Mourning the Death of Code

Welcome to Syntax. We got a potluck episode for you today. We're gonna answer, how do you pass context from one coding agent to another? You're switching from Claude to codex or anything else, and you have all this valuable information inside of it. How do you pass that from one? Are AI generated skills even good? What technical skills should I focus on? A a couple, like, questions today. I was still like, I don't Node. Sad about the future or unsure about the future? Don't worry. We're gonna make you feel all better. Focusing on high fidelity websites and UX. So, like, I... This guy gets these really high quality websites done in Figma, and then he's gotta implement them. And, like, what what is he even doing anymore? Right? And finally, we'd ask Scott about his pie setup. If you're starting from scratch with pie or pie durable, what do you even do? If you have a question, please submit it. Go to syntax.fm.

Click on potluck in the nav Bos and submit your question. We will answer it on a future episode. My name is Wes. With me is Scott. Let's get into it. First question we have here, and this was actually...

I think somebody... Like, a spam submission or somebody trying to pump their own thing because the link that they sent has UTM source syntax f m in the URL. So we're not gonna mention what it is because they're obviously trying to do something ESLint here, but I did think it was a good question. When you run Cloud Code and codec side by side on different tasks, how do you keep track of what each one is doing and carry the context from one to another? Notes, files, a repo per agent, something else. For context, we are the blah blah blah. I'm not gonna mention who this team is because they tried to social engineer us. Yes. We build... I don't know. They have some crappy assistant that that does this for you. But the question essentially is is when you have even multiple conversations where, like, you need to start a second conversation or you want to...

I don't know. You've used up all of your cloud code and you wanna switch over to Node or or etcetera etcetera. I even have this as well where, like, I I will have a Scott I'm like, you know what? I can use one of the cheaper or free models for this thing, but it still does need to understand where I'm coming from, and I have this other long conversation. So the question is, like, how do you pass that context on from one to another? Because, Node codecs, they're getting really good at, like, understanding what you've done in the past without having to explicitly reference it. But I almost always will say just, like, write these findings to a markdown file Mhmm. So that we can pick it on up or write comments in the code explaining why we've done this. I know people are are whining about writing comments and whatnot, but comments are a great way to add context, as long as that doesn't drift from your actual, implementation from that. And and quite honestly, the... Past that, these agents are very good at just picking up all of that context from where you are. If there's something that cannot be lost in your chat when you close it down, then that needs to be logged somewhere either in the code or in some sort of markdown file, which is usually the simplest way. I don't know. Scott, do you have any, like, memory things that you like to use? Yeah. So I use something called dex at dex dot rip to kind of keep track of where I'm at. What I like about this... This was made by, my boss, actually, David Kramer. And what... What's nice about this is that, like,

it it saves kind of, like, pro... It's project management that's kind of saved into, an actual JSON file.

So that way Yeah. Like, in query and whatever. So it kind of keeps track of where it is on steps of things. So I just have a general, like, skill... Or a general, like, context to say, keep track of what you're doing inside of here. That that, like, keeps it out of a particular agent's to do system because that's the weird thing about these things is they all tried to build their own to do systems. And it's like, doesn't that make more sense to be included in the project? Like, something you could commit, something like... Even, like, GitHub issues, but, like, something that is, like, task manage it. And then beads was a project that did that kind... Well, did that exactly. But I found this to be much more lightweight, so I tend to use this all the time. But in terms of, like, handoff, there are handoff skills. There are are a lot of handoff skills that people use that write like a... In a markdown file in your in your docs that, like, talk about which files were being worked on and the contacts and all that stuff and give a good thing. The the... My biggest beef with handoff files is it always writes into some, like, temp directory, and then I have to, like, go and give that long temp directory path. Or if they write it into the project, then that's just, like, one more junk file that's gonna get left over. It doesn't get cleaned up automatically. So you have to say, you know, clean this up after you pick it back up. What I will do JS, typically, I... I'm on top of my agent's usage, and I'm not waiting for it to be like, oh, you ran out.

So if I'm close to running out, I will tell it to write me a a hand off... Just a hand off prompt. So that's it. Like, something I could just manually copy and paste over. I I find that works perfectly fine for me. The the bigger system gets in terms of, like, how much more hands off you're getting, you obviously want those to write to files and have a better system there. But I don't do much other than, hey. Write me a quick prompt on which files you're looking at and the process you are. And then if I have something like decks that actually JS, like, individual task based with details on what it's doing and what files it's at, it doesn't need that context from a handoff anyways. I just say pick up the next task. It goes to the next task. That task has the information about what files you're working on or what the plan is or whatever, and it can just pick it up from that directly. Some more thoughts after you said that is if it's how we are doing something, I will ask the agent to distill that down into a skill so that the next time we do that thing, it's done in the exact same way. If it is, like, keeping track of simply what we need to do,

I love using GitHub issues for that because often Wes I'm in, like, a planning stage or I need to do a whole bunch of things, I will open up, like, a cheap agent and say just, like, these are the things I wanna do, research what needs to be done, and then create a GitHub issue. And then that way, you can tackle them one by one. You can tackle them in parallel. You can have three or four running all at a single time without you having to worry about, Node. Everything is stuck in this one conversation. Yeah. And I I can't go ahead and do that. And that's great because you can have 16 pull requests open at once, and then you can have one big boy that'll say, alright. Figure out how to merge all these together. And then if it's just a hand off, like a markdown document, like you said.

Yeah. I I mean, I I I agree with all that. One thing I have noticed, though, and I I wanna get your thoughts on this. You're getting my thoughts on something. I'm getting your thoughts on something.

AI created skills, kinda tough for me. I don't I don't know if I have, like, AI created skills that I like very much.

And I actually saw some report recently about AI created skills are less effective.

