This video explores Claude Opus 4.8's new dynamic workflows feature, which enables agents to run hundreds of parallel sub-agents for larger projects, while discussing the significant cost implications of AI development tools. The hosts demonstrate building a pixel editor using HTML, Canvas, and WebGPU in a single session, highlighting both the impressive capabilities and the substantial expenses involved, with a 33-minute session costing approximately $15. The discussion also covers terminal management tools like Herder as alternatives to tmux, and addresses broader concerns about AI's impact on content creators and the web ecosystem.
深掘り
前提条件
- データがありません。
次のステップ
- データがありません。
深掘り
Opus 4.8! ⟡ Goodbye tmux, Hello Herdr ⟡ AI Killing the Small Web ⌁ Syntax Weekly ⌁追加:
What's up everybody? Thanks for joining the live stream. Claude Opus 4.8 just dropped and we have been using it for 3 minutes and we're going to tell you all about it.
Uh kind of exciting. We just like I've been researching all morning. I dove into the binaries of cloud code. I found some documentation on it and I was going to I was bringing that all to present to you today. But it just launched. They launched a blog post as well as this new thing called dynamic workflows. We're going to kind of go through it and talk about it. What's up? My name is Wes.
With me as always is Scott Tinsky and CJ. How you guys doing?
>> What's up? What's up? What's up?
>> Yeah.
>> Good.
>> Yeah. Excited.
>> I've been running uh uh Opus 4.8 now for two minutes and we're all cooked. Yeah.
>> Yeah. What's the is programming dead?
>> Yes.
All right. Wait, ETA for someone to say that this model sucks. I mean, I think the the last one had that within moment, so >> it's they nerfed it already. Yeah, it's definitely nerfed.
>> Nerfed.
>> Yeah.
Um, leave your comments in the chat. Uh, we've got the live stream going on here.
We can What? What? Focus. Is this thing working this week, Scott?
>> Uh, the ticker.
>> Yeah. Yeah. Yeah. I >> There it is.
I uh I rewrote it in uh I rewrote it in uh Cloudflare durable objects. So >> I told you to do that. And does it work now?
>> Um it works. Whether or not we have it uh loaded up on OBS is a different story cuz we didn't have time to test it because I just pushed an update just now. So yes.
>> Beautiful. Well, if you had Opus, it would have been working already.
>> That's actually accurate. Yeah, my my tabs be crazy right now. I'm just even like trying to find it like in the mess of all the stuff I have open in preparation for this show.
>> All right. Should we uh get into looking through the blog post?
>> Before we do, let's talk about what we're doing. This is Syntax Live. We're doing this weekly. Uh we run through basically three things. We each bring something that we want to share with you. This episode just happened to come out at a very crazy time just moments after this Opus model has been released.
So, we will be really digging into something that just came out for the first time like this. But again, we're going to be going weekly every single week to share with you all kinds of cool projects in this format. So, leave a comment below. Uh, hit up the chats.
We're on Twitch. We're on YouTube. We are on uh X, formerly known as Twitter.
Any of those places, leave a comment and we will get to it or we will help share your projects, whatever you want us to take a look at if it's interesting. So yeah, let's get into it. Opus 4.8, guys.
All right, so given the blog post a quick skim, I think there's kind of a few big things that they've released here. Obviously, first the new model is slightly better in several regards.
We'll get into that. Um, I think a kind of more interesting thing is they've releasing this thing called dynamic workflows, which is part of cloud code.
Um, that is used for tackling larger projects. I don't totally understand how this is different from just triggering a sub agent. Um, but we'll get into that as well. I do I do think that this is probably what Bun was rewritten in Rust for. They probably were looking for a a big project and they uh decided to rewrite Bun in Rust.
>> Um, but Opus 4.8, let's go through it.
Um, the table below shows how Opus 4.8 compares to its predecessor. So, um, Agentic Coding, what 5% better than than that? That's a a pretty significant bench. I don't know how much I I trust all these these numbers. What do you guys think? Yeah.
>> Yeah. I think the general rule is don't trust benchmarks put out by the the company that released the thing, but it's always nice to see the numbers go up and to the right, you know?
>> Yeah.
>> Yeah. I think we wait for like third party benchmarks to come out. There's some newer like like the open suite, maybe. I think it's called Yeah, >> we'll wait to see how it performs in in these benchmarks, not from the company itself.
>> The um what is it? Pelican on a bike as well.
>> Yeah. Yeah. Is it Simon Wilson? Yeah, >> Pelican on a bike. Man, Simon Wilson is great. Uh CJ, I was wondering if we wanted to talk about those benchmarks at some point in the show. I I haven't really gotten a chance to look at them, but uh might be at least a a good topic to have on our list to at least give a glance at at some point.
>> Yeah, we'll pull it up. I think Open Sweet. Yeah, we'll talk about it once we're once later in the show. Stay tuned, everybody. Yeah, >> let's keep going through this. Uh the terminal bench almost what 8% bump. Uh pretty significant financial analysis.
So, it's interesting. They they obviously are are showing stuff that's it's for like claude co-work as well, right? This is not just a coding model.
It's it's good for just computer use in general.
Um, let's see what else we got here. Myth misaligned behavior. Mythos preview.
It's less misaligned behavior. That's good. Just kind of keep it in line there. Um, I think this, let's talk about this dynamic workflow. So, dynamic workflows. This is a new feature available in research preview. Allows cla to take on even bigger tasks in cloud code. Claude can plan the work and then run hundreds of parallel sub agents in a single session. With Opus 4.8, the agents can run for even longer. Um I I have been noticing that with 47 and 46 recently is that it it tries to like >> be like no no no we can't do that. Like I was I've been like debugging these Bluetooth mesh lights. I've I've like reverse engineered the protocol. Um and there's many times where it's just like oh well like we just can't do that. And then I was like no like I have had to like keep pressing it to be like no this is possible. I know it is possible. Um don't just wrap it up and and use some sort of crappy fallback. So maybe it will be better in that regard. Um it then verifies its outputs before reporting back to the user. For example, cloud code with Opus 48 can now carry out codebased scale migr hundreds of thousands of lines from code kickoff to merge. Oh, I wonder where they tested that. Probably with fun.
You can read more about dynamic workflow available in cloud code for enterprise.
That's what sentry has. So we'll be able to use this today. Um teams and max plans. So they link off to the the other blog post explaining it a little bit more.
Oh yeah, rewriting bun. Okay. So I didn't have to and I didn't have to uh guess that they >> can we get a a vibe check? I think did did either of you get it running in cloud code or in the desktop app?
>> I got it running in the desktop app.
>> I have it running in cloud code.
>> Let me just prompt it. Let's check check the vibes. See how it's >> I I prompted it right here. This was before they announced the blog post. So it says what what do you do better? And it didn't know what it did better itself.
>> And I'll say you are now out.
>> Yeah. So we have a January 2026 cut off range for the knowledge. Yes.
>> Yeah.
>> So what what's something that happened in January 2026?
>> I I don't know. Is uh what about some real code stuff you got?
>> Let's uh ask it something about some kind of new project.
>> Can it use a new React use effect?
I like I I don't know that like I I could ask it to write an effect like we could do that like six years ago, right?
>> No, I Yes. Yes. Yes.
>> That's pretty quick. Um, interestingly, the fast mode is cheaper. So, they have this many of these models have fast mode which it runs at 2.5 times faster and generally it's like two times faster for six times the cost. And I don't know, like obviously it's really nice if it does it, but like who or what are you doing where you're willing to spend six times more? This is already the most expensive model out there. What are you doing that you need six times the cost?
>> I just always use the best model available. Uh and uh but that's I'm not paying with credits right now. I have the the max plan. So >> you don't care.
>> Yeah. until I run out and then I will uh do something else, I suppose. Yeah, I'll go outside.
>> Um, another kind of interesting thing in the when I cracked open the binary, it said it had a warmer voice. Um, and I'm curious, like that's like always been one complaint about uh, Opus versus the like OpenAI models is that it's like not very funny or not very um, not very warm. It doesn't have any to go back and forth.
So, it it looks like they've changed it.
Actually, I guess that's one thing we can >> we can do it.
>> I mean, I've always found the the Opus models to be warmer than, you know, Codex feels like you're talking to a robot. Codex feels like you're Yeah. on the slog and and and maybe that's because I used it in I used uh both GPT and uh Opus models in OpenClaw which is a very personality forward system, right? And >> Opus exceeded so much >> really at that compared to everybody else said the opposite.
>> No, I've never heard that before in my entire No, I've never heard that. I've heard the exact opposite of when you're saying everybody. I I I definitely when people say especially in the open claw community or the Hermes community, people say that uh very specifically that it's a bummer to have to use the codeex models or the the uh open AI models because of their lack of personality.
I I have to tell me a joke. Not funny at all. I personally run like a dad jokes coding repo.
>> Yeah. And it's crazy the amount of like pull requests for AI generated jokes I've gotten recently. They're not funny at all. Like the the way I tell if I merge something or not is if it makes me laugh. I haven't merged one in a while.
>> And how long have that repo been going?
That was before the dawn of like Gen AI, right?
>> Oh yeah. Um let's check it out. Dad jokes GitHub West boss.
>> By the way, Wes here from introverted bedroom on chat. Nah, dude. Claude has always been better emotionally than Chad GPT models. Ask my guy inverted bedroom knows what's up.
>> Yeah. Uh >> uh 12 years ago I did the first commit >> to dad jokes repo and yeah those are just so many poll requests guaranteed like first time contributor.
>> It's funny. I mean this is a problem in the greater open source ecosystem but your dad joke repo is under the same attack >> world but they wouldn't give me the source code.
>> Yeah.
>> It's not funny at all.
>> It's not funny.
It's It's unfortunate if these aren't AI generated. You just say like not funny.
>> Everybody, please go thumbs down this right now. Oh, this was this was two years ago.
So, we we do have we've got maintainers on the project since it's so big. So, they must they must close them.
>> Nice.
>> Um, cool. Well, I'm interested to check out the dynamic workflow type of stuff.
I don't know. I have a project that >> Well, I'm I'm curious.
>> That big.
>> There was a chat. Let me find it. I think this would be a fun prompt to leave running and then maybe we come back to it. Let's see. Um, yeah, from Charles 999. Ask it to build a fully functional Photoshop clone in a single HTML page >> in a single prompt. And then we'll come back to it in like 10 minutes or so.
