Did Anthropic just kill Figma?

I was excited to stop talking about Anthropic. I really was. But they dropped something new. Claude design. This is actually a really exciting product launch for me because designing good user interfaces with these models is possible, but it takes a lot of effort and massaging. And from what I've seen with this release, people are actually really hyped about it. I've done a couple of videos in the past about how to use models to make good designs for your applications. And historically, I've had the best luck using the cloud models. This is actually the thing that I've kept my cloud code subscription open for because I like using it for UI. All of my recent projects like lawn and shoe have used largely opus models for doing the design. I've pulled in Gemini a little bit here and there, but the majority of the design work I've been doing lately is all with Opus. And I'm very excited by the idea of a new interface or harness that should make the model better at design. It's no secret that the design skill that Anthropic put out seems to make the models way better at designing UIs. And since that's just markdown, it's crazy it has the impact it does. Just adding a few paragraphs of context makes the models better at design. So what happens when you build a whole product around it? I don't know yet because I haven't tried it yet. And I want to give my honest reaction to trying out Cloud Design for the first time. I know I've been a little harsh on Enthropic lately, putting it lightly, but I am really hyped for this and I'm going to go in with an open mind and hopefully this is a reason for me to keep my subscription. But if I'm going to keep paying for all of these subscriptions, I need to make some money. So we're going to do a quick break for today's sponsor. AI getting good has made software a lot easier to write, but it's also made software with problems much more prolific. Two of the problems I see the most are incorrectly set up billing so users can scam you, get things for free, cancel, abuse trials, all sorts of stuff like that. We're fighting it constantly and also getting secure authentication set up in a way that models understand and know how to deal with. It's way harder than I ever would have expected. Spent a lot of time trying to figure out how to make my agents do off right. And I keep coming back to the same solution. Clerk. These guys get off and they get it in a way that is super easy for devs. From day one, Clerk's focus was always make things secure as simply as possible for developers. Everything you need to do in Clerk is done in code. And the code couldn't be simpler. You copy paste the proxy in your next app. You add the clerk provider and you're done. You can then just add the sign-in button and the sign up button. You can use their show component to have different conditions, different flags, different things like signed in or out, as well as your own custom fields you can add to the users. If you're too lazy to even do that, the docs now default to the agent quick start. You copy the prompt, you go to your app, you paste the prompt, and then you're done. And adding billing is even easier. It's so cool they integrated this. Has ever been easier to set up subscriptions in your app than it is now with Clerk. Subscriptions are on the user level anyway, so it just kind of makes sense combining them. As soon as I saw this, I realized, oh yeah, that is how it should work. Remember that show component? If you have features you only want to show users if they have a certain plan enabled, it never has been easier to do that. Do you know how obnoxious it is setting things up like this normally? Do you know how much easier things are going to be for your agents and most importantly how much more secure they will be with these patterns? Because it works on the server as well. Use O your agents will love at soyv.link/clerk. Let's get straight to it. I'll read the article in a bit, but I want to just dive in. So, let's click the design button. I switched over to Helium because I don't necessarily trust a Firefox-based browser for something like this. So, we're going to go with a Chromium based one for now. You can import your team's design system. Make prototypes. Unblock engineering. Import codebase from a folder on the computer. H export to clawed code. Huh. Oh, that was a very interesting bit there. The the prompt was what are six ways we can do this? That's one of the things I often recommend when you use the models is not just having it do a generation and then retrying if you don't like it. Ask it to do multiple and tell it to be varied because if you regen with the same prompt over and over, you'll get similar outputs. But if you tell it to make a varied set of things, the difference between them will be bigger. You can do prototypes, slide decks, from template or other. Interesting. You can also set up your design system. Let's do the tutorial. Think design, engineering, product is fair. You guys mostly know me for sales and marketing, but these are the things