So I've been meaning to... And this is actually a whole another... Let me do another sidebar, a whole another conversation once you have multiple machines. So once I have this machine and this machine and they have different skills and I have them all tracked in a git repo, it still gets out of sync or whatever. Then you have global context files and stuff like that. What's in what? So, like, I've been meaning to really deeply prune all of that stuff. Sometimes I feel like the skills and default context I have are causing more problems than they're worth and that, like, skills, like, many skills could be distilled into scripts.

And then some things obviously are not, like, script based. They're, like, you know, taste based or agentic based stuff that obviously makes more sense in a skill. But sometimes a skill, something that, like, is created by an agent to be a skill, could be just a straight up script. Yeah. I find that

when I have talked back and forth with the agent for an hour or two, and you're like, I can't lose this. Like, we figured it out. You know? Like, we're there. This is how I wanna do it.

I need to, like, be able to reproduce this thing. That's at a point where you ask it to to throw it into a skill, and I wouldn't say that's, like, an agent based skill. But if you're just, like, coming from brand new and say, hey. Make me a skill that does x, y, and z, it's gonna add a whole bunch of unnecessary stuff in there. So if I look at the last skill I made, which was the photo frame, so this was a skill on how to triage all of the pull requests and issues that we were getting in the photo frame dashboard. Because what what happened is that we had hundreds of people at the exact same time submitting their logs and their issues about their photo frame and their chips and how they got around and and how they got access. Right? So, like, I'm not... There's no chance I'm gonna read all of that by hand or by eye. What what are we doing that? Reading it by hand?

Reading it by...

You got braille skills that that I didn't know about? Yeah. So what I did is I just kinda, like... Was like, alright. We need to open up a new one. We need to process all of these, and then it just asked me as many questions as you have about how to handle these things. And and we went through a couple, and I said, we don't need that, or, Node. We're not we're not having people... Like like, a lot of people are trying to, like, pull requests, like, entire frameworks that they've added. You know? And, like, no. That... That's not what this is. You know? You can start your own project. You can make your own GitHub repo. We're not adding every feature under the sun here. We're just trying to log how to do this type of thing. So I I went back and forth, put that into a skill. That skill included... Like, I I know a lot of people don't realize this. Like, your skill can be many, many scripts. And in fact, many of my skills are are more heavy on the script side than they are on the the English markdown side.

So I think agent created skills are totally fine, but you have to, like, you have to be careful with them that you don't go overboard because all the time you see people complaining about Claude doing x, y, or z, and it's because they have no idea, like, what is even being pumped into there. And then off, you'll say, oh, I didn't realize I had

I had, like, the 18 different skills loaded into the thing, and it's it's just doing what I told it. And sometimes conflicting too. Yeah. And sometimes, getting in... Sometimes, you know, I I typically write Svelte for everything, but I was, like, working on some project that was just scripts. I mean, it... It's just scripts, and there's no UI for it at all. And my agent's like, alright. Here's the plan. We're gonna build this we're gonna build this with SvelteKit and this and that. I'm just like, I didn't mention SvelteKit, and then I'm like, oh, the... I put it in global context at some point in my life, and I just forgot about it.

So you Scott go back to that. Skills, man. Such a a gift and a curse. I have that same with the agent browser. Love agent browser. It can spin up a browser anytime it wants, and it can take screenshots and and puppeteer the whole thing. But sometimes I'm writing something. It's like, well, I will construct 4,000 tests and run them all through agent browser, and it's so slow. And Yeah. And it uses so many tokens, and you have to, like, pull it back. So it's Yeah. That's one one thing I really like about OpenCode is it clearly shows you on the sidebar which, MCPs

you have, and I kinda wish I also showed you which skills you have because I would love to just be able to, like, turn them on and off per chat. I'm sure you can, but it's just... It's it's too much of a pain in the butt so that I don't do it. Maybe I need to get into a flow when Wes I'm creating a new project, I'm choosing the skills specifically for that project that I would like, and then get rid of global skills entirely. Like, that feels like maybe for that? Oh, I don't want a skill for that. What I want is a script for that. Hey. Run, what do you wanna do? Okay. I wanna pick this one, this one, this one, this one. Okay. Otherwise, I don't want any of these dumb skills getting in the way because sometimes these... They do be getting in the way. Yeah. I I... Very interesting idea I just had there. I'm sure somebody else has already done that. So we'll we'll check it out. Alright. Next question here from Weigart from The Netherlands.

Hi, guys. I get designs from clients mostly in Figma. My job is to translate these into working websites. Since most of my clients are designers or design agencies, high design fidelity is important to them. So that's what I specialize in over the years. In the age of AI, I feel like that very specialism is holding me back. I can outsource almost all dev work to agents, but the central thing I'm being paid to do and also the most time consuming part of my work, high fidelity design translation, is something that agents still kind of suck at. Yeah. Yeah. What is your experience in this area? Do you have any tips? How do you get agents to write CSS like I would do it? And do you have any model you'd recommend for this type of work? Man, I typically lean on any anthropic models anytime you're doing any kind of visual anything. I just find they they have better idea of the way something looks. Now whether or not that holds true when you're, like, deep fidelity of design and not designing itself in the agent is maybe a different story, but I tend to lean on the anthropic models anyways because I just prefer to work that way. I've been doing this because we have a a site design for the new Sanity website, which has been in progress for a long time. And one thing that I got a lot of value out of was the Figma MCP, which gotta make sure you're using the right harness because they only let you use some harnesses with the Figma MCP. That's a little bit of a rant. But with the Figma MCP, it can get the values directly from your Figma files. So it's not like measuring things. It's getting the values from them. It can read your design system. It can pull all that stuff. So if the design systems are deeply designed inside of Figma, the Figma MCP connection pulls the values out, then you can have, like, a visual checking process in the agent agent browser. Any of these things, you can check it at different browser sizes to make sure that things look correct, the values are correct, and that, therefore, you're adding in these validation and verification steps.

Yeah.

Man, in...

