Subscribe
Share
Share
Embed
Conversations with designers, founders, and builders behind some of the best work
00:00:00:00 - 00:00:06:06
Unknown
then you have this agent, which is maybe headless, completely headless, running, on the server.
00:00:06:08 - 00:00:13:09
Unknown
And for that, we have actually built Bessel clock CLI. So you can run past the headless today
00:00:13:11 - 00:00:31:13
Unknown
fully on the server. And it can obviously tap into pencil files and work with them and great. For you, it can create PDFs and screenshots and send you an iMessage with a screenshot. You can say what you don't like, what you like, but it still taps into your design system.
00:00:31:13 - 00:00:59:02
Unknown
It feels like design tools are splitting in two directions right now. And that is with Figma May earnings call coming up this week, all while Clod Design Figma make V0 lovable Google stitch all converge on what seems like the same question who is going to own the next design workflow? Meanwhile, some designers are debating how much time they should be spending touching the actual design files versus being more of a creative director reviewer
00:00:59:02 - 00:01:01:09
Unknown
and Tom Kirk has been thinking about this too.
00:01:01:10 - 00:01:27:10
Unknown
He was the creator of Adobe XD. He was Adobe's youngest flash evangelist as a teenager, and he watched as the hype for flash slowly migrated over to HTML5. And now he's hand coding pencil.dev, a startup placing the bet that agents are going to be working tasks ahead without taking the shape work out of the hands of designers. This is state of play.
00:01:27:12 - 00:01:36:01
Unknown
Let's get into it.
00:01:36:03 - 00:02:11:22
Unknown
Frankly, I remember when pencil that dev was announced. I mean, you and I spoke about it, last year when you were kind of building it and then to see the reception when it was first dropped to, I think, like, general audience was pretty astounding. And I want to talk a lot about pencil, but I actually want before we get into it, you have been around these tooling, and I even go so far as back to when you were an advocate for flash and, you know, used to talk about how that was going to be, kind of this era where anyone could build these interactive websites.
00:02:11:22 - 00:02:35:20
Unknown
What was kind of the demoing like back then when you were trying to make that argument? You know, it's really funny because so going even like a little bit into the history, my parents had a design studio and, I was always around, tried new tools, and they would teach me, CorelDRAW, illustrator. InDesign and all of these different tools and Photoshop, obviously.
00:02:35:20 - 00:03:03:21
Unknown
You know, then when I started using when I was like seven, eventually I wanted to build interactive things and HTML, you know, happened. But then I discovered flash and I was like, wow, what is this thing that, like, moves and has sound like we're talking about like 2001, 2002? Okay. Like that's a long time ago. And we were able to build these like interactive websites.
00:03:03:23 - 00:03:23:20
Unknown
And I was completely astounded by that. I was like, this is the future. And and it is the future in a way, you know, like websites are now becoming more and more interactive and so on. But you could also build games, right? And all these things. So I was like, I gotta start using this. It just like, completely like swallowed me.
00:03:24:01 - 00:03:43:00
Unknown
Eventually Adobe reached out. I was like, do you want to be an evangelist for flash? One of the first, evangelists. I was actually the youngest evangelist on the team, which was like, really cool. So I was just learning from all of these people around me. Yeah. I mean, and your story starts all the way back.
00:03:43:00 - 00:04:02:03
Unknown
I think for those who might be familiar with your background is creating Adobe XD and a lot of the products that you've been involved with probably think your story starts back in Palo Alto, but I think your story starts closer to Czech Republic, right? Software engineering. What was what was kind of the climate then? What was the culture in building software at that time?
00:04:02:05 - 00:04:18:12
Unknown
You know, I was still a teenager and I just like discovered all of these things. And, every time, every day after school, I was like, oh, my God, I'm so glad I'm done with school. So I can finally now go back and actually learn the things that I want to learn. I would go back home and I would just like dig into these tools.