that probably fit for doing this. Did my email leak at some point? Where? Oh, that's great. Yeah, thanks Anthropic for just prominently showing my email phase censor that out. There's nothing I can do about that because Anthropic doesn't understand people screen share sometimes. The [ __ ] Anthropic. Can I just like make a Chrome extension that hides my email address or something? Fixed. I love that it broke the layout of the page. This extension is not good. But uh yeah, now whenever my email gets leaked, it's going to be replaced with [email protected], which should hopefully let you know how I feel. Let's learn about this new thing. Continue. Product mockups, landing pages, interactive prototypes. It's really interesting is you can import your design system, but it doesn't seem like it will use your codebase. It's more for creating a first pass. Really is going after like the Figmas of the world, which uh when you see the stock for Figma and Adobe today makes sense. Figma's already been getting crushed. As somebody who bought a lot of Figma stock during the IPO, yeah, they're down 85% since IPO, which is insane. Like actually absurd. I feel bad for anyone who stayed at Figma waiting for the IPO and then got screwed because they couldn't sell until after. Tip one, mock it up. Design mock start with quick wireframe explorations before committing to high fidelity. Say what the screen does. Claude sketches options. Bring your context. Drop images or paste screenshots with command V. Mount a local folder from the import menu so Claude reads your codebase live. Point at what to change. Click comment in the toolbar. Click any element and annotate it. Leave several. They batch into one message when you send. Tweak values live. Knobs mode let you drag CSS values. Size color spacing directly into the preview. Then prompt Claude to apply your adjustments. Interesting. Hand off to uh hand off to Claude Code. Packages specs and structure into a dev ready folder. Download it. Then tell Claude Code create this design. You can save it as a template as well. Cool. So I know what we have to do. The T3 code site is a real website that exists, but it's far from great. So, let's fix that. Let's do our first prototype. We'll do a wireframe create. We'll attach a codebase. Let's pull out whisper for this one. They have a speech button here, too, but uh I don't trust that. I would like to redesign this page. The marketing site should showcase just how great T3 code is for any developer interested in trying it out. Some of the things I want to highlight are in this list below. One, we should showcase that all existing harnesses are usable and you can bring your own subscription. So if you're using cloud code, codeex, open code, or cursor, you're covered already with T3 code. Two, it's fully open source and really easy to fork as well. If you don't like something about T3 Code, feel free to fork and redo it however you want. Three, we are obsessed with performance and quality experiences. We are the best coding app for doing agentic orchestration and managing all of the agents that you're working with for your projects. Four, great workflows for integrating with Git and GitHub. A one-click button to file a PR after you make your changes. Five, designed for parallel work. If you're working on multiple things at one time or multiple projects at one time, T3 Code is here to make sure that all happens. I love that it successfully capitalized Cloud Code, Codex, Open Code, and Cursor. Whisper Flow is great. This isn't even an ad. I'm just very happy with it. They got the order wrong here. Like this should have been a separate point. I have to change these. Doesn't matter too much. Make sure the design is dark mode entirely. Our users are sensitive to light mode and bright colors. Let's see how it does. Chat has a lot to say. Anti flashbang gang. It's true. Yeah, we are delicate flowers. I'm happy you guys understand. Doing my best. Sorry that this is such a bright UI. I will say this UI feels very Figma, but it also feels very different from all the other UI I've used from Claude and Enthropic. Like these little tabs are super different from other things I've seen them do before. The idea of a comment section where like I can share this design or this work and others can leave comments and then I can tell the agent go fix or address all of these. That's really cool. There's ideas here that I like and we definitely need a new interface for doing collaborative design with AI. I conservatively I am hopeful so far. I do feel bad for our friends over at Tailwind though because they just started working on UI.sh which is a way to do a similar thing from your terminal. Haven't had a chance to play with it yet. I don't even know if I'm in. I know that Julia says, but I'm not. I'm very excited to try this. Let's look through what it's done so far. Design system and plan. It is grabbing things from the existing marketing site. The fonts, the colors, muted text, borders, subtle noise overlays as an accent color, a single signal color using blue