On April 14, Anthropic's CPO resigned from Figma's board.

Three days later, Claude Design launched and Figma's stock fell 6.8% the same day.

The market connected the dots quickly. Then quickly moved on to trying it out over the weekend.

I did too, and the results are delightful. Let me show you.

But first, some catchup:

Before we get into today's edition, if you're a professional who wants to go from knowing about AI to building with it, this weekend is a good place to start.

📅 24th April | 🕰 7PM IST onwards

Session 1 | Developer productivity, AI agents, and AI-led coding

📅 25th April | 🕰 10AM IST onwards

Session 2 | Agentic AI workflows, case studies & demos of agentic workflows

Session 3 | Orchestration, multi-agent systems, MCPs, and RAGs

As a bonus, you'll also learn to build over 30 agents in 2 days.

It's free. 25 spots left!

It's a new product that lets you build prototypes, slide decks, dashboards, landing pages, and one-pagers by describing what you want in plain English.

You describe, build a first version and then refine it through conversation, inline comments, direct edits, or sliders that Claude generates on the fly.

When you're done you export it or hand it directly to Claude Code.

Available for Pro, Max, Team, and Enterprise plans only, and web only for now.

When you open it, you pick a starting point: Prototype, Slide Deck, From Template, or Other.

Then you either start from a text prompt or connect your brand.

Before you type a single prompt, decide whether to set up a design system.

Without one, Claude defaults to safe and generic choices, which will obviously look like every other AI-generated thing you've seen.

With one, Claude reads your brand from a GitHub repo, live URL, or a Figma file.

It extracts your exact hex values, type scale, and spacing tokens. Every project after that inherits your brand automatically.

The trade-off: extraction takes 15 to 20 minutes depending on codebase size. For a one-off design or exploring, not worth it.

Test 1: A landing page

Prompt: "Build a landing page for a B2B SaaS tool that helps marketing teams track competitor ads. Dark background, bold typography, hero section with a headline and one CTA button, three feature tiles below, and a simple footer."

Output:

Then I asked it to make the colour scheme lighter and more minimal. It rewrote the whole page.

At the same time, I toggled on the Tweaks panel and it had: four accent colour swatches, a headline tone toggle (Default / Punchy / Formal), and a CTA label switcher (Trial / Demo / Watch).

Clicking a swatch updated every accent across the entire page instantly.

My take: The Tweaks panel is delightful.

With it, you're essentially adjusting live parameters that Claude generated specifically for this design. That's such a cool interaction model.

Test 2: Mobile onboarding flow

Prompt: "Create a 4-screen mobile onboarding flow for a meditation app. Screen 1: welcome. Screen 2: goal selection with three options. Screen 3: notification permission. Screen 4: you're all set. Soft colours, clean typography, generous white space."

Output:

The visual language was consistent across all four screens i.e. same font, green palette, and component style.

My take: Four coherent screens from one prompt is the use case that would have taken at least one hour to mock up manually.

A PM or founder could take this to a design review or a user test without touching Figma.

Test 3: Investor deck

Prompt: "Build a 5-slide investor deck for an AI productivity tool targeting remote teams. Dark minimal aesthetic."

Output:

“Twelve slides in competing tools vs two prompts here” is a claim Brilliant's team made in Anthropic's launch post.

Based on what I got in one prompt, I believe it.

Test 4: Internal dashboard

Prompt: "Design an internal content moderation dashboard. Left sidebar with filter tabs. Main area shows flagged items as cards with Approve / Reject / Escalate buttons. Clean, functional."

Output:

This was the most impressive output of the four.

It understood what a real moderation queue looks like. And the content it populated the cards with was contextually appropriate.

When I clicked the Approve button to edit it, the right panel showed full design controls, almost the same level of control you'd get clicking an element in Figma.

Chat is for broad changes for anything structural or aesthetic that affects the whole design.

Inline comments are for specific changes like: click directly on an element and tell Claude what to change.

This is good because Claude batches all your inline comments and processes them as a single coherent update rather than making one change at a time and potentially breaking other things.

So a rule of thumb: if you have five small things to fix, mark all five with inline comments first, then let Claude process them together.

Token consumption is mad.

Claude Design uses tokens from your weekly plan allowance, and it uses a lot of them.

Pro users report hitting their limit in 3 to 5 complex prompts. The dashboard and deck tests were the heaviest for me.

If you're on Pro, start with the simpler tests first.

It has no backend.

Prototypes run in a sandboxed environment so everything resets when you close the session.

It's clear that they meant this tool to be for visual work and prototyping only for now.

Generic output without context.

If you skip the design system setup and use vague prompts, you'll get the default AI aesthetic.

Specific prompts with reference material produce dramatically better results.

I went in expecting to be underwhelmed.

Every few weeks a new tool drops and the internet declares Figma dead, Dylan posts something about taste, everyone goes back to Figma by end of day.

But I kept coming back to the feeling of “delight” after testing this. It just feels good to use in a way most AI tools don't.

The token limit is funny though. One deck and 40% of your weekly allowance is gone.

I get why, I just hope they make it more accessible soon because the tool deserves to be used without rationing.

As for Figma, I don't see a reason to panic still.

They need to sprint toward the parts of design Claude can't touch, like research, exploration, and decisions that happen before anything gets built.

That's where they win if they choose to play it right.

Try it at claude.ai/design. Reply and tell me what you built.

Until next time,
Vaibhav 🤝

If you read till here, you might find this interesting

#AD 1

The AI Report distills the day's most important AI news into one free 5-minute read. No jargon, no filler — just what 400,000+ business leaders need to know before their first meeting.

Subscribe free today

#AD 2

While many businesses are solving problems at lightspeed, their payroll systems seem to stay stuck in the past. Deel's free Payroll Toolkit shows you what's actually changing in payroll this year, which problems hit first, and how to fix them before they cost you. Because new compliance rules, AI automation, and multi-country remote teams are all colliding at once.

Check out the free Deel Payroll Toolkit today and get a step-by-step roadmap to modernize operations, reduce manual work, and build a payroll strategy that scales with confidence.

Download the Toolkit for free today