Claude Design is Anthropic’s conversational design experience for creating and refining visual work such as prototypes, slides, one-pagers, and landing pages. It pairs a Claude conversation with a visual canvas. For example, a product manager can describe an account-settings flow, review the resulting screens, and request changes before an engineer assesses implementation.
Claude’s current documentation places Design within the Claude experience and Artifacts, rather than treating claude.ai/design as its permanent home. Use it to explore or communicate a design. A useful prototype is not, by itself, approval for production software.
This guide explains where Claude Design is available, how to move from a bounded brief to a reviewed prototype, when to establish a design system, and what to check before sharing, exporting, handing off to Claude Code, or migrating older work.
A prototype can clarify an idea. Production acceptance remains a separate product and engineering decision.
What Claude Design is now
Claude Design lets you describe visual work in a conversation, provide relevant context, and review the result on a canvas. You can refine the result through chat, inline comments, or direct canvas edits. Anthropic’s current getting-started guide describes the workflow, design-system inputs, sharing, exports, and Claude Code options.
Anthropic announced Claude Design as a research preview on April 17, 2026. That announcement is useful for launch history, but it is not the right source for current plan access or usage rules. The current product guide describes Design as part of the broader Claude and Artifacts experience.
The practical boundary is straightforward: choose Claude Design when a conversational brief and visual canvas can answer a defined design question. Keep requirements approval, accessibility review, technical feasibility, and production acceptance in their existing product and engineering processes.
Where to find it, plan access, and the standalone-site deadline
Current Help Center guidance lists Claude Design for Free, Pro, Max, Ultra, Team, and Enterprise plans. Enterprise administrators may control access to relevant capabilities. You can start from a Claude conversation or the Artifacts tab, and the documented workflow also includes Claude Code and Claude mobile apps. Sharing-setting changes must be made on Claude web or desktop.
The standalone site at claude.ai/design is scheduled to close on December 14, 2026. Anyone still using it should begin a migration inventory now instead of assuming that every project record will transfer.
A design system can move without bringing its chats and comments with it. Inventory projects, design systems, chats, comments, shared links, and exports as separate material. Download important project archives, preserve decision history you still need, and then verify migrated permissions and organization defaults.
Follow Anthropic’s standalone-to-Claude migration guide for the current steps and archive instructions. Material left in the standalone environment may be deleted after closure according to Anthropic’s retention practices, so treat the date as an operational deadline rather than a general product announcement.
The practical workflow: brief, canvas, review, revision
A useful first pass begins with a bounded brief rather than a request to make a good app. Name the target user, the task they need to complete, the screens required, the main interaction, content constraints, and any approved design references. The following is a suggested team brief, not a Claude-published schema:
{
"target_user": "workspace administrator",
"user_goal": "invite a colleague and set their role",
"required_screens": [
"member list",
"invite form",
"confirmation state"
],
"primary_interaction": "choose a role before sending an invite",
"content_constraints": "use plain language; show validation errors",
"design_system_reference": "approved workspace UI system",
"reference_files": [
"current-members-screen.png",
"role-guidance.pdf"
]
}
These fields make the request reviewable. They do not create an API contract, guarantee a particular screen structure, or replace the team’s requirements record.
The sequence below is a recommended working method built on Claude’s documented conversation, canvas, editing, preview, and handoff capabilities. The Claude Academy prototype tutorial provides additional practical examples.
A small review record helps separate the visual artifact from the decision about that artifact. The following is an illustrative team record, not a Claude data format:
{
"artifact_reference": "link-to-reviewed-artifact",
"artifact_revision_reference": "revision-2026-11-20-03",
"reviewer": "product-owner",
"required_screens_checked": true,
"interaction_checked": true,
"open_issues": [
"Confirm role-change permissions with engineering"
],
"decision": "revise-before-sharing"
}
At this grain, one record represents one reviewed artifact revision. If a team stores comments, exports, or usage observations separately, each should have its own stable identifier and source reference. A project name or prompt text alone is not a safe deduplication key.
When to set up a design system first
A quick internal concept can start from a prompt and a few relevant references. If the design will represent your brand, be reused across work, or go to engineering, begin with reviewed design-system material where available.
Anthropic documents inputs including codebases, design files, Figma files, slide decks, documents, logos, fonts, screenshots, and other brand assets. Claude can use those materials to generate reusable colors, typography, components, and layout patterns. The documented Claude Code /design-sync workflow can synchronize a design system from code.
