Skip to content

Podcast

What to Build First With Claude Design

The AI Daily Brief: Artificial Intelligence News and Analysis

Source ↗ ← All highlights
  • Claude Design Targets Exploration More Than Final Output
    • Claude Design looks less like a full design endpoint and more like an AI wrapper for prototyping, wireframes, and mockups that feed other tools.
    • Nathaniel Whittemore notes Anthropic emphasizes exploration, brand-system ingestion, exports, collaboration, and handoff to Claude Code rather than final polished production. Transcript: Nathaniel Whittemore Now design is pretty much what you would imagine a design tool from Claude to be. And it’s important to note that a lot of this is not some entirely net new functionality. It’s a new wrapper around design with some pretty significant UI upgrades for the design experience. In sharing design, Anthropic points to a few different use cases. First of all, one of the big value propositions they discuss is the ability to explore concepts more broadly before committing to a direction. They call it rationing exploration, and point out that basically no one has the time to actually prototype a variety of different directions for exploration, so you pretty much have To commit to a design system from the beginning before you actually dive all the way in. Agendic design dispenses with that constraint, and the interaction that it proposes is that you prompt it to build a first version or versions. As you’ll see as we walk through the process, you can get specific with Claude Design about how wide a variety of options you want it to explore, and then from there you refine it. Now what’s interesting about the user experience here, is that pretty much every way you could think of to refine the design is available in this interface. You can use natural language inputs to describe what you want to change. You can use inline comments. You can direct edit on the canvas. And there’s a feature that a lot of people are really excited about that are these custom sliders that allow you to drag from one end of the spectrum to another for a particular design Element, such as a font, a color, a shading, etc. Early testers Anthropic said, we’re using claw design for things like realistic prototypes, product wireframes and mockups, design explorations, pitch decks and presentations, Marketing collateral, and frontier design. Basically future-oriented things that don’t necessarily have a reason for being quite yet, but which could prompt the next generation of building. So two key subtexts in that set of things that Anthropic says this is for. One, there’s clearly a bunch of things for non-design knowledge workers, pitch decks and presentations, i.e. The core default asset of most knowledge work, and marketing collateral as a specific group that has to interact with design a lot but who are not themselves necessarily designers. Secondly, though, the other sub-theme that’s clear is that at least right they are not trying to pitch this as the full end product. That’s in contrast to something like Claude Code, where Anthropic is in no way at this point saying just use this to build your prototypes or your vibe-coded tests, and then you can go Back to hand for the final version. Claude Code, and Codex by the way, and every agent decoding tool, is pitching you on the full experience. And yet again, here when it comes to Claude Design, they’re talking about realistic prototypes, product wireframes and mockups, and design explorations. I don’t know to what extent that is, one, them not having as much confidence quite yet in the design product as opposed to the coding product, or two, a blow softener to other design tools In the field. One of the signals that we had in advance of the announcement of Claude Design that something was coming was that Anthropix Chief Product Officer Mike Krieger had resigned from the Board of Figma earlier in the week, presumably because they were going to release something competing soon. That also, by the way, might explain why they chose to release this on Friday. It could have been something of a magnanimous act to try to soften the blow. That might be overthinking it, but it does feel like these companies have a good relationship. And while the meme is Anthropic trying to destroy everyone, it feels less like an intentional effort to disrupt existing players and more just a matter of they’re going to build what They’re going to build. And if you’re in the way, well, that’s kind of on you. Now, going back to the announcement, another thing that stands out is that they very clearly want this to be integrated into your overall company experience. They spend a lot of time emphasizing that it can ingest your brand’s design system. They highlight that you don’t just have to start with a text prompt, you can also upload images or documents or even point it at your code base. And then there are a bunch of tools for making sure that whatever you produce can get to where it needs to go next, whether that’s collaborating with someone else on your team, exporting It to a different type of artifact, or handing it off to Claude Code. (Time 0:02:18)
  • Claude Design Feels Systems First Not Asset First
    • Nathaniel Whittemore frames Canva as asset-first and Claude Design as systems-first, aimed more at websites, apps, and front ends than single posts.
