Anthropic has been churning out new AI products at an astonishing rate, and its latest–Claude Design–is perhaps its most impressive. It’s shaping up to be impressively expensive, too.
Released just today as a research preview, Claude Design can create web page prototypes, app wireframes, pitch decks, marketing materials, and other digital products from scratch, using nothing but a prompt. You can then export the results into a variety of formats (such as ZIP, PPTX, HTML, and PDF files) or hand them off to Canva or Claude Code.
Powered by the Claude model of your choice–Anthropic’s latest, Opus 4.7, is recommended–Claude Design might seem like yet another AI tool designed to replace humans, but my guess is that its true potential could only be unlocked by a human.
Sure, you can use Claude Design the way I did–just plug in a simple prompt and see what happens–but you’ll probably get the best and most distinctive results by feeding it an existing codebase and design assets. That means if you have an existing online business, you can share your site’s code with Claude Design as well as some representative graphical files, allowing Claude to get a sense of the look and feel you’re aiming for.
To make the most of your Claude Design spend, you’d also want to give it as many specifics as possible about the result you want–and again, this is where a human designer’s eye would come in handy. The more precise you can be with Claude Design at the beginning, the fewer iterations you’ll have to go through at the end.
Anyway, back to my experience with Claude Design–and I’ll tell you right now, I’m no designer.
Starting off with Claude Design (which, during its preview period, you access on the web), you’re presented with a tabbed chatbox for submitting an initial prompt. The tabs include options for a new prototype and a slide deck, as well as the ability to start from (or create) a template. You can also add your company name, link a GitHub repo, connect a local folder, or upload design assets like fonts and logos.
I took the barebones approach, starting from zero and submitting a simple prompt: “Let’s create an interactive graphic that explains AI tokens to everyday users.”
Claude chewed on that for a bit that then peppered me with a battery of multiple-choice questions: Who’s the target audience for this graphic? (Total beginners, curious but non-technical adults, and students were among the answers I could pick.) What format might be best? What kind of interactions did I want? What should the overall style be (New York Times-style editorial, cartoonish)? How expansive should the graphic be?
Answering those questions took about a minute, and then Claude Design sketched out its overall approach — a clean editorial explainer (“NYT/Pudding feel, serif headlines, generous whitespace, one accent color”), devised a list of five project milestones, then dove right in.
Claude may have some follow-up questions for you after submitting a Claude Design prompt.
Ben Patterson/Foundry
My Claude conversation moved into a left-hand sidebar as the actual work began, while the right side of the webpage turned into a large canvas, where I could see Claude’s work in progress. Tabs at the top of the canvas let me view various versions of the project (Claude Design can create multiple variations at once) as well as browse the actual project files.
Within five minutes, Claude Design had created a draft for me–a handsome-looking webpage that takes you through AI tokens step-by-step, including interactive sections where you can type in words and watch the token count in real time. The copy was clear, friendly, snappy, and–as far as I could tell–accurate. It was pretty much just what I wanted, all in the first shot.
Overall, it took Claude Design roughly 25 minutes to complete three variations of its AI token-explainer prototype. Pretty impressive.
But then I saw my Claude usage meter. I’m on the Claude Pro plan–which, to be fair, is best suited for individual and everyday users like me, not a business looking to prototype an app or a website–and quickly realized that I’d already blown through 80 percent of my weekly Claude Design allowance. (For now, Claude Design usage appears to be counted separately from your overall weekly Claude quota.)
Claude created this cool interactive element to go along with my token-explainer webpage.
Ben Patterson/Foundry
Then came a misunderstanding plus a big mistake on my part. Trying to click from one prototype variation to another and hitting a plain-text “preview token required” warning (turns out Claude hadn’t hard-coded the variations into a single HTML file), I mistook the “undo” button for the back button, instantly zapping all Claude’s work. “The undo wiped everything. I need to rebuild all files from scratch.” Claude calmly told me.
Eyeing my nearly empty Claude usage gauge, I downshifted the model to the cheaper Sonnet 4.6 as Claude cleaned up my mess. Five minutes later though, my weekly Claude Design meter hit zero.
Luckily, Anthropic recently handed out overage credits to Claude users (following the company’s decision to block Claude subscription users from using OpenClaw and other third-party AI agents without using the Claude API), which allowed me to top-off the tank and finish the designs. But even had I not blundered into wiping my Claude Design files, I wouldn’t have made it far into the review process before my quota ran out.
The finished product looked impressively polished given that I’m not a designer and gave Claude Design only the barest of directions. But yes, we’re talking another token-hungry Claude product here, one that Pro users in particular will barely be able to use before burning through their usage limits.
