The Taste Series

Why this series exists

Every AI-generated website looks the same: gradient hero, rounded-everything, floating 3D object, fade-in on scroll, three-column feature grid, Inter.

That is not because the models have bad taste. It is because they have no taste until you give them yours.

This series is the end-to-end path from “my AI output looks generic” to “my AI output looks authored.” You read 5 guides, install 4 templates, and walk away with a portable /taste skill that runs in every project you open.

---

The 5 guides, in order

The series is designed to be read top to bottom on a first pass. You can also jump directly to the guide that matches your current problem.

---

The 4 templates

These templates are the shippable artifacts of the series. Every one pairs with a specific guide and is ready to drop into ~/.claude/skills/ or your project root.

---

The full workflow (once you have read everything)

After the series, your daily loop looks like this:

1. Write in your project: work on a project that has its own CLAUDE.md at the root. Global rules for this project.
2. Folder-scoped taste: each surface (website, docs, emails) has its own local CLAUDE.md.
3. Personal default: your ~/.claude/skills/taste/SKILL.md travels with you. It is the baseline in every project you open.
4. Generate something: ask Claude Code to make a section, page, or component.
5. Audit the output: invoke /taste, /anti-pattern-check, and /motion-audit in that order.
6. Fix and score: violations get fixed inline, the scorecard prints, you decide ship or regenerate.

Once this loop is in place, generic AI output is no longer a risk you manage case by case. It is a pattern your stack prevents by default.

---

Who this series is for

You will get the most value if:

• You use Claude Code, Cursor, or any AI tool that generates UI
• You have already noticed that your AI output looks the same as everyone else’s
• You have design opinions that you have never written down
• You lead a design system and you are thinking about how it survives the AI era

Skip the series if:

• You do not use AI tools to generate interfaces
• You prefer to critique generations manually rather than automate the review
• You already have a mature taste-encoding practice and want to compare notes (the essay is still worth 12 minutes)

---

Suggested reading paths

The “I just want to stop the bleeding” path (30 minutes)

1. Skim the AI Anti-Patterns Gallery and paste the combined anti-pattern block into your project’s CLAUDE.md.
2. Install the Anti-Pattern Check Skill.
3. Run /anti-pattern-check on your most recent generation.

You will not have full coverage, but the worst 10 AI clichés are gone today.

The “I want a real taste system” path (90 minutes)

Follow the 5 guides in order, install all 4 templates, and run the full workflow once.

The “I want to teach this to a team” path (2 hours plus a workshop slot)

1. Read the series yourself first.
2. Use the Taste Stack and Build Your Own Taste Skill as the spine of a 90-minute team workshop.
3. Ship the Anti-Pattern Check Skill and Motion Audit Skill into your system repo. The whole team invokes the same skills.
4. Use the Taste Audit Scorecard as a shared review artifact.

---

What is next after this series

If the Taste Series solves your “AI output looks generic” problem and you want the next level, these are the companion tracks I am building:

• AI-Native Design Systems: how to ship tokens + components + skills as one system
• Design System Governance in the Agent Era: the moat is not components, it is rules
• Obsidian + AI for Designers: where your taste rules live alongside your research and content

All three build on the foundation this series gives you.