Like, right now, we saw in the video where Randy, the producer, went up against me in, like, trying to recreate this button, and Randy just fed it the video of the button, and it it pooped out the exact button. And then we looked at the code, and the code was actually awful, bad for performance, but it absolutely nailed it. So tips here, Opus five five and Fable and Astra are all excellent at, like, visual inputs.

So just make sure you're using one of those models if you really care about, like, the high fidelity.

He's probably also talking here about how how, like, the interaction of a lot of this stuff really just sucks at the UI. And in that case, you just kinda need to go and work on your, like, lower level primitives JS to, like, these are the things that need to be done. These are how we do things. That's both in in code examples and JS well as in, like, an agent's dot MD or a skill that you're building so that your agent knows exactly how you want things to be done.

And then, also, I think you just have to, like, go slower with these things. Like, the interaction in the... Of UX and UI design has not been cracked by a lot of these models or or any, I would say, as as far as, say, like, any of these models. They're not very good at the interaction.

So you just need to go a little bit slower and be that, like, human in the loop. Because as you said, like, if that's not you, then then, like, you you pretty much don't have a job here JS if your entire job is just taking the Figma document and outputting it into high fidelity. Like, they're gonna figure out pretty soon that they don't need you for that type of thing. But if you really spend the time on that, the subtle interaction, just getting rid of all the crappy, sloppy interface.

The care. And figuring out... And, like like, that that... Quite honestly, that is going to be your job JS how do I build something and direct these agents in a way so that I can get the output as if I I were building it. And, like, we Wes had that with the code, I don't know, for probably at least a year or two years now of, like, yes. This is code I would have written. We said that many times on this podcast. So, like, how do you get this thing to, Wes. This is interaction, HTML, CSS,

JavaScript, accessibility, all of these things that I would have written myself. Yes. Yeah. Part of that too lines up with, like, in the same thing, like you said, with Node. How you get that trust that the agent's gonna write stuff that's good and whatever. And some of that comes from your validation and verification processes and, like, how you're doing that, whether that is, again, feeding visual images back in there, but also, like, yeah, accessibility checks, all those types of things that, like, lead to better code, in, like, more deterministic, but also still, you know, agented flow. So that's something that I... I've invested a lot of time. And it's funny. This goes back to the last answer when you were saying, like, sometimes the verification or the scripting process at the end of... You know, when it's, like, working could take forever. That's the one thing I'm looking forward to with my new beefy computer is that it's actually going to be able to run all of the tons and tons of actual verification scripts that I have running on some of my projects better than this Mac mini. Because sometimes if I'm working on more than one project at once, I have, like, something that's taking screenshots and doing this and that, and the computer just gets, like, locked and stuff. Thing just fires up. I have not heard my... Like, I've had my laptop for five Yarn now. I I don't think I heard the fan once Until the last year.

Until until the last year. I hear it all the time, and it's because it's spinning up a thousand little agent browsers, taking screenshots, rendering out three d, checking them against. You Node, it's it's quite a bit. So that's an interesting thing to talk about in terms of the compute JS it's not just about, yeah, being able to run your own models locally, but it's about being able to, like, run your verification as fast as possible as well. Yeah. That's gonna be crazy. Next question from Scott your traditional CS developer. In this AI fueled error for developers, what do you suggest people do to develop technical expertise to be better to embrace AI as a tool and become less expendable in this job market, especially for developers who do not have a traditional computer science background? No algorithm classes or theory classes taken. Yes.

This is the question that... We've probably had this question a 100 times over the last year or so. People are like, like, what what do I focus on right now? Right? Like, do I do I double down and learn a lot about, like, algorithms and get that? And the mathematicians might tell you, no. Mhmm. Because their... The AI is getting really good at that as well. So I I think it's really important to understand that it's no longer enough to say, how do we use this tech to do what we have done previously? If you... If you're looking at this and be like, this was my job, and now I'm going to do my job faster with this piece of stuff, it... It's it's not going to cut it because anybody's gonna be able to do that type of thing. I I think if you are looking at this industry right now and, like, how do I stay relevant? There's a couple Wes you need to ask yourself. And, first one is, how can I apply this to new technology either to my industry or to new industries? So the people that I see that are doing super well right now are the ones that are trying to figure out, okay, now that we have this magic genie in a bottle, what can what can we use this for? What what could we... What problems do we previously have, or how do we apply this to our industry? So I'll give I'll give you one example. Right? Pattern drafting. Right? Like, my wife is a fashion designer. She sews. She buys patterns online, cuts them out, etcetera, etcetera. And a huge portion of that industry is people who design these things and and make these patterns so they they fit perfectly. Right? And and there's, like, this new software called Clo three d that's coming out in there... That people use to, like, design clothing, and it's a kinda all three d. And you can get the two d pattern that you can cut the fabric out and sew it together. It's it's pretty nifty. Right? So, obviously, my mind goes to, well, can can AI make patterns? You know? Can can you build something like the... Like, with... Maybe even with Wes? Can you make, like, a T shirt and then say, make the neck a little bit shorter, make the arms a bit longer, etcetera, etcetera, right, or body scan myself and figure out what the perfect tee would be for me? There's a lot of people in this, like... Not a lot, but there's some people in this space right now being like, how can we apply this new tech to to pattern drafting? And and, that's a really interesting place to be for someone JS.