    • He argues the real users are Claude Code power users and marketers who need visual work without living inside a traditional design tool. Transcript: Nathaniel Whittemore Now on this question of what type of companies is Claude killing, the places where people naturally went were the other design tools in the space. Think Canva, Figma, tools like that. One thing that’s interesting to note is that Canva’s CEO, Melanie Perkins, is one of the named quotes on the launch of the Claude design page. And I don’t think that that’s just Cope. I actually think that while these tools both exist in the design space, they are a little bit different. So far in my experience with Claw Design, I think a good division is to think about Asset Design versus Systems Design. Asset Design is something like, I’m making an image, I’m making a post. Systems Design is something more like, I’m making a website, an application, a front-end. While Canva is certainly not unable to do systems design, in fact a lot of their developments over the last couple of years have been more robustness in that set of features such as your Brand guidelines and templates, a lot of the use of Canva is much more specific and discreet to an individual asset, and that doesn’t feel to me like what the core experience that they’re Going for with Claude design is. Now this probably makes sense given the tight integration with the things that people build with Cloud Code, which tend to be less individual point assets and more systems like websites, Applications, front ends, etc. In terms of who this is designed for, it doesn’t feel like designers per se to me. So far, at least, it’s much closer to a Vibe design product in the way that we talked about Vibe coding about a year ago than it is to some full-throated Figma replacement. The two audiences that stand out to me are, one, Claude Code power users who aren’t designers themselves. Witchier embodies this when they write, I live in Claude Code. The visual half has always been the break in my flow. Spec something in words, lose context, re-explain it to Figma, etc. Claude Design is the missing half. Draft UI inside Claude with Opus 4.7 Vision, iterate by talking to it, and the handoff to Claude code pulls your design systems into context automatically. Design to implementation in one conversation instead of three tools. I don’t think it would be totally off to view this as the primary intended use case for Claude design, at least initially, given how important and central to the strategy Claude code Is. But as they found, a richer UI UX around design has benefits for others as well. The most frequently named category of person are marketers, which makes sense as the knowledge workers who have to interface with design and the visual most often. Austin Lau from Anthropic explicitly calls it one of the coolest features for marketers today. Austin writes, I’ve posted a lot about building marketing tools without being technical, and this is the visual side of that. I described a landing page editor and Claude built me one. (Time 0:06:58)
  • Early Users Tested Marketing Pages Apps And Videos
    • Early users made email templates, animated social posts, agency homepages, app front ends, Shopify variations, and promo videos inside Claude Design.
    • Nathaniel Whittemore highlights moon-launch web design, a dating app mockup, and product-page variants generated without separate image tools. Transcript: Nathaniel Whittemore Marketing assets are definitely one of the big themes. Salma, who focuses on AI for brands, generated an email marketing template. Victor Adi and many others did some version of animated social media posts. Many people experimented with rich visual web design. Mark Dalla Maria writes, Claude Design is insanely good. It one-shot this Artemis II moon launch site in a few minutes. And part of what I’d point out here is that Claude Design isn’t using generated images from a model like Nano Banana or GPT Image. It’s using code and SVGs to create the imagery, which, as we’ll hear about in a minute, is a challenge for some, but also creates a lot of interesting opportunity for more interactive And more dynamic-feeling web experiences. Namia at Supafast says she just one-shotted her design agency homepage, and Justine Moore from A16Z built the entire front end for an application. She writes, asked Claw Design to make a dating app for X posters. It eschewed the swipe-based interface and chose a daily digest with your best matches. Also pointing out something which, as we will see, is one of the major frustrations with Claw Design, Justine also writes, on another note, I hit rate limits on my max plan in under 30 Minutes. Another area of marketing where I saw a ton of people experimenting was around product websites. Now you often see this as a test of new image generation models, where people will give the model a specific package and ask it to put that in various product photography contexts, or Ask the new image model to generate packaging and product photography from scratch. Now again, Claude Design is not doing that, but what it can do is what Olivier here experimented with are things surrounding that product photography, like design variations for your Shopify page. Olivier used it here in the way that the blog post discussed, not just trying to create a good Shopify page for his product, but designing a set of four variations so they could figure Out which was the best direction for the product. I also saw a ton of people experimenting with the video generation capability. People are making some pretty impressive launch and promo videos all generated with Claude Design without any separate image generation tool. (Time 0:09:30)
  • Socratic Prompting Turns Design Into Structured Thinking
    • Claude Design’s strongest UX may be its Socratic workflow, which turns vague prompts into concrete product and design decisions through guided questions.