>> Do Do we want me to do that?
>> Um, I don't have 48 on my cloud code yet. So, maybe you you forced the update, right? All right, Scott will run it.
>> Okay, >> let's let's see what else chat's saying about this. I haven't been keeping >> Where was that chat? Where was that?
That was on YouTube. Okay, I'll just copy and paste that prompt exactly.
>> Let's hope Opus 4.8 isn't $30 per million token. No, it looks like it is the same as 47.
Um, yeah, pricing for regular usage is unchanged from Opus 47. $5 per million inputs, 25 million output. Pricing for fast mode is $10, which is like 47 already, which is like stupid expensive.
>> Imagine what these >> functional Photoshop clone.
>> I There was an article the other day. I was like, like where's the vibe coded Photoshop clones?
>> I'm curious what you guys think about that.
>> I'm not going to say a Let's do something other than a Photoshop clone.
Photoshop.
>> Let's do Photoshop. I I think it's a curious task because it's very visual and like >> in order for it to test itself, it needs to maybe like spin up a browser or like control the mouse.
>> So, yeah, I think >> that's what it's supposed to be good at.
>> Okay. Yeah.
>> Computer use, right?
>> Yeah.
>> You're just going to burn all my all my uh all my usage on like a Photoshop clone.
>> Uh let's make a pixel editor.
>> Not a pixel editor. No, like an actual useful piece of software.
>> Push it to the limits. Like we're replacing programmers single >> every single every single possible feature.
>> Chad, give us a good prompt for this that's actually reasonable.
>> You're being too nice, Scott. You got to Yeah, I don't know.
>> I'm protecting my my tokens.
>> A Photoshop clone. Photoshop got all kind of stuff in it. It's and every time one of these these things comes out, I test it and try to build something and then you get all these ass hats come out being like you can't you can't one prompt stuff and expect you to like know if it's good or not. And like I like that is true, you know, like this is not about like oneshotting things. This is about taking complex infrastructure that has been around for 10 years and being able to >> Yeah.
>> Uh take on such a big project.
>> Totally. So maybe make your own Linux distro or something.
>> Make my own Linux distro.
>> Scottics. What would a Linux distro be that is made by Scott >> Alpha?
>> Yeah. Everything is >> everything's cutting edge and breaking all the time. Yes.
>> Which is not that far off from most Linux distros. Maybe not cutting edge but breaking.
>> Um on the same thought, I have another another question for you guys. So, I'm learning um Da Vinci Resolve right now, right?
And I bought like a like a paid plug-in for Da Vinci Resolve to do some stuff and I wanted to tweak it to be something different. So, I like I like decompiled it. It's it's a zip file. You like they have like a special format, so I didn't really decompile it. But, if you rename it to dozip, you can open it up. You can see all the code that's that's behind this paid plugin. Um, I then took that I dropped it into Claude and I told it add these features and like I'm I was just thinking like like I can't release that obviously like I can't I can't put that up on GitHub cuz like I just like took somebody else's work but like is that is that okay to do?
My opinion is like if you keep it for your own personal use, you're not making money off of it.
>> Yeah.
>> Yes. Also, not even not just making money off it, like because some people might do that and release it for free, but now you're basically taking income from the guy that did all the hard work and made the original plugin, but for personal stuff, I think absolutely. I think >> Mhm.
>> I'm all for it for personal stuff. I think that's going to be a bit of a problem is that like people will take paid software um you can this is a show I want to do is like reverse engineering apps because there's there's uh lots of apps where you can take like a desktop app or um what is it called gear g i r da um and there's another one for like Android apps and you can like decompile it and like look at the the >> the function calling >> um like I think that's going to be a bit of a problem where you can simply just take a paid app >> give it to it and say, "Hey, rewrite this thing from scratch, but like make sure you you copy all the features." I think that's probably why a lot of these these apps are going more towards like server side, like it hits an API and to actually do a lot of the work.
>> Yeah, I'm pretty sure Zach from Bite Dance was uh decompiling his uh Robin Hood app and then using an unofficial API that he constructed to control Robin Hood. So, yeah, all kinds of wild stuff.
No, I I I've thought about this lately, too. Like I actually use an open source Reddit app on my phone, >> but there are lots of features that I would like that now I can just ask AI to like vibe code. And it's not it's I'm not working on a product. I don't necessarily worry worry about like the code quality. I kind of just want these features to exist. So that's absolutely something that I would I would fork their codebase. I mean, I think that's maybe that's the issue though, right? is like in the past if I did want to spend the effort on this stuff, I would try to make my contributions reusable for others and I'd try to package them up in a PR because it is an open source codebase, >> but I'm now more incentivized to kind of just like work on my own stuff, not worry about quality or worry about what it looks like to others because it'll just work for me.
>> And I think that itself will be an issue. So >> yeah, let me throw up on the uh on the screen here. Let me throw up. I have I have uh um I have GPT8 or GPT8 >> my Oh yes. Um I have it running in clawed code here and I told it to do a couple of things to use type GPU. I told it to use uh HTML and canvas. I you told it to just use HTML like no JavaScript frameworks and it it I mean it went off and started doing tool calling like normal like you'd expect. It's loading skills. Um it it doesn't feel right now any different than anything. So um I think what we'll do is we'll come back and and check in on this a little bit later to see what happens. And and where the hell did that go? Uh, I'll report I'll report back as to uh what the status of this ends up being and how good of a job it does. So, we'll see.
>> Nice.
>> Yeah. Um, cool. So, folks, we are talking about Opus 4.8 right now. We're talking about GPT8, whatever the heck I said before. We're talking about Opus 4.8. We're talking about different projects. Do we want to get into our actual projects now, or do we want to hang on uh Opus 4.8 for a little Let's uh let's see what you guys brought.
>> Okay. Opus 4.8 was yours. Cool. Well, uh mine's already open here. So, I'm using something called herder, which is doing the classic Oh, the the classic thing of removing the eer. So, it's her herder like with an R herd. And uh man, so Herder what this is is an app that really uh replaces t-mucks and it lives inside of any terminal. So it's a terminal utility and what it does is it is T-M essentially but with a thousand nicities and a thousand nicities kind of crafted around working with agents. And I found this to be really super good. At first it was one of those things when people suggest me something I'm like oh I'm using T-Mox and everyone's check out harder check you know it's like okay maybe I won't because uh too many uh people are bugging me about it I don't that's a character flaw of mine I will >> if people know about it it's not obscure enough for Scott >> that's unfortunately all too true. Yes I've already lost interest and uh so I gave it a try and I got to say I love it now I I started using T-Mox only like a couple of weeks ago. So, terminal folks who are big into T-Mux and all this stuff, let me just tell you, I didn't get it. I'm like, my my terminal's got tabs. It's got tabs. This can split the window. Why the heck do I need this?
Well, I moved most of my development to this Mac Mini over here. And uh I started using something called Mosh instead of SSH, which is basically just a drop in replacement for SSH to mosh into that computer. And then I start up a new herder session. And herder functions very much like T-max where uh you can create a session like this session can run on this computer. I can close it. I can do whatever and then I can come back later and reattach to that session and it's just running. And the beauty of having it run on another machine is that I can prompt like crazy, close this computer, come back, open it back up, and because I'm using Mosh, uh it re just reconnects automatically. and Herder just has all my sessions going so that I can close this or come back to it. And so if you're wondering why T-Mox in the first place, having that ability to to reattach to sessions is like really what drew me to T-M rather than I always thought the cell was, oh, you can split tabs and stuff like that. Um, okay. So, where Herder shines is that T-Mox by default, I I know T-Mux people are going to hate to hear it. It sucks.
Uh the the clipboard saves your your copy and paste to a buffer. You'll never convince me that's something that I want. I don't want a a buffer. I want I know the Vim folks are going to be angry about that. I don't want a buffer. I want it in my dang clipboard so I can copy and paste from app to app. And so I I was getting really annoyed with like the copy and paste functionality of the default I will say the default functionality also like out of the box T-max has no mouse support and just general stuff it it it sucks to use uh for somebody who who's not into this stuff. Um and then also Wes you you'll you'll relate to this T-Mer for that matter. This is a thing I don't like about Herder either because it's an app within an app, right? It's a system within a terminal. Your keyboard shortcuts always have to be like uh what is a prefix. So I have to do controlB before I even have access to any key bindings. So oftent times keybindings are like a two-step process like B shiftnr and and it like drives me nuts.
>> There's got to like like warp has like a t-mux option you just turn on, right?
there. You got to be able to like get past that. And that that's maybe just like a >> Yeah, I'm going to fix that. I'll do some bindings.
>> Integration.
>> Yeah, I'll do something there. I just actually got a uh this is so funny.
Earlier today, um one of the Raycast devs was talking about using a foot pedal. I ordered myself a dang foot pedal. So, I got foot pedal coming. I I I'm going to be It's got three clickers on it. So, I'm going to be use I wish I had three feet, honestly. I could >> This is the Elgato one.
>> It is the Elgato one. And when when I was looking at foot pedals, I was like, we got to make our own foot pedal. We got to design. These all suck.
>> You know what I'm doing?
>> Metal hack week.
>> Go ahead. Go ahead. Yeah, go ahead.
>> Century hack week. This this I'm going to get like a like a Mac truck manual gearbox with like a like a old boys like nice hot ding. And if I want to like like you you put it into different modes, you know, like Vim, you got to like you got to hit the Jake brake to exit. That's I'm for sure doing that.
How sick would that be to have like a like a gear selector from a truck?
>> Here's what I'm doing, Wes.
>> Not a gear selector. I'm going to have I'm going to get a bunch of broken old guitar pedals and I'm going to swap out the internals and I'm going to put them in like a guitar pedal uh platform like they have. I'm gonna stand up. I'mma put stomp on it like I'm uh changing my uh effects. I'm turning on my dis.
>> You don't even have to take the internals out because like like hold on like what does a guitar pedal do? And we might need producer Randy to chime in here.
>> It takes an audio in and audio out. So I you know they're they're analog typically.
>> Yeah. But like couldn't you take the analog system? You could use them as almost like onoff switches because like that signal is going to travel over just a regular audio cable >> and then you could have like a little circuit that converts from an audio cable to like USB or something like that.
