.

4:39:14Just could you say a few words about what vibe coding actually means in 2026?

4:39:20As a middle-level developer, I see most, if not all, others using AI,

4:39:25as it's impossible to compete without it.

4:39:28But for many, vibe-coded is now a slur.

4:39:31Wow.

4:39:32Yeah.

4:39:33Sometimes unfairly thrown at handwritten projects, projects with AI by experienced devs and fairly non-coders and with expensive models.

4:39:41So vibe-coding, is it a bad term?

4:39:45Is it a loaded term?

4:39:46What does it mean in 2026?

4:39:47Is it a dead term?

4:39:49What's up with that?

4:39:49I mean, my initial gut reaction to the term was, like, it's a very specific kind of AI coding where you don't care about the code.

4:39:58It's all about the vibes.

4:40:00But the coding that I've been doing lately, I would call it vibe coding because AI is writing all of the code.

4:40:07But again, like I talked about earlier, I'm still making all of the decisions. I'm still orchestrating it. I'm still engineering it. AI engineering.

4:40:15Yeah, yeah, yeah.

4:40:16Yeah, yeah.

4:40:17I am still, like, you still have to, so yeah, I don't, I wouldn't call what I'm doing with the tool that I'm building Vibe Coding.

4:40:23I would call it AI-assisted engineering.

4:40:26But I do see, like, Vibe Coding being thrown around as a slur.

4:40:29Like, you can just, you, somebody posts their app, you're like, oh, that was Vibe Coded.

4:40:32Yeah.

4:40:32Don't care about it.

4:40:34So it's tricky because it's a tough line to walk, I would say.

4:40:38Yeah, it is funny because it does seem like there's a group of people with, if anything, had AI had a hand in it at any point, it's labeled as vibe-coded slop or whatever, toss it in the bin.

4:40:51Which I think anybody who's been using any of the models in a modern context, like modern isn't like the past two weeks.

4:41:00Yeah, who knows in this industry anymore, right?

4:41:02Anybody who's been using models in that regard knows that, hey, things have shifted.

4:41:06and the models have gotten better.

4:41:09So I don't know.

4:41:12Yeah, to me, like there has,

4:41:14like the language doesn't like kind of bugs me a little bit

4:41:17that people get so hung up on the terms or whatever,

4:41:20but I understand why they exist.

4:41:21What I don't get is like the insistence of throwing everything out

4:41:25just because AI is used.

4:41:28That said, there are some things when you have clear tells of AI that are bad,

4:41:36that show a lack of care, show a lack of taste, show a lack of control over what you're doing.

4:41:42And those types of things I think stand out in a way that really turns people off instantly.

4:41:48I would not tell you how many YouTube videos I turn on these days,

4:41:52and I get like two minutes into it, and I'm like, oh, this is a Claude script.

4:41:56Toss this. I'm done with this.

4:41:58They're reading from Claude. Their animations were generated from Claude.

4:42:01But then there was no follow-up, right?

4:42:02There was no editing.

4:42:04There was no tweaking, making it their own.

4:42:06And the script from Claude really kind of bugs me, too.

4:42:08Not that Claude's the problem.

4:42:09The script from AI bugs me.

4:42:11Because when you can see somebody reading a script that they didn't have any hand in creating,

4:42:16it's just, it feels like, why am I bothering to care about this if you didn't care about this?

4:42:21Right.

4:42:21Yeah.

4:42:23Definitely.

4:42:24Yes.

4:42:24Okay.

4:42:25So we're going to play a quick little game here.

4:42:27Let's do a game, CJ.

4:42:28Okay, I'm ready.

4:42:29Okay.

4:42:30Folks, we're going to be doing a game.

4:42:32So if anybody walks by they want to play in this game, please join in.

4:42:35If not, it's just going to be CJ.

4:42:36That's all good.

4:42:37We're going to be playing Guess Which AI Model Made This.

4:42:40You can join at any given time.

4:42:42You can join in the middle of the game.

4:42:44Let's see if you can do it.

4:42:46We got a few of these up here.

4:42:48So CJ, are you ready?

4:42:51Let's do it.

4:42:52Okay, audience, if you are out there, you all want to play, scan this QR code.

4:42:57You can join at any given time.

4:42:59We're going to open this up with the very first one here.

4:43:02Oh, brother, I lost my Wi-Fi.

4:43:04Okay, let's open this bad boy up.

4:43:06Oh, yeah, I see it.

4:43:06Okay.

4:43:07So it's slowly loading.

4:43:10Which button is Opus 5.5?

4:43:13All right, which of these buttons did Opus 5.5 make?