    • In Nathaniel Whittemore’s mobile app test, it asked about role, core flow, voice input, notifications, syncing, and number of variants before generating wireframes. Transcript: Nathaniel Whittemore First of all is something that I would call its robustness. If you are more of a consumer or prosumer type user and you’ve ever tried to create something comprehensive with a consumer or prosumer tool like Canva, you’ll know that as much as they Are trying to move into that space, creating something with a full design system can be really difficult. And that’s exactly the type of thing that Claude Design excels at. Another thing that I really like is the Socratic design process. When you start in Claude Design and give it your initial prompt, for example, let’s design a mobile app companion to the Context Portfolio Builder web app, add a screenshot. It’s going to review what you gave it and then come back with a set of questions for refining it. Now, anyone who’s interacted with AI to build something, especially Cloud Code or Codex, will find this interactive pattern familiar, but it’s re-skinned for design and at least So far in my experience has done a good job of using the questions themselves to actually help you think through what you want. In other words, when you look at a question like what’s the mobile app’s main role, it isn’t just a blinking cursor that asks you to input that, it gives you a half dozen theses that it has, Which can actually help you refine your thinking. So for the mobile companion, it asks what the mobile app’s main role is. Good question. Why do you need a mobile app for personal context building? The two that stand out are either quick capture, jot thoughts that feed into portfolio files, or notifications that are nudges to keep context fresh over time. Let’s do quick capture and select notifications as well. Next up, it asks what’s the number one flow to nail? Daily micro prompt or voice first capture. I think it’s daily micro prompt. From there, it’s going to ask you how many different iterations you want. For some, people will always want the max. For our case, let’s leave it at its suggested three. And for any question where you don’t have a strong feeling, you can simply allow it to decide for you. You can also see in this Socratic design process that Claude Design definitely has a technical bent. We’re designing an app, and it’s not just asking design questions, it’s also asking product questions. For example, how important is voice input? Let’s call it central. This is a voice first app. Yes, there should be notifications. Mobile edit should live sync and we’ll let it decide what tweaks to expose. (Time 0:11:41)
  • Custom Sliders Make Iteration Feel Like Real Design
    • The standout interaction is iterative control through per-design sliders, inline comments, direct edits, and prompt updates instead of only reprompting.
    • Nathaniel Whittemore says the generated sliders for spacing, density, warmth, and layout make it feel like a real design tool rather than a prompt box. Transcript: Nathaniel Whittemore As was highlighted in their announcement post, there is a really strong emphasis on the idea of being able to have multiple variations to test things out. One of the things that I’ve got cooking that I mentioned at the beginning of this show is I’m exploring some approaches to AI credentialing to solve a bunch of problems I see that are stopping People from getting jobs and stopping companies from finding the right people to help them with AI. When I asked for a teaser video for this, it gave me three different directions that were differences in both the editorial content and narrative flow, as well as some of the visual systems. Another really valuable feature is the way that it does iterations. As I mentioned, you can use the text input panel to update the prompt. You can edit certain parts directly. You can provide inline comments that the AI can read. Or, and this is something that I’m seeing people really respond to, you can use the individual slider style tweaks that they provide on a design by design basis. So here’s another project that I’m working on. One of the things that really bothers me about the AI jobs discourse right now is that people are so focused on the destructive side of creative destruction that they’re not spending Any time on thinking about what new jobs get created because of things that AI changes. I’ve been spending some time trying to game out where I think some of those jobs will exist and what their characteristics are and how many of them there might be, and turn that into this Draft of a World’s Fair style expo for the jobs of the future. Over in the tweak section, you can see I can change palette with just a single click, I can iterate on the layout, I can modulate how interactive it is, and I can automatically change the Fonts. And people really love these sliders. The smart ape on Twitter, in fact, calls them the killer feature. They write, everyone is talking about prompting, nobody talks about the sliders, which are generated per design. Spacing, density, color warmth, layout tightness, each one is built for your specific artifact. It’s what makes this feel like a design tool and not a prompt box with a preview pane. Now, there are some challenges, though. Frequent AIDB guest and AIDB education collaborator Nufar Gaspar wrote that the three things that impressed her about claw design were these auto-generated tweaks, saying that This type of click-to color swatch changes and other UX elements make, in her words, iteration feel radically different from every other AI design tool I’ve used. She also points out that the design system translation from examples is real, i.e. When you drop an existing branded deck in, even without a brand guide or a design.md file, Claw Design successfully extracted the visual language and stayed consistent across the New artifact. And finally, she says it keeps polishing on its own. After generating, it continues to iterate without being asked, fixing text overflows, inconsistencies, things it thinks it can improve. Feels like a designer doing a second pass before handing it over. (Time 0:17:26)
  • Exports Video And Limits Still Hold Claude Back
    • Claude Design still breaks down at handoff and rich media, with exports and video far less mature than HTML-based output.
