When Anthropic shipped Claude Design as a research preview on April 17, 2026, my feed filled with two reactions in equal measure: shock and confusion. Figma's stock dropped more than 7% the same day. Wix, Adobe, and GoDaddy slid alongside it. Anthropic's CPO Mike Krieger had resigned from Figma's board three days earlier. Yet when I finally sat down with the tool, the first impression was quieter than the headlines suggested.
I split my time between shipping iOS apps and making art, so design tools are working tools for me. Claude Design isn't abstract news — it's going to affect how I lay out a landing page tomorrow morning. This post is the report I'd want from a peer: what the preview actually does, where it's rough, and when it's worth reaching for over the alternatives.
What's genuinely new
Claude Design is a standalone web app that turns a conversation into a prototype, slide deck, landing page, or one-pager. Described that way, it sounds identical to v0, Lovable, or Bolt. Three things separate it:
- Native handoff to Claude Code. You can ship the output as a zip, but you can also expose it at
api.anthropic.com/v1/design/h/<hash>and have Claude Code pull it in directly. The same model lineage carries you from design into implementation. - Six output channels from one source. Zip, PDF, PPTX, Canva, standalone HTML, and Claude Code handoff. You don't regenerate for each destination — you retarget.
- The build process is visible. v0 and Lovable present a polished result. Claude Design leaves the build log and debug history on-screen. The output becomes something you can code-review rather than just accept.
The engine under the hood is Claude Opus 4.7, released the same day. On a 93-task coding benchmark, Opus 4.7 scored 13% higher than Opus 4.6 and solved four tasks that neither Opus 4.6 nor Sonnet 4.6 could. Vision was also upgraded, which matters for reading high-resolution mockups.
One-shot LP generation: what actually happened
I stress-tested the preview by asking it to build the landing page for a fictional AI sketching tool called "Opus Canvas." Single prompt. Requirements:
- Hero with kinetic typography, a gradient mesh background, and 3D parallax cards
- Features section with glassmorphism,
backdrop-filter, Lucide icons, hover tilt - CTA with a dark section, conic gradient, and SVG particles
- Colors
#6366F1/#FB7185, Inter + Fraunces (variable) for typography prefers-reduced-motionrespected,IntersectionObservertriggers, smooth scroll
All of that coexisted in a single generation. With v0, the same request usually takes five or six conversational turns to converge. Claude Design's "nail down the design system first, then generate code" flow seems to be what keeps complex requests from breaking apart.
It isn't flawless. In my run, mixing Japanese and Latin fonts threw off the baseline in a few places — manual tweaking required. Unused Tailwind utilities lingered in the output. The realistic flow is to accept the first draft and let Claude Code clean it up.
Who should actually use it today
The preview is paid-subscriber only and disabled by default on Enterprise plans, so it isn't universal. Three cases where I think it pays off right now:
Solo app developers who need a landing page in minutes. I used to burn half a day on hero imagery and layout for a new app launch. With Claude Design as the first pass and Claude Code for cleanup, that same work compresses to about an hour. This is where the time savings are most dramatic.
Freelancers generating client decks in volume. PPTX as a first-class output means you can build from scratch every time instead of maintaining a fragile template. Producing three variants for the client to compare becomes realistic.
Engineer-only teams that need interim design. Because it hands off directly to Claude Code, a team without a designer can still ship a working v1. You'll still want a designer later, but for MVP scope it's enough.
It's not the right tool for brand-sensitive enterprise work or pixel-precise UI polish. Figma's plugin ecosystem, collaboration features, and multi-year design-system tooling aren't replicated yet.
Treat it as a complement, not a replacement
The "Figma killer" framing in the press is too tidy. In practice, Claude Design fits alongside existing tools, not in place of them. Here's how I've started splitting the work:
- Exploring the idea: let Claude Design produce a rough draft to expose ambiguity in the requirements
- Real design work: Figma for component libraries and design tokens
- Bridge to implementation: hand off to Claude Code, build in TypeScript + Tailwind
- Social launch assets: Canva templates
Each tool earns its place because it's good at something different. Anthropic has clearly pushed into the application layer, but the more useful framing is that the entry point to design work is shifting toward conversation — not that conversation will replace the entire pipeline.
Things to know before you sign in
As of April 2026, keep these constraints in mind:
- Research preview status means UI and features may change significantly
- Disabled by default on Enterprise plans (it can read brand assets and codebases)
- Paid Claude subscribers (Pro and above) only
- Using the Claude Code handoff requires Claude Code to be set up
- Japanese UI localization is partial; some labels remain in English
This is one of the fastest-moving areas in AI tooling right now. Expect near-weekly feature additions for a while. If you're curious, getting hands-on during the preview is worth it — you'll develop a concrete sense of where it fits in your own workflow. In the next post I'll walk through a full Claude Design → Claude Code handoff on a real project, failure modes included.