Claude Design Review: Our Honest Take After a 5-Day Test

Claude Design in a web agency workflow is a rapid AI-assisted UI ideation and HTML/CSS wireframing tool that accelerates early stakeholder alignment by 60%, while still requiring custom Next.js engineering and Figma systems for production €4,500 to €9,000 corporate websites.

We tested it for five working days in April 2026, starting two days after the launch, on real briefs at our corporate website design agency. We build landing pages, corporate sites and web portals for a living, so we tried to break it the way clients break us.

This version of the review keeps what we saw in that test, clearly dated, and corrects what has changed since. Several facts in our April write-up went out of date within months, mainly access, usage limits and export options. Every product fact below links to the Anthropic or Figma page it comes from.

Claude Design is the visual side of Claude. Anthropic describes it as a way to create designs, prototypes, slides and one-pagers together with Claude, and at launch it ran on Claude Opus 4.7, which Anthropic called its most capable vision model (launch post). You describe what you need, Claude builds a first version, and you refine it with inline comments, direct text edits and sliders for spacing, color and layout.

For web work, the design system matters most. During setup Claude reads your codebase and design files and builds a set of colors, typography and components that later projects reuse automatically. The help center lists a GitHub repo, design files, raw uploads, Figma files, decks, logos and fonts as sources, plus a /design-sync command in Claude Code that reads tokens and components straight from a React codebase (design system guide). Design systems are still labelled beta.

It Is Not an Image Generator

Midjourney and Adobe Firefly give you pictures. Claude Design gives you a layout with hierarchy, spacing and components that know about each other, which you can click through as a prototype or export as HTML. In April that was the difference that struck us most. With an image generator, somebody still has to sit down and rebuild the picture as real components. With Claude Design the first draft is already the interface, so that translation step goes away.

We still use Midjourney for moodboards and early brand direction. Firefly lives inside Creative Cloud and makes visual assets such as product shots and background variations, which is a different job. Comparing either one with Claude Design is close to a category error.

Quite a lot, and most of it concerns access rather than output. When we tested Claude Design in April 2026 it was a research preview for paid plans, with its own weekly allowance and its own address at claude.ai/design. By October 2026 it runs inside ordinary Claude conversations, is open to Free users and shares one usage pool with the rest of Claude.

• April 17, 2026. Launch as an Anthropic Labs research preview for Pro, Max, Team and Enterprise subscribers, powered by Claude Opus 4.7. On Enterprise it was off by default until an admin switched it on (Anthropic).
• September 16, 2026. Anthropic's release notes add designs, decks and docs to any conversation and to the Artifacts tab, and state that Artifacts, including Claude Design, Claude Slides and Claude Docs, are available on every plan (release notes). Presentations now have their own tool, Claude Slides (admin guide).
• Usage. Claude Design used to have its own weekly allowance. It now counts toward the same limits as the rest of Claude, including Claude Code (help center).
• Exports and integrations. The launch post listed Canva, PDF, PPTX and standalone HTML. The help center now adds ZIP download, Google Slides, two Claude Code handoff routes and 15 partner tools: Adobe Experience Manager, Adobe for Creativity, Adobe Journey Optimizer, Base44, Canva, Gamma, HubSpot, Hyperframes, Lovable, Miro, Netlify, Replit, v0, Vercel and Wix.
• October 15, 2026. Design, Slides and Docs turn on by default for Enterprise plans. Owners can enable them earlier or switch them off before that date (Artifacts admin guide).
• December 14, 2026. The standalone site at claude.ai/design closes. Existing projects stay there until then, and up to 60 of them also appear in the Artifacts tab.

Some things have not changed. The help center still lists two gaps: there is no version history yet, and multi-person editing is basic and may not work reliably. For regulated work, the Artifacts admin guide says Claude Design does not currently support data residency requirements and is not covered by Anthropic's BAA, so protected health information should stay out of it.

You no longer need a separate site. According to Anthropic's help center, you can ask Claude for a design in any conversation or pick Output, then Design, in the message box. You can also start from a Design template in the Artifacts tab, run /design in Claude Code, or edit designs in the iOS and Android apps. The standalone claude.ai/design works until December 14, 2026, and Anthropic asks users to migrate their design systems before then (help center).

The working loop is the one we learned in April. Write a brief, not a prompt: who the audience is, what they already believe, and what you need them to do. Let it generate, argue with the first structure instead of accepting it, then regenerate with your disagreement written into the brief. The quality of what comes back tracks the quality of what goes in almost exactly.

Two setup decisions save time later. Build the design system first, from a repo or from brand files, so every project starts on brand. Then decide where the work ends up. You can export it (ZIP, PDF, PPTX, Google Slides or standalone HTML), send it to a partner tool, or hand it to Claude Code with "Send to local coding agent" or "Send to Claude Code Web". From Claude Code, /design can import a design into your codebase, export code as a live prototype, or have Claude build the whole thing. Anthropic's tutorial on using Claude Design for prototypes and UX walks through the product design case.