00:04:18:12 - 00:04:44:19
Unknown
How like, how is this like dialog made, you know, how is this this made? And I would be like dissect these things. And obviously that was like Visual Basic and Visual C-sharp and all these things. So I started learning about them, how to use them and how to build these things. It completely, fascinated me. I was at school and I would always so I lived in this small town in like eastern part of Czech Republic.
00:04:44:19 - 00:05:08:07
Unknown
And, I would go to the capital of Czech Republic, you know, every now and then for different conferences. One of the conferences there for, for teenagers like me was called Junior Internet. And that's where, like a lot of these folks who would, like, hang out and get together and just jam on things and everybody would show something new they were working on and so on.
00:05:08:08 - 00:05:34:07
Unknown
So there was a lot of stuff happening, but it was still like a very small group of people doing these things. So yesterday I sat down with George Hastings. He's the creator of Unicorn Studio, and it happened right, as I think on Twitter, you know, people constantly calling the death of this the death of that Google open source designer MD we have all these things happening, around supporting agents and design work.
00:05:34:09 - 00:05:59:09
Unknown
And so there's almost like a splitting of types of design work we have maybe what you might call stochastic design, right? This like probabilistic fluidity, like I have intent. It's driven by things like a design DMV. And it opens up here's kind of what I want. And let's see what the agents create and then we'll tweak it. But then you have maybe deterministic design.
00:05:59:09 - 00:06:18:02
Unknown
This is what we're used to right now. There's an ice block I have a chisel. And I'm, you know, being very specific about what I want. Do you view the axis that way? You know, it is is that kind of what you're seeing, the split. And if so, where do you fall on that spectrum? I think it's a combination, honestly.
00:06:18:07 - 00:06:42:08
Unknown
I think you just start with some ideation process in the beginning, some exploration that's more on the stochastic side. But then there's always like 70% where you're like, let me take a chisel and actually polish it up. And this is this is the pattern I have been seeing with our users as well. They lost bezel for at, you know, open exploration.
00:06:42:12 - 00:07:32:09
Unknown
They really appreciate the fact that they are on the canvas and they can go deep and they can do things themselves. So and honestly, like bring their own taste in. I think what's fascinating is I believe you've said that you hand code most of pencil.dev is that's still true. I would say like, there are areas where like, obviously you let the agents do some stuff, but obviously looking at some of the performance improvements and looking at, so the canvas work, I would say that like most of that stuff is hand-coded, by now because it just means that chisel to go in and fine tune these little details yourself because like, if
00:07:32:09 - 00:08:01:14
Unknown
you also let it go, just, you know, let Agent roll, you realize that it produces things that you're ultimately disconnected from. And I think that applies also to design, like how much do you want to be disconnected and how much of that stuff is still you and your inspiration and use solving for a particular goal versus like the AI, solving something which you are hoping will align with what you had in your mind?
00:08:01:14 - 00:08:24:08
Unknown
And what was your original purpose of, of solving? I, I love how, can, call in bulk from, from frameworks set it the other day in Solo's podcast, he said something like the I guess, like 80% there, but then you spent $1,000 and, you realize you're at 70%, though. So I think that's kind of the same thing.
00:08:24:11 - 00:08:51:06
Unknown
And that's why I think we still need code editors. And that's why I think we still need, canvases. Because we need that chisel. And, you know, it's a nuanced approach that I'm starting to that I believe as well. I see the value in this again. You know, I think on on Reddit, there was a post trending where someone's, agents had texted someone else unprompted, and a lot of people didn't believe it.
00:08:51:08 - 00:09:11:00
Unknown
And I think the greater thinking there, especially based on the data we saw in our survey, is a lot of designers just aren't aware of what's possible today. And so someone like you, who has a nuanced idea that says, you know, a genetic design makes a lot of sense in some cases, but grabbing, you know, the instruments with your hands is still necessary.
00:09:11:02 - 00:09:34:17
Unknown
What made you decide that agents from day one was a bet worth making inside of pencil? That's also how it came to be, honestly. Like the whole, the whole genesis of of pencil is that I open creator and I was actually working on a different project and I was like, these are the five, I think the agent to do the UI for me.