Like your... Like yourself, the person who's asking this question is, like, how do I figure out what new industry or my existing industry the problems that we can solve inside of that? And and the other one is just so, like, be curious about how things work. So, like, I I keep going back to this, like, photo frame that that he did. If you didn't see it, I I took a photo frame, and I hacked it so that you can run a web browser or, like, Rust code or a custom Android app on it, and it's a touch screen. It's just $50. Right? It's it's really cool. Go to the Syntax YouTube channel and watch the video on it because, Well worth your time. Yeah. Wes. Well worth your time. It totally blew up. Like, something like five, 6,000 people went and bought a photo frame and have been doing stuff like this. And it was really interesting to me to see how different people approach this type of thing. You know? Like like, a lot... Some people very quickly said, it doesn't work. Sending it back. Piece of garbage Mhmm. Doesn't work. You know? And then other people are are, like, pushing... Like, some guy got Omarji running on it, and other people figured out different methods to to to do this. And I was like, that just shows, like, the the tenacity of some people of being able to figure out Node that we have this piece of tech, what can we do with it? And I wanna share this one example just as, like, a, like, a perfect example of somebody who's like, I wonder if there's something here. So he's, like, really enjoyed... He sent me a DM. He said it's okay if I share this. Really enjoyed your post. I whipped up a business called Daywell and had Crock Scott negotiate with a Chinese manufacturer to order 250 units. That's 250 of these, like, photo frames. Woah. I got 23 preorders overnight. Whipped up some ads with 11 Labs and Magic Patterns. Honestly, love your content blah blah blah blah. So, like, it goes on to say, for background, I'm a commercial real estate broker in Atlanta.

Never had a computer science background. Always love hardware and wish I could hack thing. This whole world is amazing. Gladdifying your context, etcetera, etcetera. So, basically, what he's done here... And I don't condone this, like, entire AI driven business. I I I think there's there's probably a couple little things you pnpm exploit here and there, but, these types of things will will dry up. But the fact that he saw this, no prior experience, and was thinking, can I whip up some interface for this? Because this is what we really want. Can I figure out a manufacturer? By the way, he got the $50 photo frames for $17 a pop, which is... Just shows how cheap they actually are to build. Yeah. And and can I run ads for this type of thing? So, like, those are the questions you should be asking, and and Scott, can I build a business with this? That's just what this guy did. Scam some people. Yeah. Not that. What else is there that I can use this type of stuff for? You know? You should be looking at this technology and not just saying, okay. Can I build a SaaS with it? Can I recreate Photoshop with it? Those are cool, but you Scott be saying, like, what else can we do? What other industries are there? What are things that I previously would have never have done, which is building a hardware product when you are a real estate, developer and and sort of take that and run with it? So I thought that was a great example. Yeah. Just because you, have tenacity to push through and try something, right, rather than just, like, oh, LOL. It doesn't work. Okay. It sucks. Yeah. No good. Like, you're not you're not getting getting anywhere if you have that that attitude towards these types of things. You you have to... Unfortunately, you have to work Yarn, and and you can't just type in the Bos, expect everything to work. I guess this guy just did type in the box, though.

Yeah. Well, to some extent. Yeah. Yeah. I I I mean, I 100% agree. I don't really have too much to add on top of what you said there. But, again, yeah, I know their... The AI is swallowing up much more than one or two industries here. But my dad went to school and was doing, like, technical drafting. Like, he was... He's an engineer, and he was doing technical drafting. Imagine you you get really good at a skill like that, which is drawing, you know, parts, and then cab comes along. You know? It's like, oh, this is... Like, my whole thing is now wrapped up in a thousand other skills. And so I think about that a lot and, you know, the amount of, like, transition he would have had to go... Undergo to, you know, stay on top of, engineering in that time. So,

just the transition, it's all all feeling weird for people. It's true. And it's hard. Like, we Yeah. We feel for you. We feel for everybody who's who's having these questions right now. I feel the same way. Yes. Wes. But

keep at it. Keep at it. Yes. Next question from Syed. Hi, Syntax team. Longtime listener here. I'm a heavy hard I'm heavy harder to say this, but I think Syntax being a corporate Yarn. Property is finally starting to take its toll. Oh, no.

I've resisted accepting this fact for a long time, but your last potluck put the final nail in the coffin. Two points We're done.

We're done. This is a spooky story right here. Two points in particular rubbed me the wrong way. One, you claimed many people's livelihoods, pension investments, etcetera, are tied to AI companies, and wishing for AI to collapse is wishing bad for innocent ordinary people.

You said that writing code by hand is a thing of the past unless someone wanted to be artsy and handcraft their code.

Okay.

We did say those things.

Wes.

I think part of this is is...

Yeah. I mean, Wes, you might have some more formed thoughts on on on this in general. Yeah. Well, let's let's do first one. The first one is, like, you claim many people's livelihoods are tied to AI. That's true. There's so much investment, which is therefore in in people's pensions JS in these data centers and whatnot. And and we didn't say that they were good. In in fact, we got some really good DMs from people explaining to us how the water works in in these types of things because nobody seems to actually understand. It was it was very, well thought out.

But, like, I I think that is true. It's... As as much as I I don't want this to be true, there are many extremely large companies that are very highly leveraged in making sure that this AI thing pans out. And if it doesn't, then those things are going to spiral, and it it might happen. Yeah. I'm not saying that it it wouldn't, but, unfortunately, what that means is if this stuff were... If if it were... The bubble were to pop, then a a... Not the entire economy, but a good chunk of the economy will go with it.

Yeah.

Yeah. I mean, I yeah. I don't... It it does feel like...

And it... That things have progressed so fast.

So, like, I just don't... I don't feel like, like, if the AI industry collapsed, I mean, the the .com bubble collapsing too.

I I think after the .com bubble collapsed, it's not like computers went away or the Internet went away. Yeah. But there were a lot of people who... Just innocent ordinary people who got lost their jobs from stuff like that. So Yeah.

You know? To be clear, I I I also don't see how it is gonna pan out of, like, the amount of money that is being poured into all of the stuff, even just the amount of money that's poured into people right now. You're seeing a lot of people switch jobs right now. And and if you heard, like, some of the figures that these people are getting, it's unbelievable how much money that these big companies are are spending right now just to try to, like, win it. And I do not understand... And I'm just a just a lowly developer here, but I do not understand how a... Somebody spending, like, a $100 a month plan or $200 a month plan for power users is going to convert to the... Like, that making that much money. So I'm I'm just saying here that that comment was simply just like, this is how a lot of it is working, and and we ESLint to the YouTube video. But I I also am a little bit afraid of it as well. But point two here, writing code by hand is a thing of the past unless somebody wanted to be artsy. I I believe we said handcrafted, artisanal.

