More Bots Than Humans

Now I'm just super skeptical when I'm on Reddit and I see a post that gets a bunch of upvotes or I go to the comments and all of a sudden I'm reading comments that, like, I completely disagree with or is also just like a overdone take. We're just interacting with bots on social social media as well, which is totally another problem as well. People think that Reddit is this beautiful place, but it's not. It's it's starting to get absolutely awful. Where do we now turn? I guess it's friends. Do we need friends now? Is that what it is?

What's up everybody. Thanks for joining Syntax Weekly.

Tons of news has happened in the last seven days since you've joined us. We've got a lot to cover. We're going to cover there's more in bots than humans on the internet. That certainly happened quicker than I thought.

A couple of new agent frameworks out. Vercel has released their own agent framework called Eve and, Fred Shot, who's the creator of Astro also released kind of a similar one. We're gonna be digging into those and seeing what they're at. Deno released an electron, type thing, a Deno desktop, which is really nifty and whole bunch of other stuff. My name is Wes. With me as always is CJ and Scott. How you doing today, guys? What's up? What's up? What's up? Doing good.

Ready.

Alright. Well, let us know down in the chat if you have anything else that we should be covering as well. We'd love to talk about it after this. We've got maybe 20 different things that we want to cover today.

But should we get into the, the first one that we have here, Scott, more bots than humans?

Yes. More bots than humans. The, the bots are taking over. And, actually, real quick, guys, I wanna show you, something real quick before I even mention this chart here. I I was having a codex work on some. I haven't used codex in a little while.

And, you know, codex has a very specific design smell to it. Right? So check this out. This is the most codex ass design of all time. Like, this just, like, nails the codex aesthetic. The uppercase, the gradients, the unnecessary

labels. Letter spacing.

Yeah. Did you ask it for orange too, or did it just choose orange? It just choose orange. It was, like, teenage engineering? Sure. Let's throw some orange on this thing. But it's like... Yeah.

That's it. The cards within cards within cards within cards. It's it's just so classic here.

This is the the modern stink of, codecs these days.

I'll say I skipped over sharing some sites today because they looked exactly like this, and I was like, just put some effort in. Like Just... Come on. You know? It's it's too bad because this is not, like, a... It's not a bad looking design. Right? Like, it's not.

If if you would've shown me this maybe a year ago, I would've been like, this is a decently clean sign. It's just the fact... And I I know we talk about this so much, but it's like, they all look exactly the same. And my brain just immediately goes to this might not be very good because knowing where it's it's come from, which is kind of... We're kinda in a weird spot with all this right now. Yes. Yes. I I made this in in contrast to the

bots taking over things. This is made by a bot for a human because I didn't feel like I was getting enough out of my OPXY.

And so I have the, whatever, 100 pages of manual for it, and I wanted it to build me individualized lessons to tell me what buttons to click on. And it even does, like, the overlay on what buttons here to do what.

What's an OPXY, Scott? It's a sequencer by Teenage Engineering.

Oh, sequencer.

So it... It's it's very powerful and dope. And I'm good at it, but I'm not great at it. So I was like, I wanna I wanna, like, find into some of the functionality that's in the instructions that I... I'm not gonna read an instruction booklet. Are you kidding me? Yes. Nice. So let's talk about this bots versus humans. The Cloudflare folks have had this radar.cloudflare.com for a little bit. And recently, I think in the past couple of weeks here, they've added this bot versus human traffic to show you what percentage of traffic going through Cloudflare is a bot versus a human.

And what's shocking here is that at some points, it's as high as 63% bot getting up into...

That's 63% of the traffic coming through as just from bots. And this is in the past seven days. The chart doesn't go back that far because of Yeah. When they released this.

Oh, unknown error. That's always fun. I've actually found this radar.cloudflare.com to be an interesting page anyways regards to just general Internet traffic. But the bot stuff, the bot versus humans has been making the rounds in terms of the Internet is now basically just a bunch of bots talking to each other with 40% of it being human interaction, which feels like it's been something that people have been talking about for a long time. But, I mean, here it is in a trend through Cloudflare.

How So much of that do you think is legitimate web scraping that's just simply being piped into a prompt? Like like the other day I was driving, and my daughter asked when she could go to this, like, local ninja place. So I just, I just aft, we have Grok in the Tesla, so I just say, Hey, when is the, what's the hours for that? And it goes off, it fetches the website, it parses it out, and it tells us what the hours are. And I was like, That's like, that's a legitimate use case for this type of thing, because otherwise it would have just gone to the website directly. But how much of it do you think it is like that legitimate part? And how much of it do you think is like people just simply just scraping it to try to figure out

the training models or doing whatever other nefarious things? I mean, I think it's all interesting because, somebody asked this question directly to Matthew Prince or at least they responded to it. And Matthew Prince, the... He's the CEO of Cloudflare responded with bot, crawler, agent are all synonyms depending on whether or not you want them to be a good or bad thing. Norma... Normative...

Normatively.

Man, I can't say normative... I can't even Normatively.

That's the word I can say. Who's specificity

right here? Yes. I don't even think it's a word. Basically, the the the idea of bot crawler agents are all synonyms depending on if you want them or not. And and I think that is... Yeah.

It it is funny because people were asking for distinguishing between agents and scrapers and stuff. And it's like, it's just all... Doesn't matter. It's all

human. You know? Man, on the same thing, I was looking at my, like, bot traffic a while ago, and and I was actually seeing it sort of come in real time.

And I noticed that there was a lot of bot traffic coming from Apple.

And this was probably six months ago. And now, at the least, the most recent Apple announcement, they announced like, oh, they have, they said in a roundhouse way, we have our own web index. Meaning that Apple now has a search engine, essentially just a search engine. And I wonder how many other people are trying to build their own search engines, because not that they want to become a search engine, but everybody needs an index of the web now, because it's very expensive to have to pay Google or Bing or one of these other people that have an index of the internet and know where all the things are. Because, essentially, as smart as this AI stuff is, it's not very good unless you, like, pipe real time web sites into it that have actual

real data. Yeah. Yeah. And and and it's funny, Wes. I I set up on graffiti.

I set it up so that when when humans visit the site, it gives you just normal HTML.

But when agents visit the same URL instead of doing, like, an LLMs dot TXT, it returns just markdown for the agents.

So I essentially built a a web surface, a a client just for agents using the same

Yeah. Code. Yeah. Cloudflare even has, like, a setting for that as well. When Yeah. An agent visits your website, should it just serve markdown? I think, like, you you may lose some...

You obviously lose some interaction there. Right? It's... Like, Markdown is simply just straight up text. But for docs, that's that's perfect.

Mhmm. Yeah. I mean, for Dockside, that's really what it was. Yeah. So how are we feeling about this? I I I noticed one thing that was really funny here.

This tweet shows the interest over time in the word smoke test or the, phrase smoke test, which is a phrase I never used before AI.

And now every other chat, AI is like, why don't we do a smoke test for it? Fuck smoke test. Yeah. Smoke test, hand wavy, or, like, what are some of the other ones that it's like... That we're gonna see interest over time shoot up other than smoke test here? Yeah. What... There... I... There was a list of them I saw the other day.

I'm I'm forgetting. Oh, I'm blanking. Maybe the chat can help us out what those were, but there's certainly those sayings that you hear over and over again Over. Yeah. Are are just peaking in there. There's sort of like a smell, which sucks. It's it's it's the same thing with the em dash. Right? Like, it sucks for people who actually did use that type of writing or the the type of person who now I hear it all the time. It's not this. It's that.

And it sucks because, like, that was that was a good way of of writing, and that was a a kind of an interesting way of speaking. But now as soon as I hear that, the alarm bells in my head start going off.

And, it's the same thing with the web design. Like, it's not a bad thing. Just that the fact that it's overused Yeah. Is is making me think maybe it's not trustworthy or maybe it's not good anymore.

Yeah. CJ, what are your thoughts here?

Yeah.

Just getting back into bot traffic, like, didn't... It definitely has to do with the fact that most people's interface to the Internet now or to the web is a chat interface. So whether that's Google AI or Claude or Gemini or whatever else. So I I do think a lot of that traffic are these chat agents making requests to sites on behalf of users.

