Pod Insights Syntax - Tasty Web Development Treats 974: Clawdbot (Moltbot), Agents and the Age of Personal Software About this episode Wes and Scott talk about building hyper-specific personal software with AI. They explore personal agents, home automation, JSON-as-a-database, and how LLMs unlock fast, custom apps that reduce friction and replace bloated SaaS.
Show Notes
00:00 Welcome to Syntax!
01:53 What is personal software (and why it matters)
04:49 Using AI agents to build hyper-specific apps for yourself
Clawdbot
ClawdHub
13:43 Supercharging your dev workflow with Tailscale
19:06 Privacy when working with LLMs
MLX-Audio
21:39 Brought to you by Sentry.io
22:21 Real-world personal app ideas
39:14 Sick Picks + Shameless Plugs
Sick Picks
Scott: FTPManager
Wes: Roku Streaming Stick
Shameless Plugs
Syntax YouTube Channel
Hit us up on Socials!
Syntax: X Instagram Tiktok LinkedIn Threads
Wes: X Instagram Tiktok LinkedIn Threads
Scott: X Instagram Tiktok LinkedIn Threads
Randy: X Instagram YouTube Threads
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Scott Tolinski? Welcome to Syntax. Today, we're talking about building personal software. We're going to be talking about how you can build software that is just for you, runs on your network of things and possibly anyone else you want to give this to, and stuff that, well, is kind of disposable, but at the same time, things you might not want to spend all day, every day, building. Because now, you can just throw Opus 4.5 on it, knock it out in a couple of minutes. Next thing you know, you got a feature. My name is Scott Tolinski. I'm a developer from Denver. With me, as always, is Wes Boss. What's up, Wes?
0:35 Wes Bos? Yeah, man, like, the amount of extremely specific apps I have built in the last, I don't know, six months, a year, is unbelievable.
0:46 Scott Tolinski? It's outrageous, yeah.
0:47 Wes Bos? I know people like laugh at this whole like personal software thing, but like, it really is true. Like, normally, when I would have have reached for some, like, I don't know, crappy SaaS that covers everybody's use case, or I would have reached for a spreadsheet and had to like do something weird in a spreadsheet. I don't have to do either of those anymore.
1:09 Scott Tolinski? Or I'm spending months building stuff, yeah, just months. I worked on my habit app for a long time. And even though there's other people using that thing, I built it for myself, and I'm still maintaining it, and it's still a big old project. Where most of the stuff I've built, let's face it, in the last four days, which has been a number of things, which we'll talk about. And when I say "I built it," I mean, I project managed the building of it. The number of things that have gotten done there has been like outrageously fast. And there's been a lot of like little, even like technical things that I've learned about what makes some of this stuff easy and fast. We can get into some of those strategies for this personal software as well. Because I do think there's a lot of like really neat things that I've gained by doing some of this.
1:53 Scott Tolinski? So, first and foremost, what is personal software? Why might you care about it? What is it? And it really is hyper-specific software that is tuned to your needs. Like, I have a need in my life, therefore, I'm going to put an AI agent on crushing it. Out a quick project, I might not even care about the quality of the code. This isn't for public consumption. This could be just for you. And there is something about the fact that other people aren't using it and it's not exposed to the worldwide internet or possibly not exposed to the worldwide internet.
2:31 Wes Bos? Hopefully not.
2:32 Scott Tolinski? Yes, that makes the security or the slop nature of the code more freeing. I was definitely in the camp of I need to read every line of code that AI produces for me forever. Until I started working on personal software that like, let's face it. If it's not getting off my network, what the hell do I care if it's exposing API keys or something? It's just on my computer, right? So as long as it's not deleting my files or ruining my file system. The slop nature of the code, I don't care about the slop nature of this code in particular, as long as it's functioning fine.
3:09 Wes Bos? Especially if you build an app where you want to use API keys directly in the browser and make the requests directly from the browser. That's typically not a good idea at all. You shouldn't be putting your API keys client side. But if you're especially just in prototyping stage, you're like, I don't feel like building some sort of like proxy to send it to the backend and get the response and hot potato it back is you can just YOLO send those directly from the client. Which again, let me specify this is not a good idea in general, in general. But if you are just building something that is only for you and probably should also disable like Chrome extensions as well, you gotta think about like what else might be reaching in there.
3:50 Scott Tolinski? Yeah, what else might be, you're totally right on that aspect of things. It depends on where you're running, how sandboxed it is and whatever. I'm not even talking about like API keys though. I'm just like, just generally like talking like code quality. I don't give a shit about the code quality if nobody else is using it. If somebody else is using it, that that story changes completely for me, I'll tell you that.
4:11 Wes Bos? I care somewhat about the code quality because I've certainly built lots of stuff way too fast. And it becomes an absolute mess where I simply just scrapped the entire project. But we have an episode coming up in the future of like, like, how do you do this stuff? So it actually produces code that you you would want to be able to debug and work on. If it comes to the absolute worst case where you have to write some code yourself.
4:36 Scott Tolinski? Yeah, and I will say that AI is not writing all of this code for me. I am, I have Z open, I'm modifying, I'm writing code, I'm changing especially CSS still, man. Uh, I have some tips there for my CSS stuff, but yeah. So let's talk about like the idea of like AI and uh, personal assistant style or personal agents and building this kind of stuff. Because what I started was Claudebot. I started a Claudebot, which is kind of a personal assistant. Uh, that is an agent that runs on your machine. It just runs on a Bun process for me. And there's a Claude, uh, Claudebot gateway. That really handles kind of the connections and things like that. Like you can connect it to Discord and talk to it through a Discord bot, which I'm not doing.