>> I'd rather just doable.
>> I'd rather just swap it out.
>> Uh yeah, >> what I was going to say is I've been uh finding at I go I do a lot of thrifting and I've been finding foot pedals at thrift stores. I have like three. Um I haven't used them yet, but I found them.
And I was like, "This this is really cool." Cuz they're USB and the ones I found are like like made out of metal. I think a lot of um uh what do you call it? Like uh people that do transcription work or like work in offices.
>> Yeah.
>> Use them a lot. But yeah, hit up thrift stores cuz uh I didn't even know they existed and then I found it. I was like, "What's this weird USB device?" But yeah, it's like three pedals. It's like really hefty and then it just plugs in via USB and it can do mouse clicks or I think you can get a custom app to change what happens whenever you click on something. Oh, that's sick.
>> Pretty cool.
>> Yeah. Yeah. You know what? I uh I told my wife I got a foot pedal to to control my computer with, and you would very predictably believe the look she gave me when I told her that. She was just like, "Oh my god, what what is happening?" Um, so, okay, back to Herder. Herder is dope because, like I said, you can click on your mouse and and move uh to your sessions. So, instead of having tabs, you have your spaces, right? But then you also have tabs if you want tabs. So you got tabs and spaces, but not the uh the kind that you have in code. Uh tabs and spaces. The the killer feature of Herder beyond that that it's it's nice T-Ux with mouse support like I I can drag and I can drag and move this sidebar around. No. Uh the killer feature is really when working with agents, uh it has this down here of your agent view in the bottom left and that shows you the current status of all of your agents, whether they are currently working or idle or blocked. So there's a red status light if it's waiting for your input. And anybody who's worked in this stuff knows that if you have 10 agents running in various tabs or whatever, you're constantly kind of going to check on them. Is this one done? Is this one ready? Is this one?
And I found even if you just instead of using tabs at all, just keep spaces open. I I would keep I would use a tab for like maybe running the dev server.
Um, and then uh keep each of my agents kind of in here because then you will end up having all your agents over in the left sidebar and it just tells you their status at all times. So I have a herder that's running on my Mac Mini back here and then I have one that's running on my personal computer right here. And uh that's it. I got one on my personal computer, one on my Mac Mini. And that's my entirety of my tabs in my uh iTerm.
So, I don't have multiple windows, multiple whatever. I just got two tabs in my I or it's not even iTerm, it's a ghosty and it rips. I got to say it's made me a believer. Now, I still hate the keyboard shortcuts. It's something I'm working on fixing, but I have only been using this for two days and I really, really like it. And uh I got to say maybe I can dive into a little bit of my setup here beyond this. So, right now, what am I running on this bad boy?
Oh, I'm running uh let's say I'm running Claude Opus 4.8. Hot off the presses, folks. I'm running that thing. And I got on the Mac Mini West. You'll like this.
I have a Caddy setup where uh Caddy is is uh using with tail scale and I have a domain pointed at my tail scale IP. So I have sites like this. This exists running on the Mac Mini and this is the dev server. So when I am working on this app, whether that's agents or whatever, I have a real URL I can connect to anywhere in my tail scale network, my phone, my computer. Beautiful.
>> Um when I'm going on my uh trip, we're speaking in Amsterdam in a couple weeks here.
>> I don't need a strong internet connection on my computer anymore. This bad boy is wired to the internet and this bad boy is the one doing all of the actual stuff that needs a lot of connection to the internet where when I'm typing into the box with Mosh and Herder, it's like that that's hardly going to be using any network. So, it's going to be way easier to work on the plane. Um, so that's going to be sick because again, this bad boy is doing all the stuff. And uh I got to say I hope this >> client man, we're so back to the thin client. It makes me want to get a MacBook Air. My wife's got a MacBook Air and that thing rules, man. It It takes up no space at all.
>> Man, that's I I've was looking at the MacBooks the other day cuz mine's starting to chug and I'm waiting for the new ones to come out, but every now and then I think could I could I use like a Mac Studio? And like I often work from the car when I'm like waiting for kids stuff and I often think, hey, I hate I I love being able to like close my laptop and then open it in the car and I have all my stuff there. And like for coding, I probably could do that. For video editing, I don't think so. Maybe not yet.
>> No, I think for me like media is what keeps me with a beefy laptop because I do like music production and then video editing >> and that stuff's a lot harder, if impossible to do remotely. Do you Yeah, but do you do you do that on the go?
>> Yeah. So, like with my laptop, I do live looping. It's like I I bring it to a friend's house and like plug my stuff into it. So, that needs to be a little bit more mobile.
>> For video editing, that could just be a a Mac studio at home.
>> Yeah.
>> Yeah.
>> But, uh just on on the topic of like terminals, I I I never became a T-Mux user. I was never much of like using Terminal for everything. Um, so I have a similar setup with like tailscale, but I just use the open code web UI.
>> Um, and then that's ex once I'm on my tail scale network, I can just access that in the web browser. So, um, seeing you use her, I I might try it. It's interesting like to be able to like you basically can do what I'm doing, but I'm just doing it in the web browser because those apps expose what they do in the web browser versus having like a terminal interface for it.
>> Um, >> yeah, >> it it's smooth. Um, I Sorry, minor feature of Herder that I forgot to mention is that just like T-M or other things, you it automatically copies when you select text. So, I love that feature. That's a small little feature that I love. You just select the text and it automatically copies it to your clipboard.
>> Yeah, I feel like um Open Code CLI does that. I don't know.
>> Oh, a lot of things do. I just like >> Yeah, I don't know if they're using like Open 2 or something like that. I will say just looking at the interface like it's pretty cool that you have tabs like it keeps track of your agents. That's something I haven't seen in in a CLI before.
>> Yeah, I'm I'm behind the times on CLI. I don't use warp. I >> I'm I'm behind the times, too. And I I largely don't like twoies. I'm a gooey guy uh through and through. But you know what? This has got a gooey. I can click the menu button. I can click the settings button and I got a dang settings. Oh yeah, I got a settings in here. I can change my settings. And there's integrations into any of the stuff so that way um you know it can report correctly what's going on. I do I I do like this. And there's even sound alerts, folks. I actually had sound alerts on before and it was annoying before while we were recording here, but um really sick. Oh, look at them themes.
Look at them themes. I like terminal, though. There we go. That's a little >> nice. Can we check in on your uh uh claw opus 4.8 still spinning? Yes, it's still >> going for 14 minutes. Wow.
>> It's been going for 14 minutes. It's honestly only bugged me a couple times.
That's a huge complaint I have about cloud code. It's like trying to run bash scripts instead of uh using its own read file capability tool and stuff like that.
>> It's using jQuery.
>> Is it using jQuery?
>> Oh, no. It's it's just it's short forming the dollar dollar sign to document.query selector cuz like jQuery doesn't have text content.
>> Okay. made its own jQuery.
>> Well, it's made this so far. Um >> Oh, >> let me even let me even give this a refresh. It's got zoom in the canvas. I just did that kind of intuitively. So, >> yeah. So, if you're just tuning in, we we were trying to somewhat oneshot creating a a a Photoshop clone using Opus 48.
>> I told it a pixel editor, not a photo.
>> Okay.
>> But I I made a pixel editor for my hack week last year. We didn't even have Opus. I'm going to tell you that the challenge that I gave it is more interesting because I told it to use two things that are one one is after January 28 HTML and canvas to and then type GPU to add shader support on top of a pixel editor. So this isn't just make a pixel editor. I did that day one of React in 2015. So yes, you just you just you just hold your horses here. We can add a what is that even >> is it using your graffiti? Like I like the style so far. Like did you tell it anything about styles or do you have any skills for sure?
>> I didn't tell it to use anything. There is a chance that it did use graffiti because I have graffiti in my my system cla but I don't think this is >> no dark blue.
>> Yeah >> boxes with a gray outline.
>> Font choice is cool.
>> WebGPU is not available in this browser.
That's a lie. That's a dang lie. I bet that it just hasn't written that part yet.
>> It usually it usually implements the like UI first and then wires it up, which is >> kind of interesting because I would do the opposite.
>> Yeah. Well, we'll see. We'll come back to this in a little bit. We'll see when it what it what it cooks up. Oh, it's bugging me. Okay. Oh, it's popping open DevTools MCP, which is not going to have HTML and canvas support. I better warn it. Yes. Um it Okay. All right. Cool. Uh while I'm doing this, uh would somebody else like to uh take on actually maybe CJ hit him with what we're doing here and then uh you can get into what we're talking about with yours.
>> I got you. So everybody, thank you for tuning in. Uh we're live all over the place. We're live on Twitch, YouTube, and X. And this is our weekly live stream. This is the third one that we're doing uh that we've done so far. Um, and starting next week, we're going to go back to live every Monday. Um, but Randy, if you want to pull up my screen, if you you watching right now want to want to see us live and in person, uh, we're going to be in Amsterdam on June 10th. You go to syntax.com, there's a banner up at the top or syntax or sorry, syntax.fm, there's a banner at the top or syntax.fmup.
That'll take you to a page where you can RSVP and uh, we're going to do a live show. We'll also just have some networking. Um, there'll be free merch.
If you're going to be anywhere near Amsterdam, you should come see us on June 10th. That event's happening from 6:00 to 9:00 p.m. syntax. Meetup. And this is in conjunction with JSNation and Reactsummit. So, JSNation, one of the biggest uh JS conferences to happen in Europe. It's happening the next day. Um, and so you can get a ticket for that. If you use code syntax, you'll get 15% off.
And then Reacts Summit, which is also one of the biggest React conferences happening the day after that. If you go to reactsummit.com, get a ticket, use code syntax, you'll get 15% off. There's also like a joint ticket you can get that combines both of them for a cheaper price. But this is one of my favorite conferences. I'll be MCing both days. Uh Wes is giving a talk at JSNation.
Scott's giving a talk at React Summit.
Scott's MCing with me at JSN Nation.
It's a it's a big old fun time. Come see us in in Amsterdam.
Yeah, it's going to be a fun time for sure. Yes.
>> I'll also mention uh if you like cool stuff and like all the stuff that we're talking about, uh subscribe to our newsletter. So, this is a bi-weekly newsletter that comes out if you go to syntax.fm/snackpack.
Throw in your email. We don't spam you.