And that has its downsides because now users aren't clicking directly through to sites. Right? Yep. I mean, the example that you gave earlier, Wes, was, like, you're just looking for the hours for something.

And normally, maybe you'd pull up Google Maps and you'd see the place or you'd maybe go to their site. But now all of that's been skipped and you just got the information directly from an agent. So I think that's definitely where a lot of it's coming from.

I kind of want to talk about, like, the bots that we interact with on social media as well. Like, actually, real quick, CJ. Yeah. Sure. To to back that up, the the new Siri... I don't know if this will be able to be seen, but the new Siri doesn't... Normally, on iOS, when you pull down to do a search, Safari is, like, the number one thing. And if you were to just search and hit search, it would do a search on Safari.

But now it's not even an option. It only lets you do Siri.

So And Siri's probably searching the web.

Right. That's where a lot of that occasional traffic sources, comes

but it's not a a standard...

I hope they fix that, because that sucks. I'll just say that. Yeah. But... Yeah. But it is... I mean, it's a sign of the times, and, like, this is also why... Kind of why the indie web is dying. Like, no one is directly browsing websites anymore. We're all talking to... Mean, I'm not. I'm trying my best to... I have an RSS feed. There's no algorithm. I'm trying to click directly on things, but it seems to be the trend. And I think if if everyone's using an Apple phone or an Android phone, then that's two of the biggest companies that have control over this stuff, and they're basically our gateway into the web.

It's basically the the walled garden problem that we had in the early days of the Internet. But now they're

basically just taking content that's on the web and telling us what it is instead of showing us that content directly. So... Yeah. I I I feel that way about, like, blog posts and and things like that. I don't... I I don't put stuff on my website anymore hoping people will stumble across it. I I put it out there hoping that it will be consumed by these agents. And there's... There will probably be good output of that at some point. You know? Like, if you are the definitive resource for something specifically, I'm sure that we can get like listens or whatever out of that, but it is a crazy world to think about all of this.

Yeah. But tell me about the social media bots' thoughts there, CJ. And I'm sure you all have seen them too, but now I'm just super skeptical when I'm on Reddit and I see a post that gets a bunch of upvotes or I go to the comments and all of a sudden I'm reading comments that, like, I completely disagree with or is also just like an overdone take.

Yeah. And you'll start you'll start to see the signs. Like, so especially on Reddit, anybody that has, like, a a generated username...

And the thing is, like, these aren't bot generated usernames. It's like the default username that Reddit gives you when you sign up now. But it's like a noun verb number.

But you see so many of these accounts in the comments and also submitting. Like, if you pay attention to, like, r slash programming humor or even, like, r slash JavaScript, there are certain posts and certain memes that show up again every few months, and they get a bunch of traffic.

And if you go to see who actually posted it, it's an account that does nothing but, like, repost things from, like, a year ago or two two years ago. Yeah.

And then you see the same thing in in Instagram comments, like, a number of comments that are just, like, from bots telling you to follow their accounts or, like, trying to get engagement.

So, yeah, we're just interacting with bots on social media as well, which is totally

another problem as well. It's it's not... So on the Hamilton Reddit, which is like where I live, the other day there was somebody that says, I just graduated and my parents were unable to make my graduation. So I'm feeling kind of down. I'm wondering where I can get late night sushi.

And My god. And the people were chiming in, here are great spots for late night sushi. And then you went to that person's...

Oh, no. You didn't even go. You... If you went to the Toronto subreddit, six hours before that, the exact same story from a different account was like, hey. I just graduated. I'm looking for a pot... Spot to get late night sushi in Toronto. And it's like, wow. That was clearly like somebody who owns a late night sushi restaurant in both of these cities is trying to drum up because, like, all AI is is just something summarizing Reddit threads, you know? Like, that's all it is. Absolutely. And now you have to make this fodder. And if you want people to come to your sushi restaurant, you have to make this like fodder on Reddit that will show up when people search it. And it just was just like, ugh. People think that Reddit is this beautiful place of like unmolested opinions, but it's not. It's it's starting to get absolutely awful.

Yeah. Yeah. There's people bragging about creating... Like, I created a...

An open claw bot that has successfully gained x amount of karma on Reddit with, you know, whatever. And it's like, god. Do you really have to do that? Like, do you have to do that? You know?

Just... Yeah. Oh, people always use it for the awful stuff. And Yeah. I I think, unfortunately, that means that we're just going to adapt. I don't know what it will look like. But I used to be able to say, yeah, I'll go on Reddit and ask for some real people.

There's not this SEO spam that's going to be on...

If you wanted to say, I'm going to New York.

Where should I eat? And you would just Google, like, best restaurants in New York. You would get pages and pages of these, like, this blog spam of, like, SEO keyworded to the to the the ceiling. Right? Now, so I just, like, never trusted travel blogs anymore. And then you would go to Reddit and ask where where people were. And now you can't trust Reddit. So, like, like, where do we now turn for this type of stuff that you are unable to have people,

like, molest and change it. I guess it's I guess it's friends. Do we need friends now? Is that what it is? Do we need a social network of people that you know who have been to places that can recommend you things that you actually like or will like? Yeah. I I don't I don't know. It's all very, crazy to me. I wonder, as a question to both of you, how high do you think this will get? So the high that I'm seeing in here is, like, 63% bot usage over human usage, which would be, like, a 30 or, yeah, like, a a 36%.

So how high do you think this bot number is gonna go? Do you think we'll hit 70%

this year at some point? This year?

Probably.

But I think long term, if nothing changes, then 90%, 95%.

But Crazy. Yeah. Once this is in the iPhone, it's it's gonna be everybody using it. Right? Once people can simply just press a button and talk to their phone to get answers. Like, I I...

The...

How often I use it now that it's in our... My car? Unbelievable.

And I... Because, like, Siri didn't really have a good one. They're like... We ask Google Home a lot when we're at home how... What things are.

And it's gotten a little bit better now that it has, like, the the AI built into it. But, like, now that it's in my car... Because that's that's, like, a huge use case where you're just driving and you're wondering something, and you just press the button, you ask your car these questions.

Like, I think it's...

Aside from it ruining the Internet, it's a great experience, and I really enjoy it. So I I think, like, it's gonna it's gonna catch on huge. Regular people are gonna be do... Doing this, like like, crazy. Yeah. Speaking of which, guys,

before we move on to the next topic here, things getting better. You're just talking to her browsing. I have a robot coming. So if you see me sprint out of here, it's because my robot has arrived. It says it's going to arrive sometime before noon mountain time here. So there's a good chance that a robot will be delivered to me live on stream here. So just Pass meeting. Yes. Are we gonna get any, hints? Is it a life-sized robot? It... Does it It's it's the Ricci Mini. So I think they're about, like, this this big, and he he has a camera. It's like a Google Home where you can hook it up to any AI provider. You can build apps for it, and it has a camera, and it can move and look at things. So it does not have arms, and it does not have legs. Oh, good. But but it can Don't give robots arms. Don't... At least don't give them thumbs.

Don't give them thumbs. Yeah. They can... Yeah. And certainly don't give them spider legs or or don't do that.

I don't want that.

I don't even mind spiders, but I don't want a robot with spider spider legs. And that's right. Yeah. Alright. Let's move on to the next one we have here, which is Before we do that, I will say,

proof that we... There are still real people on the Internet. We got rated by the AOA channel. So shout out to AOA, and thank you. They brought over 19 viewers over on Twitch, so we appreciate it. And if you are a real person watching in chat, type type one in the chat right now.

Because If you're a robot type zero. Yeah. If you're a robot type zero.

And so, yeah, thanks to AOA, bringing over real people, not bots. We appreciate you. And if you're just joining us, this is syntax weekly. So we go live every Monday to chat about news, new libraries, the world of web dev, things we're interested in, that kind of stuff.

And, yeah, that's what we're doing. We just talked about how bots are taking over the Internet, and one bot that I will let take over the Internet is Seer bot because everything we do here is brought to you by Sentry.

So if you've got code running in production, throw Sentry on there. Any errors that happen will get reported back to the Sentry dashboard, and Seer can chime in automatically and fix those bugs, automatically add a PR, and, you don't have to do anything. Just set it up. So, sentry.io/syntax.