5:20 Wes Bos? Let's explain what this is because this has popped in the last couple of weeks. Claude, C L A W D, like a, what, like a lobster? This has, this has very little to do with like Claude code and things. It does use that, but this is a totally separate project that allows you to create your own personal assistant?
5:41 Scott Tolinski? Yeah, it is really, it is a personal assistant and, and I personally have, uh, not understood. Why like a, a personal agent or assistant in AI would be super useful. I couldn't think of any use cases for it until I started using this. And I do wanna stress with Claudebot, some of this is, you know, if you're sending your data anthropic or whatever, you wanna be careful about what you're connecting to this where. But the cool thing about Claudebot is they have connections to WhatsApp, Telegram, Discord or whatever. So you don't have to use their chat UI, which I don't like at all.
6:19 Scott Tolinski? I ended up building my own even, or the AI built one for itself. Or you can use WhatsApp, Telegram or any of that stuff to interface with it. So you don't even have to have a chat UI for it. You can just do it through Discord. And then there's all these kind of extensions. So I connected mine to like Home Assistant. I connected it to my calendar and things like that. I connected it. Uh, I'll talk about some of the Home Assistant stuff because that was crazy. It's, it's done a lot of cool stuff with my Home Assistant. There's a lot of neat connections you could make and it's up to your own discretion. As to what kind of sensitive information you want to give it because there are extensions to connect it to your email inbox and then it could send emails as you and stuff like that. So there is like a degree of what you want to, you know, curate with this thing.
7:02 Wes Bos? But the whole idea is that you have this bot, you can connect, obviously you've connected chat to talk back and forth with it, but then you also connect it to like tools, right? Like you said, you can connect it to GitHub and your email, iMessage, your home assistant, you can have it control your computer, run a browser, all of these types of things.
7:20 Scott Tolinski? Yeah. And it has a browser so it can look things up. You could have it spawn coding processes and things like that. I have mine connected to my beads so it can project manage things. Yeah.
7:32 Wes Bos? People are probably thinking like, okay, well, like what, what's the use of that? Is that that now that you just have a natural language interface via a text message so you can do anything? Like you're, you're away from your computer and you could ask it to code a project or go through your, your iMessages or like filter through your email inbox and reply to things for you, right?
7:56 Scott Tolinski? Yeah, yeah. And I even wrote a custom skill for it to connect to Tweak, which is my to-do list. And so every morning, it has cron baked into it. So every morning at eight o'clock, I get a message from my personal assistant with the items in my to-do list and then it will like start to even work on some of them. So like if I put, like I need to do something and like I need to research something or I can do this, it can start to give me links or create documents for that or like help that process along. And it can notice if like something stayed on my to-do list for several days and be like, you should really do this, you know, just giving me that, that encouragement, which is like kind of ridiculous. But the amount of things you can connect to it are really great. And where
8:40 Scott Tolinski? I found, one of the, the cool use cases that I found with it is in Home Assistant, which, yes, you might think natural language, okay, I can tell it to turn off and off my lights, whatever, right? But the coolest thing I was able to do was say, Given what you know about all the devices on my Home Assistant, what are some automations that I'm not doing that you think I could have? And it gave me a whole bunch of ideas. One of which was that I have an air quality sensor in my office. And I don't have, I just have it here and I never check the data on it or anything. And it was like during business hours, you know, nine to five, Monday through Friday, I'll check your air quality four times a day. And if anything is off, I'll send you a message.
9:21 Scott Tolinski? And the message is in a Home Assistant prompt, like a Home Assistant alert. So that cron job is basically just saying, check the levels. Is there anything off on these levels? If not, and yesterday at 12 o'clock, it was like your CO2 in your office is four times higher than it should be. You should open a window. And that led me to realize my office isn't venting correctly. The mini split doesn't vent. There's nothing venting air in this office. I'm just breathing in the same air all day. So like, that means I gotta figure out venting or I gotta just crack a window. But I had no idea that that was going on. So just that, that alone was really great. And another thing, I was having a, an automation in my home assistant that wasn't triggering correctly
10:05 Scott Tolinski? or reliably. And I said, can you investigate this? And it did, and it fixed the dang thing and gave me some suggestions. And I was like, sure, do them. And it modified my dang automations on Home Assistant. So.
10:17 Wes Bos? Man, this.
10:18 Scott Tolinski? Awesome.
10:18 Wes Bos? This LLM stuff is going to make the home automation stuff way easier because so much of home automation is people just don't do anything with it because it's, it's a whole, it's a full-time job. Having to like make all these scenes and make sure they turn on. And then if something stops working and you know, it doesn't work. And then people manually turn off the lights. It's really frustrating. And unless you're willing to invest a whole bunch of time in this type of thing, it's, that's the reason why a lot of people don't do it. So I, I got Gemini access on my Google Home.