4:43:16Okay, so option A, I see, like, the green gradient.

4:43:20Decent font choices.

4:43:22And then option B has a bunch of extra stuff on top and bottom.

4:43:25I'm going to go with A.

4:43:26Opus did A.

4:43:27You think Opus did A?

4:43:30Okay.

4:43:30Yes, yes.

4:43:32Really?

4:43:32You're wrong.

4:43:33Opus did B.

4:43:34And the A was actually Grok 4.6.

4:43:38Okay, Grok.

4:43:39It's got some interesting font choices.

4:43:41Yeah.

4:43:42All right.

4:43:43You asked him for a button, and it gave you three, like, taglines below it.

4:43:47Yes.

4:43:47Like, that's – I don't have that experience with Opus.

4:43:49And this one is kind of similar.

4:43:50So which of these is Sonnet 4.5?

4:43:54Okay, Sonnet 4.5.

4:43:54This is a year old, a year ago.

4:43:57Yes.

4:43:58Again, the prompt was asking for a button.

4:44:02I mean, again, I'm going to go with B because I think wasn't Sonnet that decent back then?

4:44:07Sonnet 4-5.

4:44:08Yeah.

4:44:08Okay.

4:44:09Let's see, CJ.

4:44:10You're going with B.

4:44:11B.

4:44:11Okay.

4:44:14I am horrible.

4:44:15Oh, no.

4:44:16I am now I'm horrible at this.

4:44:17No.

4:44:17Sonnet 4-5 was, in fact, the one that – and the other one was Gemini 3-1, which did a pretty decent job.

4:44:24I'm just learning that I like other models, I guess.

4:44:29Yes, yes, yes.

4:44:31Folks, you can join at any time if you're walking by.

4:44:34We're playing Guess Which Model Made This.

4:44:37So the prompt for this one was also a button.

4:44:40We asked this to make a button.

4:44:41Which of these did Muse Wonder 1.0 make?

4:44:44I haven't used Muse Wonder at all.

4:44:47It's got the name Wonder in it.

4:44:48I'm going to go with B.

4:44:49Yes.

4:44:50It looks wondrous.

4:44:52Is that a word?

4:44:53Two models also kind of failed the prompt of a single button.

4:44:57Yes, we need a button here.

4:44:59one of them has this awful overflow issue.

4:45:02Oh, yeah.

4:45:03Oh, yeah.

4:45:04Yes.

4:45:05Muse.

4:45:05Nice.

4:45:06Muse decided to make a whole-ass card for this instead of a button.

4:45:11All right.

4:45:11Next one here.

4:45:14Which one of these flyers did Opus 4-5 make?

4:45:184-5.

4:45:19Again, so not 5-5, 4-5, which is like 8 or 9 months old, 10 months old at this point.

4:45:25This is 4-5.

4:45:26Yes.

4:45:28I mean, I kind of associate the really decent backgrounds and font choices with the newer ones.

4:45:33So I'm going to say, but there's a lot of emojis.

4:45:37There's emojis in both of them.

4:45:41Four, five.

4:45:42There is emoji.

4:45:42Yeah.

4:45:43This is hard.

4:45:44Okay, I'm going to go B.

4:45:46Oh, four, five did B.

4:45:46You're going to say four, five is B.

4:45:48Okay.

4:45:49Four, five.

4:45:50Months ago, I don't remember.

4:45:51Is B, yes.

4:45:51Okay, sweet.

4:45:52The A was, of course, five, five, so quite a bit better.

4:45:56Yeah, yeah.

4:45:56The emojis are certainly a tell of an era, right?

4:45:59They're very a tell of an era.

4:46:01Folks, you can join at any time if you want to play.

4:46:03We got the next one here.

4:46:05Which one of these flyers did Grok 4.6 do?

4:46:09Okay.

4:46:09We learned a little bit about Grok earlier.

4:46:11Are they still loading for you?

4:46:12They loaded.

4:46:13Okay.

4:46:14There we go.

4:46:15Again, so this one has emojis.

4:46:17I want to think like Grok sets themselves apart so they don't use emojis.

4:46:21I'm going to go with A.

4:46:22Grok 4.6 is A.

4:46:23You're going to say it's A.

4:46:24Yeah.

4:46:25Okay.

4:46:25Which one of these?

4:46:26did Grok 4.6 make?

4:46:31And it was a, yes.

4:46:32I got three right.

4:46:33Yes, the other one was Sonnet 4.5,

4:46:35which you can tell from two and two here,

4:46:37Sonnet would be loving those emojis for sure.

4:46:39Those emojis, yeah.

4:46:40It's a tell at this point.