This comes out bi-weekly, like I said, every other week. And it just includes cool links. Uh so, stuff like we're talking about today will include sometimes include just fun cool sites that you can check out, latest news, that kind of thing. So, syntax.fm fmn snackpack.
I think that's all the stuff to share, right? Anything else?
>> Yeah.
Um, no, you know. Oh, yes.
>> Oh, yeah.
>> Yeah. Ry's in our ear right now. But everything we do is brought to you by Sentry. Uh, if you head to centry.io, you can learn more about it. Uh, they are the world's best um, uh, application monitoring, error monitoring platform.
Basically, throw it on your app that's running in production. if anything goes wrong that gets logged back in the Sentry dashboard and you can uh have a full trace of what was happening on the back end, what was happening on the front end. Um they also have session replay so you can see exactly what a user was doing when something went wrong. We use it on the syntax site.
It's fantastic. If you go to century.iosax, you can also get I think is it two months free on the team plan?
>> Two months for free on the team plan.
Yep.
>> Yep. Yeah. So we love Centry. We we couldn't do this and we wouldn't be doing this without them. So, thank you very much to Sentry.
>> Sick. All right, CJ. What What did you bring? You brought a whole bunch of stuff to this.
>> I brought a lot of stuff. I I'll I'll bring up one thing. I think Wes probably has quite a bit to chime in on, and then I've got a whole bunch of links that we'll not talk about a ton for each, but some some fun stuff to share. So, first tuality tuality.com uh was a blog that has now been shut down. Um I'll just read this to you because it's very short and then we'll get into what what does this mean? And this is in the title uh AI is killing the small web. But it says, "Dear visitor, two things happened recently.
The income from my book sales went from being enough for me to live off of in 2024 to zero in 2026. The traffic to my blog and my books, which were free to read online, increased beyond what I could currently afford. Virtually all of it comes from AI crawlers. So, there's no ad income. Therefore, I'm taking my blog and books offline so that I can decide what to do next, especially with regard to AI companies stealing my work.
>> So, uh first of all, if yeah, if you don't know who Tality is, you've probably seen this website. This was his his classic blog, uh Axel Rashmire. He's done tons of really great posts on like intro to JavaScript, diving deep into like JavaScript internals and he's published several books. Um I will mention that his books are still for sale. You can't access them for free online anymore, but if you go to payhip.com/rousma, you can um buy his books if you want to support him, >> but this is really just kind of like a sign of things to come, right? like it in in the past it was possible to have a blog to basically be your own self-publisher and make enough income off of ad revenue, right? Because people would find you via search engines or maybe people share and then uh they get ad revenue when you visit their site.
But with everyone leaning on AI instead of using search engines, AI is basically just regurgitating answers from his blog or almost even regurgitate regurgitating info that he published first or that he published in his books and no one like the AI companies aren't reimbursing him for that, right? And so this is one of the casualties of of of AI where um it's not necessarily like copyrighted material like like books or movies, but it's it's the indie web. It's people just make being able to make a living off of publishing stuff for free on the web, but now AI companies are making that not as possible. So I'll pass it over to you, Wes, because I know you have a lot of online content stuff like this, too. I don't know if you have any thoughts on this or Yeah.
>> Yeah.
it it's too bad um that this type of stuff happens because like as as an author of this type of content, you know, like I've spent thousands of my own dollars having people take my videos and um turn them into blog posts. I've written many blog posts. I've put tons of content out there. Um, and like I see it as well like the the traffic of people to my my uh site is is going down and and that in turn means that people don't don't buy as many courses because they don't either don't necessarily need to to learn that thing because I could just prompt the the thing for it. So that's a shame. Um, I don't know that I would take my stuff entirely offline. Um, because I don't know that I I think it's it's cool that he's doing this and and I applaud him for sort of like stepping up for it, but I don't know that I would would do that myself.
Um, but I think it's just like I I don't know that it you could also change this either. uh you kind of just need to figure out a different way to make money in the current landscape. You know, >> I think a new model of some kind, right?
Like a lot of people are using Patreon, so you can get direct support or you have like a paid Substack. That seems to be >> the way to get your audience to support you when you're putting out like free or premium content like this. Um but it it is a shame that it's kind of a sign of the times. Yeah, I mean >> Charles um >> Yeah.
But yeah, >> I asked Chris Coyer about this. Um, and he wrote a post on his own blog about it because I I essentially said like I I I kind of worded it incorrectly. I said, "Oh, did you get out of CSS tricks in time?" Um, meaning that like like he sold CSS tricks and that was like maybe like a year or so before or a year maybe two years before a lot of this stuff happened. And um and he had some really good points of like I like yes you do this for money and obviously you need to support yourself but I think a lot of people do it simply just because they love sharing this information. Um and that's what the early web was. The early web was simply just people who really love this stuff and we're still sharing it and you're still seeing people who really love this stuff and are are sharing it out. Um, it does like like if we're looking at like a positive of this, like I hope it like SEO had had ruined the web before all of this AI stuff happened as well.
People were simply just putting out content for the sake of putting out content and it really hurt a lot of the like good stuff and you're not able to surface some of the really good stuff as well. So, I think that's maybe a bit of a a positive on there. Yeah. Well, I'll tell you what I pay for. I pay for Patreon of podcasts that I like um and enjoy and get value out of because I know uh these are real people. I had a real big pet peeve about like celebrities with podcasts and stuff like this. Get out of here with this. Like we we don't need the money. You got all your other stuff. But like if there's real people doing funny or interesting or or highly, you know, whatever, like real podcasts, real shows, I'm like very happily subscribe to those on Patreon every single time.
>> Yeah. I I want to push back against some comments that we're getting in the chat.
I'm going to I'm going to get up on a get up on my soap box here, though.
>> All right.
>> Yeah. Because like the mentality is like okay like so yeah this this this comment says um the question is how to monetize in a different way for people to have content because nothing changed uh give people the info people use it the medium changed now it's AI I think sure I think that is that is one one thing to do about this is find a different way to make money because the world changes I think um right like like you were saying like when SEO became a thing people had to change how they published content and how they got people to look at their content but I think all all of this conversation is missing the point that these big AI companies are basically training on this stuff for free.
>> Yeah, this stealing his work. He did not give them permission, >> right? They're stealing it and they're the ones that are making the money. And basically, we're shifting the conversation like, oh, figure out a different way to make money. But I think the conversation also needs to be like if if AI is training on your stuff, it you need to be compensated in some way.
And I think especially like we're seeing this even with like the the new release of how AI the Google Google search with AI where they're almost like hiding the sites like in a lot of people that are using the latest uh Google uh AI search instant search results. They don't click through to sites anymore.
>> Google summarizes it and then that site doesn't get traffic anymore. Yeah. So, >> I do think that it's causing us to have the wrong conversation like like we need data providence. Like we need to know when an AI is giving an answer, where did that come from?
>> And that that's the issue. I I do agree like we could find other ways to monetize, but I I don't think we can just gloss over this and say like change your ways. Like they need to be held accountable. That's all I'm saying. I think there there there are like Brave tried to do this when when the crypto thing popped, which is like if you visit a website, you put like $10 into your Brave wallet and then it'll take that $10 and split it up amongst everyone.
And that never really took off. And uh now there's this new thing, X42, which is a sort of a standard called payment required. Um here I'm sharing my screen right now. Randy, if you flip that on.
X42 is an open neutral standard for internet native payment. Um, it absolves their internet's original sin by natively making payments possible between clients and servers, creating a win-win. So like ideally what would happen is similar to AdSense whereas every time an agent scrapes it, I don't know, you're paying 200 bucks a month for for Claude or whatever. Imagine 50 of that would go towards the the websites that you scraped. Like similar to how like AdSense worked, whereas like you had got more traffic, more people enjoyed your your site, you would make more money. Um you post a really good um reel to Tik Tok, then you can you can make a couple thousand bucks if you have some like good insights and info.
Obviously, there's a big problem with a lot of that being gamed as well, but like wouldn't wouldn't that be nice?
Will that ever happen where like, oh, a bot scraped my website because I I spent three days writing about JavaScript closures and now now cloud code just scraped it and put it in their training data. Wouldn't it be nice to get a couple bucks from that? Do you think we'll ever see something like this?
>> Do I think we'll see it? No, they don't care. They're only going No, they're only going to do it if they're forced to do it and they're not going to be forced to do it. So, no, we're not going to Yeah. Yeah. There's 0% chance of that happening in my mind. They like money too much. And honestly, they probably uh because of the the cost of all of this, they're probably not exactly raking it in in terms of uh you know, they're they're subsidizing everything already.
So, I would love to see the like shady stuff that these companies do to get training data. Um because like obviously they're using Google and looking for websites, but there there's a lot more, you know, like I even look at the traffic of people watching my courses and videos and I'm pretty sure some of those are bots. Um, and there they probably have these like autonomous things that go and sign up for every single closed doorored forum and course and every book and every PDF on like a small town you can only request, you know, like I'm sure they have all these different systems for getting access to information that is not widely available on the web because that's that's very valuable. So, I'd love I'm sure they're very tight lipped about how how they do that, though.
>> Yeah. Yeah. Word.
>> Yeah. So, something to think about. Um, >> I don't know if we should call attention to uh people trying to blame us in the comments, but I'm going to do it anyways. Uh, stand out says, "I'm not sure why I click on syntax. You guys sound like junior devs arguing about new coding standards."
>> What do you guys like to say about that?
>> Sound like junior devs. Hold on.
>> No chance that we got a sunglasses guy.
>> That's the the the com worst comments always come from wraparound sunglasses guys. Yeah, >> except for the thing. Well, I'm sorry, wraparound sunglasses guy.
>> I'm not a new dev. I'm not a junior dev.
>> Yeah, I have a full head of hair, but I have more experience in coding than you do. So, let's keep moving. I don't have a full head of hair, but I still have more experience.
Uh, cool. In other news, um, I have a couple of stories here talking about, uh, what MPM is doing to try and stop, uh, supply chain attacks. And one of the things they've done is, uh, oh, I need to share my screen. Here we go.
>> There we go.
>> Here we go. Here we go.
Is uh, they set up uh, stage publishing for MKM packages. So basically you can set things up so that if you're using a GitHub workflow to publish your package um which does it automatically like when you merge into the the main branch this instead will put it into staging and then you can use two-factor authentication to approve that publish.
