AI

Claude Design is the best feature Anthropic has launched in a while, but it has some serious limitations

At a glance:

  • Anthropic's new Claude Design tool lets paid subscribers generate full-fledged presentations, social media graphics, UI mockups, website wireframes, and infographics via text prompts — available now in Research Preview at design.claude.ai.
  • The tool's editing workflow silently counts every manual tweak as a prompt against a separately metered weekly limit, with Enterprise users getting roughly 20 typical prompts for onboarding — causing Max 5x subscribers to hit the wall by mid-week.
  • Despite the friction, Claude Design produces noticeably better-looking outputs than dedicated AI design tools, and inline commenting on the canvas for targeted edits is its smoothest feature.

What is Claude Design and how does it work

Anthropic Labs quietly rolled out Claude Design, a full-fledged design tool embedded inside Claude web, putting it on par with tools like Figma and Canva for everyday visual work. Instead of spending hours dragging shapes around a canvas or hunting for the right font, users describe what they want in natural language and Claude builds an editable design file on a proper canvas. The feature covers presentations, posters, social media graphics, UI mockups, website wireframes, and infographics — essentially anything you would normally open Figma or Canva for.

You can access Claude Design by tapping the Design button in the left sidebar of Claude web or by heading directly to design.claude.ai. It is currently in Research Preview and is limited to Claude paid subscribers. The standalone nature of the tool means design work does not eat into Claude Code limits, and vice versa, which on paper sounds like a win. In practice, it introduces yet another metered quota to manage, and one that runs out faster than any other.

Why the tool stands out from competitors

The author has been testing AI design tools since the earliest days and says Claude has consistently produced the best-looking UI output. She describes "vibe-coding" her presentation slide decks using a Claude skill she found, and the results have been better than anything from dedicated AI design tools. When Claude Design launched, that pattern held — going from prompt to a full-fledged usable design within minutes, with the ability to iterate without starting from scratch or re-prompting.

That iteration loop is where Claude Design pulls ahead. Other AI design tools make the process painful enough that you are better off accepting the first output or starting a brand-new prompt. With Claude Design, you can edit the design after the fact directly on the canvas using an Edit button, or use the Chat panel to highlight specific elements and leave inline comments — similar to dropping a note on a Google Doc. Claude picks up the comment and tweaks just that element without touching the rest of the design, which the author calls the smoothest part of the experience.

The hard limits that kill the workflow

The biggest complaint about Claude Design is not quality — it is quotas. Anthropic does not publish exact prompt counts for Pro, Max 5x, or Max 20x plans, but it tells Enterprise users that their one-time onboarding credit covers "approximately 20 typical prompts." For an entire onboarding period, that is twenty prompts. The author, on the Max 5x plan, hits the limit by the middle of the week using Claude Design only for slide decks, Instagram posts, and occasional website mockups. If a casual Max 5x user burns through the allowance that fast, a single pitch-deck attempt could eat an entire weekly quota on the Pro tier.

Anthropic did announce on X that token limits across every plan had been doubled, but the author argues that doubling only matters if the original limits were generous to begin with. She does not believe Claude Design is impressive enough to justify upgrading her entire Claude subscription just to keep using it past mid-week.

The hidden cost of manual editing

Here is where the workflow breaks down in a way the marketing does not hint at. Every manual edit — nudging an element, resizing something, tweaking text — is silently being implemented by Claude through code. The tool quietly regenerates the design in the background to "apply" the change, which means every tweak counts as a prompt against the weekly limit, exactly the same way a chat prompt would.

This adds unnecessary friction. After each change, you wait a few seconds for Claude to catch up and reapply the edit through code before you can move on. If you are making a bunch of small tweaks in a row, which is exactly what manual editing is for, those seconds add up fast. By the time you have moved three elements and resized two, you have spent more time waiting for edits to be implemented than actually editing. The Edit button is not what it looks like — it is a prompt disguised as a UI interaction.

Where Claude Design still fits in a workflow

Despite the quotas and the clunky editing, the author considers Claude Design one of the best AI design tools available right now. The ability to go from prompt to a polished, editable design in minutes is genuinely useful, especially for non-designers who need quick social media graphics or slide decks. The inline commenting system for targeted edits works well when you want Claude to change just one element without disturbing the rest.

The tool is also convenient because many people are already using Claude for presentations, infographics, slide decks, and websites — so consolidating that into a proper design canvas removes the step of exporting a static preview or markdown outline and taking it elsewhere to polish. The question is whether the quota structure and the silent-prompt editing behavior are sustainable for power users, or whether Anthropic will need to rethink how Claude Design consumes tokens.

What to watch next

Anthropic's pattern of shipping features in Research Preview with tight limits and then adjusting quotas based on user feedback suggests the numbers could shift. The doubled token limits across every plan were a response to criticism, but they did not eliminate the underlying tightness. If more users on Pro and Max tiers hit walls within days, Anthropic may need to either increase the separate Claude Design quota or decouple manual edits from the prompt meter. Until then, casual users should expect to plan their design sessions around a mid-week cutoff, and power users may want to budget their weekly allowance carefully before diving into a multi-slide deck.

Editorial SiliconFeed is an automated feed: facts are checked against sources; copy is normalized and lightly edited for readers.

FAQ

What is Claude Design and where can I access it?
Claude Design is a full-fledged design tool from Anthropic Labs currently in Research Preview. It is available inside Claude web by tapping the Design button in the left sidebar or by visiting design.claude.ai directly. It supports presentations, posters, social media graphics, UI mockups, website wireframes, and infographics.
How many prompts does Claude Design allow, and why do users hit limits so fast?
Anthropic does not publish exact prompt counts for Pro, Max 5x, or Max 20x plans, but Enterprise users are told their one-time onboarding credit covers roughly 20 typical prompts. The author, on the Max 5x plan, runs out by mid-week just using the tool for slide decks, Instagram posts, and occasional website mockups. Every manual edit — nudging an element, resizing text, swapping colors — is silently counted as a prompt because Claude regenerates the design through code in the background.
Can I edit designs Claude generates without losing progress?
Yes and no. There is an Edit button that lets you make manual adjustments directly on the canvas, and an inline commenting system where you highlight an element and leave a note for Claude to tweak. However, every manual change triggers a background regeneration via code, which counts as a prompt against your weekly limit and adds a few seconds of wait time per edit. For multiple small tweaks in a row, the friction adds up quickly.

More in the feed

Prepared by the editorial stack from public data and external sources.

Original article