Claude Design: What It Is, How to Use It, and Why Anthropic Built It

Anthropic shipped Claude Design on April 17, 2026, and here's what's actually new about it: it starts from your existing codebase and exports straight to other tools, instead of trying to be the place your files live.^[1] You onboard with a one-time upload of your front-end code, not a live GitHub link. You finish by exporting to Canva, to PowerPoint, to a Claude Code handoff bundle, to a downloadable zip. The product is engineered as an on-ramp, not a destination.

That asymmetry is the whole story. Every AI design tool in the last three years has tried to own the full loop, blank canvas to final asset, inside one walled tool.^[13]Claude Design doesn't. It ships without a shared multiplayer canvas, without documented project persistence, and, the detail that lit up the launch discussion on Hacker News (Y Combinator's tech-news forum), without a Figma export button.^[10]Anthropic didn't forget. They chose not to ship one.

The product lives at claude.ai/design, gated to paying Claude Pro, Max, Team, and Enterprise subscribers, off by default for Enterprise until an admin enables it.^[1]It's powered by Claude Opus 4.7, the frontier model Anthropic released the day before, which the company describes as “the best model in the world for building dashboards and data-rich interfaces.”^[3]I'll walk through what it actually does, how to use it, why Anthropic built it this way, and why Figma's stock (ticker FIG on the New York Stock Exchange) closed down 6.89% the day it shipped.

What Claude Design Actually Is

Claude Design is the first product Anthropic has shipped that isn't just a model. Everything the company has launched before, from Claude Sonnet to Claude Opus to the cloud-hosted Claude Code routines released three days earlier, either improved the underlying model or wrapped it in an agent harness.^[12] Claude Design is a separate product surface at a separate URL, with its own usage quota, its own canvas interface, and its own onboarding flow.

What it produces, by Anthropic's own list, is polished visual work: “designs, prototypes, slides, one-pagers, and more.”^[1]Under the hood, the outputs are HTML, CSS, and JavaScript; the canvas renders live in the browser. One designer writing about it the day after launch put it plainly: “Claude Design, for all its roughness, is at least honest about what it is: HTML and JS all the way down.”^[11]

Design is where AI has been losing for three years. Every code-generating tool defaults to the same Tailwind and shadcn/ui (shadcn/ui is the open-source component library that ships with most v0, Lovable, and Bolt outputs), so the results converge on the same gray-card, left-sidebar, neutral-palette dashboard.^[14]Nielsen Norman Group, the usability-research firm that has tracked AI design tools since 2024, called the category “marginally better” in a 2025 status update and flagged design-system fidelity as the place AI still lost to a human in a good mood.^[13]Opus 4.7 is Anthropic's answer to both problems. It accepts images up to roughly 3.75 megapixels on input, and Anthropic claims design-taste improvements that are finally “choices I'd actually ship.”^[3]Claude Design is the product surface built to put that model in front of users who aren't going to install a model via an application programming interface (API, the developer-facing interface for talking to a model directly).

How to Use It in Five Minutes

Four steps, web interface or command line. The catch is how fast the quota evaporates.

1. Upload your front-end codebase, or skip. During onboarding, Claude reads the codebase you upload plus any design files you add and builds a design system (colors, typography, and components) that every future project inherits automatically.^[2]There is no live GitHub sync at launch; you are giving Claude a snapshot, not a connection. Upload a subdirectory of your front-end rather than a whole monorepo, since larger uploads cause lag and browser issues per Anthropic's own limitations note.^[2]
2. Write a prompt.The hands-on guides converging on Claude Design recommend a four-part pattern: [type] + [subject] + [style] + [key constraints]. Example: “Prototype a mobile meditation app, calming typography, subtle nature-inspired colors, clean layout.”^[9] You can also feed it an image sketch, a PDF brief, or point it at a live URL with the web-capture tool to pull elements from an existing site.^[1]
3. Refine with edits, comments, and sliders. The canvas lets you click any element and edit text directly, leave inline comments for Claude to address, or pull on dynamic adjustment knobs that Claude generates for things like spacing, color, and layout.^[2]Expect a few round-trips; PCWorld's tester hit 80% of a weekly Pro allowance after roughly 25 minutes of iteration, and zero a few prompts later.^[8]
4. Export, or hand off. Download as a zip, export to PDF, PPTX (PowerPoint), or standalone HTML, send to Canva for further editing, or package everything into a handoff bundle for Claude Code.^[1,2]Anthropic's exact phrasing for the handoff: “Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction.”^[1] The contents of that bundle are not publicly documented.

