Skip to content
Back to articles

Claude Design: A Conversational Design Studio

April 17, 2026/11 min read/2,103 words
AnthropicClaudeAI in DesignGenerative AI
Promotional artwork for Claude Design by Anthropic Labs: overlapping soft gradient panels in orange, cream, and blue behind the Claude Design wordmark, suggesting a visual canvas

In Brief

Claude Design is a new product from Anthropic, announced today. It lets you make designs, interactive prototypes, presentations, and one-pagers by talking to Claude instead of clicking through menus. You describe what you want, Claude builds it on a visual workspace, and you refine it by leaving comments, dragging sliders, or just telling it what to change.

It runs on Claude Opus 4.7, Anthropic's most capable vision model. A vision model is an AI that can look at images and understand what is in them, not just read text. That choice matters more than it sounds. Most AI design tools work from words only: you type, they generate code. Claude Design can look at a screenshot of a competitor's product and understand it spatially, including where the buttons sit, what the colors are doing, and how the layout breathes.

Anthropic launched Claude Design as a research preview inside its Labs program. It joins five other Labs products that all shipped in the last four months.

What Anthropic Labs actually is

Anthropic Labs is a separate team inside Anthropic, announced on January 13, 2026. Its job is to incubate new products at the edge of what Claude can do. Anthropic President Daniela Amodei framed it as a place that "gives us room to break the mold and explore."

Think of Labs as Anthropic's workshop. The main product organisation scales the things that already work, like Claude the chatbot, the API, and the enterprise plans. Labs takes bets that might not work yet.

The track record since January is real:

ProductWhat it is
Claude CodeA version of Claude that reads and edits files on your computer
MCPA shared protocol that lets tools like Slack and Google Calendar talk to AI systems
SkillsSmall add-on packages that give Claude new abilities
Claude in ChromeA browser extension that lets Claude control your web browser for you
CoworkA place where Claude works on long tasks while you walk away
Claude DesignThe new one. Visual design and prototyping.

Six products in roughly four months. Claude Design is not a one-off experiment. It is part of a pattern: Labs ships something new every few weeks, and the ones that prove useful stay. Mike Krieger, Instagram's co-founder and Anthropic's former Chief Product Officer, runs Labs alongside co-founder Ben Mann. Before Labs, Krieger helped take Instagram from zero to a billion users. Worth holding in mind when reading these launches: this is a product shop, not a research lab.

What Claude Design does

At its core, Claude Design turns a conversation into visual work. You do not start with a blank Figma file or a blank PowerPoint. You start with words.

Here is the range of things it can make:

CapabilityWhat it looks likeAnalogy
Interactive prototypesClickable designs that behave like real appsA paper menu that actually takes your order
PresentationsPitch decks built one prompt per slideDescribing slides to a designer in real time
One-pagersSingle-page summaries for marketing or partnersAn email that arrives already styled
Wireframes and mockupsRough sketches of a product's layoutNapkin drawings, but clean
Brand-consistent outputEverything matches your company's existing visual styleAlways writing in your house voice
Handoff to Claude CodeA working prototype becomes real running codeA blueprint the builder can actually follow

The common thread: these are not six different tools. They are one tool pointed at different goals. The interface is always "describe and edit". What changes is what comes out the other side.

How it actually works

There are four steps, and they loop.

Step 1: Describe what you want

You type a prompt. It can be vague, such as "a landing page for a dog-walking service", or specific, such as "a four-slide deck for a board meeting about Q2 revenue". You can also attach context:

  • Screenshots of competitors: "make it feel calmer than this one"
  • A codebase: Claude reads your existing components so new designs match
  • Design files: import what you already have
  • Images or URLs: paste a page you want to riff on

Step 2: Claude fills the canvas

A canvas (the visual workspace) opens and fills with the initial design. You see it appear in front of you, roughly how the final thing should look.

