Claude Design Launches: Anthropic Labs Turns Opus 4.7 Into a Prototype, Deck, and Wireframe Surface

Anthropic launched Claude Design on April 17, 2026, a new Anthropic Labs product that treats polished visual work -- prototypes, decks, wireframes, marketing collateral -- as a first-class Claude surface rather than a side effect of a chat window. It is powered by Claude Opus 4.7 and ships in research preview for Pro, Max, Team, and Enterprise subscribers at claude.ai/design. The rollout is gradual through the day, and access uses your existing subscription limits with an opt-in for extra usage.

This is the deep-dive on what shipped, how it actually works, where it slots against Cursor canvases and traditional design tools, and what to try first.

Key Takeaways

• What shipped: an Anthropic Labs design surface at claude.ai/design that produces interactive prototypes, wireframes, pitch decks, and marketing collateral from a prompt, uploaded image, document, or codebase.
• Model: Claude Opus 4.7, the most capable vision model in the Claude family, accepting images up to 2,576 pixels on the long edge -- mockups and dense slides no longer need pre-processing.
• Brand-aware by default: during onboarding Claude builds a design system from your codebase and design files; every new project inherits your colors, typography, and components automatically.
• Import from anywhere: text prompt, DOCX/PPTX/XLSX uploads, a codebase pointer, or a web capture tool that grabs elements directly from your production site.
• Refinement controls: inline comments, direct text edits, and adjustment knobs for spacing, color, and layout, plus the ability to apply a change across a full design.
• Org-scoped collaboration: private, view-only, or edit-share within your org, with group chat against the design.
• Export everywhere: internal URL, folder, Canva, PDF, PPTX, or standalone HTML -- plus a handoff bundle to pass into Claude Code with one instruction.
• Availability: research preview on Claude Pro, Max, Team, and Enterprise; Enterprise admins enable it in Organization settings.

What Claude Design Actually Is

Claude Design is a browser-based design surface where you collaborate with Claude the way you would with a designer: describe what you want, watch a first draft appear, then refine. The difference from a general chat surface is the output. Instead of a wall of markdown or a code block you paste somewhere, the answer is a live, editable design -- a prototype you can click, a deck you can page through, a landing page you can capture.

Three shifts make that feel different from previous attempts at AI design:

The first is the model. Claude Design runs on Opus 4.7, the same model behind the latest Claude Code improvements. Opus 4.7's vision accepts images up to 2,576 pixels on the long edge and its instruction following is substantially tighter than Opus 4.6. That combination matters for design work specifically -- the most valuable design prompts are half words, half reference images, and a model that can read a dense mockup cleanly and follow precise "move this, shrink that" instructions is the difference between a novelty and a tool you use.

The second is the brand-aware onboarding. Most AI design tools start every project on a blank canvas. Claude Design starts every project inside your design system, built from your codebase and design files during the first-time setup. You do not re-prompt for your palette on every deck.

The third is the handoff bundle. When a design is ready to ship, Claude Design packages it into a format Claude Code can consume with a single instruction. The prototype you explored in the design surface becomes a front-end build task you hand to an agent, without the usual back-and-forth of "here are my Figma specs, please implement."

Who It Is For

The launch post names three audiences, and the split tells you how to read the product:

Designers get room to explore widely. The ration that experienced designers usually impose on themselves -- two or three directions instead of twelve -- drops because each exploration is a prompt away. Realistic prototypes without PRs, interactive versions of static mockups for user testing, and wide-net design exploration are the obvious wins.

Founders, PMs, and marketers without a design background get a way to produce visual work at all. Pitch decks on brand, feature flow mockups to hand a designer, landing pages to hand marketing. The bar the launch implicitly sets is "competent and on-brand," not "award-winning," and for the work this audience usually outsources to a designer or ships off-brand, that bar is the right one.

Frontier-curious builders get a playground for code-powered prototypes -- voice, video, shaders, 3D, and built-in AI in the generated artifact. This is the closest Claude Design gets to Cursor canvases territory, and the distinction is still the output: a polished, shareable, exportable artifact, not a developer-facing inline UI.

How It Works

Claude Design follows a natural creative flow: ingest context, generate a first version, refine, share, export. Each step has a concrete affordance.

Your Brand, Built In

Onboarding is the most interesting design choice in the launch. Claude reads your codebase and your design files -- Figma libraries, brand guidelines, existing design systems -- and builds a design system that every subsequent project in the workspace uses automatically. Colors, typography, spacing scales, components -- all resolved once, applied forever.