4:46:41It's those emojis, yes.

4:46:43All right.

4:46:44The next one here is these two flyers.

4:46:47Which one of these did Gemini 3.1 do?

4:46:51Gemini 3.1.

4:46:53That's the latest one.

4:46:55That's modern Gemini.

4:46:56Like, I would hope it was B because A looks very basic.

4:47:02I'm going to go with B.

4:47:03So CJ is saying B.

4:47:05The votes are coming in.

4:47:08And, yeah, A's got a bunch of emojis.

4:47:10Which one did Gemini 3-1 do?

4:47:12Does Gemini use emojis?

4:47:13I don't know.

4:47:13I don't use Gemini.

4:47:14Oh, my God.

4:47:15Apparently, Gemini uses all these emojis.

4:47:17Yeah, sadly.

4:47:18Get your stuff together, Gemini.

4:47:19This is a frontier model.

4:47:20Yeah, yeah.

4:47:21Well, yeah, Muse Wonder actually did a little bit better of a job than the last one we saw here.

4:47:26Okay, yeah.

4:47:27Okay, next up we got some inboxes.

4:47:30Ooh.

4:47:30So these ones are a little bit tighter.

4:47:32They're a little bit harder to see.

4:47:33Which one of these did Gemini 3 one did?

4:47:36It's got to be A.

4:47:37Like B has either ChatGPT or Claude.

4:47:41It feels like ChatGPT-ism is on the right, like the font choices.

4:47:45So I'm going to go with A.

4:47:46Gemini did A.

4:47:47Okay.

4:47:47You think Gemini did A and Gemini did A.

4:47:53Grok 4-6 did B, yes.

4:47:55And that'd be looking like last year's Opus or something for sure.

4:47:58Yeah, yeah,

4:48:00You could have told me that was Opus 4 or 5 or something.

4:48:02I would believe you.

4:48:04All right, next one.

4:48:05Which one of these is Opus 5-5?

4:48:09A or B?

4:48:10This is modern Opus, but again, both of these have tons of emojis.

4:48:16Both of them have emojis, yes.

4:48:18We thought the emojis were gone.

4:48:20No.

4:48:22I want to say A.

4:48:26No, I'm going to go B.

4:48:27Opus did B.

4:48:28Okay, CJ says that Opus did B.

4:48:31Just a gut feeling.

4:48:32I don't know.

4:48:32It's a gut feeling.

4:48:33Which one is which?

4:48:35And actually, what do you think option A is?

4:48:41Some model I've never used.

4:48:43I'll say that.

4:48:44Looks like you're right on both.

4:48:46So Muse Wonder did option A,

4:48:48and Opus 5.5 did option B, so you were right on that one.

4:48:52Cool.

4:48:53Okay, let's see.

4:48:54We got one more here for you folks.

4:48:57Which one of these was Sonnet 4.5?

4:49:00Again, Sonnet 4.5 is 10 months ago at this point.

4:49:06This is a toss-up because they're both pretty bland.

4:49:08There's emojis on the right.

4:49:10I'm going to go with, like, AI was using a ton of emojis like a year ago.

4:49:14I'm going to go with B.

4:49:15Okay, you're going to say B.

4:49:16What do you think option A is?

4:49:19I may hold you to that.

4:49:20Gemini.

4:49:21Gemini, okay.

4:49:25So option B was Sonnet 4.5, and option A was Opus 4.5.

4:49:31So those models were out at the same time.

4:49:33Yes.

4:49:34And then, I mean, the one on the left was, like, very bland.

4:49:39Yes.

4:49:40No character, I guess.

4:49:41I mean, how much character do you need in an email box?

4:49:44I guess I'd need some character.

4:49:45What was it?

4:49:46Google inbox was my favorite email app of all time.

4:49:49Nice.

4:49:49Sick.

4:49:50All right.

4:49:50Well, that's it for which LLM was this?

4:49:53Let's get back.

4:49:54And we also have some audience questions in here.

4:49:56So next question was going to be, now

4:50:00that no one uses Google as a search term anymore, will the word Google stop being a verb?

4:50:06No one is saying, let me Google that.

4:50:08If search isn't a thing anymore, how is Google going to stay in business?

4:50:12Listen, Google, like most people using Google now use the AI answers, right, the AI overviews.

4:50:18So I think that term is still there, like that verb is still there.

4:50:22Like, let me Google it.

4:50:23And then they don't look at the results.

4:50:24they just look at the AI answer.

4:50:25Yes.

4:50:26I use, I still say Google and I don't use Google.

4:50:29I use DuckDuckGo.