10:54 Wes Bos? Man, it is so much better. I love it. Like, there's one little command that I say every night, which is turn every light off except for my wife's bed lamp. And I previously would like go and turn every room off and then turn hers back on because she likes to go to bed a little later. And now I can just like say very obtuse things like turn every light off except this one and turn the white noise on. And it just like figures it out. It's a little slower. You can tell that it's like sending it to an LLM to process, but I don't necessarily care because it's so much better. And I've been so frustrated with my Google Home over the last year or so, and it's finally getting better.
11:32 Scott Tolinski? Yeah, that actually, that rips. I had to write an automation that if it detects presence from when we're not home to when we're home, because Home Assistant can do that. If it detects presence in the home, like if you're arriving home, it's going to turn on the lights automatically before you get into the house. It's like, oh, okay, I wouldn't have thought of that. You did. Okay, it's not motion sensing, whatever. It's interesting. So Clawdbot has been really cool. It does this gateway. You can connect it to various LLMs. The idea for me here eventually, right now I'm having it connected to Claude Opus. Claude is an Anthropic Claude. But the idea here will eventually be for me to get a beefy computer like a Mac Studio.
12:14 Scott Tolinski? CJ actually just got a little integrated GPU PC. That wouldn't be a bad idea to get either, but I think I want to get a really beefy Mac Studio and then have that run a much better local LLM than what I can run on my MacBook Pro and then do everything through that instead of going off to Anthropic. Because the moment that you do that and the data is not leaving your computer, then you can give it more sensitive information, maybe even like financial information. And it can do, you know, help you with your money tracking or stuff like that, or even email, you know, maybe give it read access.
12:50 Wes Bos? There's one thing I have to do every four times a year and that's like go into every bank account that I own and export it all. And everybody tells me to use Plaid or whatever, but it's like, I'm not paying for that. And some of my bank accounts doesn't work. And like the second I can automate that entire thing easily, I probably could automate it right now with a whole bunch of puppeteer and clicking buttons. But it feels too... Dangerous and like fragile, like it could break when they move a button or whatever. But as soon as I can hook it up to something like this, especially with a local LLM. And I think we're probably a bit of a ways away before we can get like decent quality LLMs running locally without having to spend like what, four grand on some hardware?
13:34 Scott Tolinski? Oh, I'd probably be spending more than four grand. I'd probably be getting into sixth territory if you're looking at the Mac Studio. But like, then again, that becomes, there's part of this personal software thing that was kind of flipped in my head about how I'm running my dev process in general. So I'll talk about that right now, which is that I've moved everything to a tail scale network throughout my house, which has become really, really awesome. So I have my phone, I have my Mac Mini, I have my computer, my tablet, all connected to the scale. And tail scale is basically allows you to have like a personal tunnel throughout your entire network. And so what it is, is it becomes the VPN on your computer, your phone,
14:14 Scott Tolinski? whatever. And when I want to access things on my Mac Mini, it is via a tail scale IP address. And that means that like anything that's running on my Mac Mini via its ports can just be accessible via that IP without having to turn on any sort of like global file sharing or stuff like that. It's just being done through tail scale. The way I would have like done some of this before, like if I had a site running on something, I might throw it on a CloudFlare tunnel and then you're putting that out to the wider internet and you could protect it with like a zero trust on that thing and have it be like a password that sends you an email. But what's so cool about tail scale is that
14:56 Scott Tolinski? like it's so seamless on my phone, my computer or whatever. And then all of these things just, I can access my dev sites really easily. So I have like a VIT dev server for a UI running on my Mac Mini and then I tell tail scale to serve it as HTTPS, because if I want HTTPS and all that stuff, I want SSL certificate. And then tail scale makes an actual domain for it with. machine name dot tailnet dot TS, whatever, and then a port on whatever port you've assigned it to. And now I have a secure location on a dev server on my own network that I can access from any device. I can test it or use it on my phone, my computer, or whatever.
15:37 Scott Tolinski? And it took two seconds and it costs free 99. I didn't pay a cent to be able to do that.
15:43 Wes Bos? And then on your phone, what's the process of setting that up? Do you put like a VPN on your phone then?
15:49 Scott Tolinski? Yeah, you really just install the TailScale app and you turn it on. You log in, which I log in via GitHub, you turn it on. But it basically just shows all the devices that are connected through your tail. They don't have to be on the same Wi-Fi network, which is awesome. If I'm out of my Wi-Fi network, I'm on, you know, 5G, I'm out and about, I still get access to all of my devices easily through the TailScale VPN. And then I was typically running like a Molevad VPN on my stuff before, but there's actually a Molevad add-on to this. So all outward traffic to the wider internet goes through a Molevad VPN through the TailScale app itself.
16:32 Wes Bos? What is Molevad?
16:34 Scott Tolinski? It's just a normal VPN. It's just like a, you know, Nord VPN, like one of those types of providers, yeah.
16:40 Wes Bos? Okay. We talked about this on our Synology server show, but I'll quickly say how I'm doing this on myself is I'm running the CloudFlare tunnels on my Synology. And then what I can do is I can selectively opt into what like you are doing with TailScale is you can opt into which machine on the network and which port you would like to then expose to the internet. But then you have the problem of like you've just exposed your local computer or local server to the global internet, right? So then you have to then put some auth in front of that. And in my case, I just use GitHub. I sign in with GitHub to be able to access any of my stuff. And it works super well. It's also free as well. But if there's something on my
17:24 Wes Bos? Local network, like for example, if I'm away from the house and I don't feel like setting up like a forwarding or port or something like that, I can also use Unifi Teleport. So this is Teleport runs on my router, my Unifi device. And what that does, you just on your phone, you flip on the app or on your desktop, you flip on the app and it makes a VPN right back to your home network. And then you can access anything on your local home network, which is good because I have a like a 3D printer I was running some custom firmware on and I needed to be able to look at the camera and I forgot to expose it to the global internet. So I could look at it when I was away from home and I just jumped on Unifi Teleport and it did that.