Generated material still needs review against the actual product and brand. Incomplete or inconsistent source material can produce an equally inconsistent system. For Team and Enterprise organizations, Anthropic’s admin guide covers controls for publishing, organization defaults, and deletion. Assign an owner before making a system a broad default.
- Are approved colors, typography, components, and product references available?
- Has a designer or brand owner checked the generated system against those authoritative references?
- Is one person responsible for publishing, revising, and answering questions about the system?
- Do distinct brands or products need separate systems instead of one shared default?
From prototype to Claude Code: define the handoff gate
Claude’s documentation supports bringing design context into Claude Code and continuing implementation. It does not guarantee that generated code will fit a particular repository architecture, framework, dependency policy, or production standard.
Before merge, the receiving engineer should inspect the repository’s conventions and dependencies, responsive behavior, keyboard access, content, authentication and authorization, data handling, error and empty states, tests, and deployment requirements. The design owner should resolve visual or interaction ambiguities. The engineer owns technical fit and code review.
A useful handoff record can distinguish the source design from the implementation decision. For concurrent systems, enforce uniqueness in the database or use a transactional upsert. A lookup followed by create is not race-safe. The following fields are proposed team metadata, not vendor-published fields:
{
"artifact_id": "artifact-123",
"artifact_revision_id": "revision-03",
"design_system_reference": "workspace-system-v4",
"repository_reference": "team-repository",
"review_owner": "receiving-engineer",
"unresolved_design_issues": [
"Confirm keyboard focus order"
],
"merge_decision": "engineering-review-required"
}
The important decision is not whether Claude can produce code. It is whether the receiving engineer can verify the code against the actual repository and requirements before it enters the team’s production path.
Sharing and exports: verify the destination, not just the button
Current Claude guidance documents ZIP download, PDF, PPTX, Google Slides, standalone HTML, third-party destinations, and Claude Code handoff. The exact behavior of a connector can depend on the destination, account authorization, plan, and organization settings.
Artifacts are private by default and can have view, comment, or edit permissions. Sharing availability varies by plan. Team and Enterprise administrators can restrict organization and external sharing, and external links may require recipients to sign in. Check the current artifact-sharing rules before sending a link outside the organization.
Before sharing, confirm the audience, permission level, and policy. Before exporting, identify the expected result:
- Presentation: choose a format such as PDF, PPTX, or Google Slides when the recipient needs to review or present the work.
- Prototype: use the visual artifact or standalone HTML when the goal is to demonstrate a flow, then test the actual output.
- Further editing: verify that the selected destination supports the object types and editability your team needs.
- Implementation context: use the documented Claude Code handoff, followed by repository-specific engineering review.
Claude Design supports standalone HTML export, but treat exported HTML as a prototype or handoff artifact until it has passed the project’s engineering, security, accessibility, performance, and deployment review. Do not assume that every connector produces an identical or fully editable result.
Usage limits and what Claude Design does not promise
Current guidance says usage across Claude.ai, Claude Code, and Claude Desktop counts toward shared limits. It does not describe a separate Claude Design allowance or a fixed percentage consumed by each prompt. Usage can depend on conversation length, task complexity, model, features, and effort level.
Monitor current usage in Claude settings rather than relying on older standalone-product estimates. Eligible paid-plan users may be able to continue with usage credits after included limits, which can add charges. Review Anthropic’s usage-limit guidance and usage-credit instructions, and set spending controls before enabling credits.
Claude Design documents visual exploration, canvas editing, sharing, export, and handoff capabilities. It does not guarantee design quality, accessibility, production suitability, identical behavior across destinations, or a particular repository implementation. Use deterministic checks for requirements with a clear answer, such as required screens, approved color values, permission rules, file format, and repository restrictions. Use Claude’s judgment for alternative layouts, draft UX writing, and qualitative critique.
Who benefits most, and when to use another process
Developers can explore an interface before implementation. Product teams and founders can communicate a feature concept and review a proposed flow. Designers can generate early alternatives or drafts. Teams working on slides, one-pagers, or landing pages can use the canvas to organize visual material.
These are bounded uses, not a replacement for the team’s source of truth for approved requirements, brand assets, or production code. Professional judgment remains important where brand differentiation, complex interaction design, accessibility, security, or production specifications matter.
Choose Claude Design when a conversational brief and visual canvas will help answer a defined design question. Keep established design and engineering reviews for approved product work, and name the person who decides whether the output is an accepted prototype or a production handoff.
