Claude Design opens research preview with PDF, PPTX, and HTML export
Anthropic opened Claude Design in research preview for prototypes, slides, and one-pagers with brand-system extraction and Claude Code handoff. Try it if you want HTML prototyping or AGENTS.md-backed workflows that push mockups toward production.
TL;DR
- Anthropic opened Claude Design's official announcement in research preview for Pro, Max, Team, and Enterprise users, and the launch summary says it is built for prototypes, slides, one-pagers, and other visual work.
- The useful part is the export stack: according to the launch summary, Claude Design can ship work out as PDF, PPTX, standalone HTML, Canva assets, org links, folders, or a Claude Code handoff bundle.
- Anthropic's pitch is not just prompt-to-mockup. As the core Claude Design evidence notes, the tool can build a design system from an existing codebase and design files, then reuse brand colors, type, and components across projects.
- Early HN users immediately pushed it toward production workflows. the discussion roundup surfaced design-system templating, HTML prototype generation, and even an AGENTS.md-backed backend trick for wiring mocks to live behavior.
You can read Anthropic's launch post, skim the main HN thread, and check Anthropic's Team and Enterprise admin guide. The oddest reveal in the thread is that people were already treating Claude Design less like a canvas app and more like a front door to Claude Code, with the discussion roundup collecting examples that jump from mockup to backend hookup.
What's in preview
Anthropic says Claude Design is a new Labs product for visual work, powered by Claude Opus 4.7, with research-preview access for Pro, Max, Team, and Enterprise plans. Enterprise access exists, but Anthropic's admin guide frames it as an admin-controlled feature rather than a default-on rollout.
Introducing Claude Design by Anthropic Labs
Anthropic has launched Claude Design, a new product designed for collaborating with Claude to create visual assets such as prototypes, slides, and one-pagers. Powered by the Claude Opus 4.7 vision model, the tool allows users to generate initial versions through descriptions and refine them via conversation, edits, and inline comments. It supports team branding by building a design system from existing codebases and files, enables collaborative sharing and editing, and offers export options to formats like PDF, PPTX, and HTML. Additionally, completed designs can be packaged into handoff bundles for use with Claude Code. The product is currently in research preview for Claude Pro, Max, Team, and Enterprise subscribers, with Enterprise access disabled by default.
The feature list in Anthropic's launch post breaks down cleanly:
- prompt a first version in chat
- refine with conversation, inline comments, direct edits, and Claude-made sliders
- share inside an org with view or edit access
- export to PDF, PPTX, Canva, folders, or standalone HTML
- package finished work into a handoff bundle for Claude Code
Design system onboarding
The most concrete workflow hook is the onboarding step that reads your codebase and design files to assemble a reusable design system. According to the core Claude Design evidence, that system can carry brand colors, typography, and components into future outputs automatically.
Claude Design
Claude Design is relevant as a creator-facing visual tool for making mockups, slides, one-pagers, and prototypes faster. The discussion emphasizes using it to communicate intent, explore layout options, and iterate on aesthetics before handing work to a designer or developer for refinement.
That matters because it shifts the product from generic vibe boards to company-shaped drafts. Anthropic also says teams can maintain multiple design systems over time in the same workflow, which makes the tool look closer to a brand-aware generator than a one-off mockup toy in the official announcement.
HTML and Claude Code handoff
Anthropic buried the sharpest creator feature in the export menu. The official post says a finished design can move out as standalone HTML or as a handoff bundle that Claude Code can pick up with a single instruction.
Introducing Claude Design by Anthropic Labs
Anthropic has launched Claude Design, a new product designed for collaborating with Claude to create visual assets such as prototypes, slides, and one-pagers. Powered by the Claude Opus 4.7 vision model, the tool allows users to generate initial versions through descriptions and refine them via conversation, edits, and inline comments. It supports team branding by building a design system from existing codebases and files, enables collaborative sharing and editing, and offers export options to formats like PDF, PPTX, and HTML. Additionally, completed designs can be packaged into handoff bundles for use with Claude Code. The product is currently in research preview for Claude Pro, Max, Team, and Enterprise subscribers, with Enterprise access disabled by default.
HN commenters immediately filled in what that means in practice. firasd on interactive prototype workflow described feeding Claude Design some ChatGPT iframe HTML and getting a question-driven setup flow for a hi-fi interactive prototype, while stopachka on backend handoff described a two-line prompt that gives the design a backend by routing cURL requests through AGENTS.md.
Those are two distinct lanes:
- Standalone HTML for demos, clickable prototypes, and shareable artifacts.
- Claude Code handoff for turning the approved mock into implementation context.
- AGENTS.md glue for teams that want the prototype to talk to live services before a full build.
How early users are working
The early thread reads like a workflow lab. Discussion around Claude Design bundles three uses that are already more specific than Anthropic's launch copy.
Discussion around Claude Design
Thread discussion highlights: - martinald on design system workflow: I wrote this approach up ... extract design system -> make templates -> export ... I've found it unbelievably powerful ... with Claude Code. - stopachka on backend handoff: Here's a 2-line prompt that gives Claude Design a backend ... Every cURL request to AGENTS.md spins up a new backend and splices in the credentials. - firasd on interactive prototype workflow: I gave it the ChatGPT iframe html ... it popped up some questions in a form ... 'Hi-fi design / Interactive prototype / html' ... and Claude said it could see the html/ folder.
- martinald on design system workflow described a loop of extracting a design system, making templates, then exporting into Claude Code.
- firasd on interactive prototype workflow used imported HTML as context for a hi-fi interactive prototype flow.
- Growtika on agency workflow said the tool was useful for expressing intent inside an agency workflow, without treating it as a Figma replacement.
That last point is probably the cleanest read on the preview. The thread mostly treats Claude Design as a compression tool for briefs, variations, and handoff, not as the end of design software.
Creative quality still runs through the model
Six weeks later, Anthropic's Opus 4.8 launch post was still selling creative progress through model behavior, not just UI. the Opus 4.8 launch summary called out dynamic workflows in Claude Code and new effort controls on claude.ai, while the Opus 4.8 discussion roundup captured users testing layout fidelity and long-running agent loops.
Discussion around Claude Opus 4.8
Thread discussion highlights: - simonw on API/platform changes: The new "mid-conversation system messages" thing is particularly interesting... This lets you append updated instructions later in a long-running conversation without restating the full system prompt, which preserves prompt cache hits on the earlier turns and reduces input cost on agentic loops. - senko on coding benchmark: My fav coding benchmark for frontier models is to build a simple RTS game in one file (js/html/css). Claude Code with Opus 4.8 in ultracode mode nailed it, the best result so far. - jkxyz on layout / structured output: My smoke test for new models is to get it to generate a crossword, and this is the first time it's done a good job on the layout.
Two comments are especially on-theme for Claude Design. simonw on API and platform changes highlighted mid-conversation system messages as a way to preserve prompt-cache hits in longer agentic loops, and jkxyz on layout and structured output said Opus 4.8 was the first model to pass a crossword-layout smoke test.
Claude Design may look like a product story, but the thread evidence keeps pointing back to model quality. Better layout discipline, longer conversations, and cleaner handoff context are the substrate under the preview, whether the output is a deck, a prototype, or a one-pager.