Can get two months free on the,

team plan. Yeah. CJ, I'm I'm doing a deep dive into self healing software on Sentry right now. See, you're a big part of that, obviously. So, shout out to the folks over at Sentry working on self healing software. I love it. Love to see it. Sick. Well, let's get into the next one. Who's up next? Yes. I got it here.

Last week, two agent frameworks were announced. So one from Vercel called Eve and one from, Fred Schott, who was the creator of Astro, called flu.

And I haven't tried them out yet, but from... Like, I've been kinda looking at a lot of these, like, that are popping up where people are like, we need something like Next JS or like, like we need like the easy framework for creating agents.

And these are sort of two shots at that. Right. And the idea with an agent think there's is one shot, One Fred. One Fred shot and one Vercel shot at this type of thing.

And I think this is kind of interesting because like, if you think about like like a lot of your workflows right now, you have like, you probably have like some MCP servers, right? And that's how it can like access things. You probably have some skills which sort of describe what it can do, or like how to approach things. And then maybe you have like a bunch of like scripts or CLIs that it can use to like interact and go back and forth with things. Right? But the ability to package that all up, I don't think has been totally cracked just yet into a nice, tidy, little agent.

And I I I think, like, these are these are kinda two interesting tackles at them. So the Vercel one, the agent one, they say an agent is a directory, which I really like this because, like, I was very much... I I very much like the Next JS approach of, like, just putting folders or files in the specific, folders, and then you can route them. By the way, update. Scott just went to the door. Robot is incoming.

Excited about that.

In the agent, you have an agent. Ts file, which is like kind of like your main configuration there. And I'm assuming inside of that, would also declare this specific yeah, right here the model that it runs on. So if you have a specific agent that doesn't need to use a very expensive model, it's nice to be able to pin that agent to maybe a cheaper model or a model that it's very good with.

You get your instructions. Md, and then inside of that you have tools.

So if you've got a bunch of scripts. I have seen a lot of people shipping scripts with their skills because a skill is a directory. So they'll often have like a skill. Md, but they'll also give it like, like some, some JavaScript that it can run, which is kind of neat.

Then you have your skills, your sub agents, your channels. So this is kind of where you're getting into like, almost like open claw territory of like, how does this thing connect to different, like in, in how, in what way are you interacting with it? Right? Is this interacting with over Slack? Are you hooking up with telegram? You know, did you build your own interface for it? Is there webhooks? All those types of things.

And then schedules as well. If you want to like put some cron jobs in there.

So Yeah. Pretty nifty.

I'm Yeah. I'm curious to see. What do you guys think? I think the big difference here is the open agent framework versus, the walled garden framework.

And so if you're picking one, I've been always Which gonna read which one is open agent? The one that is literally listed as the open agent framework, which is the flu framework.

That's like their heading on their site is. Oh, okay. The open agent framework. Open agent. Is it some kind of specification that makes it open? Or is like Well, interestingly enough, CJ, the standard agent spec is a agent spec, which is not related to flu, but another one of these, standardagentspec.org.

Mhmm.

And that is not public yet. I I do have access to it. And it, again, it solves a lot of these same things, which is sandboxing the connected system between agents and sub agents, easily getting it up, hosting somewhere, and just, like, the tools and how it all connects to optimizing for small tasks, for smaller models, etcetera. So...

Yeah. I I don't know. I think they're all interesting.

One thing that I I I have found is that the sandboxing makes it kinda...

It... Obviously, the sandboxing makes it inherently safer, but it also makes it inherently less useful out of the gate than something like Hermes or OpenClaw, which they're different systems for different things. Right? You would use this as your API for agents and stuff, but I wouldn't use this as a personal assistant the same way I would like a Hermes or an Open Claw.

I don't know. I I I... It's kinda interesting that we have all these different tools or your different apps for different use cases. I think we're gonna end on just like... Just... You just have one thing, and, like, sure, you can move your skills and whatever between these agents if you really want to, but I think that being able to pack them up nicely and just have, like, a single thing that you do all of your interactions through... Like, right now, I'm even today, I've used Cursor, Cloud Code, I use the Gemini chat interface to make an image, I use ChatGPT, and I'm just like, why am I bouncing around through all of these different things? You know, I really should just be doing one single thing. The reason why it wouldn't be Cloud Code is because maybe it just...

Like, that's only Anthropic models. You want, like, some open framework that can connect to all of these different things.

I... That's what I was liking about the standard agent one. I know I haven't used flu yet, but as of my experience with these is with that.

And one of the things you can do is you can create essentially your bundle of agents, sub agents, tools, and whatever and essentially zip them up and share them with somebody else.

So you could create, like, a a workflow of connected sub agents, different models, different tools, wrap it all up, drop it into somebody else's, and get them using it too. Right? Like, we could make a syntax one, and it would be effortless to

share that amongst each other. Yeah. You just... You pop your API keys in and bam, we've got the syntax agent, and then we're all using the exact same one. I think that's probably what it will be is you join a team, you got your own personal agents, but then you also have, like, agents for your team. Maybe there's a customer support one, you know, and, and what interface you use to work with that. It really doesn't matter.

Interesting thing about this is that there's, it comes with a front end UI for, for doing them. I don't think, I think that the Vercel one is simply just like a, like a software package, but flu takes it even a step further to give you a,

a UI for all of your, which I don't know. What do you think? Do we need that? Hermes has got a UI and it's pretty dope. Ended up making my own, but it still has one and it's dope. One thing I really do, I've actually been using Hermes to code lately. And the thing that I really appreciate about it is like the memory and things that you can get with agent systems like this. Yep. But flu, yeah, I don't I...

Yeah. I don't know. What what were people... You... What would people be using flu for? Would they be using it to code with? Are they using it to have as their their AI service that is being interacted with, for products? Are they... Like, what are what are people using this stuff for? Is it for personal use? Is it as a business use case? Like, what is it? I I think I think the idea is that it it will work with everything.

I think the idea that you would have a single harness that you're using for coding, and it's integrated with all of your things, and you sort of have this, like, computer almost that just hooked up to everything that you need, and you can go ahead and do it, and you've just got it all dialed in.

Whether that will happen or not, because, like, I I just... With all of this AI stuff, I just keep going back to, will regular people use this? Or are regular people just going to simply type in the chat GPT boxes? I even think about that with how much better using Cloud Code or Codecs or any of these CLI ones that have access to everything, how much better those are than just what people are using in ChatGPT. Because often you hear people being like, I don't really like AI. It's not very good. And it's just like, well, you don't... You're just using the, like, 10% watered down version that's in this walled garden of ChatGPT,

and you don't realize how actually awesome it can be if you get these things dialed in. Yeah. That's why I like my Hermes setup that I have set up for Courtney. I mean, it is. It's using models for specific things, but it's all very tuned.

And so now when she uses it, it feels like she's using ChatGPT, but it is. It's using the models or the the prompts or contact seating or anything that I've set up for it. Yeah. It has all the tools that I've given it, and it does things really well, like, way better than ChatGPT

in my mind. What models are you running that on? Do you have, like, specific models for specific things, or are you just using them all on one provider?

Yeah. It depends. I am using... Right now, I have open router set up, so I'm still experimenting.

The main agent for her currently is using GPT 5.5 as the the model just because you can get OAuth set up with it really easily through the codex plan and then I I, you know, for like small little tasks and things like that, man, the Grock stuff is cheap through open router and there's like somewhat reliable or or I've been even using smaller anthropic or smaller... Well, some of the older anthropic models on OpenRouter. Sonnet. Cheaper. Yeah. Yeah. Haiku, Sonnet.

Yeah. I haven't gotten too into the the niche models. I haven't.

Nice. Yeah. I've been I've been meaning to, like, dig into a lot of those, like, cheaper ones as well just because it's expensive now. Right? And people are looking for alternatives, and they seem to be fairly decent. Even, like, a lot of the the grok build ones are, if you pay for like the check mark on Twitter, then you have access to super Grok. I forget the names of all this stuff, but you may have act, you may have like, like, a, like, $30 a month worth of grok that you can just use for specific things. I just hooked it up to my warp because warp just stopped, like, the free inference.