00:09:34:19 - 00:09:59:18
Unknown
And I feel like writing these essays to just describe a navigation bar, sidebar and so on. And I was like, why isn't there like a canvas directly inside cursor where I can draw something right now and tell the agent build need is. So I build that, and once I hit it, I was like, why don't I? I just asked each and feature design something for the other way around.
00:09:59:20 - 00:10:25:19
Unknown
And it worked. And this is like when I had this, head exploding emoji happened to me. And and the rest is history. Then we just kept digging into this. There is like so much to build, honestly. And I feel like we just scratched the surface. Of this whole thing. Now, George told me something that he believes yesterday, which he said the designer has to stay the final authority.
00:10:25:21 - 00:10:59:21
Unknown
And your swarm mode, really emphasizes how designers can become orchestrators. But how do you kind of parse the two? Is it the same thing to you? How do you view where the designers discernment remains? I think it's just like being in a design agency, honestly. Like maybe, one of the leaders is more in a managerial role where, you know, brings an inspiration to the table, but like the actual work is done maybe by your colleagues.
00:10:59:23 - 00:11:21:17
Unknown
But there are also moments where you just want to go in and, like, do the tweaks yourself. And this gives you that optionality where you don't have to start from blank canvas. You can start from an inspiration that you added up in a screenshot or a photo, or maybe you just write something. This is the starting point.
00:11:21:19 - 00:11:45:01
Unknown
And then that's maybe like empathized. Fine. But it's already reflecting what you had in mind initially. And then you go in and maybe you just grab it and you were like, actually, I don't want this. And, this is by the way, we're like, having the swarm of agents is really brilliant because you can just spit out multiple variations of things.
00:11:45:03 - 00:12:18:01
Unknown
Maybe like ten, 15 variations. And you can be like, okay, I don't like this, but with every iteration, go and bring me something new, like look at the previous iteration, find out what is interesting about that and apply that one to another iteration. And then it just goes and goes and goes. Now, what's been fascinating, talking to those of you building what I think are the future tools of design, is to really try to pinpoint what you think the future of software building looks like.
00:12:18:01 - 00:12:36:17
Unknown
And I spoke with Steven Haynie, who thinks there's going to be three main tools. I believe he thinks a canvas, a sort of coding QA tool and an agent. What do you think is going to do you think there's going to be a consolidation? How do you think we're going to be building these things, both from software development and the design piece?
00:12:36:20 - 00:12:55:05
Unknown
I think, yeah, there's going to be canvas. There's going to be a tweaker. Tweaker is a tool that you maybe have running on your existing code base and allows you to tweak particular things like existing production code base, maybe in a more visual way, and then
00:12:55:05 - 00:13:01:06
Unknown
you have this agent, which is maybe headless, completely headless, running, on the server.
00:13:01:08 - 00:13:27:07
Unknown
And for that, we have actually built Bessel clock CLI. So you can run past the headless today fully on the server. And it can obviously tap into pencil files and work with them and great. For you, it can create PDFs and screenshots and send you an iMessage with a screenshot. You can say what you don't like, what you like, but it still taps into your design system.
00:13:27:07 - 00:13:51:11
Unknown
It also has to still have access to your code base and draw information from like the CSS or react components. I've applied it and back and forth. So that's kind of the, interesting part about that. You know, and I think, by the way, like one last thing I want to say here is that I don't think is necessarily about like one super tool that's going to solve everything.
00:13:51:13 - 00:14:20:21
Unknown
I honestly think that we're getting into this era of like, everybody's free of hacking on everything and assembling the workflow specific to themselves. We will provide some of these components. You will get some of these other components somewhere else. You will combine them and create your own bespoke workflow that works within your company. I think that we're seeing that now with how people are building their own custom agents.
