How I AIClaude Design is slow and I love it anyway (plus why I love ChatGPT Images 2.0)
CHAPTERS
- 0:00 – 1:45
Spring 2026 AI design drop: Claude Design vs. GPT Image 2.0
Claire frames the episode around a busy week in AI-for-design releases, focusing on Anthropic’s Claude Design and OpenAI’s new image model. She sets expectations: what these tools can realistically replace today and which workflows they best support for teams.
- 1:45 – 3:05
Claude Design: what it is and why design systems matter
Claire introduces Claude Design as Anthropic’s web-based tool aimed at wireframes, hi-fi prototypes, and even slides/video templates. She argues the differentiator is making the design system a first-class input so prototypes match real brand standards.
- 3:05 – 4:06
Importing a design system by scraping Lenny’s Newsletter site
She demonstrates importing a design system using Lenny’s Newsletter by saving HTML, logos, and assets and uploading them into Claude’s design system form. Claude analyzes the materials over several minutes to extract colors, typography, and components into a usable structure.
- 4:06 – 6:41
How Claude structures design systems + Google’s DESIGN.md standard
Claire highlights Claude’s breakdown of a design system into kits, typography, colors, components, and brand marks—useful even beyond Claude. She connects this to Google Labs’ DESIGN.md proposal, aiming to standardize how agents consume design system metadata.
- 6:41 – 9:44
Building “Lenny Doc”: a hi-fi landing page from a prompt + guided Q&A
Using the imported Lenny design system, Claire prompts Claude Design to create a marketing landing page for a PRD generator. Claude asks clarifying questions (name, audience, sections, interactivity, hero style) to shape the output before generating the design.
- 9:44 – 10:20
Why three variations is Claude Design’s smartest UX choice
Claire explains that defaulting to multiple design variations reduces slow prompt-iteration cycles and helps users who can’t articulate precise changes. The variation approach mirrors A/B testing thinking and makes selection-and-refine faster than repeated prompting.
- 10:20 – 11:05
Limits and pricing friction: hitting the cap and paying $200
Mid-build, Claire hits Anthropic usage limits and gets blocked for days, forcing an upgrade to continue. She calls out rate limits as a recurring pain point with Anthropic products and part of the overall cost/latency tradeoff in AI design tools.
- 11:05 – 16:19
Where Figma still wins: speed, direct manipulation, and fast feedback loops
Claire argues that design work benefits from instant, model-free edits—dragging, resizing, and changing fonts without waiting on an LLM call. She notes Claude Design’s generation loop is slow (minutes), reinforcing why Figma remains a superior iteration canvas.
- 16:19 – 17:57
Turning an OpenClaude article into a branded slide deck
Claire demonstrates Claude Design’s slide-making by feeding an OpenClaude PDF article and applying the design system. The tool produces a cohesive, brand-matched deck and even renders code/terminal snippets with playful design details, reinforcing her view that slides may become “code-driven.”
- 17:57 – 19:44
GeoCities throwback: “Lenny’s Product Zone” and the copywriting edge
With no design system constraint, Claude Design produces a wild ’90s GeoCities-inspired redesign that’s intentionally chaotic. Claire calls out a ‘secret power’: excellent copy generation for headlines and page text, and the value of using reference styles to steer creative extremes.
- 19:44 – 20:15
Claude Design recap: best use cases and current tradeoffs
Claire summarizes three standout workflows: importing a design system for marketing pages, turning content into branded decks, and generating playful redesigns. She reiterates the main downsides—slow generation and usage limits—while emphasizing the upside of design-system-first generation.
- 20:15 – 21:25
ChatGPT Images 2.0: why it feels like the first ‘thinking’ image model
Claire introduces GPT Image 2.0 as a step-change for image generation, especially for reliable text rendering and object/layout accuracy. She frames it as a model that can reason more effectively about composition, typography, and structured visual outputs.
- 21:25 – 23:43
Multi-page brand kit for ChatPRD + iterating with reference images
She tests a popular prompt to generate a multi-panel brand kit for ChatPRD, then critiques the first result as ‘not us.’ By adding reference images (brighter, pinker, pixelated), she iterates to a new kit that better matches the brand direction, showing a practical marketing workflow.
- 23:43
Personal color analysis demo + final recap of the week’s tools
Claire uploads a photo for a personal color analysis and gets a full layout with palettes and outfit previews, then corrects the result (dark winter) and iterates to a better match. She closes by recapping: Claude Design excels at brand-consistent assets but is slow/limited; GPT Image 2.0 excels at layout and typography; DESIGN.md signals standardization ahead.
Sponsor: WorkOS and the enterprise-security problem for AI apps
A sponsor segment explains why AI copilots and chatbots need deep access to company systems, creating enterprise security and compliance hurdles. WorkOS is positioned as a drop-in platform for auth, access controls, and audit logs to help apps pass IT scrutiny.
Reviewing the generated landing page + AI-assisted tweaking and commenting
After a long generation wait, Claire reviews the “Lenny Doc” page and finds brand adherence strong (colors/logo) with a recognizable Claude quirk: italic serif styling. She showcases built-in tweak controls (headline styles, hero layout, background, CTA, pricing section) and component-level commenting/drawing for targeted improvements.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome