The Signal

What changed.

Replit shipped Agent 4 with an infinite Design Canvas. Anthropic shipped Claude Design. Figma released an MCP that pipes your design system into Claude Code.

Per Stanford / Imperial / Internet Archive research: 35.3% of new websites in 2025 are AI-generated or AI-assisted, up from approximately 0% pre-ChatGPT.

In short, AI-powered design is going mainstream.

The Hypothesis

The emerging thesis we're exploring.

Anyone with a $20 subscription can now ship a clean, professional-looking homepage in an afternoon. That's a legit unlock. The problem is that everyone else can too. And as we explored last week, AI is a consensus machine: really good at converging on the average, not so good at producing anything that stands out.

We think the edge going forward is design care. The small things. The little details that resonate with your customer base and tell them this was made for them. When everyone else is shipping clean-but-sterile, same-as-everyone-else designs, the opportunity belongs to brands that leverage AI's efficiency in order to spend more time on the human touches.

The Big Picture

The structural shifts behind the hypothesis.

1. Audiences have learned to feel AI consensus output. Per Digiday, consumer preference for AI-generated creator content dropped from 60% favorable in 2023 to 26% in 2026. Klaviyo (Dec 2025): only 7% of consumers trust brands more for visible AI; 31% trust them less. Adobe (2026): 33% stop interacting with a brand once they realize content is AI-generated.
2. The differentiation moves up the stack. When fonts became free, typography became the brand signal. When stock photos became free, art direction did. Now that professional design is one prompt away, care, intentionality, and weird-on-purpose become the signal. a16z's October 2025 Design Engineer Fellowship made it the explicit thesis: "Vibe coding gets you to 80% fast. Judgment, systems thinking, and craft become the whole game."
3. Design as a core differentiator is becoming democratized. Linear: $100M ARR with ~100 employees, $1.25B valuation. The Browser Company built and sold Arc to Atlassian for ~$610M with around 50 people. Cron, a two-person team, became Notion Calendar.
4. Care compounds. A mascot. A custom motion system. A hand-drawn illustration style. These get reused on every surface and ladder into brand identity. Polish is a one-time output. Family's wallet design became the acquisition thesis when Aave bought the company. Linear's interaction details are why competitors keep cloning.

What's Working

Real examples from the frontier.

• 1 founder, 4 hours, 40% lift. Dhravya Shah (solo founder, Supermemory) rebuilt his product in a single Claude Opus session, replacing months of agency work on Framer. His claim: 40% conversion lift over the previous agency build.
• Daniel Kuntz's Little Guys. Tiny digital companions built with the kind of hardware-level obsession people screenshot. You can see the level of care put into it just by looking at the product.
• Tony Dinh's TypingMind. Solo founder. Crossed $1M lifetime earnings in 2025 selling a better-designed UI on top of commodity LLM APIs.
• Cofounder. Andrew Pignanelli's agent orchestration platform. What separates it is an obsessive focus on design care. Andrew ships "design highlights week" threads about micro-decisions like "a beautiful border that animates upon approval."
• The Browser Company / Arc. Roughly 50 people built one of the most carefully designed products in their category. Acquired by Atlassian for ~$610M in 2025. The exit wasn't about features. It was about care that mainstream browsers couldn't replicate, even with infinitely more resources.

Frontier Players

The people and teams building at the edge.

• ​Josh Puckett — Former design lead at Dropbox, now writing about uncommon care, taste, and craft.
• ​Ryo Lu — Leads design at Cursor (joined from Notion in early 2026 after five years there). Now leading what is arguably the most influential design role in AI-native software.
• ​Benji Taylor — Head of design at X. Founded Family, the self-custody crypto wallet widely cited as one of the best-designed products in consumer fintech.
• ​Emil Kowalski — Design engineer at Linear; the components he ships get cloned and called inspiration.
• ​Brian Lovin — Designer at Notion building AI products. Built the Prototype Playground that lets Notion designers ship in code via Claude Code.
• ​Ridd — Host of Dive Club, the podcast going deep on design engineering.
• ​Daryl Ginn — Designer documenting the small typography and layout moves that separate AI default from finished work.
• ​Rauno Freiberg — Interaction designer at Vercel; ships the marketing pages and dev tools that other designers study as references.
• ​Haya Odeh — Co-founder and design lead at Replit; built the Agent 4 Design Canvas.
• ​Daniel Kuntz — Little Guy creator, building tiny digital companions with the kind of attention to detail people obsess over.