from the app. That's fine with me. mood, dark, minimal, high contrast, dev tool feeling, no gradients, no emojis, hairline, borders, and mono eyebrows a structure. And then it has the sections it's going to make. Now, it's splitting this into logical files, making it icons, JSX, and style CSS. So, it is writing code. It's just not writing the code that you're supposed to integrate. So, again, to be very clear, the point of this product isn't to use it on your codebase. It is to mock things around your codebase. And if you haven't worked at big companies with crazy design systems, you might not know how far they will go. But when I was at Twitch, a lot of work was put into keeping the system that they had in Figma in sync with the system we had in the actual codebase when working on the site. Trying to make sure that the designers had the exact same components and options that they were designing with so that we could mostly replicate it in code without having to design new things. We could just use the component library that we built at Twitch and the UIs that we were given from Figma mapped really well to that component library. Something that has historically been challenging is taking those designs from Figma and getting an agent to implement them correctly. There have been a lot of attempts, but if the design is not properly using your design system and your component libraries and the relationship between that and your actual codebase isn't clear enough, agents would just struggle. So, I'm very curious to see how this goes. Both how good the design comes out, but also when I try handing it off to cloud code, how it does there, too. definitely taking its time expected, but also I've worked with designers and it takes even longer. I'll talk a little more about why I'm excited about this. I've talked in the past about a designer I really loved working with and I've always felt that the best designers are the ones that have the best relationships with all other sides of the stack. I'm sure we've all thought of the full stack spectrum where you have backend on one side and you have front end on the other. But the applications we build go way further. On one hand, there's the infrastructure, there's the chip design. There's a lot of stuff that goes further that direction, but it goes the other way, too. So, let's just delete backend because we're not here to talk about that part. And we'll pretend front end is where this is cut off. And on the other end here, you have your user. But the front-end developers aren't always just building the thing directly for the user, much less talking to the user about it. Usually, there are other steps between. You have your designer somewhere between the front end and the user. And you'll often at big companies also have a product manager of some form. You might even have a support person that goes between the PM and the user as well. But for now, I'm just going to make this as similar to my experience as possible at Twitch where the users would have issues. The product managers would identify those issues, go through all the reporting and the other things that users are experiencing, reach out to a handful of them to talk with them, and once they have collected their proposal or their user stories, they will bring that back to the designer and work back and forth to create designs for what this product could look like if they were to solve the problems the user's having or build something the user wants. Once they've done their back and forth, it goes to the front-end developer often. And the front-end developer and designer will go back and forth about what is and isn't implementable, what does and doesn't work, all of that stuff. But if I've learned anything throughout my career, it's that the more you can bridge the gaps between these areas, the better off you are. So, as a front-end developer, it's really, really beneficial if you can do a good job talking to design and also talking to backend. One of the things that elevated my career the most when I was at Twitch was that I was one of the few front-end engineers when I moved over to front end that still cared deeply about the backend and put a lot of time into thinking through the APIs, interfaces, and working with the backend team. I remember when I was talking with the guy who at the time ran the backend API, so the entire GraphQL layer that powers Twitch, and how disappointed he was that so few front-end developers were talking to him and how shocked he was that I had reached out to do a call and catch up with him because he just thought the front-end devs didn't care at all because for the most part, they didn't. And one of the reasons I got to have so much impact at Twitch is because from this position, I spent a lot of time helping and working with backend and also a lot of time trying to bridge the gap with design. It even went as far as me working directly with product managers a lot too and eventually I was starting to talk to users directly to try and fix their stuff. This is a T-shaped thing to be clear. So