Claude Design has no price of its own. It is included in every Claude plan, Free included, and design work draws on the same usage limits as chat and Claude Code (help center). So the practical question is which Claude plan gives you enough room. On Anthropic's pricing page in October 2026, Pro costs €20 a month (€17 a month billed annually), Max starts at €100 a month, and a Team Standard seat costs €25 a month (€20 billed annually).

Complex projects with large codebases or many iterations use more of your limits. When you reach them, Claude Design is unavailable until the limits reset, unless you have usage credits enabled. On Team and seat-based Enterprise plans, usage comes out of each seat's session and weekly limits and admins can buy extra credits. Usage-based Enterprise plans are billed at standard API rates, with spend limits per organization, group and user (Artifacts admin guide).

Free access is enough to try the tool and form your own view. Whether it carries a full project depends on how many iterations you run. For a business buying a website, the subscription is rarely the number that matters: set against a €4,500 corporate build, €20 a month is rounding error.

We started on April 19, 2026, two days after launch, and used Claude Design for five working days on real work rather than in a sandbox. Five designers ran roughly six client briefs through it: landing pages in industries we know well (fintech, logistics, healthcare), UI component systems, navigation architectures, design system foundations and brand identity directions. Generate, evaluate, push back, regenerate, until somebody got hungry.

What We Tested

Landing pages came first, because a landing page is the sharpest test of whether a tool actually thinks. A good landing page isn't pretty by accident. It's built around a sequence of things the visitor needs to believe before they'll act, and Claude Design reasoned about that sequence. The first time we saw it structure a cold-traffic layout differently from a warm-traffic one, from the same product brief and without being told to, the room went quiet for a second.

UI components were next: component libraries, data visualization sets, form systems, navigation patterns. Consistency within a single session was better than we expected. Things felt like they came from one system, not from a pile of unrelated Pinterest boards.

Day Five Compared to Day One

On day one we were poking at it, looking for edges and trying to find what it couldn't do. By day five we were directing it. We'd picked up how to write a brief so the first pass is already close, how to push back without confusing it, and how to use constraints to speed things up instead of treating them as guardrails.

By our rough count at the time, the step from brief to a first wireframe with complete information architecture took 45% to 60% less time than our Figma-only workflow. Treat that as a five-day estimate from one team, not a controlled measurement.

We've used a lot of AI tools in real workflows, and most of them speed up one task while leaving the thinking alone. In April, Claude Design was different in four ways: it reasoned about structure, it pushed back on bad briefs, it changed direction in minutes, and it kept a session consistent. It also has limits, mostly ones Anthropic documents itself and ones our short test never reached.

Structural Reasoning

Ask for a SaaS landing page aimed at an audience that has never heard of the product category, and it puts education before social proof. Ask for the same product aimed at people already in the category, and the structure moves around. It behaved less like a visual tool with a chat box and more like a design-thinking tool that happens to produce visuals.

The Pushback Is Real

More than once during the week, Claude Design flagged that our brief was ambiguous or contradicted itself before it generated anything. It asked instead of failing. That's closer to working with a senior designer than with a production tool, and it's not always welcome when you're in a hurry. The results were better every time we slowed down and answered properly. Annoying, and also correct.

Iteration Speed and Consistency

Changing a layout direction stopped being a matter of moving elements around. We rewrote the brief and a new layout fell out of it. Inside a session, components shared typography, spacing and interaction patterns. Consistency across a design system is one of the most labor-intensive parts of our work, and having it handled without babysitting was new.

Where It Stops

The help center lists no version history yet and only basic multi-person editing (help center). Design systems are in beta, migrated ones may need some cleanup, and live component previews can't be reached with Tab, so they can't be operated from the keyboard (design system guide). That last point matters if accessibility review is part of your process. And the output is only as good as the brief: Claude Design will produce a credible structure for a wrong brief just as readily as for a right one.

Five days in, we hadn't seen the ceiling, and we said so at the time. We had not stress-tested Claude Design on the hardest work we do: multi-stakeholder discovery, large portal architectures, or brand systems that have to hold together across 12 different touchpoints. Everything we did was exploratory by the standards of a real client engagement, so read the April observations as first impressions from one team.

We have not run or published a second test since, and this update doesn't pretend otherwise. What we can describe is where Claude Design sits in our process, as our service pages and process guide already set out. On a corporate site, the UX lead approves the information architecture while Claude Design generates three to five candidate structures to test it against. On a landing page, it proposes three to five UX directions and a senior designer narrows them to one. Every AI-generated artefact passes through a senior designer before it leaves our hands.

Claude Design and Figma start from opposite ends. In Figma Design a person makes every visual decision, and Figma's AI helps inside that file. Claude Design starts from a goal and constraints and generates the structure first. Figma's own prompt-based tool, Figma Make, is the closer comparison: it turns prompts and existing Figma designs into interactive prototypes and web apps with editable code (Figma help center).