18:05 Scott Tolinski? Yeah, well, that's cool. And so you have to have a Unifi router to do that?
18:08 Wes Bos? Yeah, yeah, you gotta have a Unifi router. So it's probably not most people. It's probably better to use something like Tailscale or Cloudflare Tunnels.
18:18 Scott Tolinski? Yeah, I kind of avoided the Tailscale thing because I don't know why I was like, oh, I want to expose it to the internet. And then through this stuff, I'm like, no, wait, I don't. I don't really want to do that. I want to have it all just on my devices. And as long as I'm running Tailscale on those, it's all good.
18:32 Wes Bos? Yeah, you want your devices to go back home, not have your home devices exposed to the global internet, right?
18:39 Scott Tolinski? Exactly. Yeah, exactly. Especially for this stuff. Now, there's some things that that's probably different, like my audiobook shelf or my Plex, like if I want to give like my brother control or access to those libraries as well, then, you know, I don't want him to connect to my Tailscale network. So those make more sense for Cloudflare Tunnels or to expose that port via that. But I'm not going to, I'm not going to open the ports. Okay. So let's talk about like privacy stuff because you do need to make sure that whatever you're sending to the LLMs is still something you want to be sending to a third party. Again, I've experimented a bit here with local LLMs. I have Ollama running. I haven't found anything that I love yet.
19:24 Scott Tolinski? I think my computer needs to be a bit beefier. Now, the thing about Claudebot is, is that the assistant part of things doesn't need to be that good. It needs to be good, but not like. Not OPUS 4.5 good, like, because you could have that assistant spawning OPUS agents to work on code stuff, or you could just work on it via your, your normal Claude code or whatever. So the actual personal assistant really doesn't need to be that good. It just needs to have a decent amount of context. So when I have a better computer, I will run, I think, all of the personal assistant stuff on Alama and be totally fine with that. Another thing I did is I, I've been getting into other local processes for like processing like text to speech.
20:07 Scott Tolinski? I, I use one called Kokoro. Co-ko-ro.
20:11 Wes Bos? Kokoro, yeah.
20:12 Scott Tolinski? Yes, I use Kokoro and the way I have it set up is through something called MLX Audio. It's a Mac specific thing for Mac Silicon. Basically just runs as a server on my computer. Then my Mac Mini hits that endpoint through the tail scale address, generates the audio as an MP3, sends it back to the Mac Mini where it's stored and then served via a Svelte Kit app. So what I was able to do with my chat assistant, my chat UI was build a really quick text to speech speaker button. I click the button and it reads any message. That also enabled some other neat ideas, which I'll talk about a little bit, but you can run text to speech entirely on your local machine.
20:52 Scott Tolinski? And what I found is especially with this Kokoro, the voice quality is super good. I started with Piper and it was like very like 1990s kind of computer voice. But with Kokoro, man, the, the quality is, is way good enough that I could have this read me an audio. I could have it generate an audiobook, honestly.
21:14 Wes Bos? Yeah, I've used it and you can, you can run it in the browser with Transformers JS as well. I've used it in the browser before. You have to use a slightly smaller model and it's, it's not as good, but it's still extremely impressive of how good it is. You know, we've gone from like, we had like Bonzi Buddy in like 95 and then we went to like 2021 with very similar quality. And then it's just in the last couple of years, it's gotten so much better.
21:40 Scott Tolinski? And if you want to see all of the errors in your application, you'll want to check out Sentry at sentry.io/syntax. You don't want a production application out there that, well, you have no visibility into in case something is blowing up and you might not even know it. So head on over to sentry.io/syntax. Again, we've been using this tool for a long time and it totally rules. All right. Yeah, so for me, I'd like to move as much of this stuff locally as possible. And Kokoro runs totally fine next to anything running on my laptop. And since it's all on the same network, it's just communicating back and forth via the TailScale URL. So really nice there. Let's talk about app ideas.
22:23 Scott Tolinski? I've kind of dropped some of the app ideas or stuff that I've been working on or talking about through here. But like some of the app ideas. So we talked a lot about giving this like personal assistant stuff, like what can a personal assistant do? But what about like, I have my personal assistant sending off agents to build personal software. And I did that through a like a mothership app, which is basically I built a chat UI that I can chat. I can spawn new agents. I can work with my beads and projects and then I can add on to that with individual personal software things. So one thing that Courtney and I have been talking about having forever is that anytime she asks me, "What do I want for dinner this week?", I say, "I have no clue." My brain can't think of a single thing you've ever made in your entire life.
23:07 Scott Tolinski? And so to augment that, I'm going to be taking photos of the meals she makes when she makes them. And it auto categorizes them, it auto catalogs them, and it puts them into a bank. So I could say, "Give me one of each style of food," or "Give me food ideas." And it can give me food ideas through that bank. So I have meal tracking in there now and it's just building this database. Now, how that's working is it's just taking the photo from my camera. And since it's all running locally, it's just throwing in my file system of the app. And then for database stuff, let me tell you, the number one database for this stuff is just a stinking JSON file.