With love.

Yes. We used to say that all the time for for not using, off the shelf stuff.

Yes. Yarn. Handcrafted artis and all.

And, unfortunately, I think that is is true as well. I I wouldn't say all of my code is is, AI generated these days. I certainly have to hop in there and and do quite a bit of stuff, especially for a lot of the UI interaction stuff.

No worries. But I do believe that that was a a fair point as well. But he he he goes on to say is that given the controversy around the whole end AI industry, I would say the responsible approach for the most popular web dev should be neutral, if not slightly on the defensive side. I, Said, I don't think you get to tell me how we feel about this type of stuff. And, in fact, like, I'm... I feel like I'm a fairly balanced guy on this podcast, and I feel like I have a a decent take, and I'm able to change my mind when when people will message me and say, hey. You you weren't right about that. I don't... I disagree. Sometimes people have much better informing opinions there, but I'm not going to intentionally just take a a side... A defensive side on this because we we have to be. I'm I'm I'm trying to spit it as straight as we can here. Yeah. And I also think,

the thing, you know, obviously, you all get defensive about stuff because I'm a sensitive guy. But the, Syntax being a corporate property is finally starting to take its toll. I wanna say, we have complete creative control and complete ownership over Syntax.

The only thing that Century has ever said to us is talk about Century less because you're saying Century too much.

They they are so incredibly hands off and supportive. They give us a budget. They give us, no direction whatsoever other than opportunity to try things and to get good people on the team. So any change that you've seen in syntax is coming from Wes and I, I mean, and the team overall. It's not not... It's like, this isn't this isn't a corporate thing. No. So I don't I I understand why you feel that way. But Yeah.

Anyways, you should check out Sentry, though. Yeah. Sentry.io/syntax.

Right? You should check out Sentry's SEER. Let your issues fix themselves.

They will find the root cause, and you can at 2AM in the morning, it'll submit a pull Wes. So the thing will be fixed. You wake up.

Your your code has been self healed. So thank you, Said, for the,

the opportunity to to plug century at century dot I o forward slash syntax. Yeah. And I do be feeling a little bit like Garth in Wayne's world right now because I'm saying that, and I'm wearing a Century T shirt. I took this... I took off my Century sweatshirt to to reveal a a Century T shirt.

So I understand, but, they're just a really cool company, and they help us do the cool stuff. So, Child of Century. Yes. Okay. Evan Wills says, how do you handle reconnects and web serial? I thought this was a really cool question. So I'm building a browser ID that compiles firmware for an Wes 32 smartwatch.

But, essentially, it's taking a tiny little ESP 32 microcontroller and and popping that into a screen vibration motor color display and making kind of like a DIY smartwatch, which I I think a lot of people would be interested in. Yeah. It flashes over web serial. Your episode six twenty eight with Nick Herr on JavaScript and the ESP is partly why I'm asking you. When the s three resets into a bootloader, the connection can drop and come back. Yeah. So when when you put a new firmware on the Wes 32, it has to go into a special Node, which can then accept the new firmware, and then it will boot right back up again. The problem with that is every time you connect to something on web serial in the browser, you get the stupid little pop up where you have to, like, select it. It's the same thing, like, you know, when you have to select your webcam from the drop down.

There's no way to automatically reconnect to that type of thing.

So what are your options here? A, you would have to, like... This is why a lot of people go electron. It's like you still wanna use the tech, but you you want a different experience, you'd have to ship, like, a custom browser that just automatically reconnects without the the security thing of web serial popping up every single time.

The other option is you... Can somebody just run this... Tell the users to run this type of thing in in the terminal? Make make a desktop app that somebody can download. It's so nice to do web serial in the browser because then you can just reconnect to that thing, or you you could probably solve this with UX as well. Mhmm. But the nice thing about doing this flashing stuff in the terminal JS, like like, I've got a badge right here that you can connect to over USB, and I just have that thing plugged in all day long. And it's just constantly throwing new firmwares on it Yep. Running logs on it, checking, oh, what's... What what what is the shape of data that comes off of this sensor? And that's just not possible in the in the browser. So web serial is more just for, like, I wanna flash this thing quickly. But if you're just, like, constantly iterating on something, I probably would look at some desktop app or, like, a CLI two e.

Yeah. Yeah. Oh, man. I... Wes, my my prop boy, which is my ESP, 32 s three device with the screen on it, I I had to really quickly take it apart. It's such a tight unit.

It got really hot while I was working on it, like, a very concerningly hot. And so I was just like, I have all these little screws, and I'm like, I gotta get this thing open. You know? It wasn't plugged in even anymore.

So... Yeah. Yeah. That was a lot of fun.

I was a little concerned for a minute.

Probably... Yeah. I've I've made a few of them very hot, and it's usually a short of some...

Something is shorted to the wrong pin, and you're sending voltage. Oh, yeah. Maybe something got pinched or something.

Yeah. That's probably what it was. And, unfortunately, I've... You have... Sometimes you have to chuck those in the bucket because they're they're pooched.

Chuck it in the bucket. Yep. Chuck it in the bucket. Yes, sir. Alright. Next one from Reina.

I'm a longtime listener of Syntax from Vancouver, Canada. Oh, welcome. I didn't really love... I wanna go to Vancouver. That sounds like so much fun. I've never been to Vancouver. That's crazy. You're from Canada, though. I mean Yeah. You're right there.

Traveling inside of Canada is very expensive, so it's just not something we do. Yes.

That'd be cool, though. I feel like I... Yeah. I'm closer to Vancouver than you are. From the very first episode... Oh, thank you. And have grown as a developer alongside the podcast. Going from not understanding what React was to building full stack production applications, you guys have been a source of invaluable information. I also bought courses from both Scott and Wes. Well, thank you. Oh, man. Laying it on thick, laying it. Yes. My question JS, I have spent many years building an internal React web application that runs our entire family business. Over the last few months, I've really ramped up coding with AI, currently using Cloud Node in Versus Code.