The Stack

The tools doing the heavy lifting on the frontier.

Reference

• ​Mobbin — 142,200+ mobile and web flows from real shipped apps, organized by pattern.
• ​Refero — Curated screen library for components, systems, and onboarding flows.
• ​supahero.io — A supa cool library for landing page hero sections.
• ​bentogrids.com — The go-to resource for bento feature grids.
• ​cta.gallery — Pattern library for CTA design, button states, and conversion.

AI Builders

• ​Aura.build — AI landing-page builder. 166k+ users, exports clean HTML and Figma.
• ​Lovable — AI full-stack builder. Best when "homepage" is really an MVP that needs auth, payments, or a dashboard.
• ​Replit Agent 4 — Design Canvas + code + deploy in one workspace.
• ​Bolt — Speed-to-prototype builder with native hosting, databases, and SEO. Best for the working demo you need by Friday.
• ​Cursor — Code-first AI builder; production React/Next you can extend and own.
• ​v0 — Vercel's builder. Production Next.js with the clean developer handoff.
• ​Base44 — Built for non-technical folks. Removes every config decision: DB, auth, hosting handled automatically.
• ​Claude Design — Anthropic Labs research preview. Prompt to prototype, deck, or one-pager from text or codebase.
• ​Claude Code — Terminal agent that writes production code from natural language. The way Notion's design team prototypes now.

Polish

• ​Iconify — 200,000+ icons and brand SVGs in one API call.
• ​Unicorn Studio — No-code WebGL animations. Embeds in Framer, Webflow, or Figma.
• ​Jitter — Browser-based motion design. Great for landing-page micro-animations without After Effects.
• ​transitions.dev — Copy-paste library of production-grade UI transition snippets.

Motion

• ​Motionfly — Prompt to product launch video. Generates 3D device mockups, animated text, voiceover, and sound design from a text description.
• ​Remotion — Render React components into MP4. Works inside Claude Code.

Skills & MCPs

• ​Tasteskill — Leon Lin's open-source anti-slop framework. One install, works across Claude Code, Cursor, v0, Lovable, Gemini CLI. Free.
• ​Impeccable — Curated design skills library purpose-built for AI harnesses. Drop-in taste defaults.
• ​UI Skills — TypeUI's registry. 48 design skills, each with preview and install path.
• ​Awesome Design Skills — 67-skill curated list for Claude Design, Stitch, Codex, Cursor.
• ​Make Interfaces Feel Better — Jakub Krehel's popular essay on UI micro-details, packaged as an installable skill.
• ​Emil Design Eng — Emil Kowalski's philosophy on component polish, animation timing, and invisible details.
• ​Interface Craft — Josh Puckett's toolkit. Motion DSL (Storyboard Animation), live controls (DialKit), and a design critique framework.
• ​Mobbin MCP — Gives AI agents access to 621,500+ real app screens.
• ​Figma MCP — Finally, the official Figma MCP. Generates from your real Figma source instead of screenshots.

Foundations & UX

• ​Laws of UX — Jon Yablonski's free archive of 30+ UX principles.
• ​Design System Checklist — Arda Karacizmeli's free open-source checklist for building a real design system.
• ​Growth.Design — Comic-format teardowns of how the best products design for psychology, not aesthetics.

Signal vs Noise

The counter-arguments, and whether they hold up.

💨 Noise: "AI made everyone a designer. Visual quality is dead as a moat."

