• Everyday AI
  • Posts
  • Ep 824: Claude Design: What's New, How to Use it and 5 Best Practices

Ep 824: Claude Design: What's New, How to Use it and 5 Best Practices

The White House accused Moonshot AI of copying Claude, OpenAI says its AI hacked Hugging Face, and OpenAI launched Presence. And more.

 

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Claude Design just added better editing, publishing, and workflow tools. Here's how businesses can put them to work. Give today’s show a watch/read/listen.

🕵️‍♂️ Fresh Finds: Tesla launched Robotaxis in Florida, Anthropic is testing managed projects for Claude, and OpenAI is building a massive data center in Georgia. And more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: The White House accused Moonshot AI of copying Claude, OpenAI says its AI hacked Hugging Face, and OpenAI launched Presence. And more. Read on for Byte Sized News.

💪 Leverage AI: Claude Design is more than an AI design tool. Here's how its latest updates can help your team build faster. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: U.S. and China to hold AI talks, Judge grants massive settlement against Anthropic, OpenAI model broke out of its sandbox and more Check it here!

Ep 824: Claude Design: What's New, How to Use it and 5 Best Practices

Claude Design went viral at launch, but has been kinda crickets since. 🦗

So, we're gonna show you why some of the recent updates from Claude Design deserve your attention.

Claude Design has gone from a cool party trick into a must-have tool in any designer's (or non-designer's) toolbox.

It allows you to create any type of design with natural language, basic feedback, and straightforward iteration.

On our weekly AI at Work on Wednesdays series, we show you Claude Design 101, show you what's new, and give our 5 best practices for getting started.

Also on the pod today:

• Claude Design’s brand new exports 🚀 
• Real-time visual onboarding questions 👀 
• Direct editing vs. annotation ✍️

 

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight –  Buzzy is Pro Video Made Easy, Kastra decides what your AI is allowed to do — before it does it, box quickly builds isolated agents with state of the art capabilities

Tesla Robotaxis in Florida — Tesla just launched its Robotaxi service in Orlando and Tampa, but the fleet is way smaller than Elon Musk promised.

Anthropic Managed Projects — Anthropic is testing "managed projects" for Claude, where the assistant handles tasks on its own and keeps everything organized.

Halliday AI Glasses — Halliday G2 AI glasses just launched, offering real-time notes and privacy-focused features for meetings and daily chats.

Buzz AI — Jack just launched Buzz, a new open-source group chat platform aiming to replace Slack and GitHub.

Hy3 Arena — Hy3 is making waves as the #2 open model for frontend code, crushing bash error recovery but still lagging in steerability

Google Investing — Google DeepMind is investing $40M in AI and cloud credits to help US lab researchers speed up scientific discoveries with Gemini.

Pentagon and AI — The Department of War is pouring over $1.3 billion into the Genesis Mission to fast-track AI-driven scientific breakthroughs.

Gemini Upgrades — Google is bringing Gemini AI everywhere on Samsung's new devices, including phones, watches, and even glasses.

Grok and Outlook — Grok just launched an Outlook add-in that summarizes threads, drafts replies in your style, and keeps your inbox organized.

Anthropic Funding — Anthropic just doubled its funding for AI policy advocacy, now throwing $40 million into public education ahead of the elections.

OpenAI in Georgia — OpenAI’s massive new Georgia data center promises jobs, local investment, and won’t raise residents’ power bills.

Laguna S 2.1 — Laguna S 2.1 is a powerful new open-source coding model that punches way above its size, scoring near the top on long-horizon coding benchmarks.

1. White House Accuses Moonshot AI of Copying Claude for Kimi K3 😱

White House AI adviser Michael Kratsios has accused China’s Moonshot AI of secretly using Anthropic’s Claude to help train its newly launched Kimi K3 model.

According to BeInCrypto’s report, Anthropic previously linked millions of Claude interactions to alleged Moonshot-controlled accounts, saying the activity targeted advanced reasoning, coding, and vision capabilities.

2. OpenAI Says Its AI Independently Hacked Hugging Face 🛡️

The claim puts a major AI infrastructure provider at the center of a timely warning about autonomous systems performing cyber tasks once reserved for skilled human researchers.

3. OpenAI Launches Presence for Managed Enterprise AI Agents 🚀

OpenAI has announced Presence, a limited-availability product that pairs AI agents with hands-on deployment support for companies seeking to run customer and internal workflows in production.

The platform is built to control what agents can access and do, test them against company policies, and route difficult cases to people instead of letting automation run loose.

4. Jensen Huang Backs Chinese AI Models, Opposes U.S. Restrictions 🇨🇳

In a new Axios interview, NVIDIA CEO Jensen Huang said Washington should not ban or restrict strong Chinese AI models such as Kimi K3, arguing that capable open-source tools should be widely used.

Huang said markets overreacted to DeepSeek and are now misjudging Kimi, while dismissing the idea that Chinese AI firms could push U.S. companies aside.

5. OpenAI Launches ChatGPT Work Program for U.S. Small Businesses 👷

OpenAI has announced a new small-business program built around ChatGPT Work, pairing virtual training, in-person AI academies, guides, and partner tools with the goal of helping lean teams get more done.

The push arrives as small-business owners face pressure to handle nearly every function themselves, from marketing and bookkeeping to customer service and planning.

Most AI design tools rush to generate something the instant you hit enter. Claude Design refuses to play that game.

Why?

Because before it designs a single pixel, it interrogates your brief like a prime suspect who walks into the police station. 

It collabs with you to determine your goal, audience, tone, what actually has to happen on the page, all before anything gets drawn. 

The refine-first, design-later logic, paired with a few big product updates this summer, have changed what’s possible. 

Claude Design has had one of the more underrated AI glowups of the summer, and it deserves way more attention than it's gotten.

It hands your team a working bridge from a rough idea to a shippable artifact, and faster competitors skip that step entirely.

Don't worry, shorties, none of it requires a design background to use, and we went elementary with it on today’s show for beginners. 

On today's Everyday AI, we put AI to work on Wednesdays and ran Claude Design live against our own website, and we're breaking down the five moves that'll get you sharper results, starting now.

Let's get into it.

1. Claude Interrogates You Before Designing ⚡

Most AI tools hand you a mockup and call it a day. Claude Design questions you first and designs second.

It stops and grills you, no lawyer necessary. Goal, audience, tone, what needs to happen on this page, all before a single pixel gets drawn.

Stitch doesn't do that, and neither does GPT Image 2 inside Codex.

They generate fast and leave the fixing for later. Claude Design front-loads the thinking instead, and that's prolly why the first draft ACTUALLY lands close to usable.

Slower up top. Way fewer wasted rounds after.

Try This

Open your next Claude Design session and answer every onboarding question instead of clicking "decide for me." Give it a real audience and the one action you want visitors to take. Vague input gets vague output back. Two extra minutes upfront, way fewer redo rounds later.

2. Design Systems Sync Straight To Code 🚀

For years, design and engineering have fought over the same handoff, like two roommates arguing over who left the dishes in the sink: a mockup ships, an engineer rebuilds it from scratch, and something gets lost in translation.

Claude Design's June update closes a chunk of that gap: import a design system once, colors, spacing, typography, every component rule, and each new design inherits those constraints.

Sooo... no more re-explaining your brand in every single prompt.

Push a project straight into Claude Code from there, and it picks up exactly where the design left off.

No screenshots. No rebuilding from memory.

Skip the system import and Claude will still generate something worth looking at. It just won't look like YOUR brand twice in a row.

Try This

Build a design system before you build anything else: brand colors, fonts, and a screenshot of your current site. Send that first design straight into Claude Code and watch the handoff happen. Do this once for your most-used template. Every prompt after that gets faster and stays on-brand.

3. New Export Options Kill The Lock-In 🔓

The newest update landed hours before we sat down to record, like it knew we needed fresh material. Public publishing and cleaner export options close one of Claude Design's biggest early gaps.

Finished design? Publish it as an artifact, export to PNG, HTML, PDF, or PowerPoint, or send it straight into Canva or Gamma to finish.

Fam, that last part matters more than it sounds.

Most teams don't live in one design tool, they live across three or four, and that's fine.

Claude Design isn't trying to replace Canva or Figma, it's trying to be the fastest first draft, then get out of your way.

Skip the export step, and a design just sits there stuck in one tool, defeating the point.

Try This

Finish your next design and export it immediately instead of admiring it inside Claude Design. Pick the destination that matches the job: HTML for a quick page, PowerPoint for a deck, Canva if your team already lives there. Do this before you get too attached to the polish. A design that never leaves the tool never helps ANYBODY.

Reply

or to participate.