Teams can maintain more than one system, which maps to the real-world case where a company has a product UI brand, a marketing brand, and an internal-tool brand with slightly different conventions. You pick the system at the start of a project and the rest of the session stays inside it.

The system is refinable over time. As the product's brand evolves, you update the design system and new projects pick up the change. The explicit alternative -- pasting your brand guidelines into every prompt, hoping the model remembers by the end of the session -- is what everyone has been doing, and it is exactly the friction Claude Design is built to remove.

Import From Anywhere

Four entry points cover the input side:

• Text prompt: "Landing page for a B2B MCP server directory, clean editorial style, one-column hero, two-column feature grid below" produces a first version in the design system you already configured.
• File upload: DOCX, PPTX, XLSX. Upload your board deck and turn it into a polished on-brand version. Upload a PRD and generate feature mockups from it.
• Codebase pointer: aim Claude at a repo and generate designs that look like the product already does. Useful for marketing collateral that has to feel consistent with the live app.
• Web capture tool: grab elements directly from your production site so prototypes carry the actual look and feel, not a reconstruction.

The web capture piece is the pragmatic one. A common failure mode of AI design tools is that the prototype looks roughly like your app but subtly different, and the subtle difference is the part that matters for user testing. Pulling real elements from the live site closes that gap.

Refine With Fine-Grained Controls

After the first draft, three refinement surfaces do most of the work:

Inline comments let you click an element and tell Claude what to change about it. The same affordance a senior designer uses when reviewing a junior's mockup.

Direct text edits let you fix a headline or a caption without prompting -- type the correction in place.

Adjustment knobs are sliders -- built by Claude for the specific design -- that tweak spacing, color, and layout live. When the knob you want does not exist, you ask Claude to build one, and it materializes for this design. This is a quiet but interesting interaction pattern: the tool tailors its own UI to the artifact.

Once you have dialed a change at one spot, you can ask Claude to apply the pattern across the full design. Edit one card, propagate to all cards.

Collaborate

Sharing is org-scoped. A design can be private, shared for viewing with anyone in your organization who has the link, or shared with edit access so colleagues can modify the design and chat with Claude together in a group conversation.

The group-chat model is the part to sit with. Two or three people in the same Claude Design session, all prompting the same artifact, is a materially different workflow from passing a Figma link back and forth. It resembles the collaboration model Cursor has been building around canvases and the multi-session desktop redesign in Claude Code: the agent is a shared colleague sitting with you in the artifact, not a one-to-one chat.

Export Anywhere

The export surface is deliberately broad:

• Internal URL for sharing within your org.
• Folder for organizing related artifacts.
• Canva for marketers who will do the last-mile polish there.
• PDF for review and sign-off.
• PPTX for pitch decks that need to land in a format executives already consume.
• Standalone HTML files for prototypes that need to live somewhere outside Anthropic.

The PPTX export in particular is the concession to real workflows. Decks get forwarded, annotated, and presented in PowerPoint or Keynote, and an AI-generated deck that only lives inside the AI tool does not graduate to the meeting it was built for. PPTX means it does.

Handoff to Claude Code

The design-to-build handoff is the feature that connects Claude Design to the rest of the Claude Code surface. When a design is ready, Claude packages everything -- components, layout, interactions, assets -- into a handoff bundle that you pass to Claude Code with a single instruction. Instead of describing a mockup in chat or pointing at a screenshot, the design itself is the spec.

This is the piece that differentiates Claude Design from a general AI image tool most sharply. The output is not just a picture. It is a build-ready artifact that the rest of the Claude toolchain already knows how to consume. Pair it with a Claude Code Routine -- "on every new design bundle dropped into this folder, open a PR that implements it" -- and you have an end-to-end loop from napkin sketch to draft PR without leaving Claude.

Five Use Cases That Ship Day One

The launch post calls out six use cases. Five of them are the ones most teams will feel immediately:

Realistic prototypes. Designers turn static mockups into clickable, interactive prototypes for stakeholder review and user testing -- without filing a Jira ticket, waiting for a front-end engineer's sprint, or opening a PR. The prototype carries the real design system, so the test results reflect the real product, not a stand-in.