23:48 Scott Tolinski? Man, I get instant saves, I get all this stuff. I don't have to log in, I don't have user accounts or user access, it's just one person. But it's just writing it to a JSON file every single time you even type a character. It is awesome that it just does that. And then now like, it's trivial for the LLM to read through that JSON file. Or what you can even do is I Slash commands to my chat UI that can actually just run script. So it could pull certain ones with JavaScript instead of having to do an LLM call for that.
24:25 Wes Bos? Yeah, that we talked about that a couple of episodes ago where I was like, I just want a brain that I can throw things into. And surprisingly, I didn't get any, any suggestions from the audience of like, I just want something where if I have something that I'm remembering, you know, like the lug nuts on the tractor at the cottage are this weird 516 spine thread. And I just want somewhere to put that or I measuring windows in our house or measuring a gap where we need to find like a dresser or something in our, in our house, like, like all of these little things. Where I just need to remember, I always think like, where should I put that? Do I put that in an Apple note?
25:06 Wes Bos? A lot of people email it to themselves, you know, and like, I just want to be able to very similar to how you use like a reminder, just be able to, to speak that into some sort of brain and have it come back. I found this thing called Sublime, which is unfortunately not sublime tech, sublime.app. I haven't tried it yet, but it seems to be
25:29 Wes Bos? that, but also a great way to like manage inspiration bookmarks, you know, like almost like a, like a Pinterest where you would dump things into.
25:39 Scott Tolinski? Dude, my, my chat UI could build that into the chat UI in an afternoon. That's what's so crazy about this stuff. And then you don't have to do some third party service. You can have it do exactly what you want. So the meal tracking thing was great. Another thing I did was fitness tracking and insights. Right now I'm just uploading a screenshot of my tonal workouts. It's doing OCR on that text. And it's cataloging that workout. And then if I give it my goals, it can come back and say, hey, I noticed you're doing a lot of traps and you, your, your traps are fine. Yeah. So like, like cut the trap workout and do something else here, like just to be able to get that insight over time.
26:19 Wes Bos? Like Apple Health as well, you know?
26:21 Scott Tolinski? You could hook it into Apple Health, yeah.
26:23 Wes Bos? And push and pull data. I think a lot about that with software that I pay for yearly or monthly. That I don't use all of the features. So macro factor.
26:36 Wes Bos? Probably one of the best apps in the world for tracking calories and all of that like health and fitness. You know, if you've used like daily burn in the past, um, macro factor, Scott recommended it, what probably two or three years ago. And I've been paying like, like 200 bucks a year for that thing. And I mostly just use it for tracking what I'm eating. And there's some nice features, right? You can scan a barcode and you can change how, like if you're having one and a half servings of something and they also have this new AI input where you can take a photo and describe what it is and it will break it into. But I, I just keep looking at that being like, this can't be that hard, you know?
27:16 Scott Tolinski? Yeah.
27:17 Wes Bos? Couple graphs of where you're at. There's gotta be APIs for all of these foods, um, available somewhere. You know, I'm, I'm sure that exists. So like 200 bucks a year, that's quite a bit. And same with like, um, I use Strong Lifts, um, which is like, like a, like a workout tracking app. And I look at that every time. It's like 60 bucks a year, which is not too bad. Like, I don't mind paying. I like that. The developer's kind of independent and whatnot. But I often think like, I'm just pressing a button every set.
27:48 Scott Tolinski? Yeah.
27:49 Wes Bos? It's tracking how much weight there is, you know? And, uh, there's an Apple Watch integration, which is an absolute must. So I'd have to build something like that. I don't know if 60 bucks is worth, uh, not use, moving away from that. But there's a lot of little things where I'm just getting nickled and dimed every single year. Every single, I'm paying for it. I'm thinking, I wonder if I could just build this myself.
28:11 Scott Tolinski? I know. I know. I was thinking about that. And then some of the stuff I just like, I just threw a subpage on and say, hey, can you throw this together? And then like 15 minutes later, it's like, done. I'm like, what?
28:21 Wes Bos? Mm hmm.
28:21 Scott Tolinski? What?
28:22 Wes Bos? To be clear, sometimes the answer is absolutely not. And it just
28:25 Scott Tolinski? Yes.
28:25 Wes Bos? will kick out a whole bunch of garbage and you, you wasted a bunch of your time. We're not, we're not being those guys that are just like, you can just think anything into existence, you know, like,
28:35 Scott Tolinski? No, but I also have good bones.
28:37 Wes Bos? Sometimes it just duplicates all of my CSS. Like, we got bigger problems as well, but I think we're on a good path right now.
28:44 Scott Tolinski? We'll, we'll talk about this in the using AI effectively, but I have good bones. I have a good system. I have good files, good agents to work on this stuff. And I use my CSS framework Graffiti, which has good AI documentation. And so like the agent says, do not add a new class unless like this UI element does not exist already in Graffiti. If, like, do not, like, all those things. And it actually has been really adhering well to that. So that, that's been like positive in terms of keeping the structure. That said, I'm sure some of these files are monstrous. Another thing I added in here is like journaling where like it will custom prompt me every day at like one o'clock and like ask me a question where I can like respond to it.