And I was like, alright. Well, I'll just I'll just hook it up to my grok because I'm I'm not not necessarily always using that board.

Yeah. But getting back to the frameworks, I... Yeah. You guys talk too much. I just don't know when I can shut you up. Here here I can. You can jump in. Just talk over lion's den. Yeah. I'm not used to it. When when we were... I'll say when we were in person, it was so easy to, like, maybe talk and... I don't know. It's harder. We're all remote. But just talking about, like, agent frameworks, this is early days. Like, you could you could potentially compare this to the early days of JavaScript frameworks. I mean, it's a little bit later just because people are already have framework brain and and AI is already guiding them towards, like, building framework like things. But I think anything that makes it easier for people to create agents without having to learn about, like, workflows or durable executions or retries and all of this stuff that I could probably do and build manually.

But if you have it built into the framework, it's going to make it a lot easier, especially for companies that are trying to create agentic flows. Like, I I haven't gotten into personal agents or, like, home agents yet.

I'm thinking about trying it out with local AI, but I'm still not sold that it's going make my life any better.

But I do see it for companies that are trying to build agents into their company and their products.

Any sort of standardization, I think, is better than none because then people can move from one company to the next and already know how to use a particular agent, that kind of thing.

Yeah. And you got to use something that has, like you said, all of this retry and the durable, like, workflows built into it. Otherwise, like, you... People are gonna be like, oh, like, our company has agents, but it sucks. You know? I can't drag and drop an image into the UI or, like, I can't pay something or I can't. You know, it's like the people are used to this, like, chat GBT, and then you come into this company and, like, we have our own version of it and it's not very good. It's so nice when it... You're... It's built on top of something that has all of those pieces that come for free. And then you simply just need to hook all your, integrations up, and you can let rip.

Exactly. And I think that's why seeing a framework from Vercel... So they built Next. Js. We can kind of trust that they know about infrastructure and how to structure things.

Yeah. At the at the same time, I've seen some some comments in the chat of, like, it's potentially vendor lock in. Like, if you use the Vercel agent framework, now you have to deploy your stuff to Vercel.

Yeah. So that's why I would lean towards a more open agent framework, because maybe I want to deploy it wherever I want.

But the people building these things, I want to see that they have a track record of good architecture and and good design decisions because anybody can vibe code themselves up like an agent harness or whatever else. But all of these core engineering problems are the things that I wanna see talented people work on, I guess. Mhmm. Agreed.

I love it, guys.

I I... CJ, I'm I'm gonna... I can't wait for more local AI stuff from you. Because I'm eagerly awaiting the new Mac Studio release just to to throw on some some local AI stuff going on for my my base system. But because I'm big on privacy, and and I'm also trying to figure out, like, how can I

have cool, interesting daily summaries with my calendar and my email, but not give an agent access to absolutely everything? So, like, how can I solve that problem of, like, selectively making my data available, but it also not sending all of that data all at once to these AI providers? Like, Google already knows everything about me. I don't necessarily want Anthropic to know everything about me or have all my data in there in the prompts as well. So I think that's a problem that needs to be solved is how do we do this locally with local models or in a much more privacy conscious way and in a way that doesn't just give an agent access to your entire life so that it can, like, really mess things up.

Yeah.

Very much agree. I I think about that a lot as well. Like, oh, I could, I could use some of these like new cheaper models for my, like, I, I desperately need something that will tackle my email. I just get so much garbage, you know, it's like 99% garbage that comes in. I desperately need something that will help me process and filter and whatever, but I get so much that it it it... It's too expensive to run all this stuff through,

like, one of the... Those other models. And then I think like, okay. I'll you I'll pause you really quick because our stream went down. So I think Randy's working on it. Alright. We're back, everybody. Just say hey in the chat if you can see us.

Streamlabs' fault. We blame it on Streamlabs. We're using we're using their multistream feature, and everything looked good on our end, but all the streams went down everywhere else. So now we're only streaming on YouTube, unfortunately.

But

we are back. I wonder if the Internet's going crazy right now because the steam machine, folks, the Steam machine was just announced for pre sign up, join the list, whatever, and, like, Steam as a a site is like...

I don't think it's just their traffic, but their their

login is down. There's some major core of... What's the website? I know people hate down detector.

People who run infrastructure hate down detector.

But personally, I very much enjoy down detector, because it it tells me when websites are down. Know, like, the status page is never as fast as actual down detector, and it's it's just, like, actual people clicking. Yeah. I have a problem with that. If if we're on down detector right now, it does...

Doesn't seem like anything is...

Anything about Streamlabs?

No. I don't think enough people use Streamlabs,

to... It's not my fault. See it? That's all I know. Yeah. It's not our fault. Any of our faults. We didn't do anything. We didn't do anything. YouTube had a uptick

in the last... Yeah. YouTube did have a uptick in the last little bit, so that may be something has nothing to with Shopify.

Had a little uptick as well. So perhaps something happened. But we are back. What we were talking about before we left was, like, running these agents. Right? You wanna be able to, like, pipe all of your stuff through. And I was saying, like, yeah. I'm like, my email is personally a really sore spot for me right now because I get so much of it. And, and like spam detection is no longer, no longer works. Right? Like it, every, I get all these emails that are like cold intros or, just people like writing like sales pitches.

They all, they all slip through it. It's so obnoxious because they, they put me through these like ringers where they'll follow-up in, in three days, whatever.

So I need something that's going to parse every single email. And I had it hooked up. I use Missive as like an email client. I had it hooked up with like an API where every single email that came through, it would like parse it. But it's just, it's too expensive to put all of your email through it. I get way too much. So we're talking about like, what do you do with that? Can you run it through a local? Or, could you put it through a cheaper LLM? Like, I don't necessarily want to be piping all of my email through like a Chinese LLM.

So I think the local is the move there, isn't it CJ? I'm sorry. I'm just... I'm try... I'm not listening. Okay. You know what? I'm I'm trying to include CJ for once. We keep cutting him off. We're not letting him talk.

This is why. CJ's not paying attention. Scott, do you... What do you agree? No, CJ. I'm trying to fix it. Yeah. You you fix it. I'll talk to Scott here. Yeah. As long... And the local stuff does feel like that's the the key for a lot of this stuff for me, especially if you have personal assistant stuff. I certainly don't wanna be sending every single thing that, you know, you wanna be able to use these tools for what they're good at without them sending your Social Security number and whatever to OpenAI.

Not like they actually don't have it. But... Yeah. This is why why I'm so interested in getting CJ's take Yeah. On the season, and getting him him do it. Because, like, like, are they gonna be faster? Because, like, I was using... There's this, like, model for flux, for generating image to image transformation. I was taking like a video of myself and then transforming it to like me with muscles.

And if you host it on, if you use like the API, WebSocket API, and stream it, you can get like 20 frames a second.

And it's it's sub second generation, meaning that you're able to give it an image and say, put muscles on me. And it gives you an image back in like, I don't know, four or five hundred milliseconds, which is super fast. And I was like, I wonder how fast I could run that on my MacBook Pro M1 Pro Max 64 gigs of ram. So yeah, five years old, but four or five years old, but still a pretty beefy machine. Right? And that went from using the API, four hundred milliseconds, to myself, eighteen seconds, you know? Like, like, free. Like, probably 40 times slower to run it on my own hardware. And it made it made me think, like, are we are we anywhere near? And I... If I had, like, maybe the newest MacBook Pro, maybe it would be even faster. If I have a $4,000 graphics card, could I ever get that to be that fast?

Yeah. It's early days. And I think especially for image generation and video generation, for local, you need really beefy setups.

But, for instance, like running the Jimma 27 b or the Quinn 3.630 b, I can get between, like, forty and sixty tokens per second running those things locally.

And that's slightly faster than even ChatGPT or Gemini in a lot of cases. So we're we're closer with text and for smaller models.

Even...

So if... My rig has a 128 gigabytes of unified memory. And so these 30 b model... 30 b and 27 b models are anywhere from 20 to 30 gigabytes.