if you're really deep on front end, it's very valuable to also be a decent bit deep on back end, a little bit deep on design, be involved in product management even if not quite as much and still spend some time talking to users. So, if each of these bars is how much time I was spending on the things and talking to the people in the space, I think this is a relatively good representation of what my experience at Twitch was like for my second half. I'm bringing all of this up because of the designers that have been awesome that I worked with. One designer in particular I mention all the time. Her name is Iris. She was the designer that kind of inadvertently made me become a more front-end person because at the time I was backend but just dealing with front end things. We get the new Twitch app out. Iris was so deep in everything she touched. She didn't have development experience. She wasn't a developer at all. So when she gave me some designs that were going to be really annoying to implement, I pushed back really hard. The problem was that the designs she gave us used rounded borders and pop outs heavily. So we had a card like this where the corners were rounded, but we also had a button in here that would pop out when you covered over it. And this was in a time before things like popper were reliable. So the problem here is if this was a hover button and this would appear outside and you have an overflow rule on in order to round these corners with a background that this would break because that overflow rule would keep this from being able to break out vertically. Obnoxious. And this is very hard to explain to non-devs. I'm sure some of the devs here don't even understand what I'm talking about here. Overflow rules and CSS suck hard. I I have so many fun facts about overflow rules. I won't bore you guys with the details there because we're not here for that. We're here for cloud design, but this is a problem that's very deep in the technical details. And Iris had a bunch of questions. I was a little annoyed because Iris's job was to take products things and then make a design and then fix it when I tell her what she can't do. So, I entertained her for a bit, was a little frustrated, but was like whatever. She then came back a few hours later with designs that fully addressed the concerns I had. She added more layers with more curved corners, but didn't use background colors. So, I no longer had to turn on the overflow rules. And then I could break out the pop out stuff again. And I was blown away when I realized she perfected the art of asking the right questions to make the design meet any set of needs across different people. I was blown away because she wasn't just doing that with the developers. She was doing this with the product people. She was doing this with the users. She just wanted to solve the problems that were given to her. Similar to what we do as engineers. If she could collect the problems and then understand the problems, she could solve them. And when I realized design is similar to what I did on front end or even backend, where a backend developer's job for most places is connecting the database and the data representations with the applications users use. She was the bridge in a similar way between what I had to build and what the users needed. And this rewired my brain on a fundamental level. And I'm so thankful for Iris for taking the time to indirectly teach me this by just being really [ __ ] good at her job. Iris was unbelievable. About a year or so later, she asks me another random CSS question out of the blue. I'm like, "Wait, what? Where does this matter? We don't even use this style of thing at Twitch." And this was like over Slack. And all of a sudden, I hear footsteps. She had gotten up, walked over to me, and handed me her laptop. And was like, I really want to build a prototype for this thing so that we can show it to users before we finalize the designs and make it your problem. I normally wouldn't bug you with this, but I just want to make sure the prototype works well. And she showed me a mocked up prototype she had built herself. Reminder, no code experience. And what she showed me was the first prototype of what ultimately became Mod View, which is still my favorite project I've ever built. She made a functioning prototype, not like it worked with real back end, but all of the elements were resizable. It had like the resizing stuff like this, the ability to move stuff around. And she built it herself in vanilla HTML, CSS, and jQuery. Unbelievable. This was way before AI and vibe coding [ __ ] This was like 2019, maybe 2018. Unbelievably cool. So much respect for her for that. And this is why I am excited about what Anthropic is cooking here. If you give a motivated person like her the tools they need to make something useful and playable, like something they can test out themselves, show to users, show to developers, and they're in between role between the user and me as the programmer can be done in a more collaborative and flexible way. That's magical. And I know