Nik: Welcome to AI and Design, where we explore artificial is reshaping the world of design. I'm Nik

Dan: And I'm Dan Safer, and we're faculty at Carnegie Mellon's Human Computer Interaction Institute. week we break down the latest AI developments, dive deep into topics that matter to designers, and with fascinating guests who are right at the intersection of these fields.

Nik: Whether you're a designer working with AI or an AI practitioner interested in design, we're glad you're here.

Dan: We're catching up on some news this week, including the launch of Claude Design 2.0 that happened right before config.

Nik: Also, friend of the pod, MC Dean asks who sets the quality bar now?

Dan: Emily Campbell lays out the layers of AI experience.

Nik: And finally, Perplexity launches Brain, a self-improving memory system.

Dan: Alright, well before we dive into all that, Nick, talked extensively about this. My is full of this, but anything you want to note about what ⁓ Figma at Config twenty six since ⁓ I was and you were

Nik: so I felt like a lot of the announcements were a bit muted to me. And ⁓ maybe that's just because I am not as big of a Figma user as I once was. maybe it's because there other big announcements that are happening and news stories in regards to like Claude design and other things. and so some things can get overshadowed. It felt like there were a lot of nice feature enhancements with things like Figma motion. I mean, that looks cool. At the same time, framer motion has been around and other there's been other motion tools. I think putting it into Figma makes a lot of sense for the users. I think it's a great enhancement for power users of Figma. ⁓ the shader stuff was fun. I've never been a big shader designer, but I think that's it's cool. actually felt like the thing to me that was the most exciting was the generative plugins. And there's two reasons for that. The first one, I love design tools. I love building design tools. I love thinking about design tools. And the idea of giving your users more ability to design their own design tools, I think that's great. the ability to do that with generative AI now, where you basically can have the system code a Plugin for you that's going to work well for your workflow, I think, is awesome. Also, as someone who's tried to design Figma plugins in the past, I've felt sometimes limited or sometimes it was just tough. Like you gotta learn their API, and when we were hand coding that, it wasn't super easy. And so I love that idea. I think one of the things that maybe isn't as much, but I thought was really cool in some of the fine print was that when you Create a generative plugin, it costs tokens to generate it because basically you have to have AI generate the plugin for you. But when you run that plugin afterwards, it doesn't cost you any tokens to run. Probably because it's using the API tools, it's using the the capabilities that Figma has to do stuff. ⁓ thought that that was really cool because it means that things that you might have done where you ask AI to do all these things. And this is how I am working in some cases now. Like I am asking. asking Claude to do the most ridiculous simple things now. And I know I'm burning tokens, but it's like just it's a path of least resistance. building these plugins that leverage the capabilities and then now can work for free effectively for you and your team members. I that that's really cool. And I also think that's something that could really differentiate Figma in regards to tools that designers want to use because you can think about your workflow, you design your plugins, and then you have now all of these specialty tooling that it works for your workflow and how you like to do things. Also, the Figma plugins are probably deterministic as opposed to sort of like, well, the AI might do it well or it might not do it well this time. I actually think that that was a really cool thing and something that could help Figma to really differentiate itself if people dive in and like build a bunch of really cool plugins for their workflow so that Figma becomes a very customized platform for their team.

Dan: Yeah, I hadn't really thought of it like that, but you're right, in that it does a couple things at once. One, helps you customize ⁓ you're burning more tokens, two, yeah, because you're customized and you can able to build this thing for your particular workflow, that's more of a lock-in for them, right? That where you're like, okay, well I'd I've customized this thing to be all set up for me and I don't wanna leave the platform that I know. to go somewhere like Claude Design, for instance, where I don't have access to my plugins that I may have spent, months and months building.

Nik: Right, and in addition, I don't have the way to call those types of workflows or plugins right now Claude design, but we'll we'll get to we'll to talking to Claude design and we'll see how long lasts. but I think given the fact we for for example the AI and report, the 2026 that we talked about a couple episodes back, there's a lot of designers who are using code development tools, to create their own design tools. So I think that this feels like something that users are gonna want as long as they actually ⁓ know that there and they use it. And another thing that I think is going to be potentially very interesting is Figma figure out how to recognize what users are trying to do with the AI? ⁓ And then say, wait a minute, we can build a plugin for that. We actually have the API calls for what you want to do. We can make that so you don't have to burn any more tokens to do that type of workflow. And we'll just create or recommend that you create a plugin for this. That's where I think it could get also really, really exciting by them analyzing, looking at it what it is that the users are trying to do with, say, Figma agent, and then saying, wait a minute, that could just be a plugin.

Dan: and if they notice you doing the same thing repeatedly, it's like, well, how about if you make a plug in for this?

Nik: That could potentially help them to improve the Figma API itself and the Figma platform to for more capabilities. And so they can basically keep improving the Figma platform based on that. I could totally imagine that as at certain points, if you recognize that there's all these people creating these plugins, and I don't know, they have to probably think about the user privacy aspects of things, because maybe those plugins might represent proprietary or like I mean, that's some secret sauce workflow that your team has to work faster. I don't want the world knowing that. ⁓ at the same time, there's a lot of people who share plugins ⁓ freely online. So maybe there'll be options to say, ⁓ I generated this plugin and like share it with the community or share it with Figma. So it'll be interesting to see how they use this. so that's my thing. I think that that's probably one of most interesting things that I saw from config and I think that feature could be really cool and could be used to great effect to great return on investment actually for Figma power users who leverage it and actually build things to make Figma work even better for them without always having to be like cool I'm just using AI to make it work faster.

Dan: I think there was a handful of things that you and I and probably our audience are were really interested in. And this was what this was the most baked of those things. The other things like code layers skills were of like, ⁓ those are those are coming. Code layers was like, here here it is and it it looks but don't look too close under the hood. And skills was just kind of like, yeah, that's coming. We're we're excited about it. And no more details on that. So looking forward to seeing that that appears.

Nik: I agree, but that actually leads us into our next story, which is Claude 2.0 and the fact that they are basically able now to do a full round trip between the sort of Claude design Claude Code and being able to go back and forth between the two systems.

Dan: Right, so we were out and right before config, anthropic Claude Design two zero, a major upgrade, it introduced things like canvas editing, ⁓ design system importing, shared usage pool, and this yeah, the two-way integration with with clawed code. And The thing that I was like super excited about, because I know this is the hobby horse that we ride on this show is yeah, that that you no longer have to rely on text prompts for these tiny tweaks. You can just Drag and resize and align elements direct manipulation for the win. Exactly. so the stuff that we really like and that designers really like and just and makes life so much easier is now much more functional. And the joke I was saying at ⁓ config was

Nik: Direct manipulation.

Dan: That Figma is trying to move more into the code space, and and Google trying to move a little bit more into the Canvas space. And so I guess we'll s we'll see who grabs it first or who grabs the most first. couple other cool things about Claude 2.0. you can now pull design system straight from a repo or a Figma file or a code base. ⁓ So instead of Claude guessing this is what your sliders look like, it actually knows what it looks like. Thought that, of course, is gonna be beloved by everyone. other thing about having those components in ⁓ system now is that Claude will do a self-correction your design rules before it even shows you anything. So it does this kind of verification loop before it presents its output to you and not afterwards. So that's I think great. A really nice addition. And the last big thing was something that everyone who tried Claude Design in first round was like, ⁓ my god, I just used all my tokens in like three rounds of this. And now your entire pool of of Claude Design doesn't have a separate one, it's part of your subscription now so that I think will be a big bonus to people actually starting to adopt this tool a lot more.

Nik: The Claude design updates you they're coming fast and to what was speaking about earlier, The plugins are cool and I just wonder how much longer be before there's plugins for Claude design because it's a good idea. And you might want to do stuff in a more deterministic or things that are like workflows that shouldn't cost as many tokens to do.

Dan: Well here was my question about that. the thing about plugins on Figma Is it kinda makes sense because You kinda get this like easy access to the General Figma plumbing but I'm guessing the plug in is mostly about getting access to sliders and having some kind of your own direct manipulation for something that you want to do rather have a skill that does it the same way all the time.

