Loading summary
Transcript63 lines
- [00:00]
A
My goal for 2026 is to fully transition into a builder with AI. Even for me though, it's really hard to keep up with all of the latest tools and tactics that are popping up every single day.
- [00:12]
B
We're in a really Magellan exploratory era with all this stuff. Nobody really has definitive answers.
- [00:20]
A
So what are some of the right steps that you can take to strengthen some of these design engineering muscles?
- [00:26]
B
Everybody's gonna start having their own workbench and that wall at the workbench is going to be a little bit different for everyone.
- [00:35]
A
What trends should designers jump on and which ones can we safely ignore?
- [00:39]
B
Claude Code can be so many things that aren't coding. Claude code is just Claude Computer.
- [00:48]
A
Welcome to Dive Club. My name is Rid. And this is where designers never stop learning. This week's episode is with Kyle Zantos, who's leading the new UX Tools labs, where he's responsible for figuring out what matters when it comes to new design tools and new work workflows. And as a result, he's pretty quickly become one of my go to resources as I navigate my own journey building with AI. So this week's episode is a deep dive into very specific tactics that you can use on your own design engineering journey. And to start, I wanted to get his take on some of the new Cursor updates.
- [01:26]
B
You know, last time I was on here, we were talking about Warp and using cloud code inside of Warp and stuff. And that is still a great way to go. But Cursor, they have an in app browser, which is cool. What they added recently was this visual editor and this got a lot of people talking. There was a big debate about it. That was spawned essentially from Cursor's positioning, was like, hey, look at all these tools. Aren't these familiar to you as a Figma user? And isn't it cool that we have these? There shouldn't really be a divide essentially between like design and engineering and stuff. And so I, I started experimenting with it. We didn't just get Figma, but for code that's still not, still not happening. But when you're doing tweaks and stuff and want to just play with like spacing or arrangements or something like that, it's a pretty easy win because this is an app that I made for my friends and I. We've had a long standing NFL pick em pool. We used to use Excel spreadsheets and then we were using the Yahoo Fantasy app this year. They were like, could you maybe make one for us? And I said, why don't we try? So I did, and we've used it all season and it's great. But I might want to tweak. Like, this has some extra padding that seems unnecessary, and I might not know the exact value that I want to change for that. So it's nice to just be able to do subtle tweaks. Kind of go like, yeah, that's a little better. Maybe I could, like, bring it in a little bit. Okay. Okay. You know, so I can. I can do little tweaks like that. Like. That's terrible.
- [03:15]
A
That was terrible. But also really cool to be able to do that in the ui.
- [03:19]
B
Exactly. So it's bad. Bad design decision, but nice to just throw it in there and see it immediately. Right. And so it's, it's keeping you out of the prompt loop that I think designers can get stuck in or annoyed by, self included. And it, and it lets you play and experiment quickly. Right. And then once you dial those in and you can go like, as nerdy and as engineer deep as you want in here, you know, with both the design and the css. And then you just hit apply over here to have the cursor agent. So I still don't even pay for cursor right now. It just still isn't my, like desired AI. But from quickface standpoint, I'm using Claude code in a regular terminal in here in cursor again, and I'm using the crap out of the in app browser and the visual editor to just kind of imagine, tweak, experiment and get it in in real time. So I think it does matter to designers from that. Let me do some, some tweaking and some kind of experimentation, but it's not, oh, this isn't the full like figma is dead and vectors are dead moment. If there ever will be real quick.
- [04:35]
A
Message and then we can jump back into it. Fun fact for you. I'm right on the cusp of the top 1% of granola users. So when I say I cannot get enough of this product, I mean it. It is that good. But after seeing their end of year crunched report, I have even more respect for this product. I never have a conversation about design without grunting granola. And somehow it perfectly, perfectly organizes all of it for me, whether it's chatting with guests for Dive Club or doing user research, or keeping track of tasks for inflight. And then Crunched gave me this beautiful synopsis of all of the key themes and even some funny moments from the past year. It's Just a good product. So if somehow you're not already using Granola, head to Dive Club Granola to spin it up for your next meeting. Another week, another huge lovable release. They now have connectors with products like perplexity, 11 labs, fire crawl and even Miro. So you can build apps that pull live data from anywhere on the web or maybe even talk to you. And it's as simple as writing a prompt like, read this page out loud when users tap play connectors totally blow the roof off of what you can create with natural language. And you can start today, just head to Dive Club Lovable to start building. Okay, now onto the episode. Is it enough though for you to confidently say to a designer who hasn't adopted any of these tools yet, hey, Cursor probably is that first real code editing tool working with production. Because it's kind of, at least for me, like I'm in warp right now. Like, I left Cursor to use work with cloud code and now I'm looking at this like, man, I got some fomo. I'm not gonna lie.