29:26 Scott Tolinski? It will be able to pull out and tag things and like mark things as wisdom. It has like 200 prompts that I haven't yet to see because I haven't, I just told this to do this last night. And then it's going to do it via audio. So I can just push record in the app and speak. And then it's going to do that speech to text and then auto categorization, cataloging, and then putting them in as markdown files into a folder. This is a good one because it has a little bit of a background about like my concussion and my recovery. I had it build a personal guided meditation. And using the text to speech, the voice is so good. It actually puts in ambient noise beyond it. So it's using FFmpeg to layer an ambient noise with the text to speech.
30:11 Scott Tolinski? And it is as good as Headspace or anything, but it's saying like, Scott, like, recovery is a journey. Like, like it's talking to me exactly about my issue and my like health thing with my concussion. And it's talking to me directly. It's, it's like, man, I've never wanted to do meditation. And I'm so like, now I have guided meditations that are like really tuned to my exact situation.
30:40 Scott Tolinski? Man, that was really, really like kind of an eye opening thing for like, wow, it can do this? That's crazy.
30:45 Wes Bos? Yeah, it's very impressed with all of the stuff that you can do. I built something where my kids have this like Yoto and I needed to like upload a whole bunch of MP3s that I had. Um, and then set icons for each of them and set some album art for, for them and like this, like they have a really nice. UI for doing it, but it's just like slow, um, being able to do it, you know? I wanted to upload like 20 of them at once. So I just boop, boop, boop, just figured out the reverse engineer the API. And built a little script to be able to find them in the thing. AI is so good at doing fuzziness, um, where things might not be named exactly as you would expect it. And it's so good at just being able to like, look at the directories.
31:31 Wes Bos? Oh, he, here's the one that you're looking for. And here's what the names of the actual files where this one had a dash in it. This one didn't have a space in it, you know, like so much of my like script writing over the past 10 years has been so built on rigidness. And then maybe writing some like reg access to make it a little bit more flexible. And the like fuzziness of data and data being a little bit weird has been so much better with all these AI agents.
31:59 Scott Tolinski? There's something really powerful about software that like you wouldn't have the patience or time to write. But like now that you can throw an agent on it and have it do it in no time. It unlocks a bunch of access to like, like little helpful things that you just straight up wouldn't have. I, for me, I've just been thinking about like, what can I augment about the, the friction in my life here? Um, whether that's reminding me about to do's or, or again, like helping my concussion recovery, et cetera. So that's where my brain's been with a lot of this stuff lately.
32:32 Wes Bos? I was doing my Christmas lights, uh, this year and I went all out on building, um, window frames. So all of the windows in our house, both of the garage doors and the entrance to our house. I built out of PVC pipe, I built these, um, window frames that have every about three inches. Um, I drilled holes in them and you poke these little LED lights in, and then our, our house was just like lit up from the sky. It was, it was amazing. And in order, in order to do this, I needed to make sure that the spacing of the LEDs was exactly the same because I wanted, you drive by people's houses and you can tell that
33:13 Wes Bos? they, they don't exactly have the LEDs in the corners of the windows. And what I wanted was just like CSS grid. I wanted exact even spacing all the way across. And if you just, most people in the industry just say, ah, three inches is about good. But the thing is, is that like over a run of, I don't know, 20, 30 lights, if you're off even one or two millimeters or half a millimeter, over 20 lights, it really adds up. And then you have a, you have an LED that's too close to the edge or not far, or too far away from the edge and it just looks weird. So what I wanted is, is to plan exactly, not three inches, but however many millimeters it was going to be.
33:55 Wes Bos? So ideally three inches. And then what it did is it figured out if it needed to bump that value up or down a little bit in order to get even spacing. So I figured that out. And then it also provided me with a cut sheet, meaning that I could cut the PVC pipe exactly. So it told me exactly how long each piece of PVC pipe should go. I told it I had, I bought it in 10 foot lengths. So what it did is it figured out the minimum waist, meaning that like, okay, well, like you take the long piece from this window and the short piece from this window and those add up to like 98% of the actual thing. So you're not just cutting them and ending up with a bunch of pieces that you can't use. It thought about it all, all the way ahead of, ahead of it. And
34:38 Wes Bos? then what else did I do? One of ours had an arch in there. So I, I had it figure out, I gave it all the measurements of how tall the middle of the arch was and how, how tall the sides of the arch were. And it figured out how much of a degree I needed to angle the connectors, which was really good. And then what else did I do with it? Oh, I had it, I had an output, a 3D printable jig so that I could just put the PVC in. Drill a hole, move it forward, it would click into place, drill a hole. That was not perfect, so I ended up modeling something myself in Fusion, but it did a pretty good job at making a little jig that I used for the first couple ones.
35:18 Scott Tolinski? Can I say I accidentally killed two hot ends on my 3D printer?
35:22 Wes Bos? Did you really? Doing what?
35:24 Scott Tolinski? Dude, I don't, there's, I'm trying to do the Gridfinity base where it does a PLA and then PETG and then P- whatever, it stacks them.
35:32 Wes Bos? Oh yeah, stacking, yeah.