But if I wanna run at, like, a 70 b model, that I'm usually only getting, like, 10 tokens per second. Like, it still runs, but it's a lot slower. So I don't I don't see it happening anytime soon, especially because hardware is more expensive. Like... And to to get the same speeds, you really do need so much more hardware.

So using smaller models, maybe not have... Using or be... Being able to do image generation locally, I think that's where we're at right now.

Yeah. I don't see it happening anytime soon of getting, like, cloud level speeds from your local home. Yeah.

You know what is one interesting... Oh, I I... This is kind of related. Yeah. Let me, let me show you. I'm sharing my screen. Randy, you throw that up.

And beware. I think that this is a lot of crypto bro stuff, which is not my intention here. So please be careful, but this AcuRist has this like app that you can install on your phone and it lets people run compute on your phone. And if you're thinking like, if you're doing like weird stuff and you need like, like clean IPs or you need like, just like a whole bunch of like compute very quickly.

I know this is like the, the plot of Silicon Valley, but it's kind of interesting to think like, yeah, I need to run a whole bunch of compute right now. I need to I need to maybe run it in some sort of private sandbox.

Is it better to to to run it on a 100 people's random Android phones that are in their pockets? And I don't know that what the answer to that is, but I thought it was kinda an interesting take to see that this is actually possible.

Think there's a number of these things that are are being announced like this or released for either sharing compute or hosting your own models or hosting your own compute and then making money off of it so you can, if you have g... Beefy GPUs, you can essentially rent out that compute and, essentially get some of that money back that you spend. I'm actually very upset because I was going to be getting a beefy GPU, and then it was $8, so beefy.

And then I... And then it went up to, like, $9, and I was like, okay. I'll just wait for it to go back down to 8. Obviously, it's not going to do that.

And it's now, like, what, $12,000 for a...

An RTX 6,000 Blackwell.

Oh my gosh.

Yeah. That same GPU was $8,000 and I was thinking that was a little, you know, that was a little crazy. I coulda had... I coulda been selling that compute, Wes. Yeah. Action in. Eventually,

it's gotta get cheaper and then, like, will people be putting, like, compute boxes on their house? And then when power is cheap overnight, like, selling that compute?

I'd love to see it. I mean, that's kind of what people do with electricity already. Like, you get solar solar panels or whatever else, you can feed back into the grid.

I think if we can figure out a nice and secure way to do it...

I mean, I'm sure Accurast or whatever this company is has encryption or, like, you also... If you... If you're running your workloads on regular people's phones in this case or whatever computers they've made available Yeah. As somebody that gets added onto this network, I also don't wanna be liable if somebody is doing some Yeah. Weird Right? So if if they're distributing the compute and then it just so happens that my computer is used to do nefarious things, that's not great. Yeah. You you think about that. Like, where are all of these agents getting?

Like, how are they all bypassing these like captchas and stuff? And, and one of the ways they're doing it is they're using residential proxies and they're running it and, they're running it on like your TV when you're not using it. You know, you install these sketchy apps on your TV to get free compute, or you have like a, you go to the convenience store and the guy sells you a $100 IPTV box. Those things are for sure, also using your residential IP, to like proxy nefarious traffic through.

Not necessarily even nefarious, but like scraping. Right? Yeah. So there's a really good Dark Debt Diaries episode that talks about exactly that. Based in, I think even so Jack Reider, his dad, I think even bought the box and, like, put it in their house that it gives you a bunch of free TV.

But a whole bunch of people did research into it. And, it gives you free TV, but it also becomes a residential proxy. And it tries to hack all the devices on your on your home network. So, like, it's a thing. It's definitely a thing.

Too good to be true.

Yeah. Put it on a subnet or, like a... Like, not even a subnet, but, like, a private network? Yeah. I... Would you do that for free TV? Maybe. I think your ISP would probably start getting complaints.

Yeah. I think if if you're technically savvy enough to set that up, great. And and I've just been learning how to do that. Like, I've got an access point, and I'm setting up VLANs, and I'm trying to isolate my networks, my home automation network from my... In my cameras. I'm not trying to isolate that from my main network. But

it's not easy. You have to, like No. Regular people. Settings menus. Yeah. This... Everything we talk about on this podcast always comes back to regular people ain't gonna do that.

Yeah.

Yes.

Thick. Alright. CJ, what do you got for us?

I got some things I'll just mention to everybody that's watching. This is Syntax Weekly. Welcome in. Sorry. The stream went down, but I see a lot of people have trickled back in. So welcome. Welcome.

We've got...

I got one more thing for you, and then we're just gonna bounce back and forth. We have a bunch of various news and and stuffs like that.

But what I have for you today is Deno desktop. Let me share my screen real quick. So if you're familiar with Electron or Tari or any of these ways to create cross platform desktop apps using web technologies, Deno has released, currently in Canary, Deno desktop.

And so if you're not familiar, Deno is a JavaScript runtime, essentially like a replacement for Node. Js or replacement for Bunn, but it's from the original creator of Node. Js. And they've been working on this a very long time, and they try to position themselves as, like, the more secure Node. Js. There's a lot more ways that you essentially have to opt in to network access or access to your ENVs or access to your file system, that kind of thing. And they're also Node. Js compatible. So they've been working really hard to make it so that you can run apps within Deno, but you can use all of the same Node. Js APIs that you're used to. So Deno desktop is one more thing that's just built in. So from the command line, you can do Deno desktop, and it can turn any project into a desktop application.

If you wanna try it out right now, you do have to install their Canary build because right now, Deno is version 2.8, and desktop won't be available until version 2.9.

But the idea is you can have a very simple entry point. So I have an example app here. Just a very simple main dot t s, and it's a Deno serve that serves up some HTML.

That's literally all the code that I need here. And then from the command line, I can do Deno desktop, and it will launch a desktop app. Let me drag it over here.

And so now this is the experience you're probably used to if you've used anything like Electron.

And so I can go back. I can edit the codes. Let's make this font color white.

CJ loved me saying the word codes.

I I do.

I used to not say codes, like, because it sounds weird.

But in the research community, they say codes a lot, and so that's why I started seeing it. Yeah. So you get you get hot reloading. You can edit the text here. This example is super contrived because I literally have an HTML string that's being served up in the desktop app. But the cool things about this are out of the box, it can use just the system web view.

And so these are known as back ends. And this is one of the reason something like Tari got really popular is they ship with the system web view by default. I think in the latest version, maybe, Scott, you know, are they switching to Chrome embedded? Yeah. You can change it now? Okay. But Deno is shipping with that by default. Essentially, if you're on a Mac, it'll use the Safari web view. If you're on Windows, it'll use the Edge web view. If you're on Linux, I think you have to install, like, Chrome embedded framework or something like that.

But you can do that. But you also can use the Chromium embedded framework, which makes it so that on every system, it uses Chrome, and it... It'll automatically bundle that in.

And so this is nice because then you know every desktop that's running your app is running Chrome, so you can be sure that for the most part, everything is gonna... That works in Chrome is gonna work on everywhere you're deploying your app to. But it's nice that they're shipping this on day one. You basically get to choose your back end.

And the other aspect of this is you can use frameworks as well. So I just showed a super simple example where you just have an HTML string and you're serving that up with a Deno serve.

It also has framework detection.

So if you're in a folder that has a Next. Js app or an ASTRO app, the docs here say that they support others.

Right now in testing with Canary, I could only get it working with Next. Js and ASTRO. But the idea is if you already have a SvelteKit app, let's say you wanna turn that into a desktop app, you can literally just c d into that directory, do Deno desktop, and it will detect what your framework is and automatically set it up so that it has, like, the embedded local view, and it starts up the server whenever the desktop app runs.

So this is something else that...

I I feel like where Deno has gone with a lot of these things is is trying to be all in one and require way less config. Right? If you wanted to do the same thing with, let's say, Electron, there's a million and one starter kits.

Like, there is, like, the Electron build framework, but that doesn't come by default, and you have to find the docs on it to figure out how to build for different platforms.

Or if you wanna get Veat going, with Deno, they're shipping out of the box framework support, choose your back end. And so this kind of seems like another great unifier of being able to build desktop apps.