- [06:21]
B
Yeah, you probably should have FOMO because it's pretty convenient. There are other ways to do this kind of, but as an entryway cursor, as an app, like even on the free plan. Yeah, you know what? This is the other reason that I'd say it is a good. This is my tool for using Claude code and other CLIs as a designer starting out. This is the other reason that I'd say that there is something to be said. I still like spinning up CLAUDE code or Droid or codecs or whatever in the cli. I think it works the best. But from a stuff can be scary standpoint, the CLAUDE code extension, it's the VS code extensions which again Codex, Factory and Cloud Code all have. Those are all the CLI based AI agents that I use for coding. They have these interfaces essentially via these extensions that might feel a little bit more friendly to designers that are getting started and you can do all the same stuff and you just get it in a layout that is a little bit more friendly for folks that are still a little bit scared of the terminal. Plus the other thing is that, you know, if you have a bunch of text and stuff in the terminal, you can only use your keyboard to navigate button. Here I can do.
- [07:36]
A
It's a big deal actually.
- [07:37]
B
And, and amazingly that's kind of a huge deal. Even I still go back and forth. Like I kind of feel like even certain plugins or workflows still have a few Things that only work well or do what it's supposed to do in the true terminal CLI version, as opposed to the extension. But if you're starting out, I think that it's perfectly fine to use the extensions and give yourself a more friendly UI for a designer to get used to.
- [08:06]
A
Okay, So I want to keep tapping into your perspective as somebody who is a serial tinker. And you're always, like, playing with different workflows and trying these tools, and you're just learning a lot that I think designers can benefit from. So I kind of asked you to come to the table with just some of these little experiments and lessons and insights. And really the goal, at least for me, is to try to help designers who maybe feel a little bit overwhelmed by the amount of stuff going on, stay up to date with what actually matters, and come out of this with some practical tactics that they can immediately start incorporating into their own workflows. So given that, what are some of the things that you've been adding to your own toolkit lately?
- [08:50]
B
Yeah, and I gotta be honest too. Like, I play around with tools, I tinker a good amount, but I've figured out how to limit myself better. And that's been important because I really, I'm. I'm skill building right now. So much of my time is spent trying to fill gaps that I have in my knowledge to become closer and closer to being able to call myself a design engineer. You know, exploring micro animations and things like that, and just learning code things in general. Right. So, like, there's so many tools, and we're going to be in a tool explosion right now. So there's a lot that you can ignore. The stuff that I have today is more, because I tried to not be uber specific about this tool and this tool and this tool and more like a, a philosophy and approach in a way that if you kind of are thinking this way, you can just keep applying it to your set of tools because I think everybody's going to start having their own workbench and that and that wall of tools at the workbench is going to be a little bit different for everyone. And that's fine. You can spend too much time getting to, like, what exactly is the right setup. One of those things is something that's come up on your show a few times recently. MDs. A little bit further back, Nick Pattison was mentioning they're doing this, and you're starting to see this a lot more around. But if, if you haven't started doing this, I really recommend it which is make a playground for yourself whenever you have something in your mind that you want to execute in your design and you can't find the existing version of it. For my personal site that I'm slowly but surely actually building, I have this like, you know, this kind of retro, futuristic world that I've been building in like AI imagery and video and stuff. It's sort of a little nod of like this, this alien material that's arrived that I'm exploring. And so, so I have this hover video, right. And I wanted to put some music to it and I finally found like the, the. I found an old sample chop that I did that, that I thought would work nicely with it, but I couldn't find. So you see how I have like my sort of ghosted buttons over here. I. I knew I wanted to put the waveform, like sound on, sound off thing probably in the bottom right. But I couldn't find an animated like version of it that I liked. And so I just ended up building a studio basically that gave me a bunch of options to play around with. So this. I can actually try this.
- [11:39]
A
That's so cool.