00:14:20:21 - 00:14:41:00
Unknown
If you were to look at anybody's agent set up, with something like Hermes or Open Claw, it's it's like, the spaceships in the game, No man's, No Man's Sky, where it's everyone can just have a combination of whatever they put together that works for them. And you use the word that I'm really interested in exploring, which is headless.
00:14:41:02 - 00:15:03:22
Unknown
And so, you know, if we think about headless content management systems, well, those things kind of separate themselves from the presentation layer. What does it mean if there's a headless design tool? Is it is it tool agnostic. What is what does that mean to you. So essentially that means that you can run it from terminal. And that's that's the whole point.
00:15:04:03 - 00:15:26:19
Unknown
So you don't have to launch the canvas. The canvas is there. It's just off screen rendering things for you. It can still read these files are MDF files or pen files or a combination of these files or bitmaps and so on. And honestly, you know, then file on the disk, it's just a text file is Json file.
00:15:26:19 - 00:15:50:00
Unknown
So it's a matter of like working with context that he provides to it and spitting out something that you're hoping to get out of it. And by combining these pieces together, you create something that fits what you imagined initially. So you're not, like, limited by this tool and having something you can extend it. You can you can make it yours.
00:15:50:00 - 00:16:10:15
Unknown
And I think that's the most exciting thing about this whole thing. It's almost like a safe bet to consider. The thing that becomes consistent in the future is the the briefcase that designers lug around that has their context. Right? And they just kind of take that briefcase into whatever tool or workflow they're doing. So that might be customer feedback.
00:16:10:16 - 00:16:29:08
Unknown
It might be, you know, whatever the product specification is, it might be the linear JIRA tasks. And as long as they have this briefcase, they can kind of move into different tools or workflows. That might change, but they always need to have context. Is that does that sound accurate about how do you think about something like that?
00:16:29:08 - 00:16:57:17
Unknown
It's less about like us working on the same canvas. It's more about us working with the same context and and refining that context for the agents to work with. So that's why I'm like foreseeing a future where like, we will need new tools because it just opened up so many possibilities that we would have never even imagined back in the days.
00:16:57:19 - 00:17:23:16
Unknown
Now, you know, pencil.dev is a design tool. What's the languaging that you're using right now? An agent powered canvas and agent first. Canvas. Agenda canvas. Yeah. Agent. The canvas, I think, is the best reflection of what we're building. But I would even say, like, in one of those videos at the end, we had like Autonomous Design Agency and, it's kind of like that thing.
00:17:23:16 - 00:17:51:09
Unknown
It was something that I find incredibly profound, which is like, what if you have this context, right, that you have built internally, but suddenly, like everybody within your organization can sort of hire an AI designer for a particular job, you know? And that doesn't mean that, by the way, it doesn't mean at all that like, you still don't need to go in and like, do things.
00:17:51:11 - 00:18:21:08
Unknown
Maybe it's again about a starting point. Maybe it's again about just like getting the ball rolling or like solving a specific particular task that can be fully done by an agent. This idea that an agent is doing design work is sometimes triggering for people. And until you actually see, like no, like solving the the, the blank canvas problem is actually incredibly helpful, even if you end up throwing the work away.
00:18:21:10 - 00:18:40:12
Unknown
There's something to be said. Why? We always have done Crazy Eights, right? Why? Why did we do Crazy Eights and sketch out a bunch of bad ideas until we it can, because it fleshed out some of the thinking and it put us on a path that approximates in a direction we might want to go. And sometimes seeing bad ideas come to life allows us to.
00:18:40:13 - 00:19:07:20
Unknown
Okay, cool. I can eliminate that. And it and it puts me, you know, I'm focused more on things that I think are better. When you think about who you're building pencil for, who who comes to mind is like your ICP. It's really funny because like looking at the current user base, like there is a big portion of developers using it and then there is designers, then there is design engineers graduating sort of, into this like mixed role.
