Design.md: Portable DNA for AI Design Consistency
Greg Isenberggo watch the original →
the gist
Meng To demos how Google's Design.md captures design essence (typography, colors, effects) in a markdown file to maintain consistency across AI-generated landing pages, motion, slides, and apps, solving 'design drift' from one-shot prompts.
Design.md as Portable Design Blueprint
Meng To explains Design.md, Google's open-sourced markdown format, as a 'recipe' for design systems—encoding typography, colors, spacing, effects, and WebGL animations into a single file attachable to prompts. This 'design DNA' ports seamlessly across tools like Aura, Codex, Cursor, and platforms from web to motion design, slides, and mobile mocks. Unlike rigid Figma templates, Design.md provides foundational flexibility, preventing AI from defaulting to generic purple gradients. To explains: 'The HTML is more like the finished dish and the MD file is more like the recipe.' He shares free resources from communities like V0, Lovable, Framer, and his own Aura gallery, where users download Design.md + HTML pairs to jumpstart custom work.
Greg Isenberg probes how non-designers source inspiration without copying cookie-cutter sites. Meng advises remixing in tools like Variant.com: generate variants, stack 'skills' (prompt ingredients like lasers, skeuomorphic, 3D, copywriting), then extract Design.md. This avoids 'design drift,' where page one shines but subsequent pages genericize. Meng: 'You start something you're kind of satisfied with but then you get into the other stuff and it becomes completely different.' He demonstrates carrying Design.md between local tools like OpenClaw or Cloud Code, embedding it in folder contexts for agentic generation of entire app sections.
Taste and Skills: The Moat Against Generic AI Output
Meng argues taste is the solo builder's moat in an era of high baselines but rampant homogeneity. Humans adapt by curating 'second brains' of inspirations—surrounding themselves with niche products, committing great designs to agent memory. Skills act as 'ingredients': lasers for scroll-stopping effects ('everyone clicks on it'), 3D for depth, skeuomorphic for tactility. Stacking them atop Design.md elevates from vibe-coded to custom. Meng: 'Nowadays just having the typography just having the colors is not enough... you need a moat.' Greg notes urban parallels: homogeneous cityscapes mirror generic sites, urging unique modes like local-first apps outpacing Notion/Figma.
On Google Stitch, Meng sees token efficiency as key for startups: local tools like Codex generate nested MD files for full workflows without cloud costs. He contrasts iteration (90% of work: refining within a medium) vs. remix (10%: porting DNA to new mediums). Craft shifts from pixel-pushing to 'judgment per minute'—agents handle mechanics, humans curate. Meng runs four products solo by queuing Midjourney-like flows, iterating 1,000+ prompts.
Live Demo: Landing Page from Design.md + Skills
Meng demos in Aura: attaches downloaded Design.md (lasers skill) + HTML reference to prompt: 'Create a landing page for Aura, a chat app using AI that ships to email.' Aura generates a consistent, animated site instantly. He iterates sections, remixes for variants, ports to slides/promo videos. No Aura needed—works in any prompter. For motion/slides, same DNA yields cohesive outputs. Meng emphasizes pairing Design.md with HTML for full fidelity (e.g., WebGL lasers), hitting '50 to 80' uniqueness fast.
Solo Scaling: Agents, Local Workflows, and Distribution
Meng shares his evolution: first podcast with Greg boosted MRR from $3K to $15K via exposure. Now, local agents in Codeex process podcast folders holistically: 'Prepare 10 sections or notes... it has all of that knowledge.' He builds 'vibe-coded' Notion clones with hyperframes, Reotion slides—all AI-generated. Advice: Use every niche product for taste; fast edge iteration beats teams. Greg highlights distribution's underrated hardness. Meng: 'Going to Greg's podcast is like the bucket list... number one thing.'
'Taste is the real moat right now, and you build it by surrounding yourself with great design and using every product in your niche.' — Meng To
'One-shot prompts collapse on page two; a design system carries the soul across every medium.' — Meng To (paraphrased key point)
'Judgment per minute as the new craft.' — Meng To
'The shift in craft is from moving pixels to making judgment calls per minute, with agents handling the mechanical work.' — Meng To
Key Takeaways
- Download free Design.md files from Aura, V0, Lovable; pair with HTML for prompts to ensure consistency.
- Remix in Variant, stack skills (lasers, 3D) to avoid generic gradients; extract MD for portability.
- Build taste: curate second brain of niche designs, commit to agent memory via 'remember this.'
- 90% iterate within medium, 10% remix to new (web→motion); use local tools like Codex/OpenClaw for scale.
- Judgment/minute > pixels: queue flows, focus moat on taste for solo 4-product operation.
- Start demos with foundational systems over specific screenshots for AI flexibility.
- Local folders + MD enable agentic full-app generation without token limits.