Nik: yeah, I think it's really just like The plugins are really about controlling the API. which is nice because those are deterministic functions that Figma's already provided to you. Whereas there's there is I mean, I don't really know. There is no API for Claude design. You just ask it to do stuff. so I arguably it would just be, yeah, you have skills and so you could just write skills and I think that's maybe something that I don't see as as well integrated into Claude design as yet. Like actually as I think about it, There isn't actually an easy way to a attach our own skills. ⁓ Like you can add your design system and you can add it from a bunch of different places, but when you go to skills, there's a create and it's make a deck, make a dock, it's it's all these things that are just fixed and it just says use, but there isn't a give it another skill. Like it has its actually its own front-end design skill, but like you can't give it your own front-end design skill. You can attach a code base, which I'm sure you could put skills in, but it's not so easy from what I can tell here well, I imagine we'll see long it takes for them to add that. But so here's something, right? Right now, don't really have the ability to update and change Claude design to work how I want it. Which is also kind of similar to like the way Claude code works. And this is where the come with say people who are designing their own harnesses and they're using open code or pi So these are editors That manage agents and they do agentic coding, but you define a lot of how they work. you define how it iterates and how it looks at things, it does say self-correction. Whereas Claude design, I don't get to change that. And actually it's a it's an interesting thing because there are that are trying to mimic the open source kind of you can design your own coding harness. open design is a project that is of the open source equivalent of Claude And the cool about it is that you can go in and start doing you can muck around with it a lot more and you can define it to work the way you want it to work. And it's interesting where actually now that I think more I know that we're keep talking about the the Figma plugin thing, but that's sort like a middle ground here where Figma's a lot of good decisions for you, but you still have this ability to say, I want to make Figma the tool work in a certain way. for me. And I don't really have that here with Claude design. that's something that maybe is bit limiting. But I for most people, especially if you're using Claude design as say someone who's maybe software developer and you're mostly in code all day, like Claude design is an awesome way to say, cool, now I've got like a front end interface to actually come up with how I want something look as opposed to always having to write it in code and then just debug it in a browser.

Dan: Mm-hmm.

Nik: Right, this is a way faster iterative loop to figure out, a front end.

Dan: again, like you're right there, you're already in the tool. Why wouldn't you just flip over and do it that way? But I mean designers say the same thing. Like, hey, we're already in Figma. why am I gonna jump out to Claude? But also, it also seems like one of the things that said is that you can also you can hook it up to your AI of choice.

Nik: and I'll say, I know I our listeners are probably like, you guys are really Claude fanboys here. But part of it is that everyone figures out what they work with, but you know, codex is pretty good and I know a lot of people who shift back and forth to codex, So while I think that Claude design is exciting and it's interesting. I don't know how much lock-in it's actually gonna give given the fact that the whole ecosystem changes fast and every week some other coding agent is maybe getting better. and thus that's where companies like Figma or open source projects like Open Design ⁓ have a ⁓ leg

Dan: And let's not forget that Claude Design didn't exist two months ago. this a new product. it's literally fresh out of the I mean they did kind of a pretty big update like two months later, that's pretty impressive. So who knows what it'll be like in Several months from now, I mean the speed of these things, like you were saying, it's it's incredibly fast. So stay tuned for more on the Canvas wars. Dent dun. So let's yeah, let's move on. our friend and former on the pod, MC Dean, maker of many design skills, has a new article out called Who Sets the Quality Bar? And what MC is talking about is the gap between ⁓ the demo that you make the like shipped user experience and it's this pretty large design gap and the it exists right now is that usually nobody has any standard before the building started. So she basically answers her own question. She's like, nobody owns this. And this is kind of her response to the AI and Design 2026 report that Designer Fund put out a couple of weeks ago. And about just this gap between, these unreliable, quality these models put out. where you would like to go and there's no one that owns that and ⁓ some of it does come out of a good problem here and the is that designers actually starting to build Tools for other people to use. are design system components and brand guidelines, and they're putting those into coding tools. And so anyone can use them starting from this shared quality baseline. So the intent is really good here, and it's and it's the right way to go. But in the end between that where the designer is setting up all these things and and something going live things get lost and so that's what this article is talking about.

Nik: So a question I have is, isn't is it that nobody owns this? Or is this what the product manager or the team lead owns? I mean, when I'm looking at the markdown file, these are really great questions and things to think about, who's the user? What's the purpose? What does this solve? But They also do feel to me like things that team with a mature design process would be managing. guess is it that just that people now are building so fast that like they're not thinking about this? and this is why you gotta have the the skill file to kind of like, let me help slow you down there a little bit, make sure you got this.

Dan: Right, exactly. I think that's it. Is that some of the stuff a PM or a designer is not looking at? Maybe not even an engineer. Again, these are tools that maybe someone from HR is like, hey, I need to fix this widget in the HR site or whatever it is. And something gets shipped that doesn't meet the quality standards. And it's like, well whose fault is that? Is it the people who who never even saw it? Is it the designer up front? Is it the HR? It's like it's nobody. It's kind of it falls between the cracks. And so of course, MC being MC, she actually does create a a template for setting up quality criteria. And and you can it, we'll again link to this. in the show notes, but she sets up a quality bar dot md file where people can just use that before you build making sure that some quality bar has been set before something goes live.