Product wireframes and mockups. PMs sketch out feature flows in Claude Design, hand them to Claude Code for a rough implementation, or pass them to a designer for polish. The tight integration with the rest of the Claude toolchain is the unlock here: a PM-produced mockup does not have to bounce through three tools to become a pull request.

Design explorations. Designers run many directions in parallel. Ten variants of a hero section in the time it used to take to build one. The ration on exploration drops, which is an unusually direct productivity win in a field where "more options to choose from" is itself the output.

Pitch decks and presentations. Founders and account executives go from outline to on-brand deck in minutes, then export to PPTX or Canva for polish. Claude Design is aimed squarely at the moment before the meeting where you realize your deck looks like a Google Doc.

Marketing collateral. Landing pages, social media assets, and campaign visuals that match the product brand because the product brand was ingested on day one. Marketers loop in designers to polish, rather than starting from scratch.

The sixth use case -- frontier design with voice, video, shaders, 3D, and built-in AI -- is the "see what this can do" category. It is the closest analog to what readers of Cursor canvases already use agent-generated UIs for, and it is where the most surprising outputs will come from early adopters.

Claude Design Next to Cursor Canvases and Artifacts

Three products in the current market share surface DNA with Claude Design: Cursor canvases, ChatGPT canvas, and Claude's own artifacts. The key distinction:

Cursor canvases live inside an IDE and are optimized for developer-facing inline UIs -- dashboards, triage views, eval-failure explorers. The canvas is a pane in the Agents Window. Target audience is developers mid-flow. Our Cursor canvases piece goes deeper on what that looks like day to day.

ChatGPT canvas is a chat-adjacent writing and code surface. It is strong for documents, weaker for visual design deliverables, and has no first-party handoff to an implementation agent.

Claude artifacts live inside a regular Claude conversation. Great for a one-off HTML prototype or a single React component rendered inline. Not built for brand-aware multi-artifact design work, not built for org-scoped collaboration, no PPTX export.

Claude Design is the dedicated surface for visual deliverables -- prototypes, decks, wireframes, marketing collateral -- with a brand system applied, org-scoped collaboration, and export paths into the tools teams already use. The overlap with artifacts is real at the single-file prototype level, but the differentiation starts the moment you want the output to look like it came from your company rather than from Claude.

The comparable move from the coding side is Cursor's composer-to-canvas flow. Both shift the agent's output from text to durable interactive artifacts. The difference is where the artifact lives and who it is for: Cursor puts it inside the editor for developers, Claude Design puts it on a shareable URL for the whole org.

Pricing and Access

Claude Design is included with Claude Pro, Max, Team, and Enterprise plans. No separate subscription. Access draws on your existing subscription limits, and if you run out before you are done, there is an opt-in to continue beyond those limits with extra usage billing. Free-tier Claude accounts do not get access to the research preview.

For Enterprise organizations, Claude Design is off by default. Admins enable it in Organization settings, which is the expected move for a research-preview product in a regulated environment. If you are on an Enterprise plan and do not see claude.ai/design working, check with your admin first.

The rollout through April 17 is gradual throughout the day. If you open claude.ai/design and do not see the product, refresh later. Anthropic has said integrations with more third-party tools are coming "over the coming weeks" -- expect the export surface to grow from its launch-day set.

What to Try First

Three concrete first sessions that pressure-test what Claude Design is actually good at:

1. Onboard your real brand. Point Claude Design at your codebase or upload your existing design files. Verify the generated design system matches what you expect. This is the single most consequential step for every subsequent session -- if the system is off, every project in the workspace will be.
2. Rebuild a real deck. Take a deck you have already presented and redo it in Claude Design. The quality check is whether the on-brand output feels ready to present without a second polish pass. Export to PPTX and open it in the tool you would normally use.
3. Run a Claude Code handoff. Generate a small landing-page mockup, export the handoff bundle, drop it into Claude Code, and ask for an implementation. The end-to-end flow is the whole bet Anthropic is making with this launch. If it works on a small piece, it is the flow for everything after.

Claude Design is the latest beat in Anthropic's week of product expansion -- Opus 4.7 yesterday, the rebuilt Claude Code desktop and Routines the week before, and now a design surface that plugs into both. The underlying move is consistent across all four: Claude is growing beyond a chat box into a set of specialized surfaces that share a model, a design system, and a handoff protocol. Claude Design is the visual-work edge of that surface. If the handoff-to-Code bundle lands as advertised, it is also the most important one yet for non-engineering teams to start using Claude at all.