A running reading list

How to make a design system machine-readable: structured specs, token layers, and audit scripts so AI agents stop hallucinating styles across sessions. The article that reframed how I thought about this problem.

Introduces a Claude skill that uses structured, relentless questioning to stress-test any plan or idea. A good example of how a well-scoped slash command can become a thinking tool, not just an automation.

A GitHub repo of AI agent skills for designers to use with Claude Code and Cursor. Encodes structured design processes across eight aesthetic philosophies, with full workflows from brief through implementation and review.

A practical walkthrough of using Claude Code to run A/B design tests. It covers how to set up variants, prompt the agent to evaluate them, and use that loop to make faster design decisions.

A public Notion workspace documenting a design sprint process. Useful as a reference for how to structure sprint phases, share artifacts, and document decisions alongside an AI-assisted workflow.

Flora Guo, founding design engineer at Paradigm and formerly at Vercel, on what it actually takes to design AI-first products. The part that stuck with me: treating AI as a learning tool rather than a task executor, and how she rebuilt her workflow around Claude Code over a year.

Anthropic’s first technical PM on the eval-driven loop that replaced the PRD, and how Claude Code, MCP, and Skills got incubated. A useful look at how product decisions get made when the underlying model keeps moving.

Designs and sometimes codes — cyberdecks, custom keyboards, circuit boards, creative coding experiments. I follow her for the builder mindset more than any single project: she just makes the thing, and it’s genuinely fun to watch. Also on YouTube and TikTok.