    • Nathaniel Whittemore cites broken slide transfers, Canva errors, SVG-only image creation, harsh rate limits, and a 4.5 out of 10 score for video creation. Transcript: Nathaniel Whittemore However, Nufar points out that there are some challenges as well. One of the big ones for her was the exports. She had a ton of problems moving particularly slides into other formats. PowerPoints that didn’t have the fonts that Claude used had significant quality degradation. She got error messages trying to move it over into Canva. She could export things as screenshots, but it was slow and not editable. Basically, the only export methodology that actually worked well was HTML. Simon Smith also pointed out that the lack of a native image generator does have limits. He writes, it doesn’t have a native image generator as far as I can tell, so when it needs images, it will create SVGs. It’s pretty good with SVGs, but this of course limits the type of images it can create. Still, for most people, the biggest limiter so far has certainly been the rate limiting. Josh Gonzalez says, oh, so Claude Design has its own usage limit outside of everything else? And of course, already hit it, so now I can’t use it until next Friday? Okay. YouTuber Theo writes, really liking Claude Design so far, except for the fact that it just wiped out my project after burning 10% of my usage. Greg Eisenberg from the Startup Ideas podcast found that for a lot of use cases, it was really good. He gave wireframing a 9 out of 10, mobile app design an 8.5 out of 10, deck research and design a very specific 8.7 out of 10, and video creation a 4.5 out of 10. Effectively, while the video creation feature is novel and might be valuable in some specific contexts, it’s certainly not replacing other video creation tools just yet. (Time 0:20:06)
  • Claude Design Competes More With Agents Than Figma
    • Nathaniel Whittemore says Claude Design resembles Figma for people who never wanted to learn Figma, not a true replacement for existing pro designers.
    • He sees more direct overlap with GenSpark and Manus for code-powered slides and visual artifacts than with full-suite tools like Canva. Transcript: Nathaniel Whittemore Speaking of which, what are the right comparisons for Claude Design? I mentioned Canva before, basically saying that I thought that they were slightly different. I don’t know if Canva would agree with the characterization of them as an asset-first design tool as opposed to Claude Design’s systems-first design, but that certainly reflects At least my usage. And what’s more, Canva is just a much more full-sweeted tool at this point, with a lot of things that I have a hard time imagining that Claude will ever want to build into this. The more proximate comparison, and certainly the one that markets ran to, was with Figma. Now, part of that, of course, is because Mike Krieger left the board of Figma, but I still don’t think it’s exactly accurate. I think it’s playing in some similar spaces, but for me it feels closer to Figma for non-Figma users than a Figma replacement for existing Figma users. I kind of think that it brings a lot of what designers like about Figma into a package that’s for people for whom Figma might be a little bit too much program. Now maybe the market is right, and that’s just a temporary state, and the immediate seven-point drop in Figma’s share price reflects where things go. But for now, like I said, it feels to me like the two audiences that Anthropic is pushing this towards are one, cloud coders, which will of course have some overlap with the people who Currently use Figma, to the extent that those cloud coders don’t have to collaborate with designers in the same way that they used to, and two, non-designer knowledge workers, particularly Marketers. There is some overlap with Google Stitch, although I do think that Stitch is a lot closer to an AI-native Figma experience than is Claude Design. Where I really see the most comparison, at least in terms of direct challenges, are with the general agent tools GenSpark and Manus. Pretty much my main use case for these tools has been code-powered slide and visual design, and that’s pretty much exactly what you can do with Claude Design now. Now, I certainly don’t want to argue that just because my particular use case for GenSpark and Manus now has competition from Claude Design, that A, everyone’s use cases are disrupted, Or B, that those are even the most important use cases for GenSpark or Manus. It just does have enough overlap that it’s worth calling out. (Time 0:21:35)
  • Looser Briefs Sometimes Produced Better Designs
    • Nathaniel Whittemore used Claude Design for mobile wireframes, retrofuturist slides, a teaser video, and a World’s Fair style jobs page.