Step 3: Iterate

This is where Claude Design parts ways with a regular Claude chat. You have three ways to change things:

MethodWhen to use itExample
ChatBig structural changes, tone shifts"Make the whole deck more playful"
Inline commentsTargeted fixes on one elementClick a button, write "bigger padding, darker text"
Direct editsMoving, resizing, draggingThe way you use Figma or Keynote

Together, the three methods cover the range from "rewrite the whole thing" down to "move this button two pixels left".

Step 4: Export or hand off

When the design is ready, you can send it somewhere else:

DestinationFor what
PDFSharing outside your company
PPTXEditing later in PowerPoint or Keynote
CanvaFinishing touches in a familiar editor
HTMLPublishing on the web
ZIPAll the assets at once
Claude Code handoffTurning the prototype into real running code

The last one is worth sitting with. A prototype built in Claude Design can be handed to Claude Code, which implements it as real software. You can go from a sentence to running code without leaving the Claude ecosystem.

Why vision is the hinge

Many AI tools can generate designs. What separates Claude Design from most of them is the model underneath: Claude Opus 4.7, a vision model.

A vision model is an AI trained to see images and understand them, not only read text. When you paste a screenshot of a competitor's landing page, Claude does not just register that "an image is present". It sees where the heading sits, what colour the call-to-action button is, how the spacing works. Roughly the way you would.

Compare that to other popular AI-in-design tools:

ToolPrimary inputWhat it makes
v0Text promptsReact code components
LovableText promptsFull working web apps
Figma AIFigma files plus textDesign-system-aware edits inside Figma
Canva AITemplates plus promptsPolished marketing assets
Claude DesignText, images, code, files, URLsDesigns, decks, prototypes, one-pagers

The others are powerful, but most of them expect you to start from text or from inside an existing design file. Claude Design accepts any of the above, including a photo you took of a whiteboard. That is the vision model doing the lifting.

Claude Design vs Artifacts

If you already use Claude, you probably know Artifacts: the side panel where Claude puts generated code, diagrams, HTML pages, or React components. On the surface, that sounds a lot like Claude Design. It is worth drawing the line clearly.

DimensionArtifactsClaude Design
ScopeOne item per artifactA full project with many pages
IterationChat onlyChat plus inline comments, direct edits, sliders
Can importNothing, you start from a promptCodebases, design files, screenshots, URLs, DOCX/PPTX
Brand consistencyNot managedAuto-applied across the whole project
SharingPublic publish or linkOrganisation-scoped with view or edit permissions
ExportCopy or download one filePDF, PPTX, HTML, Canva, ZIP, Claude Code handoff
PlansFree tier includedPro, Max, Team, Enterprise only
InterfaceSide panel beside the chatDedicated canvas workspace

The short version: Artifacts showed what happens when a chat makes one thing. Claude Design shows what happens when a chat makes a whole project.

What early users say

Three partner companies gave testimonials in the announcement. They are worth reading for where Claude Design already helps.

Melanie Perkins, Co-Founder and CEO at Canva, focused on the collaboration itself: Canva and Anthropic have an existing partnership, and Claude Design exports directly to Canva. That detail reads two ways: Anthropic is handing finishing work to Canva at the end of the pipeline, or Anthropic is eating Canva's top-of-funnel by taking the "first draft" part. Probably a bit of both.

Olivia Xu, Senior Product Designer at Brilliant (an online education company), described a concrete speed-up: complex pages that used to need "20+ prompts" now need "only 2". Same result, a fraction of the effort.

Aneesh Kethini, Product Manager at Datadog, talked about the meeting-room effect: "from rough idea to working prototype before anyone leaves the room". Prototyping used to be work you scheduled for later. It is becoming work you do live.

Read together, the three testimonials point at the same shift: speed, in-meeting prototyping, and handoff-friendliness with tools teams already use.

Limits and caveats

Claude Design is a research preview, which means three practical things.