So, yeah, curious your guys' thoughts on this, and maybe we can dive into it a bit more. I can show you some of the stuff that I've seen. My first thought was like, like, why do we need another one of these?

But I'm just going through the documentation, and it says bindings. So, like, bindings are... You're able to call back end code from the front. And and that's really cool because you can, like, have, like, a database save method on the server, but you can just call it from the client, you know? Like, every every framework has that.

But it says bindings are not IPC.

That's what you use in Electron.

The Deno runtime and the rendering backend run as threads and processes inside of the same address space or coordinated process groups web view.

Calls go through the same process channels and the back end dispatches them from the run live. In practical terms, bindings are fast enough that you do not need to worry about call frequency for typical app or the work workload.

So there's no, like, there's no socket where things are are being called back and forth.

I guess they run-in the exact same process. I'm not exactly sure how that how that works. It'd be kind of interesting to talk to Ryan about this, but that is interesting.

Yeah. I think that's one of the other selling points. So they do have a comparison table that talks about all these other frameworks. So you would compare this to Electron or TARI or Dioxus, But all of the other ones have, like, a multiprocess model or, like, an IPC model, which is interprocess communication, where you have one process that's the renderer, that's the thing that people see, and then one process that's the back end, that's the actual thing running that you can maybe write custom Rust code or custom Node. Js code. And then what we're seeing here with Deno is it's all in process.

So you can use their bindings API that basically makes it so that you can define functions that can be called from your front end.

But when it's running all of this, it's all happening in the same process. Those things, it's going to be faster communication because it doesn't have to hop between processes, potentially easier sharing of large amounts of data, and be... Also being able to use native things like a like a Node. Js worker thread...

Or sorry, like a Deno thread and have all the benefits of that, but still be able to communicate from your front end codes.

So aside from all the niceties and built ins, I think they've they've they've settled on a pretty cool architecture as well.

I wonder...

Does this also, like... It just doesn't share memory, does it? Like like, imagine you had a massive JSON object, and then you need to send that to the front end to, like, display.

One thing is that, oh, now now you have duplicated that object in memory. Right? It looks like a server and on the on the front end. Is is that something that can be done with this? I know people were looking into that with web workers, where, like, there... The memory is shared. Like, you have a... Again, if you have an object in a worker and you have an object in in the browser DOM, then you're simply... You're duplicating it twice. And that that can be... That's usually not a big deal if it's just an object or whatever, but if you have very big payloads, that can certainly be an issue. Now the browser has this idea of, like, a shared memory array or shared buffer. I forget what the exact name of it is. So I'm wondering if that plays in to this as well. Yeah. I'm curious about how it connects to macOS APIs directly.

Like, the Tory space is always kinda crazy because you're needing somebody to create some kind of translation layer between Rust and the Apple ecosystem.

Like, things like in app payments and Tory and stuff for macOS are are just a a giant pain in the ass. So I, like, wonder what that story is like. So... Because those are the things that nobody thinks about when they think about, like, oh, the Electron system is so... The Electron ecosystem is so mature.

They have those things all figured out. Like, if you want to connect directly to Apple's in app payments, etcetera, then, it just works.

So, like, what is that story, the deployment and stuff? I I since this is my first time looking at it, I mean, they do have stuff about deployment and and and things inside of here, bundling and all that stuff. So I would love to dive into this. I I had no idea this was a a thing, but man, I I love I love options in here because I certainly don't love electron. It's just the best of the experience given the ecosystems of it all.

And that's the crazy thing about AI right now is that people can spin up these things fairly quickly because they have ideas of how they should be implemented.

And you have... We have all of these learnings from Electron. So, like, I don't... I think, like, probably the days of Electron are are likely numbered. There's a reason why so many people are on Electron. Everybody says it.

I I hate Electron, but the ecosystem's the best. It's the most battle tested, you know, etcetera, etcetera.

Those are maybe not benefits anymore now that you can use an agent for a lot of this stuff. Yeah. And people saying that they wanna use the in app browser or the

the Safari, the CEF, what... The people wanna use the the browser that comes onto the machine.

I would urge you to think very carefully about that because if you don't control the environment that your front end code is running in, then it makes debugging and testing very difficult because even the version of Safari that ships with my computer is different than the version of Safari that's on Wes' computer right now. It sounds like web development, Scott. It's... But it doesn't have to be because with Chrome, Wes, you can bundle it all together Yeah. In a nice packaged version of the version of Chromium you wanna use

and Or like a custom bug build as well. Like, you wanna turn HTML and Canvas on because you're doing some cool things? Like, Remotion does that. Remotion has a custom build of Chromium with all the flags and every... All the knobs turned exactly how they need it for their product. Yes.

Yeah. I will say Leo is in the chat. So Leo worked on this feature over at Deno. Oh, okay. He's eating dinner, so he can't follow along closely.

But he he did say that there is no shared memory.

So I... I'd I'd have to dig deeper to see how really all of this is working because I guess, technically, Chrome is running as a separate process. But I guess maybe what they mean is, like, the code to actually communicate with that Chrome is still inside of Deno. I I I don't know how all of that would work, but apparently the memory isn't shared directly. So there is gonna be some some hot potatoing of of data Which in maybe you also want. Like, shared memory is is probably one of the biggest vectors for attack as well. Yeah. And there is a question from Acid Spark that says, how is the built size or bundled size? If you look at their comparison table, you can see that, if you're using the system web view, then it's about 40 megabytes. And that is because they bundle Deno with it. So, like, the actual Deno binary is gonna ship with this, so that doesn't have to be installed on people's machines.

And then, like... Because if you compare that with Tari, they basically ship completely built apps. So, like, they're not shipping the Deno runtime or anything like that. They're... There's... Your your built app is being shipped with, like, Rust modules themselves.

So it's definitely gonna be a bit smaller.

Yeah. Compiled to to native codes. Yeah. So this is, like, a good middle ground. Right? It's not the... By by default, that 40 megabytes is not gonna be the same 100 megabyte payload.

A little bit bigger than Tari. But if you do use Chrome embedded framework, you're looking more so, like, electron sizes, like, 150 megabytes or so. I like that.

This gives me a lot of hope. Yes. Yeah.

And I would say check out their docs, install the Canary to give it a try.

But they have all the APIs that you would expect. So they have ways of opening new browser windows. Like I said, they support frameworks. So out of the box, you can get, like, hot reloading with Vite and things like that.

This was always hard for me to find in the world of Electron, but they have ways of creating app menus and status bars status bar icons.

And also how... You have control over what's in the dock on a Mac or in the tray on Windows.

So they've thought about all the stuff. I'm excited to to dig deeper into it. You know what I don't see, CJ? The same thing that we had an issue with in Tori when we were building the syntax production assistant in Tori was deep link support. And we had to do all this crazy hacky stuff to get GitHub

auth working inside of the app and YouTube auth working inside the app. That might actually be platform specific. So for yeah. May...

Yeah. If Leo is listening, that's a feature request is we want deep links.

Otherwise, you're probably gonna have to add platform specific code for, like, Windows or or or Mac to support deep linking. Yeah. Because especially, like, OAuth flows. Like That was crazy. To.

Yeah. Like, open the browser, but then that should relaunch your app so that they can be logged in. We had to jump through a lot of hoop hoops to get that going. That's for sure.

Should we get into some of the other news? Because I have another one that's that's related to Yeah.

Colin Hacks is is the guy who created Zod. He's the guy who created standard schema, tRPC.

Also guy, I didn't realize this, worked on Bun.

He's released Nub, which if you are keen, you will realize that that is Bun backwards.

And so he's releasing, I know you love a nub, Scott.

He's released this thing that is essentially adding all of the bun, like Deno nicety features to Node JS. So what it does is it augments Node JS by adding in everything instead of having a separate runtime.

So nub uses these to add buns best feature on top of stock Node JS, like non erasable TypeScript syntax, meaning that like, yes, node has TypeScript right now, but it only has like type stripping. It doesn't have, like enums, which I don't know, jury's out whether you should want that or not, but has it, TypeScript friendly module resolution JSX, you know, like that's another thing about the Node JS TypeScript support is that type JSX is is nowhere to be seen. So you just, you simply can't use it.