    • He found looser prompting sometimes produced better design choices, like a webpage-slides toggle and stronger unconstrained slide treatments. Transcript: Nathaniel Whittemore So let’s go back now and actually look at a few of the things that I designed to get a sense of what you can do. Remember, we were designing a mobile app companion to the Web Builder context portfolio. Claude Design organized the mobile app wireframes into four categories. The first was the daily prompt, which it compared to Duolingo for self-knowledge. And you can see that in addition to showing the wireframe, it also has an annotation explaining what it was doing. Although in this bunched up view, that’s a constraint of the video, it gets a little squeezed. You can still see that it’s trying to provide you the context you need to give it better feedback to do the next version better. It also had three other sections, a voice scratchpad, a portfolio glance, and the ambient things like notifications. A couple other things that I built that I wanted to point out. The first is that I gave it GPT 5.4 and Opus 4.7’s analysis of the most recent AI usage pulse survey and how to turn that into a slide presentation that could also be published as a website. It actually gave me three different visual versions and this is the one that I selected. I also felt like I might have overly constrained how it approached this one with the onboarding questions. And so for a second version, I just gave it the 1950s retrofuturism as a design brief and let it rip. And I think what this shows to me is that to the extent that your usage and rate limits allow for it, there is something valuable about being a little less prescriptive, especially if You are open to different directions. As always, with design, better or worse is going to be a little bit in the eye of the beholder, but there are certainly parts of the slides that it did unencumbered by my design prompts That I like and think are better. For one, it has this little toggle up here where you can view the whole thing as a single webpage, or you can view it as slides. I already shared the AI credential overview video and you can see here again another great example of the UX that Nufar was talking about where the tweak section allows you to very quickly Change and view things in a way that feels very different than other AI design tools. And lastly, going back to the Jobs of the Future World’s Fair page, the thing that I would point out on this one is that I think one of the areas in which Claude Design will shine is when you’re Trying non-standard things. (Time 0:23:35)
  • Best First Projects And Prompting Tips For Claude Design
    • Try Claude Design first on slides, web projects, simple SVG-friendly launch videos, brand systems, mobile wireframes, and Shopify-style product pages.
    • Nathaniel Whittemore also advises slowing down for icons, naming, and key details, and explicitly banning generic SaaS defaults like Inter and blue-purple gradients. Transcript: Nathaniel Whittemore So if you want to go experiment with Claude Design, here are seven things that I think you should try. First of all, definitely do a set of slides. And keep in mind that when it comes to doing slides, you don’t necessarily need to hand it the entirety of the information there. You can take advantage of Claude’s reasoning capability and hand it something that you want it to turn into slides, as I did with Claude and ChatGPT’s assessment of this monthly survey Data. I would also definitely try a web project. I think this is going to be one of the core use cases is basically Claude design complementing Claude code for the copious number of websites and web apps we all build now. I think that you should try a simple launch video. I would think of something that’s well suited to the SVG word heavy kind of design that this is going to excel in as opposed to something that needs native image generation in a different Way. But since this is a pretty new capability, it’s one that I think that people might find themselves attracted to just because it’s not replacing something. It’s a net new opportunity for them. I think that you should try to develop brand assets. Although my guess is that this is going to be one area where claw design might struggle in terms of getting you what you need. Although I could be wrong about that. Where an area that I think it will do better is around an overall design system. For this, what I would suggest is take some slide deck or presentation of yours that you like, or the best part of your company’s website, and then either ask it to improve that and turn It into a design system that’s transportable. Or if you’re not confident that it’s exactly where you need it to be, ask it to build on what you like. Maybe even give it some example of some other brand that you like and see how it integrates the two. Finally, for some more advanced and bonus usage, try wireframing some mobile app, even if it’s just one that you’re thinking about just for demo purposes and from your personal life. The reason to do this is to see Claude Design’s product thinking in action. And finally, if you have any context for it, I certainly think that trying to build a Shopify or product site could be really valuable as well. Now, it’s only been a few days since this has been live, and so, so far there aren’t all that many tips and tricks. Two that stood out to me as I was reviewing, Ryan Mather writes, know when to slow down and do things by hand. He writes, new icons, spot illustrations, naming, some details will always make an outsized impact. It can be easy to get sucked into the hyperspeed of agentic design. Knowing when to slow down is an art form of its own. I think this is a great point, and one of the ways to invest the time that you are saving with Claude Design is to be able to focus on those handful of details that really matter. Another one again from the SmartApe. They say the default aesthetic is generic sass, ban it explicitly. They write, without constraints, Claude Design defaults to inter, roboto, aerial, and predictable gradients. It’s the YC batch aesthetic. To get anything distinctive, you have to ban it in the prompt. No inter, no generic gradients, no stock blue to purple. This isn’t in any documentation. It’s the first tip every serious user discovers. (Time 0:25:48)