Nik: Yeah, one of the things that I think is interesting about this, and again, because it's an encoded is that this gets shared with the team. Maybe it gets put into, whatever that you're using, it always accesses this. And it's potentially been by maybe team members who've thought about it, or it gets called by your AI. ⁓ coding assistant and says, Hey, before I move forward, we gotta talk about this quality stuff and ⁓ fill this out. I could actually see this as a really good way, but simply just even saying like every time, you know, we start a new project, no matter who you are in the in the company, you've got to fill this out before ⁓ the agent moves forward. And this would do a lot to help improve the quality of things. And then also if you have any kind of institutional sharing of these things where it could say, hey, here's what other people wrote, or here here's our template within our company, especially, ⁓ know, if you're working on say a any feature effectively could leverage the same thing. But I really like your idea here of like, ⁓ people in, HR, people in business operations being able to build their own tools internally and hey, also those should be of a certain quality for our company.

Dan: Mm-hmm. Yeah. I mean, as we start democratizing and engineering, further away from designers and engineers where all that stuff is now scaling in such a weird way. we need to figure out the work when there is that weird gap. between the people who used to, gatekeep who gets stuff put in the product for good reasons, one of being quality. you know, now we're starting to see that, these tools anyone can contribute to pieces of the product for good and for ill.

Nik: Yeah, and this brings up something that we've talked about on the show before of this sort of future or probably forming role of an AI design ops. ⁓

Dan: Yeah, it was, how do we change the design ops role to be more this? Where's how do we scale the design so that it then touches all these folks? because the design ops used to be how do we have The design team make most impact. And now it's well, how do we make design more broadly have a bigger impact across the company? it's a twist on the design ops role.

Nik: Yeah, I'll say if there's anyone out there, I know we've talked to some people who kind of have talked about that and like they're trying to do it but if there's anyone out there, in that role or you're like, ⁓ yeah, that's what I'm doing now, it'd be cool to talk to you and hear about how you're implementing it at your company or in your organization.

Dan: All right. moving on. So our next topic is an called Layers of AI Designing Beneath the Surface. And this is an article Emily that I happened to meet two weeks ago now at config. And basically, what this is sort of a riff ⁓ on my old friend Jesse James Garrett, his elements of user experience diagram that every design and every designer ⁓ post 2000 has had to look at and learn and think about. And her take on this is that the elements of user experience basically works when you a deterministic system. And it a very kind of deterministic model of design where you did this and then this happened and then that happened, and you made these stacks that all kind of built on each in a predictable way. What she is trying to say is that now that we're in kind of the era of generative AI, our product systems have become a lot more complex. so we need to actually start to think about what the layers ⁓ AI experience actually are. because the layers actually become not as much fixed as they are in Jesse's diagram, but they become much more interdependent on each other. So you may have user interactions that then go all the way to the like the governance layer. up to the model and through the harness so that and maybe they come back in a different way and change the user interface at the top of the model. based on the behaviors that it's doing constrained by the harness or the context that it's in, or the or the governing model. And so the layers of user experience start to really shift in a of a profound way. so that they are just much more interconnected and much more fluid in a way that we have not ⁓ prior to the last couple of years in this kind of realm of generative AI.

Nik: Yeah, this article quite comprehensive. I really like the breakdown here. because it take you through sort of all the areas. I think most people, ⁓ maybe our probably you're gonna be pretty aware of the Two and maybe the third one, you know, of course the user interface and the context. I think a lot of people are understanding how context driving things. I I think more and more people are kind of thinking about harnesses and sort of how you have agents that are orchestrated working in the background on things. And I that you're gonna see more design work in the harness layer. One thing that I thought was interesting and I'm glad was brought up was the discussion of the layer. And I'm ⁓

Dan: Yeah, I was gonna talk about that.