35:33 Scott Tolinski? And in that process of switching the two materials, I've gotten two clugs so bad that a hot push doesn't work, extraction doesn't work. I gotta take apart the whole thing. I had to take apart the gears because the gears were stripping the filament.
35:47 Wes Bos? Oh man.
35:47 Scott Tolinski? Couldn't even pull it out. And two hot ends. And I know granted they're like 35 bucks or something, but like, two hot ends dead in like three prints maybe. I was so frustrated when the second one died. I was like, this is, oh, this is gonna kill me. And I'm trying to do the hot push where you get the thing you hold in your pushing and it's like billion degrees and you don't want to ruin anything.
36:08 Wes Bos? Did you do a hot pull?
36:09 Scott Tolinski? There's not enough in the, it's like right, it's flush to the top of the thing. You couldn't do a hot pull.
36:13 Wes Bos? I did a hot pull. I have the same one as you. I had, um,
36:17 Scott Tolinski? No, I'm saying like the, uh, the amount of filament coming out of the top of the hot end is flush with the top of the hot end. So there's nowhere to even reach in to pull. So.
36:26 Wes Bos? Oh yeah. And since you can't push it out, you can't. Oh, I see. Yeah.
36:31 Scott Tolinski? Yeah, I had to take the whole thing apart. I had like a thousand screws and everything laid out. I like got really familiar with the.
36:38 Wes Bos? I once melted the end of some of the filament and stuck it on a little piece and let it cool. And then you heat up, then you heat up the nozzle and then you, then you do a pull. I had a little fleck of, as a marble filament, I had a fleck of the marble stuck in the thing and that, that killed it as well.
36:57 Scott Tolinski? Yeah, I could still try that. I don't know if it's just the filament that I bought. It was the same filament that killed it both times. So I'm like,
37:03 Wes Bos? Oh yeah.
37:04 Scott Tolinski? Throw that filament out.
37:04 Wes Bos? It's probably a bad, yeah, bad batch. Anyways, we should do an updated 3D printing episode because we're coming up on a year now.
37:11 Scott Tolinski? What?
37:12 Wes Bos? Uh, or maybe not that year. I'm coming up on a year on mine and all the comments on our 3D printing episode where. See it and talk to us in a year, it's gonna be collecting dust.
37:23 Scott Tolinski? I could use a second one.
37:24 Wes Bos? Honestly, I have a second one and I'm thinking about like, hmm, I could use a bigger one.
37:28 Scott Tolinski? Yeah, right. Whatever. Yeah, whatever the next bamboo one they come out of is bigger. Yeah. I know. I just like built my whole workshop over the whatever.
37:35 Wes Bos? Looks good. Yeah.
37:37 Scott Tolinski? Thank you.
37:37 Wes Bos? Yeah, that's anyways, that's another show. What else? Deal finders is something I've written several times over. I've used this to find like filament is a really good example where like sometimes filament will go extremely cheap, especially on like Amazon returns. You know, people get something on Amazon and it doesn't work. They return it. And then they put it. Sometimes it's like marked down like half off. So I bought a little bot that would just scrape the Amazon API and then alert me if something pops up, things like that. I've done that many times over the years and it is so much better when you can use like a proxy man and just give AI the actual output. Be like, this is what a request looks like, you know, here's the curl command.
38:19 Wes Bos? Here's the actual return values and then it can write types for all the return data and then go ahead and re replicate all of those requests.
38:28 Scott Tolinski? Yeah. Yeah, it's a wild world.
38:30 Wes Bos? Yes, I'm, I'm gonna install this cloud bot. I've been, been holding off because everybody's talking about it and everyone's all excited about it. And I was like, yeah, I, I probably should spend a little bit of time on that. I think it'd be fun, but I gotta figure out this. This like, but the personal brain thing, I'm gonna figure this out because this is high on my list of things that I wish I had.
38:50 Scott Tolinski? Yeah, it's amazing what the, you can do with just markdown and JSON. If this isn't like public facing software. So that low friction of it has been one of the most revelatory things for me is like, oh, right. When you're just saving everything to a JSON file, everything gets way more simple, right? So that's all I got. Wes, do you have anything else?
39:14 Wes Bos? That's it.
39:15 Scott Tolinski? Cool. We want to do sick pics and then get out of here?
39:18 Wes Bos? Yeah.
39:19 Scott Tolinski? Cool, uh, sick picks. Do you have any?
39:22 Wes Bos? Yeah, I'm gonna sick pick a Roku. So we are mostly, in our house, we are a a Google TV family for the most part, meaning that we have a lot of these like-
39:33 Scott Tolinski? Do you have a yard sign?
39:34 Wes Bos? Chromecast. Sorry?
39:36 Scott Tolinski? In this house, do you have a yard sign?
39:37 Wes Bos? In this house, we are a Chromecast family. No, like, we have a bunch of these like-
39:43 Scott Tolinski? Cloudflare.