We like Figma. It changed collaborative design in a way that was different in kind, not a small upgrade, and that contribution is real. The table compares the two only on what each company publishes, using Anthropic's help center and pricing page and Figma's help center and pricing page.

Is This Figma's Nokia Moment?

In April I compared Figma and Adobe to Nokia and BlackBerry: companies that were excellent at what they did, whose advantages belonged to the old way of working. Writing this update in October, I'd put it more carefully. Figma's moat is that every designer knows it and every handoff runs through it, and that moat only weakens if the early work stops starting in a Figma file. Figma also already has a prompt-based tool of its own in Make, which the analogy doesn't account for. How much layout work moves into tools like Claude Design, and how much stays in Figma for collaboration and handoff, is an open question. Nobody knows the answer yet, including us.

Adobe Is Now an Export Target

Adobe's position looks different from how it looked in April. Firefly still generates imagery rather than interface structure, but Adobe Experience Manager, Adobe for Creativity and Adobe Journey Optimizer now appear in Claude Design's list of partner tools (help center). For now, Adobe is one of the places Claude Design work gets sent.

Claude Design suits people who need to see an idea before they commit money to it, and teams that already have a design system or codebase it can read. It suits them less where its documented limits bite: regulated data, data residency rules, or work that depends on version history and several people editing at once. The split below draws on our April test and on Anthropic's own documentation.

Good Fit

• Founders and marketers who want a concept to react to, an internal template, or three directions to compare before they brief anyone. The Free plan is enough to try this.
• Product teams with a React codebase, since /design-sync reads tokens and components directly.
• Designers who want several layout directions quickly and would rather spend their own time choosing between them.
• Teams that already publish through Canva, Lovable, v0, Vercel or Wix and want a direct handoff.

Poor Fit

• Healthcare teams and others handling protected health information, because Claude Design isn't covered by Anthropic's BAA.
• Organizations with data residency requirements, which Claude Design doesn't currently support.
• Teams that need version history, or several people reliably editing the same design at the same time.
• Anyone who wants a finished, revenue-carrying website without a person checking whether the structure is right for their audience.

No, and we cover the wider question in a separate piece on AI vs human web designers. The short version from our April test is that Claude Design compresses execution: the first layout pass, the iterations, the component variations. What gets more valuable is judgment. Is this structure right for this business problem? Is the brief wrong? What did the stakeholders never write down?

The designers who will struggle are the ones whose professional identity rests on production speed rather than judgment, and I say that with sympathy, because the industry told them for years that speed was the job. The ones doing well treat it as a lever for their own work: they write better briefs, review output faster and spend their energy on conversion logic and information hierarchy.

Most people who land on this review are not designers. They run a company, they need a website, and they want to know whether a tool like Claude Design changes what they should pay for or who they should hire. It changes who produces the first draft. It doesn't change who has to be right about your audience, your offer and your numbers, and that's still where most of a website's cost and value sit.

What Changes for the Buyer

You see real layout options early, while changing direction is still cheap. In our process, Claude Design produces candidate structures and a person decides, and our published windows are 5 to 10 business days for a landing page, 2 to 4 weeks for a corporate website and 6 to 8 weeks for a portal, or 8 to 12 weeks for compliance-heavy builds (website development timeline). You can also opt out of the AI-assisted workflow: a multilingual corporate site built fully by hand moves from 4 to 8 weeks to 8 to 14 weeks and costs roughly 25% more.

What Does Not Change

Who decides. Claude Design will produce a credible structure for any brief you hand it, including a brief that is wrong. It has no way of knowing that your highest-value visitors arrive from a comparison page and want pricing first, or that your sales team keeps losing on one objection that belongs in your hero section. Those calls determine whether a site earns money, and they still come from a person who has read your numbers. Our website development process puts that reading before any generation happens.

How to Tell Which One You Need

Try the tool yourself before you call anyone. If you want a concept to react to, an internal template, or three directions to compare before committing budget, Claude Design on its own is enough, and paying an agency for that step would be waste. If the site carries revenue, has to rank, or has eight stakeholders with opinions, the generated structure is the cheap part of the job. Our landing pages start at €1,500, corporate websites at €4,500 and web portals at €9,000. What a website budget actually buys breaks down the line items, and the pricing page lists the tiers.

In April 2026, after five days, our verdict was that Claude Design was the most significant design tool release we had seen since Figma itself, mainly because of the way it reasons about structure. Six months on, the product facts have moved in its favor for most buyers: it's in every plan, it works inside ordinary conversations, and it exports to more places, including Claude Code. The gaps Anthropic lists (version history, multi-person editing, data residency) are real, so check them against your own needs.

Figma and Adobe have the money and the talent to respond, and Figma already has Make. Whether early layout work moves out of Figma for good is still open, and I don't think anyone can call it yet. Our own answer is practical: Claude Design proposes, a senior designer decides, and we run AI as our core operating model on that basis. If you want to judge the kind of sites that process produces, our web design portfolio is the place to look.