Nik: Because this is something where it's actually it's fun actually when you scroll down the emergence layer in the little cool graphic actually becomes this weird amorphous cloud instead of like, it just highlights the the layer block, which I was like, that's cute and I think representative of emergence. Because it is this weird thing of these things that come out of the AI models and the AI systems that just were not really predicted and it's to control. A good example that gets brought up in the article is the open AI goblin instance. For those who might not know, all of a sudden started talking about goblins in all kinds of weird places where it didn't make sense. And basically what OpenAI kind of determined was they tried to give a little bit of a training incentive to make the personality of ChatGPT a little more geeky. And then that just like led it to sort of this space within, the embedding space of like, goblins and fantasy stuff. And so it it started mentioning goblins all the time. And they've eventually fixed it, but it it was just sort of this thing that came out of some of the ways in which they were say trying to train and tune how it was behaving. And so there's all kinds of things that will happen like this that we don't really understand. And I think that understanding that that's a thing, aware of it and probably being able to monitor for that kind of stuff is gonna be really important the future of AI driven product design.

Dan: Yeah, I mean there's all kinds of emergent behavior, everything from the way that sentences are phrased when they're by AI to six fingers ⁓ on on You know, was a big thing for a while too. And I think, having that as part of the diagram one of the things that makes this a probabilistic setup versus something that is very deterministic. Not that not the deterministic things also can't have emergent properties, but it really has emergent properties when we start talking about probabilistic systems as well. Like the weather is a another great example of a probabilistic system. All kinds of stuff. That's that's what we can't. Have super accurate models because

Nik: think ⁓

Dan: There's a lot of probability in complex systems like the weather or systems.

Nik: the complex systems is an interesting segue into our last article for the day in thinking about memory, it adds another layer of complexity. It's memory systems are gonna also become an important part ⁓ how AI products probably start to work as they work more and more with people ⁓ people expect AI products to know more about them ⁓ to be able to remember things when they come back to using those products over and over. And does us into our last story of the day, which is perplexity announcing brain, which is their to ⁓ AI memory systems. brain builds a context graph ⁓ work that computer, which is their AI agent system, it what it what that system does. And the that's interesting here is that oftentimes memory in a lot of AI systems is about y the person, the user, like their preferences, their thoughts, like things about them. Whereas ⁓

Dan: Right, their per a personal context.

Nik: Exactly. whereas this is actually specifically more about the work that the agent does. And it tries to effectively remember more about worked, what what corrections got made, so that it can learn to do better, which is actually, much more in the line of ⁓ I'm using AI as a worker. This is a an employee and should be learning and remembering when I've given it feedback, when I've said, hey, that doesn't work so well, or that was good, and yes, that let's accept this type of output. And so it's it should be learning from that.

Dan: Right, it's the as as teenage idea, right? But for the longest time the teenage intern does not learn. Like you sense no, stop doing that. And it it never learns. It'll always keep coming back and doing that. And now if you're doing some kind of corrections hopefully the agent can start to self-improve on the corrections that you're giving it. just to make itself smarter. And what I thought was kind of cool about this is or maybe kind of funny about this is that it starts building basically a wiki of what of what it's doing and the context and Presumably you can go in and edit the wiki and and do stuff just like you were, you know, editing your company's intranet in what what was that Two thousand five, two thousand ten? ⁓

Nik: Yeah, I'm just imagining our listeners here shuddering at the thought of editing a SharePoint, and actually it's kind of fun because we have had company intranet wiki and they are a great way to organize ⁓ knowledge about certain things ⁓ they're just hard to edit. And so now when the AI editing it's sort of effortless. And the fact that the model still works, what works for human teams if kept up, if maintained, also works for an AI, you know, that's kind of cool.

Dan: Well, I just happened to see the ⁓ that Claude has been keeping of me today. And speaking of these emergent properties, Claude thought that we had renamed our podcast here. He thought he had we had renamed it Magic. That it used to be called AI and Design and now recently it's called Magic. So ⁓ Yeah, I don't know how or why, but welcome to Magic, everyone. You cannot ask you you cannot that's a good question. Let me s let me see. Why did you think

Nik: Can you ask it wha in what chat? Yeah, try it.

Dan: My podcast was renamed to Magic. Should I spend a whole bunch of fable credits finding this?

Nik: Yeah, you only have ⁓ for another day.

Dan: I might as well burn ⁓ yeah. Let me look back to where that service is searching now. ⁓ I have it, I have it. So the transcript fought back in mid-June I uploaded an episode transcript Magic Episode 2. And that comes from is that Riverside creates magic episodes where it does its own little take of it.

Nik: So while we're searching for dance ⁓ yep. Okay.