And so this is one more gate to prevent automated publishings when things happen in CI or or other stuff like that or even if a token gets stolen. This would be one more step because you'd also have two-factor authentication before the final publish. Um, so they released this. I encourage pretty much everyone to um do this for their repos uh because it's just one more step to prevent your your your stuff from getting automatically published. Um, and then one other thing that's currently a request for comment is making install scripts optin. So, this is currently a thread over on GitHub. And basically, I think all of the other tools like PNPM and yarn and bun, they don't run uh pre-install or post-install scripts by default. You have to like approve them.
>> And this is just a request for comment to make that happen by default in the npm CLI tool. uh and that'll basically make it so that if a package does get compromised, if its execution mechanism is via pre-install or postinstall, this will stop that from automatically running. So the question with this is will the the agent simply just do what it needs to do to make that thing run because I think what's going to happen is that oh this doesn't work therefore let me post install run this script.
Yeah, >> I do think well obviously yeah you have to be specific about okay do I want to run this pre-install or postinstall script.
>> Yeah, >> and if you're letting the agent make that decision then yeah absolutely that could happen but then that also comes down to well maybe you should be running your code in a container or in a VM that where if something bad does happen then it doesn't compromise your whole system.
Uh but you're right like you still at the end of the day have to decide if you're going to run it and if it was compromised at that point you're poned.
So yeah, I think it's one it's one more one more check, one more step in the process >> uh to prevent things like this happening and and I think that's kind of all they can do. Like just the past year we've just been floundering in supply chain attack news >> and they've got to do something. I think stage publishing >> is a good way to go. I I I think that like the top I don't know any package that gets more than x number of downloads and installs in a day probably needs to have this enabled by default.
Like whenever uh chalk got compromised and debug got compromised like those are some of the most installed packages on the entire ecosystem and there was no checks and balances for publishing new versions. So when the entire ecosystem depends on a package, it should have more checks and balances to prevent stuff like this from happening.
Yeah.
>> Yeah. Yeah. There's that. Enable it if you haven't. But yeah, >> this is one of those things I saw happen and was like, "Okay." Um I don't I don't Okay. I don't I don't know. Like Wes said, I don't know if this is going to move the needle or what. It seems like a good step. Seems like something that's nice to have. But yeah, >> certainly needed to happen. Yeah.
>> Yeah.
>> Yeah.
>> Yeah. Man, these uh these supply chain attacks are >> I don't know. I don't know what the what the actual way to fix any of this stuff is.
>> Well, like it certainly like I think this is one step. Another step is to do something similar to like what Socket is doing where like >> just run AI against every single thing that is published. It's probably not cheap, but it certainly needs to happen.
Um, every single time something is published, can we just wait 2 minutes for you to like run your checks on it before it gets published? It's probably a bit of a pain in some regard, but >> like it seems like these the socket and there's a couple other companies similar to them. They seem to be catching a lot of this stuff. Not everything, but they're catching quite a bit of it. So, why is npm not just doing this themselves? It's pretty fascinating because yeah, it's cool to see like Yeah, I think socket.dev and uh Sneak and I think it's Whiz is the other one like they publish blog posts like 10 minutes, 20 minutes after these things happen because they have automated systems that detect >> uh malicious packages that got published. Um and so yeah, I mean at this point npm is owned by Microsoft.
They have the money to do it. They they might as well be running a similar system. Um yeah, it it's unfortunate because like it used to like before all of these checks and balances, you know, it used to be like publish anything, but >> yeah, >> now people are realizing that they can take advantage of that. And um >> it it probably is time for Microsoft to step >> npm can't be cheap to to run either though. you think like I I downloaded the claude binary and it was I think it was like 400 megs or or megs or something like that and like there's some big stuff on npm and the the the bill the bandwidth bill alone for npm must just be insane.
I have to think about that with like hugging face, you know, oh, let me just download a 90 gig model, you know, who's paying for >> stuffies, I guess.
>> Every single time I go to like the playground of any of these apps that aren't charging me to like try out a new thing or something, I'm like, who is paying for this? That is crazy. Yeah, >> we are with our with our data, I guess, with >> and the VCs are.
>> I don't think anybody's making any money yet.
>> All right. In other news, uh Express has gotten a facelift. Um and uh you all you maybe you can tell me what this design this I see this design and I think Versel this is like >> I mean props to them though because the ExpressJS website has been the same for years and and they've they've gotten a facelift. They have new docs. But I will say the design just looks like every other >> and the sidebar the sidebar. What is going on with that sidebar? the spacing >> the sidebar is functionally excellent.
>> No, well >> to be able to click try click to any method.
>> Okay, in there it's fantastic. Try switch between different versions of express fast as hell.
>> Okay, that's all great, >> but come on. What is this? The spacing and the hover status. The sele I mean I'm talking from a visual UX perspect.
Yeah. Okay, great. It's fast. It's rules. I actually like all that >> smell on that. I don't know. You could I mean my I I think I could take a 5-second pass at that and and just make it a little bit better. That's crazy.
>> I'm curious if it's be like to help with having a better click box like why these are so separate. But if that's the case, then maybe you have different styles.
Like this could be a good style for mobile because your finger wouldn't fat finger clicking on one of the other items.
>> But >> yeah, >> but why wouldn't you highlight the entire >> Yeah, I don't know.
>> Oh, sure. The icon, not the text. I see what you're saying.
>> The spacing is just wonky. in in general.
>> Well, I have good news for you.
>> It's it's fast and it's open source. So, if you want to fix it, make a PR, maybe make a suggestion.
>> Use some of your Opus 48 on this. If you're if you're going to whine, you got to do the time.
>> I My agent will do the time. Uh >> yeah. Okay. Express. What's up, Express?
Yeah. Um I guess >> people are using Express. Yeah.
>> You might remember if you're still Yeah.
Like I haven't used Express in a long time cuz I'm on Hono, but a lot of large enterprise are still on Express or maybe they're using Nest.js and that's using Express under the hood. A lot of people are still using Express. Let's look on um yeah, is it npx? NPMX.dev.
>> For for me, guys, when when something becomes out of vogue, that means nobody uses it and I should not be using it.
>> Express gets 102 million downloads per week.
>> Huge. Absolutely huge. Yeah.
>> Uh yeah, >> it's Express is like will never go anywhere. I I know people are always like, "Oh, don't use Express anymore."
And like I'm very much on that page because it's not it's not web standards, right? like Express was built when the standards were something called connectjs and and that was like the the signatures of request and response but um we now have web request and web response and uh a cyclical storage and all these things which is what like hono and all the modern frameworks are built on top of so >> I wouldn't start anything new but I'm certainly glad that there are still people putting time into like maintaining this thing because it is so widely used >> definitely and this is this is the old site if you don't remember that's what it used to look like this is the new site um and it was migrated from jeekal jeekal to Astro so there's another uh reason to to use Astro for your static sites or your doc or your documentation sites um express chose them as well um they added new content added more translations um and oh I didn't even realize it's a new logo too yeah for the longest time the express logo was just the word press and now I like that. I'm a fan of this logo. Great.
>> I've been trying to get um TJ Hollow Chuck. He's like the creator of he he was the creator of most of early node, you know. Um >> he was the creator of like everything.
>> Stylist uh Pug, which what was it called before Pug?
>> Jade or is Jade something else?
>> It was Jade and got renamed to Pug. um Express. He He's a creator of like absolutely everything new. Every now and then I He's like a photographer in London. Like I think he went totally away from coding. Every now and then I try I message him on Instagram, see if he'll come on the show and talk about the old days. He >> went from Yeah. the most prolific person ever to just a a ghost.
>> Um yeah, totally. Hey guys, >> happened with a lot of these guys.
>> Do we want to check in on uh Opus and see where we're at? Yes.
>> Okay. So, folks, if you're just joining us, we are live talking about all kinds of different stuff, but of course, Opus 4.8 just dropped and well, we got access to it right before the show and we put it on a task and uh I'm running it in Claude Code here. I put it on a task that is to uh build a pixel editor. And that pixel editor needs to um Sorry, I don't I'm not sharing screen yet, Randy.
uh that pixel editor needs to use type GPU and HTML HTML in canvas to be able to do live effects. And so this is quite it's not a basic ask, but it's not like a crazy ask either. So here are the current results. I had to prompt it twice in this process. So here's what we have. I had to prompt it twice. The first thing I had to tell it to do was to fix this dang sidebar because there was that overflow text issue we saw earlier. It did not get that on its own even having Chrome DevTools MCP. Um, and then the other prompting I had to do was just to tell it to add more effects, but other than that, um, let's add some color in here and I'll show you guys what the end result is. Maybe we can say um so that was like really uh two prompts really uh a top of the initial prompt. Okay, this kind of says syntax but it doesn't really cuz the pixels are small. And if we go there's a layers so it has layers. You can hide them. You can do all this. I didn't tell it to do this. Um and then you have the effects of which you can tell it to show on canvas. So it's using type GPU and HTML in canvas to render a CRT effect. You can bake it into the layers to apply multiple effects. You Let's see. Look at that one. We got some motion in this bad boy. Um, and the coolest thing is I didn't I had to point it to HTML in canvas docs, but I didn't have to point it to type GPU. It did use context 7 automatically, and it just used context 7 for that and seemed to handle type GPU pretty dang well. Um, oh, another thing I had to prompt it was telling it what the flag name was for Chrome uh to get the HTML and canvas support. And obviously in a pixel editor, like some of these effects like outline and posterize are stupid. Um, but the CRT one, man, I think this one really shows a little bit of coolness that it was able to do this this easily. So, >> yeah, really interesting. Can I still draw on this thing? It looks like I can still draw even with the effect applied.
That's cool.
>> Yeah. Those are those so those are just shaders that are being loaded. Is that what that is?
>> Yeah, it's So the way this works is that there's an HTML in canvas that wraps around the So this whole thing is HTML.
Um and then the canvas element wraps around that and draws it and then we're applying a shader to that canvas. So that shader sitting on top of it. So >> why does that need to be HTML though?
>> Cuz I told it to.
>> Oh, okay. Good. That's why it doesn't know you could do it all in canvas. It was part of the challenge. It was part of the challenge.