4:50:31So like, it's going to be a while before I shake it as a verb.

4:50:34I don't know, what about you?

4:50:35Yeah, what is it going to take?

4:50:36A whole generation before Google is gone as a term

4:50:39when people are going to be,

4:50:40it's going to be that iOS computer,

4:50:42that iOS commercial.

4:50:43Like, what's a computer?

4:50:44Like, what's a Google?

4:50:46Yeah.

4:50:46I mean, I don't know.

4:50:48I don't really say Google anyways as a verb.

4:50:50So I think it's mostly done.

4:50:53And I've heard people say, like, let me GPT that.

4:50:56So that's like the new verb.

4:50:58Yeah.

4:50:59People actually, I have been hearing that.

4:51:02Yeah.

4:51:02In the wild, like at the airport and stuff like that.

4:51:05And not in California's airport, in Denver's airport.

4:51:09Yeah.

4:51:09Probably got time for one more.

4:51:10Yeah.

4:51:11Okay.

4:51:11So we got one more question in here.

4:51:14Which one do you want to pick, CJ?

4:51:16You got a question you want to pick in particular?

4:51:18We got one about WebMCP.

4:51:20We got about.

4:51:20Let's do the one about apps

4:51:23because this is very relevant.

4:51:24Like anybody trying to build something right now.

4:51:26Yes.

4:51:26So why release an app?

4:51:27Okay, so this question came in and it says,

4:51:30we used to build apps for ourselves

4:51:32and then release them to the world

4:51:34in the hope that someone may find them useful

4:51:37and then maybe develop a reputation

4:51:39and make a few dollars.

4:51:41Why release an app nowadays though

4:51:43when if it's any good,

4:51:45there's going to be 100 slop cones within a week

4:51:47and it'll all get lost in the mirror.

4:51:50Yeah.

4:51:50Yeah, my initial thoughts are like if you actually put care and thought into it and don't just like vibe code and then throw up a link to a GitHub repo or whatever else, like people are going to care.

4:52:02Like when people become an actual user of an app and come back to it, I think it's because they know that there was care put into it, right?

4:52:12Yes.

4:52:12Like it's a nice user experience.

4:52:15or like so right now I'm working on a developer tool.

4:52:17It's not exactly like this, but I put so much effort into the docs

4:52:21and so much effort into making sure that people understood what this tool was.

4:52:24Some TLC.

4:52:25And what it could do.

4:52:26And like a lot of people are just like they build the tool

4:52:29and then like let AI write the readme and don't put any thought into

4:52:32it. Yeah, yeah, yeah.

4:52:33So I think there's still space for it, but you have to care.

4:52:35Yeah.

4:52:36Like because you'll just be one more app in the sea of apps

4:52:39that everybody just slopped together.

4:52:41So I think care is what it comes down to.

4:52:43Yeah, care is what it comes down to.

4:52:45I think the perfect example of this is Rex, the Rex terminal.

4:52:51My gosh, the Rex terminal.

4:52:54Hex or Rex?

4:52:55Rex.

4:52:55Rex, okay.

4:52:56Rex.

4:52:57Yeah, Rex is the terminal.

4:52:59Hex is the voice.

4:53:01Voice,

4:53:02And then Dex is the issue tracker.

4:53:05My gosh, okay.

4:53:06So Rex is an app that's built with a lot of care.

4:53:10And you can tell it's built with a lot of care because everything, even the little notification that says you have an update, is beautiful.

4:53:18It's got a marquee.

4:53:20It's designed well.

4:53:21And it is like thought in care is put into every tiny corner of this application.

4:53:27And when you have things like that and you compare it to a slopped alternative, it makes it very clear that the people who are putting the care into it are still going to get attention.

4:53:41They're still going to get users, and they're still going to get downloads.

4:53:44There's people who are making a lot of money building small developer tools these days, releasing them to an audience of people who like their work, and then still happily pay for that stuff.

4:53:56Right.

4:53:56So you do hear everybody saying they're all vibe coding their own personal software these days.

4:54:00I am too.

4:54:01Trust me.

4:54:02There's all kinds of stuff I'm vibe coding as personal software these days as well.

4:54:06But the cream still rises to the top.

4:54:08I agree.

4:54:09100%.

4:54:10Yes.

4:54:12CJ, what are we at for time here?

4:54:14I mean, I think technically we're at our limit.

4:54:16Okay.

4:54:16Folks, this has been it from AIE.

4:54:19We're in San Francisco.

4:54:20It's going to be a great conference here.

4:54:22My name is Scott Talinsky.

4:54:23This is CJ Reynolds, and we're Syntax.

4:54:26Peace.