Decorators dot ENV auto loading. That is something that Node has, but you have to, throw a flag in there. Right? YAML, TOML using, etcetera, etcetera. And he claims it is what, 24 times faster. So it also includes like a...

Just one of the items that includes is a pnpm replacement.

Seems like everybody's building a, what do you call package manager these days. So nub run. Oh, I guess.

Wonder if it uses,

like, I wonder if you, when you say run, does it use pnpm instead? Like, is it almost like a Frankenstein of like, No, you say nub

nub install is three times faster than pnpm install. So it doesn't. Right? And I think I read somewhere its own installer with a pnpm shaped CLI.

I A pnpm shaped

CLI. Yes. Npm, pnpm, bun, round trip, yarn, speed up. It. He slop fork pnpm and said make it faster. And I No, no more I mean that with all... Like, I like... This is this is really interesting to me. Right? Obviously, he he releases something I I pay attention because he's a super smart guy, and it's changed so much. You know? Like, a lot of this AI stuff is all built on top of of, Zod.

So this is pretty interesting.

Yeah. This showed up on my TL, and it's one of those things that's like, I ain't got time for this right now, but I do think it looks interesting. Yeah. Are you doing? I was just throwing that out there now. Yes. Throwing on my TL, my timeline. Yes. My FYP. Yes.

No. It's exciting. I think anything that unifies the ecosystem... And and the idea here being that still use Node for all the stuff it's good at, but then give all the extra little features that Node hasn't taken the time to add, I think it's I think it's a good direction. Mhmm.

So we nubbin'? That's what we're doing? It's a version man... This this kinda also encroaches in the, like, the v plus area as well. Right? Like, it's a it's a version manager.

Everybody's competing for that CLI script. So it's it's not a total replacement for v plus, but it does overlap quite a bit of the the features.

Yeah.

No

linting or that kind of stuff that you get No. V No. None of that. Formatting. And, like, no front end either. Right? This is just a node thing.

So I probably stick with...

It's hard.

Can I v plus a nub? Like, can you can you nub to v plus to Node? You know? Can I go three levels deep? Is that possible?

Perhaps. Brand new brand new sentence. Can I v plus a nub?

Yeah.

Alright. What else we got? Well, first of all, shout out to Zebazella in the YouTube chat who says, I've been listening to the podcast for years but never watched any videos before. None of you look anything like the images I built in my head of your voices alone. So hello.

Welcome in.

Oh, that's great.

Cool. Alright. I I got something to share.

I don't know. Have you guys heard of foresight JS? It essentially predicts user intent from mouse and keyboard cues.

So Oh, yes. If you're just on their landing page, so it's foresightjs.com, as you move your mouse around, you can see that it projects this vector of, like, where your mouse is headed to.

I think it's as hell.

So... But that information can be used to say, oh, the user is, like, 10 pixels away from this button, and the mouse is headed in the direction of this button. So I'm gonna go ahead and preload the content for that that link that they're about to click on. This is rich. So... Yes. Yeah. Their their tagline is prefetch smarter. So Foresight predicts...

Foresight JS predicts user intent to prefetch content before it's needed. It's fully configurable and traceable using different strategies for desktop and mobile users. So, like, on mobile, you would care more about, like, touch, and so you can tap into things like on touch start.

And... Wow. Okay. Yeah. So there's this playground down here. You can see it happening. Like, as I start to to scroll... Like, move my mouse to the left or to the right, like, these individual elements are are being fetched according to, like, where my mouse is moving before it gets there. Yeah. So this is one way to make your site feel faster.

And the the reason this is news is they just released version four a couple of weeks ago, and they added React and Vue components. So on their home page, you can see you can install the React or the Vue version, and that will let you do this stuff directly inside of React.

And then if you are using any other framework like SvelteKit, you can still use their JS. Their engine. They basically just give you events that you can tap into. Like, you subscribe to specific events, and then you can set it up for whatever framework.

But, fascinating stuff. Slop Porkus Svelte version. This looks dope. Yeah. Yeah. Yeah.

Let me let me show you a place where this will be killer. Because, like, the prefetching stuff, very cool.

But the simple mouse trajectory stuff needs to be in the browser. I think so. So like, look at this. I'm on Maker World, and this happens all the time. Oh, have to see your browser. Oh, no, wait. Sorry, we do. I'm done. No, I'm done. I'm looking at the wrong screen. So Maker World has this, like, little drop down where you can, like, click on your user profile. And if I wanna go to my likes, I gotta, like, click here. And then for audio listeners, I'm just, like, hovering to a nested drop down. Oh, it sucks. And and I wanna go to my likes. And, like, what you have to do is you have to, like, like, make sure your mouse doesn't go outside the button. And then Yeah. There's a little bit of gap in between, which will trigger off.

Oh, no.

So it... Because there's... I'm sorry. There's margin in between it. So that that trigger... That triggers the mouse off. Right? Yeah. So you gotta, like, move very quickly. And like like famously, Amazon did this, like, probably ten years ago as they figured out what the, like, hover triangle will be.

But this is this is even one step further of, like, if I just want to go very quickly to a nested menu, you shouldn't... That's so frustrating that I have to, like, like, the plank to get to the button that I want. This foresight JS, pew pew pew pew pew. Like, knows exactly where I want to go. And just...

Fetching, yes. But, like, not closing the menu if something...

If if you have the foresight to think that somebody is going to hover over top of it. That makes perfect sense. So, like, instead of having, like, a hit area, it's more so, like, a hit area plus mouse direction plus a certain amount of And easy enough, now the bottle just stays open instead of auto closing. So, yeah, love it.

Will be very expensive.

Who who does this? The McMaster Car. I did a video on this McMaster Car aggressively pre loads and pre renders a lot of their pages, which is why the McMaster Car feels so fast.

And they only do it on hover.

And the other options are you can do them when the links get into the, like within the viewport, you can, you can figure out what are commonly used ones.

There's a whole specification for preloading links based on if the browser thinks you're going to visit them.

But all of these things aggressively prefetch. And if, if you're using bandwidth or compute anytime those are, you may be just throwing that in the garbage. So it's a, it's a delicate balance of like, am I okay with my hosting bill going up? As... If if it means like I'm selling more widgets or or people are are are having a better experience.

But like Hey, Wes, you see you see this? This get predicted events from Alka TV posted, get predicted events was in baseline 2024.

A method on the pointer event interface returns a sequence of pointer event instances that are estimated future pointer positions. What? Have it open on my screen. Yeah. It's baseline 2024.

I have never heard of this. Never heard of this. Alka, thank you. This is crazy. Anything's Alka? Shut up, Alka. Git

predicted events.

It... And they even have a little demo on MDN if you scroll to the bottom that will show you kind of similar where it shows you...

It doesn't... It's not as cool as the little arrow example. They need MDN. You do the little arrow example. I thought that was pretty pretty nice way to say it. You gotta move woah. Faster Interesting. Mine looks different. Mine has, like, red circles on it. Yeah. Mine's red. No. It's just... Oh,

okay.

Interesting.

It doesn't work as well as I thought it would. That's the frustrating part about a lot of these specs.

It's like, yeah, that's what we want, but that doesn't actually look very good. I don't think you know because you haven't tried it. You just I should try it. But like, I'm looking at where the circles are, and it it just puts a bunch of random circles up or below my cursor.

I'll try it. This is... I had no idea that this was an API. And I've... And I love obscure APIs. And this is in every single one. Yeah. AlkaTV, this is crazy that you knew this.

Shout out. Nice work. I wonder if if Foresight is using some of that under the hood. I I did a search for this API in the repo, and I didn't see anything, but it could've just been me searching I could've had a a typo. Who knows?

Interesting. It's a- and it's pointer as well, meaning that it's- will likely work for touch as well. Or gamepad, you know? It's everything.

Yes.

Okay. Sick. What else we got?

What we got for us? Wes, you got something?

What have we got here? Flexport CEO calls work from home fraud or white collar fraud.

That's stupid. I I want... So if you didn't didn't see this, the Flexport CEO, which is like a shipping logistics company, he says, like, like, I can't work from home. I have a huge house too.