>> Okay, good. Okay, good. I had somebody on on Twitter yesterday that was showing me what they built in HTML and canvas and I was just like there, you don't need HTML and canvas for this. HTML and HTML or canvas and canvas.
>> No, great. This is good.
>> And it looks like um it looks like you can So, let me put a layer on here.
Looks like I might be able to bake effects into layer. So if I Yeah, this is interesting that it it did a good job. I will say that. Did a good dang good job on this. You can bake effects into layer. And now I can reorder effects.
>> Reorder your layers. Did it do drag and drop?
>> Yeah. Let me uh turn these off and on.
Um Oh, there's just buttons. It didn't do drag and drop.
>> Never ever does interaction. It always gives you buttons to go up and down.
>> Yeah, up and down buttons is such a thing.
Man, the UI is so far behind the actual functionality because like the like this is amazing that it could build this. I know it's probably not amazing because we've had six months of everybody building crazy stuff, but like like think about like even just a year ago.
This is amazing that this happened in two prompts.
>> I would say the UI is just so crappy still.
>> Yeah. And this is honestly better. This is better than what Codeex would give me. And this is better than what uh probably what Opus 4.7 would have given me cuz Opus 4.7 not good at this stuff either. Um but this is essentially oneshotted. And I think the most impressive thing to me is that it's using HTML and canvas and it's using um the uh type GPU because type GPU is not like a it's a spring. It's not like an established library. Um, it's cool ass library, but it's not like ubiquitous.
It's not express here. So, did a really good job with that.
>> Yeah, I'm I'm I'm impressed. Oh, yeah.
>> There's a comment though like we should see how much did this cost. So, can you do from it? Can you tell us?
>> Do I want to?
>> Yeah.
>> How what what is it you forward slash what >> for/cost or for usage?
>> Is that just in this session though?
Yeah, it should be just for these these prompts.
>> Okay. So, it used the Opus u front end design. It used grill with docs. I didn't tell it to use grill with docs unless it just loaded that.
>> That's what I was asking. If you scroll up, I think it gives you the >> Yeah. Well, how many tokens?
>> Keep going. Yeah. So, for 33 minutes cost $15.
>> $15. little man. Put that >> I would not give $15 for that piece of software.
>> You wouldn't.
>> What's funny? I mean, the thing is Yeah.
So, imagine AI didn't exist and like you hire a consultancy to build this app.
>> It's going to cost thousands of dollars, but the fact that you got it in >> in 30 minutes for $15. Sure. But like, who's going to use it? Like, there's a lot more work to do to prompt it to make it >> bracing for the Kimmy comments. Here we go. Chat, tell me how we could have done it for Kimmy and for $3. It's not my fault that uh I'm not good at pixel art, folks.
>> Yeah, the thing is Yeah. Yeah. So, um Yash is saying $15 for a session is nuts. When you're paying API pricing, I find it to be pretty common. So, I've been using Opus 446 for most things. And um I'm on the enterprise plan, so Centry is paying for it. Shout out Century.
Centry.io/sintax.
Get all your air monitoring there. But uh very regularly I see like $15 usage for like a 30 or a 45minut run of an agent.
>> So that's I I don't think people realize I know that we keep talking about this.
>> I don't think people realize how expensive this stuff actually is. And the the $200 plan that everyone's paying for. Like I guarantee Scott uses $200 a day.
>> Oh, brother. Yeah.
>> Oh, this kind of looks like a poop, but it's supposed to look like the the little sententury guy.
>> Yeah.
>> Recent colors >> of AI usage. Um, if you're on Copilot, your usage based pricing ends in 3 days.
Um, so get on it now if you don't have any unused credits because the price is about to 10x. Um, I don't know if you guys use Copilot, but I I they did a thing where this month was the last month of usage based pricing and then next month they're moving moving to token based pricing, but they show you what your bill would have been.
>> And so in the in the month of April, >> my overall bill was $140, but with tokenbased pricing, it would have been $1,400.
And that was that was a month with like less AI usage than I've than I've done before. Like March was probably my highest AI usage because I was doing a bunch of agentic stuff and that was like a $400 bill um usage based pricing. And if we're using the 10x amount, that would be $4,000 >> um >> for that amount. I'm I'm not ready to pay to pay those amounts. So, I'm going to um I'm trying to get all my agented coding out in the next three days and then going back to mostly manual stuff.
Yeah, >> that is nuts. Nuts. Nuts. Nuts. like everybody's like, "Yeah, it'll get cheaper." Like, do you think it will? Um I saw earlier today um what was it? Somebody from XAI was saying that they are writing like C that runs directly on the graphics cards to hopefully do it like faster and I'm assuming cheaper as well. But like $4,000, imagine you you got that 10 times cheaper. It's $400 a month.
There's a lot of people are not spending $400 a month on this stuff. But in the same breath, I'll say I'm not going back to writing stuff by hand because it's $400, you know?
>> Yeah, a weird spot. I think maybe you'll see people shift like I think Wes you're already used to using open router so like it's possible that maybe you can switch to Kimmy K2 or um >> some of the others uh I forget all the names but that usage >> is really good >> and yeah and they're getting better so I mean it could be that these massive models just like aren't worth it in terms of like pricing for what you actually get >> but but that does mean you still have access to dumber models or like older models that are like 10 20 time like 20 times less in terms of price. So >> yeah.
>> Yeah. Someone says, "Is China subsidizing AI coding tools?" Um probably. Everybody's in a war right now and everybody's trying to make them as cheap as possible. Um and the the one thing China has is they got cheaper power. Uh so probably I don't know if they actually are, but >> yeah.
>> What else do we got here? What sub would you guys recommend? Man, I I I've heard a lot of people liking like the open code go sub. This is like kind of some like lower power model, but still pretty good. I think that that's probably a good move for that. I know a lot of people are into Kimmy. Sorry. Go ahead.
Well, I'll try I'm going to try Open Code Go because like right now for my personal stuff at work, Century gives us a cloud subscription, but for personal stuff, I was using um and like I mentioned, I was using Copilot, but I'm going to try Open Code Go because that gives you access to the the more open models. We'll see how it goes. But I'm going to try it.
>> You're going to try?
>> I still like >> Yeah.
>> Cursor as well. Um because the >> cursor is I don't know what the Ultra is 160 a month. Oh, it used to be 200. Oh, 160 if you buy a year's worth.
>> Um or Yeah. And like they're for sure you can use all the models on that and they're for sure not making money on that.
>> I use and also their their composer 2.5 is really good. It's it's fast and it's like it's not as good. Like there's sometimes I've had to like change it to back to Opus when it wasn't figuring something out, but for a lot of cases it's totally fine. Yeah, that I if any, you know, I get why, but they should let people use it as an API um as well.
Cursor composer. Uh I I just don't want to use cursor. I'm sorry. I that there's there's like I'm very comfortable with my tools and I I don't want to use cursor, but I I personally use the the Cloud Code Max plan as we've seen in this video. Cloud Code Max is what I've been using for a while, and honestly, I get the best results out of it. I don't know if it's just that my setup is dialed or what, but Codeex to me drives me nuts. Um, and so I have a hard time uh with the chat GPT stuff. I did have a subscription to that. I use Codex for a while as my main model, but what I tend to do is use um what does that sound >> somebody got a fidget spinner?
>> Yeah, it sounded like a fidget spinner.
Randy, >> you guys can't hear. Ry's back there with a fidget spinner while we're trying to have a live stream here.
>> I would have guessed Wes. Um, I I I really like Open Router, folks. I have an open router for small models that I use for all all kinds of stuff. And and that's a thing I would really love to get into more is having systems that really like, you know, use the correct size model for the correct size thing instead of just always going with the biggest >> that will become much more common in the future is the model will figure out oh this is the type of t they're already doing it with yeah >> like chat GPT you can you can see that it switches >> um to a model based on the type of question that you have sometimes you can manually change it but I think it' be able to like accurately pick which one is is necessary um Justin Schroeder actually did release like a a thing where you could use composer in like open code. Um, but apparently got in trouble for that.
>> Yeah, they they blocked it right away.
Yeah. Oh, we got a smoothie delivery.
All right.
>> She always shows up. Thank you, Court.
>> You're welcome.
>> That's looking straight up neon. Scott, what's in that? Oh, what is in this court?
>> Um, it's spinach, mango.
>> Spinach, mango, >> water, and protein powder. Nothing like >> protein powder. That's the spinach that's getting it.
>> Healthy smoothie.
>> Oh yeah.
She says healthy smoothie. I I love a good healthy smoothie. Especially he's got spinach in there. You don't even know about it.
Um Chad, what's up? Kimmy 2.6 in PI coding agent is pretty damn good for a lot of stuff. If you haven't tried it, get it hooked up on Open Router and give it a go. Yeah, you know what? Pi. Pi. I love Pi. I love the idea of Pi. I spend more time working on Pi than I do on my code.
So that's why I've been just sticking with clawed code even though I like I don't like the anthropic walled garden approach.
>> My setup is so dialed and it works so well for me right now that like I don't want to touch him which is abnormal for me.
>> I just searched my email for Kimmy and I have like 15 requests and free credit offerings from them. So they are certainly losing money on this thing right now. Um, most of these companies will like email me and say, "Hey, here's a hundred bucks or here's 200 bucks to to blow on something, but I get a lot of emails from Kimmy."
>> Yeah, I'll say about the like somebody mentioned, why don't I just move to a a Claude or like anthropic max plan? And that's exactly like I've I've gotten used to being able to use open code and then choose whatever model I want.
That's the That's what I got when I was combining co-pilot with open code is I could use GPT or I could use claude. Um, and if you get a Cloud Max plan, then you have to use Cloud CLI or like their remote tool. Like you can't use Open Code. So, kind of like like you're saying, it locks you into their ecosystem.
>> And so, I I just want something that lets me use the tools that I like instead of having to having to switch that. I I do use C-Pilot to bounce between models too to to try them out specifically or um when I I don't have something that I need a long large context window for um I'll use Copilot for show. Yes. Um on the note of uh models picking it I on that report that it did it did say it used Haiku 4.5 for some stuff in that process. So I didn't tell it to do that. So obviously Opus 4.8 used Haiku 405 for something. I think a lot of the sub aent stuff, especially if it's like simple stuff like um listing or reading terminals or or like parsing out a terminal, you know, it takes a huge dump from the terminal and tries to like uh tell me if this thing was successful or an error, you know, like that's dead simple stuff.