While this has allowed me to work through massive backlog of things, I've always wanted to add in the app. I find myself reading less and less of the code and in some cases, not fully understanding it.

Node my gosh. Don't tell Saeed.

Oh, no.

As the features I want to become more and more advanced. While on one hand, I'm solving very real business problems quickly, on the other hand, I'm starting to realize that writing the code wasn't really something I fully enjoyed.

I was always...

It was always the end product that I looked forward to.

I doubt I would start a new project if I wasn't able to use AI. Luckily, because I had to learn things the hard way, I understand enough about programming system architecture, UX, UI design.

My question is, how do you guys feel about this new paradigm of building? I don't mind AI doing the work. I know I can do myself, but I feel weirdly guilty about replacing the hard Node learnings with AI. Man, I feel this all the time. I think about this all the dang time. But in the same regard, you would not believe how much I learned about building websites for Internet Explorer five, Internet Explorer six, Internet Explorer seven. You would not believe all of those things that I spent time and energy into learning that I do not ever have to think about again. Like, how many of those those tools and techniques,

like sliding doors Node know Wes all the semicolons go, all the commas. Yeah. I know that. All, like the syntax.

I I I know, like, what, three different clear fixes. I know all that stuff, but there's so many of those things that I I spent real hard earned time having to understand and learn and then, like, became invalidated with other things and Yeah. Frameworks that I spent... Even even learning React.

Right? Over the course of React, you had to learn, like, 18 different ways of doing things in React. So things come and go like this. So I wouldn't feel guilty about it, but I I I understand what you're saying, and I limit the the skills that I've worked really hard for. And I've had to actually, like, learn that you, like, need to, like, kind of mourn some of this and, like, really truly, like, understand that you put time and effort into this stuff. And I I I think you can't look back on it too much as, like, looking back because you're not gonna be looking forward.

And, unfortunately, in this world and industry, you have no other option other than to look forward right now. So, if if you're enjoying the work more now, don't feel bad about it. Yeah. Don't feel bad about it.

This is also always the case with with these things JS that... Like, I saw this quite a bit because I I sold courses to tens of thousands of people, including Brag. JS that... Yes.

A very, very explicit brag there. But what I would see is that, like, half the people... And, like, even when we did, like, in person, half the people were, like, didn't care about the code. The code was just a means to an end, and they just wanted to make the product because they had it. And then other people, like, we very deeply cared about the code. Yep. And and, like, that... I I probably in in both of those camps. Right? Like, I was very excited about actually what we're building, the problems that we were solving, but I also got really excited about new CSS syntax, you know, new media query. That was so exciting to me.

And a lot... Not all of that is gone, but a a a good chunk of that is is going away.

But I think you said it said it well yourself here JS I I understand enough about programming, systems architecture, UI, UX design. Right? Like, a lot of those things...

None of those things are going away. Knowing where to put the semicolon, is simply going away. But, like, caring about the architecture, the programming, the the interface, all of that type of stuff, you still need to certainly care quite a bit about those. Also, like, if you're the type of person that is just excited about the end product, like, the the real estate agent that we just showed earlier, excited about the the end product here. Right? That's good. And we're probably just seeing a little bit of a shift in terms of, like, who is getting... Is going to be very good in this industry. We we also see it with, all the old timers. Like, I have been in this industry long enough that when I first started coding, I was, like, alongside a lot of, like, the early people at at Shopify and, a lot of these different companies. And and you're seeing a lot of those people who are... They Scott of phased out of the code into the management, into the upper c suite, and you're seeing them get back to the code now because they're able to do it. They're able to just prompt the way they want, and they are very good at product design and product market fit and all of those types of things. So I I think we will see a bit of a a shift in terms of who is really good in this industry.

And the people that really cared about the code and not so much about the end product, maybe we'll have to switch what what it is that they're excited about.

Or they could just start making their own harness. That's what everybody else seems to be doing. Right. You build your tooling, which I really love to do. I love to build my own tooling. But you know what? For me, it Wes something I always, like, really emphasized in teaching anyways, is that, like, people don't care about the, the people don't care about the the the tech as much, and not not everybody. Right? But, like, when you're first starting, especially, right, you just care about, like, the ability to make things. The output. That's what gets you excited. Right? The output. Yes. Yeah. You... And and so, like, when you Vercel attempting to learn things, you put intention behind that where, like, you're not just building it to do app. You're building, an app for your mom because she needs it for these reasons. Right? Your motivation is completely different. So, yeah, getting into your motivation. And it seems like you have the right motivation to excel in this new world, which is constantly changing.

Next one here is from develope papa.

Have you heard about AGUI for agent integration? No. I haven't. This sounds like an ad. You recently made an episode about WebMCP.

It seems like they are trying to do the same thing with totally different approaches. What is your take? AG, you... Wes, you know about AGUI with a... Okay. Here's microsoft.learn.

Oh, no.

No. There. Yeah. Well, maybe that's it.

AGUI, not to be confused with a two UI.

So this is a interaction layer from the folks that do Copilot kit, not to be confused with Google Copilot or Yeah. Copilot kit things confuses me. Yeah. They need to change that name, unfortunately. That's... Copilot is kind of Microsoft owns Copilot. You know? You can't Copilot. Node that. But Yeah. So let let... Let's talk about this space really quickly. Right? So you have your application, and then you have an agent. Right? And in order for your agent to communicate with your application or to use your application, there are many different ways. Right? There is computer use where an agent can simply just use the browser.

There is MCP, which is you set up, like, kind of a headless look at, and you can call tools back and forth and get the raw data out of it. Right? There's WebMCP, which is you have your existing website, but you can also register those tools as an MCP. And then you can do things like you can use the website yourself, but you can also tell your agent to say, hey.