that if they get this right, people like Iris are going to just take over the world with it. So, let's hope it did a good job. So, we have the T3 logo, but we still have T3 code the text there. How are we doing this right now? We're not. Is there a way I can full screen this preview somehow? Aha. Orchestrate every coding agent from one comm surface. I don't like the word wrap here with the underline, especially here where the underline is like hitting that letter. Either code is a control plane for agentic development. Bring the harness you already pay for. Cloud code, codec, open code, cursor, and run them in parallel across every project you own. download link. See how it works. Don't love this. It grabbed my screen. Did it make a new screenshot? This is not what it looks like. Oh, it made an interactive fake UI here. Interesting. Then this continue working on hero animation. Dot dot dot function you already pay for. T3 code doesn't resell tokens. Plug in cloud code, codeex, open code, or cursor with the credentials you already have. We orchestrate them. You keep your plan. Did not grab the right logos for any of these, which is funny. They made a new fake UI here. Not accurate. Bad word wrap here. This is workable, but we got work to do for sure. I like that herog grid is a toggle and I can just turn it off there. Oh, the preview tilt has a little animation when you turn on off. That's cute. I do really like the tilt for that type of thing. Okay, let's do some feedback. This word wraps on medium screen sizes. Oh, there's a comment button separate from the send to claude. Most of these header items aren't needed. Leave the comment. We don't emphasize the open source nature. Well, here also the wrapping with the underlines looks pretty bad. Oh, the tweaks is part of the page. You can comment on the tweaks panel. That's hilarious. I turn off and on. That's convenient. There's little things, but like considering how broken the experience I had with the Claw desktop app was, this is significantly more polished already. You could even draw on things. Oh, that'll be very useful for for us Excaladraw types. Trim this copy down. We should get logos for the supported harnesses here. This bar is pretty cringe. We need a better scroll hint for sure. We need a better hint that the user can keep scrolling. Where's the mock? Oh, did it get rid of the mock UI? I think it did at some point. Okay, we have all these comments. I can check them all and then hit send. That's cool. I don't know. I check all of them, but yeah, let's do that. It's not updating the UI here. Do I have to hit that to refresh it? There we go. Cool. Yeah, that fixed that. This word wrap's still broken, but I also told it to trim all that out. Refresh. Will it? It still has that there. I hate the word wrap it does there though. The scroll hint doesn't fit the page size well for this size at least. Yeah, it still is a AI doing design things here that I don't love. Top bar is a mess. The wrapping in general sucks. The scroll hint isn't visible on any screen size I've tried. Give this another shot. You know what I'm going to do? I'm gonna try spinning up a second one. Marketing site V2. Another fresh screenshot. Do I just ask it to make multiple if I want multiple? This is a questionable menu. Why would anybody want Opus 3 for this? Are you joking? Why would that even be here? Make four different versions so I can compare them and pick my favorite. I didn't give it the code base this time, which I think is the right call. We'll see how it does without it. Um, this one seems to have frozen. I sent the message. It's not updating. Uh, I was liking this. I go to the other ones. File not found. H. I haven't used much claude today. Almost none today, actually. And the current session, I'm on the 20x plan still. And for my current session, I've used 0% total. I was at I think 6% for the week, and now I'm at seven. Oh, cloud design is its own separate thing here. And I'm at 18% used for that. Oh, great. And it resets Friday at 5:00 p.m. It's currently Friday at 6 p.m. So, this is going to get burned through fast. And I don't know if this one's going or not. This is just dead. Oh, it did fix it. It just didn't show me it fixing it. I'm back on the other one. I told it to build multiple. And I can tell here it's using the frontend design skill, the same markdown file, because these are all of the designs that it uses for that. So, the V1 is the editorial refined version. Oh boy, let me rework all of this response. The response tag leaking into the output. The scroll hint is just always visible now. Great. I hate that. There's a lot of little things that make me think Anthropic's like the only company to do this. Well, one of those things is that the new model can do much higher resolution images. I can't tell you how many times I pasted an image to even like a OpenAI model like GBT 5.4. 