- [11:40]
B
So then once I find my active state in my muted state that feel good to me, I can just grab this code and then bring it in and tell cloud code. This is the functionality that I want tied to my waveform component and this is the design and behavior that I want. And so like this is something that once I. There's a few things I need to tweak, but like I'll just go put this out publicly and other people can play around in here too and get their own version and, and just paste that in. Right. Like I built in these different effects. This is all, you know, trying to fit with my sort of like imaginary retro, futuristic world thing. But it's just, it's. It's fun, it's useful. I've already built all this out. So, you know, what I can do now is not make this limited to just waveforms. I can keep all these effects. Right. Because I might want to use more of these in other places in my personal site while I'm doing this sort of branding world building thing. And I could just bring in other icons or even, honestly, I could make it accept images and stuff.
- [12:41]
A
Yeah, you could do almost anything.
- [12:42]
B
Yeah. Like another thing that I made for the same reason was a hologram projector that I can upload videos into and displays in A cool, interesting 3D way.
- [12:52]
A
How are you coming up with some of the language even for these different types of prompts. Because I'm looking at these effects, they're beautiful, but I have no idea how to describe some of these things. And so I can feel myself wanting to recreate it. And yet I'm like, oh, gosh, that blinking cursor feels a little bit more intimidating than normal when I'm trying to describe things that I have no language for. So how'd you even get to like this?
- [13:17]
B
There's kind of. There's two answers to that in my mind. One is learn the language of code and the sort of like typical developer nomenclature for it too. So you can learn descriptor adjectives that you see, like other devs using to describe certain things. Because the likelihood that the LLMs were trained on some of those words and stuff as the translations for in natural language, we're calling this effect Glitchy CRT scan lines. And it knows the code translation of that. Right? So that's one version of that answer. The other version of that answer is something I've started doing a lot more recently, which is just go, if you find a source of something you think is really cool, tell Plod Code to go figure out how to do it. And then you can learn it as a skill or just apply it to that app right away. You know, whatever. So, case in point, I saw a Tweet from James McDonald referencing this multi layered shadow technique. And I clicked through it and I just looked briefly at it and I was like, oh, my God, this speaks to me. It is like subtlety that is obvious if you know to look. And it's sort of one of those, like, before afters. Basically, the idea is instead of literally applying a border on your shape or your container or something, instead put on a few layers. In his case, he usually references 3. Three different shadows that are layered with different opacities and spreads and values and stuff, and they create a border effect. But it's so much more subtle and it feels so much more elevated. Figuratively and literally, the aesthetic I just, I love. So, like, I did it to the mobile app that a set of games that I built for my son, who's a toddler. And these are all, like, games that help him learn, you know, numbers and letters and sounds. He like me, actually. I feel like he surpassed me in his NFL fandom. He's obsessed. He can do all 32 of these, no problem. But just like, look, look at how nice these borders are. And I just, I just. I just told Claude code while I was in this repo, I dropped in this URL. And I said, this technique is awesome. I want it applied to our components. Go do that. And it literally just went looked at it, figured out the technique, ripped the snippets, and then applied it to. To my design. I even kind of started to experiment. Like, you can see I contained the app in a, you know, phone like container while you view it on desktop. And so even there, like, these are three layers to create that border effect. But if you want a stronger border effect, I was like, okay, take the same theory, but make it stronger. Maybe do like five layers. And so that's how I got a differentiation in this phone container border effect that feels different than this one. So you can take the theory of, hey, some layers of different values of shadows to make hybrid border, slash border and elevation effect. But then you can just keep. Again, with the same philosophy, which Claude Code is kind of like hip to, now that I fed it that URL, we could keep extrapolating it and using it differently in different places again. For example, these buttons have, I think, four layers instead of three. And so it's that slight differentiation I'm.
- [17:07]
A
Chuckling to myself because I did the exact same thing with Derek Briggs's tweet, back from when he was working on Campsite. And I know that him and James have collaborated on that technique as well. And so it's like, definitely a through line, but I took a full tweet thread and an image and I fed it to Claude Code while I was working on the inflight UI kit. And I was like, don't copy this as exactly. Scan through it and figure out, like, where are the nice easy wins based off of this tweet thread. Cause I want to steal some of the best parts of this style. And it just worked. And to this day, my buttons look so much better in code than they do in figma, because I haven't taken the time to go through and do all the shadows and the negative spreads, and I probably never will.