You don't need to clutter your context window that with really expensive tokens. Um, I will say one other tip I have for like saving money is if you the like the cashed tokens is something that's not talked a lot about essentially is if you are going back and forth with an agent and those tokens are are cash. You're not paying for them every single turn. Um, but I think the cash times are like some of them are only like 10 5 10 minutes.
Some of them are hours long. But if you like come back the next morning and you've got like 600,000 tokens in contacts and you send it like one more message, you're you're that's like a like a $30 message or or like a $15 message just just to say one thing. Um, so if you find yourself having to walk away and then come back, it's often a good idea to say like write your findings in a little markdown file and then start a new session up given it that and just give that little markdown file to say keep going with it.
>> Yeah. Yeah, man. I wonder, you know, how all these little tips and tricks are going to change over time. It's like >> we probably won't do that for a while.
like that will probably be a silly thing to have to like manually do that. But right now is if you find yourself just like having the same chat because it knows about everything you've been doing. Um that can get really expensive.
>> Yeah. Do we think we're getting a new GPT today as well? There's been talk on Twitter about a potential new uh GPT releasing today or do you think that Opus is going to get all of the limelight?
>> The last time they released them on the same day. So I bet I What's What are we at now? 5.5.
>> It would be 56. Well, >> the 56 would be the new one.
>> Or maybe 55 because there's 53 codeex.
They never came out with a 54 codeex, right?
>> Mhm.
>> And so maybe track all these names.
>> I don't know.
>> Yeah, just Yeah, the naming is crazy.
Just give me 1 2 3 4 5 decibb decibb goodter goodest.
>> Yes.
What if they're going to eventually go to the a uh like the Apple model of of just like naming it after the year, you know?
>> I bet that's that's what it will be. I bet it'll be like Opus or Haiku. You like they'll have the three things and like it'll always just be the latest one. They'll just be constantly updating it. you probably won't have to worry about it cuz even now we're we're reading release notes on a little point update and do we really know how better it is, you know? I don't know.
>> I don't know. It did a pretty good job on this little app that it did.
>> That's the thing to that though.
>> Yeah, it's all vibes based. Like they >> they release it but like you have no idea. It's literally just the same prompt and you get the same stuff back.
Maybe it's a little better. I don't know. This we're putting too much power in their hands, right?
>> Mythos confirmed. Is that true?
>> Mythos confirmed will be coming in public >> public release.
>> Let's see.
>> I don't know. Myth Mythos, you know, yes, security open like a Mythos isn't going to change our workflows that much besides like what uh poning things, but even the pwning that it has done and being able to find um security holes was done after a lot of time and effort in the model. It's not it's not just like you go to prompt and say, "Ooh, find me security bug in this." It's doing some complex operations uh multiplestep um penetrations there.
I I bet the days of us being able to do like kind of shady things with these models are very limited. you know, like I had it I've had it crack a couple pieces of software for research purposes. Um, and and like kind of dive in, decompile a lot of apps. Um, and I've noticed recently it's it's starting to say, "Hey, I I won't do that." Um, and I bet in the next a couple months or whatever, you're not going to be able to do any of this malicious stuff. And then there's going to be like these like black hat models that people have cracked or distilled where you can do do sketchy stuff with them.
>> Not that you should, but I'm I'm doing it for the the sake of the sake of the show, you know.
>> You're doing it for the sake of the show. Yes. Yes.
>> Yeah. I'll uh I'll take take this pause here to let you know what we're doing and then I have like one more fun thing to show you all before we get out of here today. But we are going to be in Amsterdam in person on June 10th from 6:00 to 9:00 p.m. If you head to syntax.fm/meup, you can RSVP. Uh looks like we got three or four RSVPs since going live, so that's great. Uh but you can see us in person. We're going to do a live show.
We'll also have free merch we'll be giving out. It's going to be a fun time.
Would love to see you there. And then the following day is JS Nation and I'm going to be mcing there. Scott's going to be MCing there. Uh Wes is going to be giving a talk and the day after that is Reacts Summit and uh Scott will be giving a talk there. I'll be minging there and if you want to get a ticket for either JS Nation or React Summit, you can use code syntax to get 15% off.
There's also a joint combined ticket if you want to go to both conferences. But it's one of my favorite conferences.
>> Excellent conference. Yeah.
>> Yeah.
>> We'd love to see you there. If you're anywhere near Amsterdam, take a train, quick flight. It'll it'll be fun hanging out. And even if you don't come for us, there's going to be a really cool dev community there, like at least 100 people at our meetup.
>> Um, it'll be be a good time.
>> Um, I'll also mention uh join the syntax snack pack. So, we have a buy weekly newsletter >> where we send out links to cool stuff we found on the web, very similar to this live show, but um as in link form and you get it in your email inbox. we won't spam you and also you'll you'll get to know about upcoming things like this Amsterdam meetup and various other things that we do. So syntax.fmackpack snackpack. Yes. I'm so excited. I love these conferences. They're great. I'm excited for our meetup. I'm excited for our live show. I'm just excited to be there in the house. My talk has got some crazy slides. I cannot wait to share them with people. I was showing Wes yesterday and he thought they were sick.
CJ, I'll have to share them with you while I don't want to spoil anything, folks, cuz they're that cool. We'll talk about it on the syntax episodes after I do that talk just to avoid any spoilers here. Cool. CJ, what is the next thing you have for us on the docket here?
>> This one is a fun one. Uh, I came across a thing called PolyCSS. Have you guys seen this?
>> Poly CSS.
>> Poly CSS. So, it renders >> I thought about doing stuff like this.
>> Yeah. So, it renders 3D meshes in the DOM with CSS. So, there's no WebGL and no canvas. Um, I'll show you in their gallery like some of the stuff you can load up.
>> Um, and then you have >> thought about making this.
>> Um, >> yeah, a lot of this stuff looks like very like Minecraftesque. Um, but basically like any any texture, any 3D thing. But the cool thing to see is if you look at this thing in the DOM, >> it's not a canvas. It is literally HTML elements with some transform uh matrices. So, it's all done with CSS.
You can see that like each of these these elements are actually selectable in the DOM.
>> Yeah. And that stuff is crazy. I built I mean I'll have to show this some other time. I built a thing that had like boxes and cubes flying through those boxes in 3D space and that took me forever. Now granted, I did it by hand.
Who knows what they did this with?
probably AI, but some of that stuff is is crazy to to get your to wrap your head around with especially with CSS cuz it's not made for doing a a dang deer jumping in 3D. This is crazy.
>> Yeah, in the chat they're asking about performance. I don't know if it shows up on stream, but you can see the FPS in the bottom right. So, this llama, this animated llama is getting about 32 FPS.
>> 42 FPS. Not bad.
>> Not bad. And then like some of the still models, like let's look at this avocado.
um 50 FPS. It probably drops a few.
Yeah, we're getting down to like 50, but it gets up to 60 FPS even with this. And and it does have more of a I don't know N64 PS1 type look cuz you you're literally drawing triangles like there's not a lot of smoothing. Um still >> So, who's making Golden Eye in CSS? Cuz that's a lot of Golden Eye and CSS.
Yeah.
>> Yeah.
>> Oh, that would be sick.
>> Speaking of Oh, yeah. Here's a nice little >> Yeah. Super Nintendo cartridge. Yes.
>> Yeah. So, fun little project. Um, I honestly don't know the use for it because at at the end of the day, like if you >> I don't know, maybe a show of force is what it is. Show >> we'd have to see some benchmarks for other stuff. But it's cool because they have like an importer. So, you can take any like OBJ file or like box file. So, these are things that that are exported from like 3D modeling software and it load and it can load them in. And they also released uh the live like vanillajs react and view. So yeah, check that.
>> That is great. I can't believe somebody made that. Sorry, I'm out of out of frame on camera here. I can't believe they made that.
>> Yeah, >> just can.
>> I don't think people realize how good of a renderer the browser is.
>> Yeah.
>> Yeah. I always hate when I get comments.
It's like, why are we shoehorning the browser to do things? It's like the browser's got great layout engine. It's got great rendering engine. Like browser rips, man. Love the browser.
>> Yeah, >> shout out to the browser. Yeah, >> shout out the browser. Yeah, on a related note, my my wife is uh taking an Android class right now >> and just complaining about like they reinvented a ways of do sty doing styling and like elements and like and she comes from the web world. She's like why can't they just >> oh like like development? Yeah, I thought she was like like taking a like I assumed like she was taking a class on how to use her phone and I was like >> no, she's making custom apps.
>> But yeah, so when you make >> when you make a native Android app, you have to use the uh like their UI kit and like their XML language for UI.
>> It's all like super convoluted and it's like why she was like why can't they just use flexbox? Um >> yeah, she she to learn cotlin.
>> Nice. Uh, no. I think it's an it's a university course, so like I think she's still on or using Java, but >> I I did a lot of Android courses on level up tutorials back in the day. And I hated it. I hated Java. I hated uh the entire thing. The the way you did images uh like border radius and stuff was to do all the little corner images like we used to back in uh 2010 on the web. Like yes. Uh it was no fun back then. I I hopefully it's gotten better. And even then like the you know iOS has a really good simulator and the Android like development kit and all that stuff was just >> felt like garbage to me. But um >> but in short we got it real nice here on the web like the APIs the standards >> it's good it's good time to be alive.
Yeah, I don't think I would build a native app as somebody said um outside of React Native or Expo um or links and a folks uh we had Paul uh Paulo on to talk about custom renderers for spelt and I've been seeing I subscribe to all the the the uh updates in the custom renderer support inside of spelt. So I am hopeful that one day I'll be able to build my native apps with spelt as well.
Just uh represent my t-shirt here.
>> Nice. So, that's all I brought. Scott, did you have another thing you wanted to talk about? Like Hermes, maybe? Or >> Yeah, I could talk about Hermes. Um, have you guys seen Hermes? Let me pop this open. I'll uh I'll come back.
>> Oh, yeah.
>> Oh, yeah.
>> Hermes.