Log these five expenses in in my invoicing software. Right? And then there's this whole other world of generative UI, which is you are talking back and forth to maybe an MCP, but instead of getting back just raw data, you also are getting back, like, UI that would be probably integrated inside of your chat. Right? That's what, MCP UI or MCP apps, the whole... That whole idea is. Right? And all of this stuff is just constantly changing, and and what people are using is is constantly moving as well. But that's kinda like the space right now. An AGUI is sort of a layer in between your application, whether it's a website, whether it's an iPhone app, desktop application, and your actual agent. And it it Scott does, like, back and forth. It has one really interesting thing that WebMCP doesn't have, which is it's all event Bos, meaning that, like, when something happens in your application, it then can be triggered. Whereas, like, WebMCP simply just like you can you can visit the website.

You can ask for a list of items. You can ask to update an item. Right? You're just doing tool calls back and forth, and that might change in WebMCP, but that's kind of a interesting point... Part of AGUI.

AGUI is developed by the CopilotKit folks, and CopilotKit is one of the ways that you can consume the agent data and then turn that into actual UI. So, we're talking about, like, generative UI. We're talking about, you're chatting in chat GBT, and then you have, like, an inlined, UI. Maybe it's like like a restaurant that you wanna book or maybe it's, some to do items that you wanna check off, shopping list, etcetera.

One of the ways that you can consume the AGUI data is via rendering out some HTML.

So AGUI seems to have certainly...

Has some some big momentum before them. It's very actively developed. They seem to really know what they're doing. I don't really care if WebMCP wins over this or even if they are a total overlap between the thing.

I'm just excited to see more things of of being able to use applications agentically.

Yeah. Yeah. You know what? It it is very funny. So I built WebMCP into my video app and stuff like that. And then I was chatting with my own personal agent, and it would... It suggested creating an animation in my video app. And I was like, can you do that? And it's like, I'll connect over WebMCP.

And then it just made it connected over my WebMCP and and gave me, like, a prototype of something. And I was just like, man, WebMCP rules. I I... It's... It rules because it's built into the browser. And and I don't know. To me, it it just feels, like, easy to adopt, if that's the case. But, I haven't seen this before, and now I'm on their website. It's just crazy how much stuff has just shown up, all these new things.

All of them. And it's it's also... Like, I think the hardest part for a lot of these things is just being able to explain how these things are. Like, I guarantee I'm gonna get CSS from the people at this company after this goes goes live being like, you didn't really totally explain it all that well. You know? And it just has to to fit perfectly into it. And though I'm I'm pretty excited about WebMCP as Wes. Although I do wonder about the whole events thing, because if you want to say, hey. When something changes on this website, you'd... I I think right Node, you have to sort of pull for that Yeah. Type of thing. Whereas Yeah. Being... We've learned new programming. Event based programming is very good. Wes.

Yes. Alright. Next question here from Anan. Amazing content. Can you do a Py setup video? How you use it and full flow?

Yes.

I I I put this one in there because I I don't want you to tell us your full flow and do a whole video here. But I'm I'm curious, like, Scott, if you were starting absolutely from scratch either with pi or or, like, pi durable, like making your own agent, like, what's your what's your process for that?

Yeah. Well, I think the number one process that, like, you should think about, especially getting into something like pi, is that what a lot of people find is that less is more. What's so enticing about pie is that you can extend it to do whatever your heart's content. And you go to the packages, and there's this massive catalog of packages promising you all kinds of stuff. And some of them are very good and are super necessary, and others are are less necessary, and some might even, you know, hinder things. Like, recently, there was some, evals and benchmarking around, different harnesses, and just stock pie scored better than o m p, which is basically pie, which is... What is it? O my pie? What JS o m p? O m p dot s h. Well, I mean, what is it, like, Oh My Z S H? Is that what it stands for? Oh, yeah. Yeah. Yeah. Yeah. Oh My Z S H. Oh My Pie. So that's the same thing. Node though. It's o m p. And... Oh Node. O m p's got the coolest website, bro. The... It's got the coolest website. They actually have a really good terminal called TURN that I got access to that is sick. There's a lot to like here. And what there is is that this is just a collection of really tuned in and picked out Py extensions.

But I wanna say that, like, the the thing to get into with py is to not install too much too quickly.

Install things as you need them or want them or want to try them out and experience it with them or without them because, I've gone through the flow of having a lot of stuff and a little stuff installed. And right now, I don't have very much installed in terms of, like, my PI setup.

But if there are things you want to accomplish, you you can install them. For instance, there was no good WebMCP one, so I just made a WebMCP extension for my my Py agent. And, like, super easy work Node problem. And and, like, in general, I Node an MCP extension before, but now MCP is baked in anyway. So I would say start minimal. Only add what you need to add, and don't don't go nuts unless you, like, find yourself, like, really needing extra stuff here and there. Because... Yeah. I I I think that, like, too oftentimes, we we wanna, like, get too in the weeds of, like, really customizing when the minimal setup is is very good. Yeah. Yes. And that kinda goes back to

some of the previous questions we had JS to, like like, what should I focus on? Like, a lot of a lot of people are like, maybe I should, like, figure out how to, like, rice out my system absolutely perfectly.

But in some cases, you don't need to. I I think you still do need to for for a lot of the UI stuff we're talking about, but that might not be the case in the future. Yeah. I... You know what? My my Py durable setup, Wes, has gotten a little bit more riced up. But even then, it's not extensions because it's just like

you... You're connecting tools and making it all custom. I recently... The best thing I did in my py durable setup, which py durable is essentially like...