39:44 Wes Bos? Yeah, Cloudflare. We use only Cloudflare and Chrome in this house. No, I have a bunch of these like Chromecast that you plug in, right? And they're fantastic and amazing. And of course, because they're fantastic and amazing, it seems like Google's going to phase them out. But I, we recently put a TV in our gym. And the problem with the TV in the gym was that it's, if you're on the treadmill, or if you're like blasting something, it's, it's kind of loud for everybody else in the house. So I was trying to figure out like, how do I get the TV sound through my AirPods without having to pair them with the TV? Because like, that's a pain in the butt. Pairing Bluetooth, I have this pair of headphones, which I have paired to my phone
40:29 Wes Bos? and my computer, and like switching between them are absolutely awful. And apparently, if you have an Apple TV, switching AirPods is, is a little easier. But I was, I was trying to figure out like, how do I get the audio through my phone? And Chromecast doesn't do that. And somebody on Twitter says, if you get a Roku, you can watch stuff on the Roku, but then use the Roku app to pipe the audio through your phone, which then in turn, if you've got headphones on, it goes right through that. And it works amazingly well. It works so well. There's no lag in it. It sounds really good. And now you can just like watch something on the Roku on the TV, but then have all of the audio come through your phone or, or through your, your AirPods,
41:13 Wes Bos? or whatever Bluetooth headphones you're, you're working on. And I was amazed at how well it worked. Plus like the Roku UI, very like, very responsive, very nice and fast. It's like 25 bucks or something for a 4K stick. It's a good, good product. We have a Roku TV at the cottage, but I've never like bought one of their sticks before. And I, I might be converted to a Roku guy.
41:40 Scott Tolinski? Yeah, I just, I do the same thing with the Chromecast everywhere. But yeah, it feels like that isn't one of those products that Google's like, "This is really good, we're just gonna ignore it forever." Which is, drives me nuts because some of them are just get like frustrating like, "I can't even, yeah, I can't get mine to connect to my speaker reliably or anything like that." So, yeah, yeah, it's Chromecast. I'm going to sick pick a iPhone app. If you're an iPhone user, you might know that the file app sucks dearly. And every single File Explorer app that you find on the App Store also sucks because I don't know what it is about file apps that people make,
42:23 Scott Tolinski? but every one of them is like, "You don't just want to access your files. Let me tell you what you want. You want a browser, you want a VPN, you want to connect to Google Drive and all these things and do a cloud file syncing. I just want something that is Finder-like, a file system manager that's up to date. I can connect to my servers, SSH, whatever, and just get access to my dang files on my own network." So, I actually was complaining about this on Twitter, as one does, and I was given the suggestion of the app from Booger Butt Cheek on Twitter.
43:02 Scott Tolinski? Booger Butt Cheek suggested FTP Manager. And it's a nice, simple little app, nice little files app, does everything I need it to do and nothing more, and is free, works super well, very minimal. So, shout out to Booger Butt Cheek for that suggestion.
43:24 Wes Bos? That accesses your just, your local files? It's nothing, you're not even using the FTP feature?
43:30 Scott Tolinski? It's called FTP Manager because it can FTP, but I just have my, my local things connected via SSH on here and they just show up as sources. And then, that's it. I use it for the files on my phone and then the SSH into my machines on my network.
43:47 Wes Bos? Oh, so you can access your computer's file system as well?
43:52 Scott Tolinski? Nice and easy.
43:53 Wes Bos? Man.
43:53 Scott Tolinski? You can do that in the Files app, but I found something very frustrating. It caches the life out of those files. I was like looking at a file and then looking at my computer and just being like, why is this different? There's no way to clear the cache. There's like no sync setting because it's supposed to be just accessing it. And I'm just like, why? Why are you doing this?
44:12 Wes Bos? Don't give them a button.
44:14 Scott Tolinski? Yeah. Now, they, they direct you to preview. The preview app that they now have on the phone, they direct you to that to open all photos. So I have a screenshot in my Files app. I click on that screenshot and it opens it in the preview app. And then that is like, it's like, don't just open the, just show me what the screenshot looks like. Yeah, it drives me nuts.
44:37 Wes Bos? Well, that's great. I'm gonna check that out. That's, that's a problem I have is like, I want, sometimes I want to access my computer from the go. And if that's the case, I, I usually use the Dropbox app actually. Cause most things that I care about, I'll, I'll throw into Dropbox app. Or if it's like code, I usually have it on GitHub or whatever. But I'm not a dual computer guy. I have my laptop, so I usually have it with me.
45:01 Scott Tolinski? Yeah, this TailScale stuff has changed my tune on that. I think for my next setup, I'm gonna get a big old beefy Mac Studio there. And then I'm gonna get a cheapo Mac Air, MacBook Air and take that with me everywhere. And just do everything between the two because running the dev server on that through TailScale, you still get the hot reloading and all that stuff going on. It just, the processing's being done on a different machine.
45:24 Wes Bos? Thin client. I'm in. I want the, I often, every time I take a flight, I think this, I love the 16 inch MacBook Pro, but you can't work on it on a plane. It's, it's way, it's too big. It's just like a little, the 15 inch, you could just open it enough to work on a plane. But the 16 inches is too big. Or planes are getting smaller. I think that's the same, that's happening as well.
45:46 Scott Tolinski? Oh, yeah. Yeah.
45:49 Scott Tolinski? All right, cool. Well, thanks for watching. Hit subscribe, all that good stuff. Let us know, are you building any cool personal software? Did you get inspired by any of this? Let us know what you are building, what you want to build. Or do you think all this is stupid and you hate it? Peace.
46:01 Wes Bos? Peace.
Previous 973: The Web’s Next Form: MCP UI (with Kent C. Dodds) Next 975: What’s Missing From the Web Platform? Page updated October 31, 2026 · transcript and summary are machine-generated