4 and it said it couldn't use it because the text was too small and it couldn't read the text. Opus 47 can handle much higher resolutions. So this is a much less problematic problem which is very good. Very thankful for that. And I think that puts it in a really good spot here because it can screenshot itself and know what is where. It can get more context from the screenshots I give it. And it can be more precise in general if it can see higher resolution. Here's where it left us. Don't love it. Let's let's do some more comments. I do like the fact that I can just leave comments in the UI and then have it do it all at once. The logo next to this makes it read like T3 T3 code. We should either drop the second T3 or remove the text entirely. Still in comment mode. Um I hate this bar. Just get rid of it. Users can scroll. Fine. No more status pills like this ever. Please, can you get the real logos? They should be available in the code. No worries if not. This is awful. Just get rid of it. Too much scrolling in this area. Hard to know. There is more to see. Trim bottom. So, the next section is peeking if you can. This whole section is way too big. These are cringe as hell. And if you wonder what I look like actually giving feedback on designs, it's not far from this. We don't need to number these sections. I would really like to have the logos floating around here. The logos of the other harnesses floating around here. It's a nice UI touch. We should have a screenshot of the UI pretty high up in the page so users know what this actually is. Cool. Select all these comments. I do actually really like this workflow. I don't like that I have to go manually check all of the comments to send them. Oh, it didn't it didn't clear the comments that I left before. That's actually very annoying. Whatever. I'll send the ones I just wrote. Let's see how this other one is going. So far, we have our design files. The editorial version. Oh god, I hate this. It [ __ ] up the font for the links real bad. It made a mock UI that isn't accurate. I got the terminal version. Slightly less bad, but still far from my favorite thing. That is not how the install works. I get it because it doesn't have access to the codebase like the other one did, but still not great. Still very amused by these response tags. I think that this was worked on separately from Opus 4.7, so the Opus 4.7 support isn't quite perfect yet. How's my usage doing? We're at 31% used, and I still haven't gotten a usable design out of it yet. Oh boy. Let's see if we can get a good design out before I hit those limits. Fingers crossed. I wonder if the limits are different based on different tiers, cuz this is a pretty brutal limit if not. Apparently, people who are on pro only could do two prompts before it crashed out for them and they ran out of usage. So, I'm guessing that this is not the same levels for everybody. For those who don't know, pro is the $20 a month if I recall, and max 20x, which is what I'm on as the 200 a month. So, yeah. Okay. Finish V3. This is better than the others so far. I actually like this one. Don't like the Machui that it made, but this is a direction I could work with. Yeah, that Apple logo. It hallucinated an Apple SVG there. That's one of the things that like it can't be too hard to do. Like I'm surprised they didn't add a tool for like getting brand logos to use. There's so much good [ __ ] for this. Like SVGL, which I use all of the time. It has made my life significantly better. SVGL lets you oneclick copy any SVG or download it for almost all of the things that we talk about and use. You need an Apple SVG, there you go. Need a T3 SVG, there you go. Need a claw SVG, need a Codex SVG, need an open code SVG, cursor, almost everything you'd need. SVGL is a lifesaver and SVGL has a raycast extension. So, I can just type claude, press enter, and now it's on my clipboard. Life-changing for people who use logos a lot. One more really useful thing I want to show you guys is today's sponsor. Agents have gotten really good at building apps. As long as they're web apps, I cannot tell you how much I've struggled trying to get those exact same agents to build mobile apps. it seems like they just don't understand mobile at all. Which is why having experts who do is so helpful. You might not need them all of the time, but having one or two or maybe four or five people that can come in and help when you need it to make sure you're going in the right direction with mobile has never felt more essential. And that's why I love Infinite Red so much. The whole Infinite Red team are industry experts in React Native and they can set you up right. Whether it's building your app from scratch, getting your old app back on track, or just making sure everything's set up so that you, your team, and your agents can build mobile apps successfully. That's why everybody from Zoom to Amazon to Starbucks has worked with them on their apps. While I really do love React Native, it's a little too easy to get it wrong. And that's why you'd want an expert to help and make sure you do it right. According to Max from Gas Buddy, there is no substitute for experience in getting the idiom of React Native and avoiding the numerous dark alleys around state