- [17:50]
B
Yeah, that's totally fair. I was so excited by doing this and how well it worked. I had done this before, but it just really came through for me on this. And I started exploring Yaka Webb's articles more and realized again from this sort of like, we're talking about theory right now, right? Like, it's literal, but it's also philosophy and approach theoretical and stuff. And one of the other people that, I mean, since the beginning of me getting starting to explore coding and designing and code and stuff, Jay Tompkins is, you know, one of the top people that I've been aware of. He's just so clever with what he does, but he's so minimalistic in the actual code. He's so surgical with how he uses everything and everything has a purpose and it's all so slim. It's like, it's such a. It's such an admirable philosophy, right? And so to pull the thread further of essentially referencing slash stealing, but referencing really great designers, both literal applications of techniques and philosophy, I sent Claude code on a mission to create a new skill and agent that embody the philosophies and the techniques of Jakob and Jay. So what I did was I really like using the compound engineering plugin for Claude code that is in the every marketplace created by Kieran Claussen. It's a really powerful plugin. It can be overkill, I think for some folks, but it's actually quite a bit more approachable and easy to use than you might think at first. But regardless, it has a skill that sets you up really well for creating skills for yourself of your own. And so I used that skill that it has and I said I love these guys. I pulled together probably like 15 to 20 codepens and some like tweets from J, scan through all of Jakob's site, scan through these, you know, collection of J's code pens, look through what's happening in there, gather what's useful from there from a literal sense and also process this philosophically. I want a skill that can do that well, so how should I do that? And basically what it said was we're going to make an agent that uses a skill that has a bunch of reference points. Basically I used them to make a like visual design interaction design auditor that is invoked with this skill basically and that launches the custom Claude code subagent. So now what I can do, I think I actually did this on the Lunar Lander game recreation that we'll see later. I had it where it was and then I sent the agent go do a audit on all the visual and interaction design of this app. This is the literal skill md, but it has these other pieces of the agents like knowledge base that are helping guide it through the audit.
- [21:01]
A
This is so good. I'm really underutilizing skills is what I'm realizing just looking at this.
- [21:07]
B
Dude, I gotta tell you, like the skills thing, I still kind of struggle with it. I get it now, mostly because of creating this sub agents skills, are they that different? I feel like I could do what Skillz can do with sub agents. And so honestly the easiest way for me to get past that was to use the compounding engineering plugins, create a new skill skill to build this out for me. Because I wouldn't have known to go this multi layered.
- [21:39]
A
Yeah.
- [21:39]
B
But I can tell you that having used it, I understand now why layering like this and having both the skill, these other docs and the agent all referencing each other and stuff makes a lot of sense. But, yeah, I send this out, it gives me an audit and recommendations of what I could change to make it feel better. From the collective philosophy that Claude code kind of amalgamated about what it observed from Jay and from Jakob's stuff that I sent it to learn about.
- [22:15]
A
It's so good. I mean, I'm already so glad that we're doing this episode because this is incredibly practical for me, because I've been doing this manually for me. I look at those guys, I look at Emil Kowalski's, I look at a lot of stuff. So the amount of times that I fed tweet threads or articles or snippets of his course into cloud code and then basically was just like, hey, just read this and try to find things that we can do better. I've done that so manually at least 25 times. And I never thought of investing into a system like this, but it makes so much sense and it's such an easy way to elevate what I'm able to bring to the table. As somebody who's a designer who's like tinkering in code, but definitely not a.
- [22:58]
B
Design engineer yet, that's what I loved about a Using the compound engineering skill to make this skill for me finally made me make sense of all of it. More so again, heavily recommend all this stuff is free, which is mind blowing.
- [23:13]
A
We'll put that in the show notes too, for someone who's listening, because I also haven't been there yet. So I will be clicking that link later as well.
- [23:19]
B
He should have a whole episode about compounding engineering because it as a philosophy is really powerful. I'm trying to work it in more and more to my stuff. What I really loved about it too, is that now that I have this agent and this skill in this whole, like, folder, I can add to it. So I did this not as like, this is my end all be all of my guys and this is it. It was just like, I know I really like these guys. Let's see if we can do this and make it work. And it does. So now I can collect more and just be like, hey, I want to iterate on this skill and agent in this whole situation. Here's a bunch more links of stuff I like.
- [23:55]
A
So good.
- [23:56]
B
Reassess and add it in and just like, like, and. And it's cool because you can kind of make it think about it again. So you already had these kind of. These two perspectives from Jakob and. And Jay. Let's add in emile's with like 20 links or something, right? As opposed to now add Emile's perspective. You could almost like think about everything together now. So yes, we'll have Jay's perspective, Jakob's perspective, and Emile's. But now think about, like, tell Claude Code to zoom out yet again back to the source material and think about what the general philosophy becomes with the three layers of perspective instead of the two. And so instead of just adding one layer to the cake, you're changing the whole construction of the cake when you add this other ingredient into. Or you can if you want to. And in my mind, that's where I want to go because then it's refining more and more this amalgamation of techniques and philosophies and stuff that I find really compelling.
