Claude Design supports Claude Code handoff as users rate wireframes 9/10 and video 4.5/10

• Anthropic's launch post says Claude Design is a research preview for Pro, Max, Team, and Enterprise users, built for prototypes, slides, one-pagers, and exports to Canva, PDF, PPTX, HTML, plus handoff into Claude Code, which the main HN item also surfaced.
• Early hands-on reports split cleanly by task: gregisenberg's scorecard rated wireframing 9/10, mobile app design 8.5/10, deck research and design 8.7/10, and video creation 4.5/10.
• The most interesting workflow is the design-to-code path, where the HN discussion roundup includes one report of a 76-year-old relative sketching a CRUD app in Claude Design and handing it to Claude Code for implementation in an existing stack in less than a day.
• Brand and layout quality still need cleanup, according to AmirMushich's brand-kit test, which called the output only partly usable and flagged noisy structure, weak copy, and off-brand details.
• Cost and precision showed up fast as pain points: AIandDesign's quota complaint said one session burned 67% of a $100 Max plan, while _VVSVS's post complained about spending credits just to align a text box.

You can read Anthropic's announcement, browse the huge Hacker News thread, and open the community Anthropic UI kit in Figma that one tester used as source material. The useful bits showed up fast: transcript-to-slides in moritzkremb's demo, HTML prototype export in AIandDesign's first test, and a surprisingly specific backend hack in stopachka's HN comment about AGENTS.md.

Anthropic's post frames Claude Design as a visual workspace for designs, prototypes, slides, and one-pagers, with imports from prompts, images, and code, plus collaboration, exports, and Claude Code handoff. The main HN item matches that feature list and adds the rollout detail that the product lives under Anthropic Labs as a research preview.

The launch surface breaks down into a few concrete capabilities:

• team design system integration
• imports from prompts, images, and code
• iterative visual refinement through chat
• exports to Canva, PDF, PPTX, and HTML
• handoff to Claude Code for implementation

The creator-facing angle is obvious in the first wave of demos. minchoi's launch clip shows a prompt turning into a prototype, a slide deck, and a one-pager in the same flow, while moritzkremb's transcript test shows pasted text turning into presentation layouts.

The feature that kept pulling creators and builders into the same thread was handoff. The HN discussion includes one report from DecoPerson saying his father used Claude Design to sketch a lease-management CRUD app, then handed the result to Claude Code to implement inside an existing business stack with tests and playbooks, all in less than a day.

A second HN report in the fresh discussion roundup says stopachka used a two-line prompt plus an AGENTS.md fetch to give Claude Design a backend, with each cURL request spinning up a new backend and injecting credentials for a multiplayer prototype. Another user in the same discussion set said the missing piece for a PM-to-code workflow was visual exploration first, then export into Claude Code with the design system attached.

That makes Claude Design look less like a standalone canvas tool and more like a front end for agentic product work. The interesting part is not just that it makes mockups. It is that the mockup can carry enough structure to keep going.

The first-day demos already sorted Claude Design into strong and weak lanes. gregisenberg's unscripted test is the clearest scorecard: wireframing scored highest, mobile app design and deck work landed close behind, and video lagged far behind.

Other hands-on posts fill in what those strong lanes look like in practice:

• AIandDesign's first test says importing a personal design system and generating a Bible study app produced an interactive HTML prototype in 10 minutes.
• moritzkremb's demo shows a transcript becoming a slide deck, which is bad news for lightweight presentation generators.
• Lukeedesigner's post says he built a project with Claude Design in about an hour.
• 7_eito_7's walkthrough shows a landing-page workflow that starts from a plain prompt, adds reference UI screenshots, layers in an AI-made background video, then expands sections before code export.

The 7_eito_7 thread is especially useful because it turns the workflow into steps instead of hype:

1. Prompt the site in text, or attach a reference design.
2. Add UI direction from screenshots.
3. Generate an image, turn it into video, and use it as a background.
4. Expand the same style across more sections.
5. Export the result as code.

The praise came with immediate caveats. AmirMushich's brand-kit test used a community Claude brand system from Figma, prompted Claude Design to make a bento-style brand kit, and judged the result roughly 50% usable.

His specific complaints were more useful than the headline:

• layout was noisy
• structure felt random
• copy still had obvious AI filler
• output was not ready to ship without professional cleanup

That lines up with _VVSVS's complaint about burning credits while trying to align a text box, which points to a precision problem rather than an ideation problem. Claude Design looks strongest when the task is broad composition, variant exploration, or rough structure. It looks weaker when the task becomes meticulous placement, brand fidelity, or final polish.

Cost also surfaced fast. AIandDesign's quota complaint says one session ate 67% of a $100 Max plan, which is a concrete warning that highly visual iteration may feel very different from ordinary chat usage.

The last useful finding came from people poking at the product itself, not just the outputs. The fresh HN discussion roundup says one commenter saw the system prompt exposed in network requests, and the earlier discussion set links that claim to a specific HN comment from mcrowe.

That same roundup also mentions a redirect or login loop reported by another user trying to access the product. Those are ordinary research-preview problems, but they matter because Claude Design is not only a generation model. It is a web app that now carries prompts, design systems, exports, and code handoff in the same workflow.

The loudest social posts wanted to turn the launch into an Adobe-versus-Figma apocalypse story. The more grounded first-day evidence says something narrower and more interesting: Claude Design already looks strong at wireframes, slides, landing pages, and prototype handoff, while final layout control, brand cleanup, access friction, and quota burn are still very live issues.