The Thing Claude Design Won't Do

Sit with the feature list for a minute and notice what is absent: no persistent multiplayer canvas, no real-time cursors, no documented project versioning or retention policy, no live GitHub sync on the input side, and no export button to Figma, the dominant user-interface design tool for product teams.^[7]Anthropic's announcement mentions “organization-scoped sharing” with “view or edit access” and a group chat with Claude, which is document-style collaboration, not Figma-style co-editing.^[1]

Day-one users noticed immediately. The top Hacker News thread on launch day was a running list of “where's the Figma export button” posts; one commenter called the absence “diabolical.”^[10]The designer Tommaso Nervegna's first take, published the same afternoon, drew the line cleanly.

That distinction is the product. Claude Design is scoped to one segment of the design loop, the part where you start with nothing and need a draft, and every architectural choice reinforces it. Here is the same tool against the rest of the AI design market. The column that matters is the middle one.

Anthropic Has Been Setting This Up for Months

The on-ramp-out posture only makes sense when you stack it against the rest of Anthropic's design-stack moves. It's not a single launch, it's the last step in a deliberate sequence.

Read those five dates as one product, not five. The inbound pipe, Claude Code into Figma, shipped in February. The outbound pipe, Claude Design into Canva, shipped this week. The tool in the middle is now someone else's problem. Canva's chief executive described the transfer as “seamless.”^[5]That isn't competitive language. That's plumbing.

Why Figma Fell 6.89% and What the Market Is Pricing

Figma, the design tool that went public in July 2025 at an initial public offering (IPO, the event where a private company first sells shares to public investors) price of $33 and traded to a first-day close of $115.50, has had a rough nine months.^[7] By April 17, it closed at $18.92, down roughly 85% from its post-IPO peak of $142.92.^[17] The stock had already reset hard before Claude Design was ever a rumor. A 6.89% drop on top of an 85% drawdown is a market already primed to punish anything that looks like an AI-native threat.

Is Claude Design the reason Figma fell 6.89%? Partly, but only partly. Adobe shipped Firefly AI Assistant two days earlier, on April 15. Canva shipped Canva AI 2.0 on April 16. Other design-adjacent names (Adobe, Wix, GoDaddy) all traded lower on launch week.^[17]Figma took the sharpest single-name hit because it's the closest pure-play, but the week compressed three competing AI-design launches into one tape. The market was pricing a category, and Figma caught the sharpest beta to it.

The way I read the move is narrower than the headline. Figma has its own Figma Make prompt-to-prototype product, shipped in July 2025, and disclosed on its Q4 2025 earnings call that weekly active users on Make rose 70% quarter-over-quarter and 75% of paid customers above $10,000 in annual recurring revenue (ARR, the run-rate subscription revenue a software company has under contract) now consume AI credits weekly.^[18] Figma is not a traditional software-as-a-service (SaaS, subscription software delivered over the internet) company being ambushed; it is a SaaS company being re-priced as AI-exposed before its own AI monetization, which started March 2026, lands in the income statement.

The honest read on the 6.89% is that Claude Design accelerated a discount the market was already applying. Professional designers are reading the same tape. Sam Henri Gold, writing the day after launch, reached for Sketch, the Mac-only design app Figma displaced over roughly six years by going cloud-native and collaborative.

The analogy is tighter than it looks. A new category entrant doesn't need to replicate the incumbent's feature depth to take share, it needs to be the obvious first tool for the next generation of users. The 76-year-old on Hacker News who built a commercial tenancy management tool with his son in an afternoon using Claude Design plus Claude Code is not a Figma user.^[10] He was never going to be. That is the actual demand Claude Design is serving, and it is a much larger pool than the designers who already live in Figma.

What to Watch Next

Anthropic said out loud what to watch. The announcement's roadmap paragraph reads in full.^[1]

Integrations, plural. Not a shared canvas. Not persistent project storage. Not a Figma export. That continues the shape.

Three things would change my read. First, if Anthropic ships persistent projects with versioning and audit trails, Claude Design becomes a place teams store work, not an on-ramp, and the asymmetry collapses. Second, if a Figma round-trip export ever appears, the “owns the on-ramp, leaves the craft to existing tools” framing breaks. Third, if live GitHub sync shows up so Claude Design stops working off a one-time codebase upload and starts reading (and writing) against your actual repository, the handoff becomes a full workflow and the product starts competing for the seat, not the moment. None of those has shipped. If any of them ship in the next two quarters, revisit.

Until then, the frame is the one the product itself already draws. Claude Design is a very well-engineered first fifteen minutes of a design job. What happens after those fifteen minutes, the iteration, the system, the handoff, the deploy, is someone else's product. That is a strategy I'd want to short against only if I knew what Anthropic planned to ship in the next two product surfaces, and I don't.