00:19:07:21 - 00:19:41:16
Unknown
Then there is PMS, and I've even seen like marketers and people who just basically pick it up, and create a website, create marketing assets, create, maybe like PDFs and sort of like tech sheets for their particular products for download. So it's really interesting how they're orchestrating this whole thing. Maybe they take a screenshot of their existing website, paste it in as a starting point, and then just like, anyway, fit in, but then go in and actually start making manual tweaks.
00:19:41:18 - 00:20:13:17
Unknown
And then they ask Claude or Codex to code it for them and, that's it, you know. So, so it's a mixture of things and a mixture of people and, but personas, there is also like anything which is but also is directly integrated into VS code versus anti-gravity went through roof and all these tools. So you can actually open this life canvas, within this code editor, this custom editor of sorts.
00:20:13:19 - 00:20:38:05
Unknown
And, you have the code on the left side, you have the H1 on the right side, you have the canvas in the middle. And, you can just jam with your existing code base inside pencil. I think all design tools should really try to find ways to allow their tooling to back into any shape of spaceship workflow.
00:20:38:07 - 00:20:58:21
Unknown
Because I'm finding that at least for people like me who are super, you know, into using agents, that's something that has become extremely valuable. Hey, I need I mean, really like, how do you think about building for agents as part of an ICP? This whole thing also started with us thinking about, like, openness a lot, since the beginning.
00:20:58:21 - 00:21:23:05
Unknown
So we have this, band format, which is Json based, sort of like an open design format, that describes how everything is laid out. And the whole idea there was like, anybody can take it, dissect it, read from it, generated and so on. And so there was an initial building work and now we've added scripting on canvas, which is also kind of interesting.
00:21:23:05 - 00:21:49:08
Unknown
So you can actually write custom scripts on custom components. They're like fully, generative. So you can do interface components. You can do things like waveforms and generative art directly on a canvas. You can animate it. It's really interesting. Like how many new sort of directions it's, it's opening up. And it also, by the way, can produce custom design tools for you.
00:21:49:10 - 00:22:16:02
Unknown
So for instance, I'll give you a couple a couple examples. I wanted to create a logo generator, maybe from like basic shapes from like a circle triangle star and whatever, you know, and I was like, give me hundred combinations of this. And, you can just basically ask my this agent, Jeremy, a script that generates logos, and let me edit this particular inputs.
00:22:16:04 - 00:22:36:18
Unknown
And it does that or like, I need an Ascii art generator. And I want to paste, a bitmap into it, and I want it to, to make Ascii, but also give me controls over that. So it basically built you a custom plugin inside pencil on the fly. And when I saw it for the first time, I was like, wow.
00:22:36:22 - 00:23:01:22
Unknown
Like, obviously we can build every single tool for people. But can we give everyone tools so they can build the tools for themselves? So it's tying us back to that initial discussion and like the space spaceship as well, which is like the way I think about pencil, it's like, how can we make it so flexible that anybody can take it and make it theirs?
00:23:02:00 - 00:23:25:09
Unknown
Let's put like the overarching, theme here. Yeah. And for those who aren't familiar with my analogy on the spaceship, No Man's Sky is basically like Minecraft in space. And so you can just make these amalgamations of however you want these shapes to look. When I went over to Vercel, their design team had 33 folks, five product designers, most of the rest of them were design engineers.
00:23:25:09 - 00:24:00:20
Unknown
They had, I think, two brand designers who they were now referring to as brand engineers. And I kind of dove into that, and I said, why brand engineer? And they said, well, because, you know, six months ago we weren't coding. We didn't know anything about code. And now some of the things they're doing, they're shipping tools that help their design process, their shipping, and all the tooling that they're shipping is a lot of being able to, you know, I want to create some sort of shader that I can superimpose on top of some asset I'm generating, and now they can just mess with the toggles to do that in our survey.
00:24:01:00 - 00:24:27:20
Unknown
I can't remember off the top of my head, but I want to say it was a majority of those who were using AI tooling had built a custom tool for themselves to help their design process. I was very surprised to hear that because I thought that was a more advanced thing that a lot of people, you know, okay, we're using some cloud code, but to see how many people are actually creating their own version of a tool is pretty interesting.