- [24:58]
A
I've been designing products every day for the last 15 years, but in the last six months, everything has changed. With AI in the mix, I'm cranking out ideas faster than ever. But none of that matters if I can't get the feedback that I need to get the team aligned. And right now, getting async feedback still kind of sucks. So I'm building the product I've always wanted, and it's called Inflight. I use it every day to share ideas and get feedback from the team. It's totally changing the way that I work. So I'm excited to show you. Right now I'm only giving access to Dive Club listeners, so head to Dive Club Inflight to claim your spot. I find it really encouraging. Again, as someone who. I kind of want to be more of a design engineer, you know, all day today, I have not really made that much in figma. I've mostly been in cloud code because I'm iterating on something that I already designed in FIGMA and had built with an engineer. So I want to. I want to grow that muscle. Right. Like, I want to be more of a design engineer. I have no interest in really getting into the intricacies of syntax. I do think I have decent taste, you know. Yeah. And honestly, I think you kind of do too. Like, we're obviously looking at the same types of people for inspiration. Those are who I would consider the best of the best, for sure. This creates that very clear, tangible mechanic that I can use to act on my taste, right? It's like, I know this is good, and now I can actually use it instead of just pointing at it and saying that it's good. What a fun set of next steps. I'm really excited about this.
- [26:31]
B
The amount of just amazing, like, lessons and knowledge out there for free is. Is staggering. So finding ways to actually leverage that. It's that, like, AI doing that 10x thing that you have to start thinking that way to get the full 10x out of the potential of the 10x. Otherwise you could end up at, like, 2 still, which is great. Still. But you got to start thinking in that systematizing, collecting. Now you can get big bloat too, and end up, like, with nothing usable. Also, like, don't. Don't create one skill that is every designer you've ever thought was cool. You know, like, don't do that because that'll be useless. Like, you do. You do want to be somewhat surgical and intentional and stuff. And as I go along, like, when I use this skill and stuff more, it can be overly conservative on how obvious the effect should be because it has that subtlety influence from processing Jay and Jakob's stuff. But, like, I don't always want everything to be super subtle. Sometimes you want it to be, like, big, but done in a reasonable, tasteful way. Right. So I might say at some point, take it out of our general philosophy that, like, everything needs to be subtle. Change it more to. Is this the type of animation that might be something you want the user to kind of grab onto? Or is this a little tasty, a little background for the if, you know, you know, crowd, that should take that. That really subtle approach to it. Right. And so you can just keep iterating, create the thing and then over time, give it your feedback of your taste of what to do with the taste it sort of ingested from these reference points.
- [28:17]
A
Let's keep the train moving here. What are some of the other things that you think more designers can have in their toolkit for 2026?
- [28:24]
B
Yeah. So aside from making your own, like, full tool studio thing that has the parameters and stuff. And I know you were saying you just were doing this Y. And I had seen this before, but it was a conversation I had with Will King. Shout out Will. He just was like, dude, this is such an easy win. Because honestly, what I was talking to him about was, how do I become more of a design engineer than just a designer? That's Learning some code. And one of the things he said is like, dude, this easy win is bring in this control panel. The one that I'm using is called Leva. That's L E V A. Another popular one I think is Tweak Pane. But Leva seems to just work really well with Next JS and React and all those things. And you can just tell Claude code. I want you to incorporate a Leva control panel. And literally every parameter on here is up to you. It needs to go set it up for you and assign variables and things like that. But it gives you the, you know, again, what designers feel like they can be missing when they're designing code is nailing those details, especially like interaction design and stuff. So this is, this is also for my personal site. This is going to be like a throwaway is kind of the wrong word, but kind of. I'm just going to house this in this little, like retro futuristic, like, tablet thing that I've made. It's a game. I recreated the Lunar Lander game. It was my dad's favorite game way back in the day and when we were young, like killing time in the car or something, he was. My dad was an early adopter, so he was like a. A Palm Pilot person. And, and they had the Lunar Lander game and we, we would totally just sink 45 minutes into that with the, with the stylus and everything, you know. So I went and was like, oh, this fits thematically and it's kind of a nod to my dad and my past and stuff. And I'm just making this as a little like bonus Easter egg in my personal site. But it's fully functional. Like, all of this is real. So I've got the rocket here and I've got the lander. Eventually, Xantos co will will have the fruition of all you've seen, some of the other images I've shown you, some of the visual language of that I've been creating around this imaginary world that now I also have to incorporate my son into because he saw me working with it. For mine, he's three and a half, and he was like, can I see myself in space too? And I was like, yeah, give me, give me a little bit. I've basically become a creative studio for my toddler. That's straight up what's, what's happening now. I make the best way to learn. I truly, truly, though, truly. It is building for yourself and the people immediately in your orbit is how I've learned almost everything about coding and building and Doing all that stuff. So if you notice here, I've got these corners sitting on all of the cards and the hover state draws them from the corner to the middle. So they start from each corner and they animate to meet in the middle. Now I might want that to be a little bit snappier. So I made the transition shorter or a little bit more luxurious. And there's that last. Like it. It kind of joins the initial line and then there's a following that's like the edge tracer that comes after it. So the triggering of when that secondary tracer comes in. I can, I can tweak here.