>> Um, yeah. I don't you I don't think you can call it Hermes. I don't think that is uh by the way, that acquired podcast on Hermes was super good. Uh, if uh you like the acquired podcast, Hermes is a good one. Um, Hermes agent. This is like a uh it's like a claw a clawbot or open claw. That's what they call it now. An open claw kind of style thing where you have an agent that has a uh memory a soul. It has all these things. And it runs largely without it runs largely unbound by default. It's not like in a sandbox or anything. It can be in a sandbox if you wanted to, but if you just run it on your system and you tell it to go update some files on your NAS or something, it can go do that. Um, it can yolo whatever you want as well. And I got to say, I kind of like this thing a lot. Um, it I I've been using I used OpenClaw for quite a bit, but when they pulled the anthropic support, uh, I I stopped using it and I I kind of missed it because I had a nice little setup and I built like a really nice chat UI for Courtney and she was like, "Why is that chat UI you built me so much better than chat GPT?" And I was like, "Oh, let me tell you. Let me tell you, it's finally tuned in all kinds of ways." And I've been looking at a way to bring that back. Um, and I did with Hermes. And I got to say, I really like Hermes. I am using uh different models for it. I'm trying it on open router right now, trying different models out to see what works best. I I don't know if I've like really nailed it, but it is it it it's a good project. And the thing I like about Hermes is that like where OpenClaw like OpenClaw kind of feels like a uh an experiment where they're just like having agents just yolo throw code at it left and right. This is way more of a curated development process. Uh so you feel a little bit safer about installing updates to it knowing that like there are actually people who are going to be uh keeping an eye on the development of this thing rather than just yellowing agent updates based on uh whatever chat is begging them to put into it. So I got to say uh installation process way smoother than >> before. So like >> yeah, I'm using it as a chat interface for for Courtney. And eventually I'm going to bring um uh when I get a new system, I'm going to bring uh local inference here. So that way it can use local inference uh for local chats and stuff like that. But what are we using it for? Uh Courtney uses it as her main chat UI. Whether that is for travel planning, it has access to web search, it has access to uh a shared Obsidian vault that she and I can both write to.
And um that's really it. just normal chat stuff. I have a personal assistant on mine that I've just gotten set up and I haven't really done a whole lot with, but so far it's um you know that that uh caddy setup that I have with my tail scale. It did all that for me. So, I told it to do that and it did that and then gave me a UI uh to manage all of my services. So, it's it's uh it's been nice. And there there's cron support.
Sure. I I found that the system here has been way less brittle than open claw ever was. And it anything that I would think about using OpenClaw for I use Hermes for now. And it just just kind of runs on that bad boy over there. And I don't even think about that much. It it's been pretty it's been really stable. Um which I can't say for for OpenCloud. Anytime I would install an update, it it would just uh have issues.
>> It's still running like months later and it >> I haven't updated it in quite a while.
>> This thing doesn't have heartbeats as far as I know. So you don't have to worry about it just like sucking up uh credits while you're not doing anything.
The memory system is really good, but I did replace the memory with something called Honcho, which I'm running locally.
>> Um, and Honcho, I believe, is just a database with vector search. I don't I the details are fuzzy for me. Uh, I just know it's better. Yes.
>> Opus 4.8 just landed in cursor. Cursor folks can use it.
>> You're late. Cursor, you're behind the ball. Either way, uh, Hermes, if you're interested in this type of agent stuff or it's whole world and you gave Open Claw a try, I I gotta say I like it. I think it did a great job and the UI and everything, the whole flow of it just feels a little bit nicer. I am still very early into my Hermes journey, but it works well. But people use this instead >> of like like I think the the uses of like clawed CLI or or codec CLI or open code and like these like agentic things they've sort of I think they're going to be the same thing very soon you know like you can use this to to do coding right and you can use your like cloud CLI as an as an agent in many cases.
>> Yeah sure. Yeah, absolutely. I I think the thing that's nice about this is how interconnected it all is by default and how the process for it is running a config CLI where I can uh say, "Ooh, I want this and I want web search here and I want this and I want this and I want this enter." And then it's all interconnected. And when you create profiles which are essentially individual agents that have their own memory and their own soul whatever those things all have their own access to their own private pool of skills and everything like that. So again I can have a travel planner profile that is only has access to web search and doesn't have access to my file system or uh coding or anything like that. So it can go off but it does have access to maybe like my Obsidian vault. So, it can do research and it can write to my Obsidian vault, but it can't uh go ahead and delete my entire NAS drive. So, uh you you get those types of systems going on in here. And I I like that.
I have yet to set up a personal AI agent. Um I I guess I can't think of a good use case. I mean, I I do feel like like for the home, like if I'm sharing notes with my wife or we have maybe like a project at the house or something like that, it'd be cool to collaborate on and use AI with, >> but I guess now we're kind of just doing it the old way, just in a chat window together. And >> I don't know.
>> I think like the the two killer use cases for me for these types of things is is being able to just use any of the models that you want, right? Like >> I find myself jumping between like Gemini chat for some things and chatbt and cursor and claw. Like I have like probably six or seven different things.
And with this type of thing, you can hook them all up and and you have all access to every single one. So if you want to use Gemini for image generation, but you want to use like a a cheap model for some of your chat, that's great. So that's one killer. And then the other one for me is like it's just a great spot to just dump information like a like a second mind, you know, like I was at one point I measured a whole bunch of my shirts. Um and it's great because now I have this database of shirts in an obsidian and and now when I want to buy a shirt, I just copy paste the sizing table and and say, "Hey, which one what size should I buy?" And then it'll be like, "Okay, well given all of your shirts and like I have comments on all of them being like, this one fits tight.
This one fits good. This one's a bit looser fitting." Then it will just like it will tell me which one I want. I love I love that spot of like I can't forget this information and I don't know where to put it. And I just love being able to just dump that into a chatbot and it files it away into markdown documents or memory or something like that.
>> Yeah. One of the things I like about Hermes West that uh I don't I don't know if OpenClaw necessarily has if they do they they added after I stopped using it was this idea of like background tasks.
So they have based in their own UI a conbon board and I just made one for my UI but um I instead of always just chatting directly either from my chat I can just say send to background or you can have your other profiles and agents uh pass the tasks around based on um their own roles specifically. So, I could have my personal assistant send something to my travel planner to do some work or whatever and it's all visualized in a nice way. I I found that to be very nice.
Yes. Yes. Yes. I am using her Hermes for these types of things. I do like uh like Wes mentioned the Obsidian stuff I think is a killer. Uh I I like it to manage my Obsidian stuff, but also like so people don't know, my wife has a podcast. It's called Phases. It's a child psychology and human development podcast and it requires a lot of like uh research paper stuff because she wants to site research papers and if we tell the agent to go pull research um it's able to pull research and then reference it in an Obsidian doc directly in like our show notes. So it would be like if we were doing show notes in notion and Courtney would be like I want to pull this study it would pull that study into her Obsidian give the the the footnotes of it and then link directly to the study.
So that way when she's recording her episode she has all that information right there. So again it is just like really good for those types of very specific use cases that I found Hermes to like I said just to be more reliable of that same type of thing.
>> Awesome.
Any any other stories? Should we wrap this sucker up?
>> Um, no. I don't have anything else unless you do, CJ. I uh I got a long day ahead of me here. Um, trying out Opus 4.8. I'm going to see how far I can push this thing. I'm working on another since it did so well with this type GPU project. I have another project that's like automating video with HTML and canvas and and type GPU. I'm gonna see like uh what the new brains does if I put it behind that bad boy and see uh if it can uh can handle what I'm working with.
>> Nice. Yeah, I don't have anything else.
I think I'll just do one last plug for our IRL event coming up. So, if you're going to be anywhere near Amsterdam on June 10th, we'd love to meet you. So, we're doing a in in real life meetup June 10th from 6:00 to 9:00 p.m. in Amsterdam. If you go to syntax.fmup, FM/Meteup.
You can RSVP to the event. Look at that.
We've gotten at least five RSVPs since we've gone live. Um, so we'll be there in person. We're going to do a live show. Um, we'll also just do have some networking and some free merch. So, come hang out with us. And we're doing this in conjunction with JSNation and ReactSummit. So, if you head to jsnation.com, you can learn about that conference. It's one of the biggest conferences that happens in Europe all about JavaScript. That's happening the day after on June 11th. And if you buy a ticket, you can use code syntax to get 15% off, as well as reactsummit, which happens the day after that.
reactsummit.com.
Use code syntax. You can also get a combined ticket to uh attend both of them and get 15% off your ticket. So, um, do that. We'd love to see you in person. It's one of my favorite conferences. I'll be mcing, uh, uh, Scott will be MCing with me at JS Nation. Scott will be giving a talk at React Summit. Wes will be giving a talk at, uh, JS Nation. So, it's going to be a good old syntax time. We'd love to see you in person.
>> Going to be >> stoked about it. That's going to be fun.
>> Yes.
>> And uh other housekeeping things, we have a newsletter. So, syntax.fm/snackpack.
We send this out uh every other week and it's filled with lots of cool links and sites and stuff that we found very similar to this live stream, but uh in email form. So, if you want to hear from us, be sure to sign up there. And lastly, everything we do here at Syntax is brought to you by Sentry. They are application monitoring software considered not bad by millions of developers. Um, we use it on our apps in production. Basically, if any error happens, if anything goes wrong, we can see that in the Sentry dashboard. Uh, we can see full traces of what was happening in the front end and the back end when things went wrong. And uh, they also have session replay, so you can see exactly what a user was doing when something went wrong. So, if you head to century.io/sintax, io/sintax, you can get two months free of the team plan.
So, that's all I got for you. We're going to get out of here, but I will say we're going to attempt to do a raid over on the Twitch side. So, if you're watching us on Twitch, uh I think this is our highest viewership on Twitch so far. We've gone down a little bit, but still highest viewership on Twitch. We appreciate you watching over there. Uh we're going to do a raid, which is we take all the viewers of this stream and when it's over, send you to another stream. So, stick around for that. And if you're watching on YouTube and you want to join in, head over to syntax head over to twitch.tvsaxfm.
>> Yeah, that's all I got. Yeah.
>> All right, >> cool.
>> All right, thanks for joining us everyone. later. I
関連おすすめ
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
This computer is made from real human brain cells. And you can buy it.
Talktmsmedia
3K views•2026-05-28
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29
3D Platformer Update - NO CAPES
SolarLune
294 views•2026-05-30