00:24:27:20 - 00:24:47:07
Unknown
To know that this can happen inside of pencil is even more interesting. The thing I want to know is what has surprised you? Because even me, I feel like people like you and I are very plugged in to this space. But hearing that story about the brand engineers and seeing the data on how many designers are building their own tools, that is surprising to me.
00:24:47:09 - 00:25:18:04
Unknown
What has surprised you about who or how, people have been using pencil? The one thing that definitely surprised me, and initially I was thinking that, and even how I started using pencil initially, within cursor was like 1 or 2 pages, maybe three if I, like, go Wild. But now we have this discord where we have like almost 10,000 people right now, and they started sharing their files, maybe because they wanted to show a bag or something.
00:25:18:07 - 00:25:44:16
Unknown
And, but there's like 30, 40, 50 artboards on them. And I was like, wow. So that's the unlock because you can use the agent just spit out so much. You actually produce so many different variations of things. You explore it way deeper because it's now accessible like previously. You know, that like it was in your head, but you wouldn't visualize it because you were like, oh my gosh.
00:25:44:16 - 00:26:06:23
Unknown
Like, no, I have to like reposition everything. But now it's just a matter of like asking the agent to take what do you already have, remix it or add more, or like try a whole different flow based on this? The other day I wanted to show something, to my colleagues and instead of I was like, should I go or should I use pencil for days?
00:26:07:01 - 00:26:30:16
Unknown
And I just paste this screenshot, from pencil into pencil. And I said, like, edit panel on the right side and like, do this and that in there. It completely reset. It has redrawn pencil from scratch and added, you know, those blocks, they're exactly how I want it in matter of like couple minutes. You know it would be interesting is almost agent specialization right.
00:26:30:16 - 00:26:49:14
Unknown
Little agents in pencil that actually have a brand design agent or something that had some sort of, I guess you can almost think of it like templated context. It's it's been it's been maybe a couple of months or weeks now. Where do you stand on that? Is that something that seems like it would be useful, or was that just an idea?
00:26:49:16 - 00:27:12:10
Unknown
If you're optimizing for like particle context exploration, then I would say like run sub agents, they're role based. Like the sub agent is a marketer. This other person knows everything about branding. And then there's this like assembler agent that like takes maybe input from these two agents and put it puts it all together. Maybe you have a logo designer agent and so on.
00:27:12:10 - 00:27:33:00
Unknown
So if you're optimizing for like clean context, I think that makes sense. If you're optimizing for just like parallelism, you don't want to do that because then all these agents are waiting for each other. And actually it takes longer than you would think. And sometimes it might be just easier to like, dump the whole context or like what?
00:27:33:02 - 00:27:54:10
Unknown
And, let it just figure it out. I believe you've said vibe coding kind of feels like the flash era. And I mean, you would, as we talked about, you would know best. Is there a part of that analogy that you think people get wrong? Literally, folks using flash, we're kind of like design engineers in a way that we would probably call them today.
00:27:54:12 - 00:28:11:04
Unknown
They knew a little bit about design or maybe a lot about design, and they also knew a little bit of scripting or coding, and they were able to combine these things in ways that like if you just focus on like a design you want to be able to do, and if you just focus on code, you want to be able to do it.
00:28:11:06 - 00:28:36:00
Unknown
Now with type coding, this is really interesting because maybe you're leaning more towards design and suddenly you have this sort of helper helping you to realize for vision, you know, which historically was like really hard or you would draw something in a design tool and then you were like, oh no, I want to build it. Impossible. It was like, you know, how where do I even start?
00:28:36:02 - 00:29:08:12
Unknown
But now that suddenly becomes possible. So I think that analogy is actually very, very close to the creativity that we had, but then lost. And now we're regaining it again. Now kind of following that analogy through, you know, from my understanding and perspective, flash essentially died because the web moved to open standards and mobile, and I'm not sure if you would agree with that or not, but the what what do you think is the equivalent forcing function for, for this kind of lack of a better term vibe coding.