It is live, but it is early. You can use it now with a Pro, Max, Team, or Enterprise subscription at claude.ai/design. Enterprise accounts need an admin to turn it on; it is off by default for that plan.

Known bugs exist. The support documentation lists several openly: inline comments occasionally disappear before Claude processes them, a compact layout view can throw save errors, very large codebases can slow things down, and occasional "chat upstream error" messages need a fresh chat tab to clear.

Usage is metered separately from the rest of Claude. Claude Design has its own weekly allowance that resets every seven days. It sits alongside your existing chat and Claude Code limits, not inside them, so heavy design work does not eat your chatbot quota. The allowance scales with plan:

PlanWho it fits
ProQuick explorations, one-off use
Max 5xSemi-regular use, PMs and engineers producing occasional mock-ups
Max 20xPower use, designers and creatives
Team / Enterprise seat-basedStandard and Premium tiers, one allowance per provisioned user
Enterprise usage-basedBilled at standard API rates, with a one-time credit worth roughly 20 prompts that expires July 17, 2026

Extra usage can be bought on most plans. Allowances are granted per user, not pooled at the organisation level.

Audit logs and usage tracking are not available yet. Anthropic explicitly flags this gap: because Claude Design is a Labs release, design activity is metered but not logged in the way chat and Claude Code are.

Anthropic did not give a timeline for moving out of research preview.

A Tuesday at the studio

Picture a product manager running a Q2 planning meeting.

TimeWhat happens
9:00 amThe meeting opens. The PM types into Claude Design: "Mock up a home screen for our new subscription tier, same brand as our main app." She drops in a screenshot of the current home screen. Claude opens a canvas and fills it with an initial design.
9:04 amThe engineering lead frowns at the call-to-action button. The PM clicks it on the canvas and leaves an inline comment: "More urgent, less cluttered." The button updates.
9:11 amSomeone asks in Slack what the upgrade flow looks like. The PM tells Claude: "Add three screens for the upgrade flow, tasteful animations." Three new screens appear.
9:20 amThe designer is out today but the team wants her input. The PM shares the project with her with edit access. Forty seconds later she is leaving comments from her phone.
9:30 amThe meeting ends. The PM exports the prototype to PDF for the execs, then hands the same project off to Claude Code with one line: "Turn this into a working Next.js feature branch."
5:00 pmA draft pull request is waiting in GitHub. It is not done. It needs a real designer and a real engineer to finish it. But the starting point is hours ahead of where Monday would have left things.

Half of that used to be scheduled work. Now it happens during the meeting.

The shift this represents

A pattern is showing up across everything Anthropic Labs ships. Chat is becoming the interface for creative work, not just for answering questions.

Claude Code did this for software. Skills did it for specialised expertise. Claude in Chrome did it for the web. Cowork did it for long-running tasks. Claude Design does it for visuals.

All of them assume the same thing: a conversation is a fast enough interface to replace clicking through menus. It works as broadly as "build me a landing page" and as narrowly as "shift this button two pixels left". The only remaining question is whether the output is good enough to ship. Artifacts already suggested yes for small things. Claude Design is that bet aimed at whole projects.

Glossary

TermDefinition
Claude DesignAn Anthropic Labs product that lets you create designs, prototypes, presentations, and one-pagers by describing them to Claude
Anthropic LabsA team inside Anthropic that incubates experimental products at the edge of Claude's capabilities
Vision modelAn AI that can look at images and understand what is in them, not only read text
CanvasThe visual workspace where a design appears, and where you can edit it directly
PrototypeAn early working version of something, used to test ideas before building the real thing
Research previewAn early version of a product that is live and usable but may still change
Inline commentA comment attached to a specific element in a design, not to the whole project
Design systemA shared set of colours, fonts, components, and rules so everything a company makes looks consistent
HandoffPassing a finished design to the next person to build it for real
ArtifactsAn older Claude feature that shows generated content like code, HTML, or SVGs in a side panel beside the chat

Sources and resources

Share this article