It's a durable agent where it's writing everything to essentially, like, a ESLint of what it's doing at any given point. So,

It's kinda like a like a skeleton for building your

own agent. Right? It's a skeleton for building your own agent. Yes. It's a skeleton for building your agent where, like, pie is, like, you know, an extendable harness. This is like an extendable agent type of deal. And one thing that I did with my UI that I have going that powers my my picture frame, of course, obviously, with my whole family's, like, agentic stuff JS I added a little, like, little tool that you can highlight part of the UI. It pops open a little window. This is by no means novel. Other people have made this. But you just highlight a part of it. You just say what's wrong with it, and it sends the DOM. It sends the component. It sends the the the route and then my little thing off to an agent to fix it. My god. That little bit, has been so nice. And then it it, you know, it just fires up a a durable chat. I can just go in and chat if I wanna go back and forth with it. For the most part, it just fixes things.

But, yeah, that... That's pretty much my setup. My my my py durable has been getting more use than my py lately,

but I I use them both. Yes. Beautiful. Alright. Let's move into sick pick. I got a sick pick for you today. And this is probably not for everybody, but I have lights and all kinds of stuff in my office. And I have these stupid little tripods that they they go on, and they get in the way. I trip over them all the time. They... They're very tippy because they're not super high quality. You Node? They're not like like those high end CSS stands that you would get.

So what I did is I found these brackets that you can wall mount, and they are amazing because I can wall mount all my lights. You can wall my... If you got, like, a camera or something like that, you can you can mount it on that and, basically, just swing them wherever you want. And the most important part is that they're... They don't take up any floor space, and I can walk under them JS they go high enough in my office. So just a major upgrade to this, like, kind of craziness that is an office that has to have... Like, you got lights and cameras and all that all that crap in your office. You have action too. You have lights and cameras.

Action as well. Action? Wes.

So this is the one I got. It's from Neewer, which is, one of those, like, Amazon photography brands that is Yeah. Puts out pretty decent stuff.

Yeah. One of my new... Newer, LED panels is slowly going. Like, I would... I'd say, like, 50% of the the nodes are out, but I've had it for ages now. So, so what what is that called exactly? You would say it's a a bracket. It's called a

a boom arm.

The photography stuff is really weird, man. Just trying to search for what you want. I figured out, like... I didn't know what the little stick was that lights can screw onto, and I figured out it's called a baby pin. Oh, baby pin. Okay. Baby pin. Very odd thing. Very odd. The photography world is...

Man, talk about having, like, Yarn acquisition syndrome. Every time I need something for my cameras, I'm like, oh, man. There's a lot of cool stuff.

There's a lot of cool stuff. I really wanted to get a giant, horizontal, like, bar on my ceiling. Oh, yeah. Like, a whole rig. When I was at the Microsoft Studios, they... It was like they had a, like, a grid on top, and I was just like, oh, this is so cool. The grid. I want the grid.

But, obviously, that's very, very expensive.

This thing's cheaper than I was expecting for how heavy duty it looks. $40.50 bucks. Yeah? Yeah. You can get a newer, horizontal boom arm for $36, and then I would just need to find some way to attach that. And then I could have my, because the ones I were looking at were much more expensive than this unnecessarily on b and h. So, yes.

That's good. I was also looking at the the poles that you, like, twist and, like Yeah. They're, like, friction fit. Those are really nice too because then the floor space, all they take is just, like, a pole instead of a whole tripod.

Yeah. But I couldn't find any that were, like, a reasonable price. And I almost went and bought those, like, drywaller poles that they used to, like, hold up sheets, but they were so ugly. And it's kinda... I I I opted for those. I'm very happy. I did. Yeah. Yeah. I bought a second one because I nervous. Two lights. Yeah. Attention starts... Especially with kids. You Node? Someone tries to, like,

swing on it. You got you got two of these nice lights? Yeah. I got different ones.

Yeah. The Amaran one fifty c's. So I've got my main one lighting me up. And then I have... Hold on. I can... Let me turn it off just to show you. I have, like, at the backlight. It's... You can hardly see it right now, but it just kinda fills out the back.

I could turn it up too.

Whoo. These Amaran lights are so sick. I I love this thing. I have one, one light over there that is, like, kind of like a fill, and this is, like, my key right here, which is my Amaran. And then I have one down here that's a backlight. And these ones are just, like, cheap, LED flat panels that are really harsh on your face, so I don't point them at my face. But Yeah. And then I got some some poles. I got a couple poles in here that are purple and blue to give us some color.

Lighting, so much fun. I'm gonna sick pick something that's also a lot of fun, and that is a Monoprice, ultra thin Ethernet cable. They come in all kinds of different colors to match your stuff. I got them for my rack, and, they come in... This is, you know, like, the the smallest, size one of them. It's like a tiny little... Like a a five inch Yes. A patch cable. But you can get the same cable up to 50 feet. So the same cable has all kinds of different sizes.

But let me just tell you, if you're an aesthetics person, if you're like, oh, I I gotta have my Ethernet cables be looking the same, and they gotta be looking good. My god.

I mean, one, the cape...

One, the cable is so thin, and it's not like your big old chunky Ethernet cable. It's so thin and beautiful.

I love this little cable.

And I think it will be it too because, like, Amazon is full of, like, CAT 19, even like you. And it's like, you can't trust that stuff. You know? It's full of aluminum wiring. Monoprice is very reputable.

Monoprice. Great. Yeah. So, man, these things are good. I'm actually gonna get a few more of them to, like, just standardize because I have, like, the rest of my rack. Like, the front of my rack now has all these on the front, but, like, the the back end of it is all, like, 18 different looking Ethernet cables that I've had from Aw. Over the course of my life.

I made all my own Ethernet cables for my rack, and it looks awful because it's just this blue.

And then I got these, like, orange jackets for it, and I just... I wanna go all all new. But I I can't justify it. I mean, they're they're they're... Yeah. For yours, you see it all the time. Mine mine is just in the utility room at our house. They make these in blue?

I don't want blue, though. Blue's ugly. Okay. Will they make them in red or yellow or white?

Yes.

Awesome. Alright. Sick. Well, that's all I got. Alright. Thanks, everybody, for tuning in. Please send us your questions. Go to syntax.fm.

In the navigation, there is a potluck button. Click it, submit our question, and we'll answer it on a future episode.