00:29:08:12 - 00:29:29:10
Unknown
Right. What's the iPhone kills flash moment for this generation of tools? It's just that suddenly anybody can create a little more than they were able to before, or they were blocked by having to have a team or the right people. And now suddenly you can just roll up your sleeves and, just ask the agent to believe something, and you just keep digging.
00:29:29:10 - 00:29:51:15
Unknown
And, that's what I find, like, incredibly exciting about this whole thing. And especially for people who are very creative, this open so many possibilities. How do you view kind of the happy accident territory, is it that, agents are generating 20 to 30 screens? Can an agent have a happy accident, or is it always a sample from a distribution?
00:29:51:17 - 00:30:16:07
Unknown
It can totally have a happy accident. Actually, we're now, working on this new feature, which we call, let it cook. And, in combination with real agents. It's really fascinating because it, so you pick like, let's say five agents, to work in parallel, but also you say five iterations of Let It Cook or 20 iterations of Let It Cook.
00:30:16:09 - 00:30:42:10
Unknown
And, you can decide what do you want to change in each iteration. The really interesting part there is that just let it run for like five iterations and you discover that it drifted in particular direction. That's navy. Like wow. Like that is surprising to me. Like, first of all, why it happened. Let's dig deeper into down. A second of all, this is something I would have never thought about.
00:30:42:15 - 00:31:04:16
Unknown
Yeah, there's you know, it's interesting because I personally have never been a big fan of node based canvas editors like we saw with weave or Flora. Until it started, I started to see the value in seeing like video and image generation variations from a source. And I was like, okay, I can see why that visual presentation makes sense.
00:31:04:18 - 00:31:28:18
Unknown
But but then with design work and if you think about pencil, and you think about variance, how does the how does the version control problem start to get impacted by this. So we just create a lot of content in front of you. Right. And that's why we do canvas honestly. Like that's why the tools that don't have a canvas they force you into particular flow.
00:31:28:18 - 00:31:51:10
Unknown
Then you need to actually create a different branch and so on. But then they are not connected. It's about so that it's like, this is where I think this is the power of canvas. Like, you see 20 variations and the arc angles. This, famous architect from Denmark described this as like, I need to see a crime scene in front of me.
00:31:51:12 - 00:32:21:05
Unknown
And, he basically lays down all of the ideas like a big whiteboard, and then he just, like, sits like this and, like, looks at that and thinks and just lets the sink and I think the same thing is with canvas. But, I told who are designers on the team? And, like every single file that I have seen from them, it's always like 50 artboards and is just a copy of a copy of a copy.
00:32:21:11 - 00:32:38:10
Unknown
I'd be like, possible tweak or a completely revamped things. And that's why I think you need this. Do you feel like AI is making you a better author of your work?
00:32:38:12 - 00:33:05:17
Unknown
I would say so. I think just because it allows me to do more iterations ultimately. And, historically. So so for instance, I'll give you some examples. You get a spark of inspiration and it's like 10 p.m. and you're like, I have to try it right now. So just open the laptop studio code and just like jam, like try a few things just to collect my thoughts.
00:33:05:17 - 00:33:29:13
Unknown
Even so, like, never say I wake up and I'm like, guys, I think I figure it out and like normally begin at times. What you would probably do is just like take a quick note, like jump on a call with everyone next day, but like there will be only the start of it. And then like some people will try to convince you that this actually doesn't have legs and so on.
00:33:29:15 - 00:33:52:04
Unknown
But today you're like, I have a prototype, it works. This is how it should work. Because you already did the remix. You went through multiple iterations, you check with each and you try different things. And this is, by the way, how a lot of the features in pencil happened. Someone on the team or myself, I would have a spark with inspiration, like the cursors on the canvas.
00:33:52:06 - 00:34:13:03
Unknown
Like we were working on parallelism for a long time. I always had this dream of, like, multiple agents working on the same thing, you know, and on the canvas, actually, parallelism is, like, very natural because everything is a copy of something or you have like party code works like hero section can be done by one agent for it can be done by another and they don't interfere.