management, performance tuning, and third party module selection. Our team at Infinite Red has provided this throughout our project. Your users deserve good apps. Make sure you do it right at soyv.link/infinite. This was built by Farrell B, who is another wonderful dev. Definitely give him a follow on Twitter if you haven't. He is great and I'm so thankful he made SVGL. This is one of those tools that just makes my life better. Interesting. It made an index html to tell me about the different designs it made. Preview token required. What? Okay, there's a lot of broken things here. That's not the right folder. What happened here? Refresh. File not found. Did it all just break on me? If I refresh, is it gone forever? I think they just vanished. I just gave up over 10% of my usage and the designs disappeared. Where did the files go? The files appear to be gone. Well, only the upload folder remains. Let me rebuild them. This is particularly a shame because it's clear way more effort went into this product than a lot of other recent anthropic launches. This feels more polished than I expect from them a lot, but it still has the same anthropicisms that I expect. It's just I I genuinely wish they had they would put a little more polish into these things cuz this is brutal. Now I'm like scared to click the design files button because I might just have this disappear entirely. Like that. That's the problem with these bugs is they make you fear the product instead of like the product. I went from, oh, this is a useful thing I'll use for prototypes to, if I'm not monitoring this closely, what I'm working on might disappear entirely. If I wasn't filming, there would have been no proof that I had one design I kind of liked from that. It just would vanish into the ether. NMG just said, "I'm back and Theo has his head in his hand." It must be bad, isn't it? There there's a lot of good here and it's much more polished than usual, but it's also more broken than I hoped. You know what I'm going to do? There's an assets folder. Let's grab the logos. Download the SVG. Open code. That's hilariously poorly rendered. And cursor. Like this is nice. The fact that I can drag and drop SVGs into this and it knows about them now. It puts them in the message as well, which I don't love. But let's try this. Here's the SVGs for all of the logos that should be floating. Use these without word marks. Make it pretty and friendly. logos should be much larger than they are now. As you see here, the logos are just kind of weirdly thrown around and they're a little small. See how this does. And I'm going to go back over here and say, bring them back, please. See how it does. Where's my usage now? 47. I'm I'm almost at 50%. Jesus Christ. Okay, now we're talking. Now we're making real progress. I'll do this so you guys can see it better. Oh no, when I resize, they disappeared. Preset in tab. Can I do that? Yeah, this is getting a lot closer to what I had in mind. Now we're really getting there. Actually, I would be comfortable starting with this and I don't want to run out of usage before I try the export feature. So, let's try it out because I can do the rest of the iteration with the rest of my cloud code subs since this one is uh over 50% for cloud design already. So, let's hit that export button. Hand off to cloud code. Copy command. I'll over to my terminal. Grab my custom cloud command. Paste. I'm not I should tell it which directory to go to, but I'm not going to. I'm going to just see how it does. I'm also curious where it fetches these files if it like throws them in the project directory or puts them somewhere else. Okay, it's fetching it. Now, let me explore the real marketing codebase in this repo. Cool. So, it found that it needs to go to that directory. It's making it's doing things right. this full stack experience of do this copy command drop it here and it goes like if it gets this accurately I'm going to be pretty pumped. Let's see how this one's doing with the restoring of my designs. Still going in my usage isn't moving. Hopefully it will do the right thing here. Useful info. Freeze gum with an ice cube for 2 minutes. It peels right off shoes. What? If you were wondering if this section was useful, I think they just confirmed it's not. Okay, people in chat are saying it is kind of useful, but now I'll rewrite the layout.astro with the new design system, globals, and nav. I'll skip the tweaks panel, scroll hint, and numbered eyebrows. User explicitly rejected those. Interesting that it kept that in the HTML. It seems like the HTML file it sent has other things in it rather than just the HTML itself. Let's take a look at this and see project. We got the index html. Not much here. Tether code marketing is the useful one. We also have the style CSS. It looks like it's doing traditional CSS rather than something like Tailwind, but it also inlines a bunch of CSS on the page, like a shitload of it. Very traditional HTML CSS. It's not using any libraries or anything. That's a good base point so that it doesn't get too tied to your system. But