What is a taste file? The missing layer between AI coding and good design

You've used an AI coding agent to build a UI. The code works perfectly. But the result looks like every other AI-generated interface: generic blue buttons, default shadows, safe typography choices. It works. It just doesn't look designed.

This is the taste problem. And it's not a technical failure — it's a context failure. Your agent doesn't know what good design looks like for your specific brand or aesthetic. It's making choices in a vacuum.

What is a taste file for developers?

A taste file — specifically a taste.md — is a structured markdown document that encodes a brand's complete design intelligence. Not just a color palette. Not just typography. The full visual DNA: how colors interact, what spacing philosophy the brand follows, which component patterns feel right, how depth and shadow are handled, and the emotional register the brand targets.

Think of it as briefing a designer before they start work — except the "designer" is your AI coding agent. Instead of hoping the agent guesses right, you give it the same reference material a human designer would use.

A complete design taste file includes:

• Color system — primary, secondary, accent, surface, and semantic colors with usage rules
• Typography — font families, size scales, weight usage, line height, and letter spacing patterns
• Spacing philosophy — the brand's approach to density, breathing room, and rhythm
• Component defaults — how buttons, cards, inputs, and navigation should feel and behave
• Visual principles — the underlying beliefs that drive every design decision
• Anti-patterns — what the brand explicitly avoids, which is often more useful than what it does

Why AI agents need design taste files

AI coding agents are trained on enormous amounts of code and design patterns. They know what's common. They reproduce what's common. But "common" is the enemy of great design.

Stripe doesn't use generic blue. It uses a specific blurple (#635BFF) that sits at an exact intersection between professional and modern. Vercel doesn't use gray text — it uses a carefully calibrated scale from #888 to #FFFFFF that creates depth without distraction. Linear doesn't just use dark mode — it uses near-black (#0A0A0B) backgrounds with specific surface layers that create a sense of focused depth.

These aren't arbitrary choices. They're the result of hundreds of design decisions made by some of the best designers in the industry. A taste file captures that intelligence and makes it available to your agent.

How to use a design taste file

The workflow is straightforward:

1. Choose a taste file that matches the aesthetic you're going for
2. Paste the prompt.md into your AI agent's system prompt or context
3. Build as normal — your agent now makes design decisions informed by that brand's DNA

This works with any AI coding agent that accepts markdown context: Claude Code, Cursor, Bolt.new, Lovable, v0, GitHub Copilot, and more.

The difference a taste file makes

Without a taste file, an AI agent asked to "build a modern dashboard" produces something technically correct but visually generic. With a Stripe taste file loaded, the same prompt produces deep navy backgrounds, blurple accents, layered card surfaces, and typography that feels like it belongs to a product worth $100/month.

The code is the same quality. The visual output is completely different.

Where to get taste files

Dupe extracts, structures, and packages taste files for the world's most admired brands. Each Dupe pack includes the taste.md, tokens.css, tailwind.extend.js, and a prompt.md ready to paste into any agent. Free previews available before purchase.