00:34:13:05 - 00:34:33:12
Unknown
They actually really nicely work side by side and, and we we sum it, build it. And I was like, there's something missing like the cursor, you know, and this idea that I got like Friday evening and I was like, I need to I need to figure it out. I need to like see for myself how cool that that's going to be.
00:34:33:14 - 00:34:56:05
Unknown
So just coded, coded and coded it for the weekend. And then I showed it, folks, I was like, this is this is what we need to build. And, and maybe, maybe portion of that, it was a throwaway code and maybe, maybe like 60%. That's fine. But then we did it for it for real. But everybody already knew, and everybody was very excited, like, let's get this for the finish line, you know?
00:34:56:05 - 00:35:27:08
Unknown
So like oftentimes for me it's a mean of motivating team or inspiring the team to do something. If swarm grows from, you know, six agents to 60 agents inside of the canvas, what is the designer actually doing at that point? Orchestrating an orchestra, I guess, just like you were the owner or a design lead of a design agency.
00:35:27:10 - 00:36:01:23
Unknown
But also, that doesn't mean that you're limited from going in and actually changing something. We have had image models for a long time, but they always produce a bitmap. And it is so frustrating that you can't go in and like move something in there because it feels so close, yet it's so far. And the worst thing there is, like you ask, change something and it changes like the whole thing and you're like, no, it's just what it did, this little particular thing.
00:36:02:05 - 00:36:37:22
Unknown
So I'm not saying that like a portion of this problem isn't going to get solved eventually, potentially. But I still think that either way, you need to have the control over things. And that's the last 20%. And if you don't have depth, then that's where like the slop happens and and you feel disconnected and you feel frustrated, like you mentioned that the only way you could vibe code would be to produce some assembler assembler language, and you would have to ship the app and you would never, ever be able to go into the code and change something in the code directly.
00:36:38:00 - 00:36:59:05
Unknown
Now, you know, a lot of folks today say, if you're a company and you have a roadmap that's longer than, you know, the first quarter or 30 days, you're doing it wrong. But I do want to know, let's say, in 2028. What what is the thing that pencil has done that you wanted to do that would make you feel quite pleased with what you've created?
00:36:59:08 - 00:37:18:22
Unknown
We don't necessarily have a backlog anymore. Everybody has to do lists. They, that's very immediate to things that they want to work on and that we have like sort of like ideas on the back of our mind that we want to solve. But I would say, like every week we discuss what is the most immediate thing we want to solve.
00:37:19:00 - 00:37:55:03
Unknown
And, that's just based on the feelings, honestly. Based on the discussions, we just wrap, on discord, based on what's happening on the market that we think that we should be shipping. And it's like, very fluid. So it's really hard to extrapolate, like where this is going to get us. But I think the overarching vision is that everybody has this capability to design things at a scale, to iterate as much as they can on them, and still be part of the problem.
00:37:55:05 - 00:38:22:00
Unknown
So like every engineer and every designer having like a swarm or, or even like a single AI designer at hand, they can tap into and they can that can help them create things. I definitely see motion and animations, in there. I definitely see, that the whole iteration process is going to get better and better and faster.
00:38:22:06 - 00:38:38:20
Unknown
Taste comes from participation. That is a great bar that Tom dropped. The agent can start the idea, wandering a bunch of directions and stumble upon a happy accident. But if you stop putting your hands in the clay, do you start losing touch on how to shape the work?
00:38:38:22 - 00:38:43:13
Unknown
This is the same philosophical pattern we stumbled upon in my conversation with Amelia Wattenberg.
00:38:43:17 - 00:39:01:23
Unknown
Pencil is the counterpoint I'm paying attention to. More compute is coming, but it's starting to appear that you can both adopt AI automation while preserving authorship. And that's not a very common conversation right now. I'm going to keep digging into it until then. I'll see you next time.