- [32:10]
A
What does it take to spin up? Leave it like this. Because I actually haven't used this. I've. Anytime I need a control, I actually like just custom build it. But this is way better.
- [32:19]
B
Oh my God.
- [32:20]
A
Than anything I've coded quickly.
- [32:22]
B
No, seriously, like nothing. Like just tell. Tell cloud code I want to use. Leave a. This is the stuff I want to be able to control. Make a plan.
- [32:30]
A
Like it's man. Yeah, I've been doing this all wrong.
- [32:33]
B
That's why we do this. And again. Yeah. Shout out. Shout out Will for reminding me of this. Like, I just. I'd seen some folks do it before, but reminding me. It's just the incredible power of it with total simplicity for your coding agents to set up for you. And you can tell it in your human terms. Like I even, you know, described in natural language how I wanted these corners to exist and draw themselves to each other on hover. That was all natural language. So even when it came to getting the controls around it, I still was just like I wanted to be able to tweak stuff around that effect like that. It's, you know, it's like not even. Yeah, it's not even that much more detailed than that. You could just say all the things around the hover states for the cards in the level selector.
- [33:19]
A
And that's what I did yesterday. I. I had a pulsing like animation for an idle stakes. I was trying to communicate that you can slide something and I just know I knew I needed some movement, but I had no idea. Like should I use react motion? What kind of easing curve or spring? Like, I just didn't even know. So I kind of just had it throw something and then my prompt was. And give me knobs for all of the relevant controls that I might want. Like I didn't even know what they could be. And it worked. It worked. It wasn't as good as this. I should totally use Leave it next time. But it's not.
- [33:50]
B
But it's still that simple too. It's basically what you did. But to make it a little better, you just need to know, oh yeah, the Leave A control panel, which is awesome. And this is again where it's just like collect those knowledge chunks as you're on Twitter and as you're on the web and like, be careful of amassing too much that you'll never use. But stuff that you find that's like, oh my God, that's a like no brainer. Another thing that you could do specifically for CLAUDE code, you can go into your global user settings. So I could open up a cloud code session, be like, hey, put in my global CLAUDE MD essentially. So that's known to Claude code in any project or session you do. Whenever I say, give me controls to tweak something about ui. When I say that you should know to go grab the Leave A component, bring it in to our app and ask me any clarifying questions you have about exactly what things. Even from a natural language perspective, I want to be able to tweak within the control panel so it sets up the right parameters.
- [34:54]
A
Okay, let's dig into that for a second because this is another area that I know that I'm underutilizing. So are there other easy wins or no brainers that you think designers should do to invest in that top level Claude MD file globally?