you can also bring your own design system if you want. It has all these JSX components in it. What are these uploads? The pasted images I have. Cool scraps. Then has the assets I dropped. It's cool that it puts this all in like one thing as well as the chat history that it can use to see what my intent was. A little long and this might bloat context up a bunch, but it's a good idea. I I like the idea of instead of like building a super fancy bridge between the tools that's proprietary that the solution is literally just zip all the context and throw it at the agent with a link. I I like that. I do genuinely like this way of like handing around information. But we're 7 minutes into this design and I'm scared for my limits. How are we doing on my Claude usage? Okay, Claude's only up to 2% for my current session, but Claude designs at 55% for the week. Oh boy, it's still building the new pages. We got V2 back. V3 still has to be remade, but we got V2 back. It just said all four variants are back. Actually, looks like they are. That's on if they vanish again. I'm going to keep this tab open because I'm scared. I'm still verifying and I'm a little scared about what this means. That's a cute little laptop animation, but it's not enough context on what's going on. Thinker design's done. Check out the site. That's pretty damn good. I can absolutely work with this from here. I actually think I like this product. This is absolutely workable. I'm probably going to finish this off stream. I'm going to trim up a bunch of stuff, swap some logos here and there, make things a little tighter, closer, kill sections I don't need, make the mocks a little more accurate. But god damn, this is a good starting point. Oh, you can comment on the tweaks. That's cool. We had a tweak for the logo sizes, placements, and amount of movement. I knew you could comment on it, but I didn't know you could use the comments to actually get it to do things. My designs didn't disappear this time, too, which is good. I I'm warming up to this. I like this product. There's There is value here, for sure. Oh, if you make the tweaks big, it uh makes it harder to see because it's now too big, which is funny. If I preset full. Oh yeah, I lose the that and the defaults it picked break the UI a bit. Logo size spread and they don't work either. Okay, so you can get it to modify the tweaks, but you can't get the tweaks to modify the site. I tried. I don't want to keep pushing the limits because I don't want to sour on this. I like the fact that for the first time in a minute, I have a thing I like out of anthropic because this is good. This is the best software Anthropics ever shipped. Not that that's a high bar, but it is a challenge for them. So yeah, whatever engineers made this happen, hats off to you guys. See if you can get a little more control over the [ __ ] show that is engineering at Anthropic because clearly you guys care and want to make something useful and good. And if the same attitude existed across the company a bit more, I would probably crash out a lot less. I genuinely see myself using this at least until I hit the usage limit because I'm already at 65% just from these demos. Yeah, if I was Figma, I'd be scared as [ __ ] right now cuz this is actually very useful already. With a little bit of polish, a couple more features, a little more designer focus, I could see this becoming a legitimately very good product. Will it be as big as Cloud Code? probably not, but I don't think it would be that far either. Especially if they really win in that like presentation design world, too. If this becomes the thing that gets your marketing head, your designers, and everybody else at your company to be interested in what you have, what you do on your quad sub for the company. This just made those enterprise plans much more enticing for a lot of companies. It is definitely a little expensive seeing the amount of usage I used here. I don't know what that maps to in cost, but I see a shitload of potential, and I'm genuinely excited to see where this product goes. It's nice seeing Anthropic building a product that's actually different, useful, and takes advantage of their strengths. Their design taste has always been good, both in the models and what they generate, but also as a company and the things they design. I've always been a little jealous of the aesthetic of the stuff they put out, and now I feel like I can get a little closer to it myself. This design is actually good enough for me to work on further and try to finish to make a better landing page for T3 Code. And I'm very thankful they put out a product like this. This is cool and exciting. And even though I hit a lot of hiccups and I think it's going to be quite expensive, it's nice to see him do something different and something that's actually cool and useful. So once again, shout out to everybody who worked on this. I'm really excited about this feature. I'm going to be using it a bunch and I'm curious if you guys will as well. Let me know what you think. And until next time, keep ringing.