🎯 Signal: AI raised the floor. There's no longer an excuse for having a low-quality website. But that makes taste, care, and composition matter more, not less. Even before AI, tech had a copycat problem. Delegate the bulk of design to AI and that gets a lot worse. Most companies will pocket the time savings and stop there. The real opportunity is using that saved time on what's always mattered most: building a brand that uniquely resonates with your customers, not just one that looks clean.

💨 Noise: "Care doesn't move metrics. It's polish in disguise."

🎯 Signal: Dhravya rebuilt his homepage in one Claude session with a custom animation skill and saw a 40% conversion lift. Linear's components are constantly cloned, but what people are actually copying is the interaction details (the way things animate, settle, and respond). And at the macro level, audience preference for AI-generated content dropped from 60% to 26% in three years. People can feel the difference between 'generated' and 'considered.'

💨 Noise: "Solo founders can't compete with big companies on care. Big companies have more resources."

🎯 Signal: Early-stage founders have always been better positioned for design care because they're closest to their customers. Big companies don't have that. Their markets are massive, so they compensate with design muscle instead. Design flare > care. Now, founders actually have the means to act on it. AI handles the production, which means a founder's understanding of their customer can actually show up in the design.

Opportunities

If the thesis holds, where the openings are.

• Care as the solo-founder moat. Incumbents have brand councils that strip personality out of every micro-decision. You have real taste and are closest to your customer. Use AI to communicate that through design.
• Verticals where the floor is still low. Design care lands harder in B2B verticals where audiences have lower expectations: HR tech, compliance, healthcare admin, ops tooling. Same playbook with less competition.
• Productized taste as a service. Claude Skills are open-sourcing taste. The next play is bundling them into purpose-built systems. A "shadcn for fintech." A motion system for healthcare. A vertical-specific taste pack.
• Care that compounds. Hand-illustrated icon sets. Custom motion systems. A mascot. Assets that get reused on every surface and ladder into a recognizable brand identity over time.
• Weird as a wedge. Pick one element that's deliberately strange. A doodle. A serif in one word where the rest is sans. Page transitions that feel physical. It does triple duty: a differentiator, a sharability boost, and a recall hook.
• Care isn’t just a design move. The same pattern-hitting design is queued up for copy, UX writing, email, ads, and brand voice. AI-perfect copy will become the floor on every landing page. AI-perfect email sequences will be table stakes. The first companies to ship care across all of those surfaces, not just the homepage, compound the advantage across every customer touchpoint.

Rabbit Hole

Go deeper.

"Details That Make Interfaces Feel Better"
​Jakub Krehel's working list of the small choices that separate good from great: optical alignment, font smoothing, image outlines, tabular numbers, interruptible animations. The most precise current catalog of UI craft.

"50 Design Prompts"
​Amir Mushich's public prompt library for AI design, mapped to the actual workflow. The closest thing to a public benchmark for prompting design out of an LLM.

"The Shift in Design"
​Kinsley on what actually changes when the design floor moves and what stays valuable when the tools commoditize.

"The Growing Rift Between AI Efficiency and Audience Trust"
​Digiday's data piece on the audience side of the consensus problem. Consumer preference for AI-generated creator content fell from roughly 60% favorable in 2023 to 26% in 2026, even as marketers ramped AI spend. The case for care lives in this chart.

"AI Design Slop: 15 Patterns That Out Your App as Vibe-Coded"
​The visual tells of a site that was generated, not designed. A specific checklist of the patterns audiences pattern-match even when they can't name them. Useful as a reverse checklist of what to fix.

"The Web Hasn't Happened Yet"
​Nick Jones and Devin Jacoviello at Config London 2025 on what they learned by getting progressively weirder with their design work.

"The Ultimate Resource for Design Engineers"
​Maze Heart's curated directory of everything at the intersection of design and engineering. If this issue's Stack felt like a starting point, DesEngs is the rest of the iceberg.

Wrapping Up

Designing quality, polished websites has never been easier.

And now that AI is giving some of that time back, brands can spend it on the parts of their website and product that customers feel. A signature loading state. The doodle in the footer that only your customers would appreciate.

But most won't. Most will play it safe. They'll celebrate the time savings and stop there. And that's your opportunity.