- [35:10]
B
Be slim. Only put stuff in your global one that is like truly universal to what you use Claude code for. In general too, even at your project level Claude md, it's tempting to use it as your tome, like your journal of everything that you do to your app and how everything works and all that that becomes inefficient because CLAUDE MD and Agents md, if you're using Factory Droid or Codex, they are persistent memory. So there's plenty of stuff that Claude code doesn't need to know all of the time, even if it's critical about your app. Like if I'm messing with UI stuff in my. In my NFL Pick' em app, I don't really need it to know everything about what the API is doing while I'm working on something on the front end. And if and having that in there bloats its context, making its memory worse. So in general, what I found is be slimmer on your cloud MDs, both global and local project level. And even further, you can make CLAUDE MD files in individual subdirectories that just pertain to your documentation or your backend stuff. Or your components and stuff. So being a bit more surgical about that is really helpful for. I know I want to work around the components and make our tokens and our design system more robust and stuff. And then you can have that cloud MD for just that. But then it's not staying in its persistent memory for, like, if you're working at the top level and working on the whole app in general. The reason this is a hard one to answer definitively is because go look on Twitter, YouTube, Google, whatever, and the philosophies around this stuff is a huge spectrum, right? And I've spent so much time previously reading and following everyone's advice of set up your things this way, do your flow exactly this way, use this new MCP that does this and this and, and you can learn from that. And there. I'm not poo pooing that entirely, but what I did find was it's easy to get caught up because that stuff changes so fast. Like, the recommendations from four months ago almost certainly are not ones you should follow today if it was made literally four months ago. Unless it's a philosophical thing. And that's where I'm like, learn the philosophy and the approach and the ethos of all of this more than the literal way to direct your AI to be better because they level up themselves so often. I can't tell you how many times I've built little tools for myself in the past. More in my, my earlier days, like cloud code itself just like sent out as an update like two months later that I sunk, you know, 20 hours into making this little tool. And it's like, yeah, that's not necessary. And keep in mind that all of these people, they're content creators, they get views from 10x your cloud code production by doing these three things, right? They're not the authority on all this. There's still plenty to learn. I'm not. Again, I'm not out on all of this entirely, but I think you're better off understanding the philosophy and the theory of most of this, figuring out how to implement it, sometimes experimenting. You know, again, that compound engineering plugin is one of those things that I'd say that's one of them that's like, no, do, do try that. But you also just download the plugin and kind of learn how to do the slash plan, slash workflow, slash review. But it's not a whole big setup that you see that has 10 steps on a YouTube video or something. It's kind of a vague answer for like, what to put in there. And what not to.
- [39:02]
A
I respect the answer a lot actually. And you've made it.
- [39:05]
B
There's no great answer.
- [39:06]
A
Yeah, you, that's I. And, and the reality is we're filming this video, it's going to be very helpful. And even though you're making a point of staying at a philosophical level, certain things are going to be outdated.
- [39:18]
B
You know, this is not evergreen content.
- [39:20]
A
Yeah. It's just key. They keep leapfrogging each other. And so especially as designers who are not as technical and at least for myself, if I get a piece of advice, I kind of follow it to the letter because I don't have the first principles knowledge to deviate as well or to know what little bits and pieces to take. And so keeping that more level headed approach I think is a good, it's something that I'm glad to have in this video.
- [39:43]
B
I guess anybody who's watching or listening or whatever like you know, plays guitar or something. Like how many pedals and amps and guitars and pickups and everything are there? That like there is no answer to what's the best setup. Right. Like we're basically in that territory with AI tools. Same way with guitars. Like you're always going to get a better sound out of a tube amp if you don't specifically want the solid state sound that, you know, metal players want and things like that. So there's basic tenets like that plan before you execute with your agents. That's like finally all the tools have built that in as like more of the standard. But that's like, that's been true for a while and it just keeps being true. But aside from that, when it's all these different setups and stuff like play around, you're going to keep evolving, you're going to keep learning, you're going to keep, you know, figuring out what works for you and then talk to people, read, learn, absorb new philosophies, try that philosophy out, see how it works for you. We're, we're in a really Magellan like exploratory era with all this stuff. Nobody really has definitive answers outside of some of those big core tenets of this is the best way to do this, this and this. Fill it out, find what you like to make sure that you're following and looking at people doing that kind of thing and then go play around and follow your passion there. You'll learn a lot, a lot quicker that way and become less susceptible to the loop of optimize, optimize, optimize, optimize and spending more time on, like, your. Your system and your setup than doing the actual thing.
- [41:20]
A
The importance of experimentation comes up a lot. I totally agree. That's like the era that we're in. The reality is not everybody has this blank credit card. That is the, you know, the corporate bricks that they get to just point at anything. And so there is some strategy, especially as someone who's like, charting a learning path for themselves this year, but they don't have unlimited funds. Like, how do you think about the right way to invest in these tools in a way that makes sense for, like, an individual who just wants to grow, who doesn't have this unlimited budget?
- [41:55]
B
Free trials are awesome. Use them at the UX Tools Lab. Stuff. We're going to be trying to put out ways for you to really condense. Like, should I care about this? Do I need to pay attention to this? Because there's way too many and there's so much noise and there's going to be plenty of overlap and stuff. So, you know, we're talking about setting up all your tools and your workbench and all that kind of thing, and figuring out combos for those can be really good in stretching your dollars. So let's say, you know, someone asked me a question about I hit my cloud code limits on my 20 plan quickly, regularly, and then it kind of breaks my flow because I'm like, exploring, I'm doing stuff and stuff's working. It's like, you've hit your limit now and you won't be. Your limit will reset at 2pm and it's like 11am and it's just like, it's such a frustrating wall to hit. So, like, one thing that I like to do is combine my Claude code subscription and I have a factory droid $20 subscription. And factory is nice because it's like cursor in that most of the major LLMs that you'd want to be using for coding are available in there. Gemini 3 Pro is great for UI generation. I'll go do stuff with Gemini 3 Pro through factory to go execute that stuff, then hop over to Claude code the regular cloud code to be doing bigger things, even though Claude code also exists in Factory. But either way, I could bounce one direction or the other if I'm running out of my tokens in Factory or if I've hit my limit in Claude code, you can do a add factory or add cursor. Add that $20 of one of those, and then theoretically, you could be doing a fair amount for $40. What I would say is if you're spending a decent amount of time in Claude code. That hundred dollar tier is I, I use it, that's what I have. I almost never hit the limits and I, I hit it pretty hard a lot of the time. So it feels like a lot. But you just gotta like zoom out a little bit and think about what you're doing with cloud code. Because keep in mind, like I'm not just building with cloud code like I sent you, you know the Obsidian doc of some of the things that we're thinking about talking about today and that's this. I have my Claude code sitting on top of my Obsidian vault and it helps me structure all this stuff. I can interact with linear just through Claude code to linear and just kind of like clean out these issues and let's optimize these and if these are in this kind of state, put it on hold and blah blah, blah. Claude code can be so many things that aren't coding. Claude code is just Claude computer.
- [44:44]
A
That's another one of those categories where I'm like, I know there's something there for me. It just kind of keeps getting knocked down the totem pole a little bit. But I see some of these workflows that people are stringing together using Claude code way outside of like the context of an IDE that are really, really compelling.
- [45:03]
B
That actually gives me a good reminder of a thought that I had in, in that experimentation realm. If you can block out at least a half day. If it needs to be on a weekend, fine. If it's during the week and you can do that with your schedule, great. I've started to do what I'm calling tune up days and it's specifically for. I'm not allowed to actually be productive in like a project. I can only enhance my setup, finally bring in that workflow that people have been talking about, finally install that plugin and try out the compounding engineering and finally clean up my folder structures of this and that. Like a lot of those finally things, I have content sections of it. So education stuff. If there was an interview I'd really been wanting to thoughtfully consume for a while, I'll do that on a tune up day or an article that I want to be thoughtful about playgroundy stuff too. Like go, go do something that has no like stuff for my toddler, I just, I just experiment with stuff all the time. Like right now I'm adding to his games app, basically an app within that for drawing on the iPad, both for like coloring and tracing letters to learn how to write his letters. That's a playground kind of thing. And. And giving yourself the time to play is a really good way to give yourself the license to go grab that tool, go sign up for that trial. Give it a shot. I've been hearing about this thing a million times. Let's actually figure out what that does and does this matter for me and stuff. So when it comes to the experimentation stuff and the overwhelm of all of that, I keep a running list of things that I'll want to do for my next tune up day. And then when it gets to that day, I'll sort of like, all right, I'll pick out this one and this and this and this. And then I go do those.
- [46:50]
A
Well, I feel like this entire episode has been a tune up day for me. And thank you for having me. I have. Thanks for having me. Full of, like, very practical next steps. It's corny to say, but I literally wrote out on my own intentions for this year. Like, I want to be more of a design engineer.
- [47:06]
B
Yes.
- [47:07]
A
This just got me excited, you know, like, this got me energized and gave me some very practical next steps that I could take. So I appreciate you as always, Kyle, for coming on.
- [47:16]
B
It's been fun, for sure. Thanks for having me, man.
- [47:19]
A
Before I let you go, I want to take just one minute to run you through my favorite products. Because I. I'm constantly asked what's in my stack. Framer is how I build websites. Genway is how I do research. Granola is how I take notes during crit. Jitter is how I animate my designs. Lovable is how I build my ideas in code. Mobbing is how I find design inspiration. Paper is how I design like a creative. And Raycast is my shortcut every step of the way. Now, Now I've hand selected these companies so that I can do these episodes full time. So by far the number one way to support the show is to check them out. You can find the full list at Dive Club Partners.