And I... I'm not able to, like, do work because I have my three and five year old just come in and bug me. I can't get any work done. It's... And so he he obviously doesn't want any of his employees doing work from home. Mhmm.

Obviously, that's a dumb take.

But in the same breath, I also think a lot of people working from home are just dicking around.

So Wes says people who work from home are lazy, and he thinks they don't need Not everybody. But I don't think that

a lot of people... I can work from home very well. I've done it my entire career. I've never not worked from home. But I know a lot of people are just like not as productive as the, as the hand. And yes, it is very frustrating to like have to commute in and it's very soul crushing.

But I'm curious what you guys think. Like, can you be productive doing work from home?

Yeah. I think that I think the right people can. Absolutely. I I've been working from home for a very long time. I do miss things about working in an office, but I think people like this guy should just shut up. I I mean, like, the... I think that's for maybe... You know, it's like people who are actually in the the management of directly interacting with employees and connecting with them and working remotely.

Like, I feel like I feel like that is the the type of opinion I I value more in this instance. But, yes, there are certainly people who dick around, when they work from home and that are probably not cut out for it. Yeah. Using those jigglers.

Yeah. Exactly. Like, Work dot JS, and it keeps your screen going.

But I I think Is that what you use, CJ?

I do. Yeah. Because that's what I was gonna say. Like, there's no waiting on your AI video to come out, CJ. So Exactly. This is what I'm saying. I just I'm basically not doing any work. No, I'm working hard. That's coming out soon. It really is.

But I think it comes down to the person and also the environment. Right? Like, I have a separate workspace in my house. Like, my wife doesn't bother me when it's work hours.

And I am working from home. I have worked at places that had a work from home policy, but people working there didn't treat it in the right way.

Like, one time a dev was like, I have to run to do some laundry. I'll be back later.

Just because you're at home, like, you have, like... So, like, maybe you could throw some stuff in the dryer because it's like you're on the way... It's near your desk or or something. But I I think taking time out of your workday to do chores isn't, like, the definition of work from home. I think sometimes you can make that happen because it's, like, easier, or maybe you are gonna take, like, a ten minute break. But there are people that don't understand that work from home means sitting down and working at home. Yeah. It means to be, like, super flexible or whatever else. So I I see the argument, but there is nuance. There are people like us. Like, our entire team is remote, and we get stuff done. So...

Somebody asked my wife the other day. She's like, it must be so nice to have lunch with your husband every day. And she's like, I never see him, my husband. And, like, even when I first started... Like, we were first moved in together, she thought, like, yeah, we'll have, like, a lunch hour. You have a lunch hour, and you sit down and you hang out. Was like, I like one hand a sandwich on the stairs back down to my computer. I don't take a lunch break, you know? I'm just like, when I'm on, especially now with kids, you know, like the eight hours that I get to work is secret to me. And I don't have, I don't work weekends. I don't work evenings.

I need, when I have that time, I just want to go, go, go. I don't, I'm not taking a lunch hour and hanging out with people or doing laundry or whatever. Yep. I'm a little golem in my office on my computer, like, get

my work done. And then, again, like, my lights have to shut off at 05:00. Otherwise, I would would not even go in the house. So I, you know, I... I'm...

I actually do. I have them shut off because I was like... I would, like, look down and be like, oh, shoot. It's time for me to leave work. Oh, I need to do that. My wife will often text me and be like, it's 05:00. Where are you? 505.

Can you please come up? Yeah. Kids are killing each other. I I get the the, you know, the neck around the the... Or the thing around the neck from my my lights in my office from my home assistant. Alright. Time to get out of here.

Yeah.

And I I actually did talk to some people at JS Nation, like, in person that talked about, like, they really prefer going into the office because they can't focus at home. And I think that for certain people, that's that's that's fine. That that seems to be the way it is. Like, they can't find quiet or because they spend all their time at home, it doesn't feel like they can get work done there. So, yeah Mhmm. They're absolutely people that, like, need to go into the office to be productive. But, again, there's nuance. I don't think it should be a blanket rule. Yeah. I mean, I personally prefer to be in an office even though I have worked from home for the past ten years.

But there's benefits either. I mean, all my kids get sick or whatever. I can go get them. So, I mean, there's a lot. And and certainly, being in the the the realm of audio recording twenty four seven, being in an office is just not practical to me unless I did have an office that was just a studio. So Mhmm. From that point, it might just work for my studio. Yeah. That's all the links I have. What... Scott,

CJ, do guys have other stuff?

I have one fun thing to end on if that's cool. Okay? Let's get it. Get some fun stuff. This was shared over on r slash web dev by user Cookie Delivery.

They built a three d bookshop experience.

You can walk around, pick up books, read the cover, a sample, or the full book.

And I've thought about building something like this before, but for, like, for movies. Like, imagine tapping into, like, AI and DB API or open movie database and then making it look like a blockbuster.

Imagine that your Plex is just a blockbuster. Exactly. So so this is the app. You can you can, you can walk around.

And for any book that you you stumble on, you can... Woah. Gotta gotta control it here. But any book you stumble on, you can click on it. It's got, the summary on the back of the book, and then you can click in to read the preview. So this is like a similar sample that you might get on Amazon, how sometimes it shows you, like, the first few pages or whatever.

But then for some books that are, like, in the public domain, you can read their full contents like this. Pride and Prejudice.

See, I don't even like books, and I like going into a bookstore to pick up the book. So... Yeah. Exactly. Like, it's it's a fun little experience.

So, yeah, this is fun. I think, like I said, I wanna build this for movies.

I think, like like, what you were saying, Scott, is, like, this could have, like, your plex library or your jellyfin library.

Yeah. But then also, like, you could tap into some categorization APIs that give you... Like like, I would probably tap into Letterbox because Letterbox... People create custom lists of, like, top movies with aliens or, like, if they have, like, very specific lists.

So that'd be cool to see Imagine this, CJ. Imagine you tap into the Prowlr or Radar or whatever, and then you see something directly from there, and you say, oh, I would like to, download this Linux ISO and then have that in my personal library.

So this reminds me of, I think it was Shopify has an internal, like, poster board or like a utility pole where you can, like, staple, like, poster to it. And I I just look for it. I can't find it. I think it was just a video I saw on Twitter at one point, but I love those things where it's like real time and like anybody can go in and staple up a poster.

I just love those. Like, the cursor camp was another one where you could just, like, walk around and see all the other people. I I need to build something like that. Think those are so fun.

The other thing I'll say here... So I actually went to the Van Gogh Museum when I was in Amsterdam, and they have one of Vincent's paintings that I saw there. Yes. They have the fly. Yes. I did see that painting as well. That museum was fantastic if anybody's ever Yeah. So I think having an experience like this... I I guess people have probably built a virtual museum that you can go through, but I would almost prefer it because, like, I don't like crowds. And, like, I was in... The museum was great. To be to be able to to be able to see the paintings up close, awesome. But there are literally, like, 30 people all standing around with their phones out taking a picture of a picture. Oh, yeah. Right? Yes. Yes. Yes. Yes. If give me this digital Yeah.

Don't want a picture of a picture, but allow me to have a VR

experience. And then I can, like, zoom in 50 x times and like see the actual paint strokes and stuff. You can get your nose up in there. You could add some smell o vision and smell the oils, ancient Yes.

Yes. Yes. Yes. I actually really like this idea as a person who does like, VR and AR experiences.

So for me, this this kind of, hit something for me that I'm like, oh, this is this is kinda sick. So, nice. So this is fun. Great. That's all I got. Cool. Well, now I'm I'm really inspired to go build my, my blockbuster store because I I do think that would be fun.

Definitely.

Sick. Alright. Well, anybody else have anything, folks?

That's it. You wanna wrap it up, Scott? Yes.

Thank you so much for joining us for Syntax Live. We are here every single week, every single Monday.

Next time, we will not have any technical difficulties. I can promise you that.

And then check us out on, we have a podcast every single week. It's coming out on Wednesday, and then we do videos on Friday. And CJ has a banger of a video coming up that I'm gonna mention here just so he gets it out.

But that's all I got. Check it out. Check us out at sentry.io, all that good stuff. We'll see you in the next one. Peace. Peace. My