Dan: So a file called Magic Episode 2 look like a rebrand. Funny, right? who would have ⁓ so from a name, again, I could not have predicted that when working with this. And so it interesting the thing, the kinds of context that is being picked up and possibly used.

Nik: That's hilarious. when you told me this, ⁓ was like, ⁓ I want to know what it knows about me. And I opened it up and realized that from for my Claude, I had turned it off at some point. I guess I never let it have any memory and so that it has no memory for me, which funnily, ⁓ the that you're using of like, ⁓ it's like a high school intern that's like super eager and doesn't always learn. I've also noticed I'm like, yeah, you know, it doesn't always figure out like this is how I like to work or please do this. I really feel like actually when I when I fire up a new ⁓ chat, like this is like a new Claude. know there are other things for example for Claude code like claude.md files or other types I can keep my skill and things. And and those of course I put the skill files in and then it's remembering certain things about how I like to do stuff. But it's always felt like a very manual process. And this is something that of course I think that is cool about what Perplexity is trying to do here. Probably what Claude is trying to do with the memory and what OpenAI is trying to do with any of the memory systems is how do you make this a self-improving system? How does the thing learn on its own so that I don't have to go in and always be like, ⁓ let me edit this, let me include a skill file, let me update your memory or what you want. Like you just learn from interacting with me.

Dan: And Gemini has a lot of this well. And it it a leg up on all these systems in that it's got access to my Gmail It's got access to my search history. It's got access to my calendar. It's got to Google Photos. so Gemini actually knows quite a bit about me. So it knows like for instance that I rescued a dog a couple months ago. so now when I asked for restaurant things, it sometimes will say, I know you have dog. Do you want to look for dog-friendly restaurants and stuff like that? Sometimes negatively, where I'm like, No, I just want a restaurant. but I think that is is an other source of personalization and memory that I would guess the Gemini of all of these different AI knows the most about me right in in the three-dimensional way because it has both my personal stuff and all my work kind of life. Now I ⁓ won't I have been working on this other book project and at point I was researching like

Nik: Hm.

Dan: Hairstyles for older black women it thought that I w is a fiction book. ⁓ not nothing weird. And like, you an older black woman? Shall I point you to different stylists in the area? And I'm like, No no no no no no no. So but for half a minute I was like, ⁓ could I with Gemini and be like, Why yes I am to throw some chaos into the system but didn't want the algorithm to get all weird on So I decided not to do that. But yeah, I think this idea of memory and knowing context and remembering that and changing things is is really interesting, whether it's personal or behavioral, like from an agent taking action on my

Nik: Yeah, one of the things you make me of here too is also the ability to go in and edit, but also the ability for it to forget. How does it forget effectively? And then also how can I go in and you know, eternal sunshine of the spotless mind it and remove certain things because there's definitely stuff I can imagine I want to pull because it's not working for me anymore and I recognize that I mean I'm thinking in the way like I every so often I like go into my YouTube history and just, delete it. And I just start over because ⁓ I into some kind of like it's trying to learn me and like it gets me into some weird hole that I'm like, nah, I'm not liking this anymore. I'm not liking the recommendations. You know, I don't remove my subscriptions, but whatever it was trying to do to do dynamic recommendations is just not working. And I just have to hard reset it. But I feel like there's not as fine fine-tuned abilities for me to in and update my preferences and say a recommender algorithm system. But here it's a little different because arguably if it's just a text file, I could go in and read that. I guess yeah, maybe interesting thing for our listeners here, like open up your memory file.

Dan: Yeah. Well, I was gonna say some of my students back fall actually dealt with the problem of memory and had really interesting concepts around how to do memory management. And I'll post of those in the show notes as well. I think that's probably our show for this week.

Nik: Wait a minute, but Dan, I'm sorry, did you say you were have a new book?

Dan: ⁓ I do have a new book that is coming out in October called The Restoration. And it is a book about what happens if the United States fell apart in the next 30 or 40 years, and were recolonized by the British. and the British are governed and and overseen by a hyperintelligent ai known as the sovereign and so this is first book of a planned trilogy coming out in october twenty-seventh at a near you Yeah, if you're interested, I've got the sample chapter up and bunch of the glossary and a map and some other stuff. So if you're interested in AI and science fiction, perhaps this this book is for you. I'll post the link again in the show notes. And on that note, we will really wrap up the episode. Thanks for listening to AI and Design Pod, aka Magic. And see you back here for next week a great summer. Stay cool.