"Layers Gives AI the Design Judgment It's Been Missing"

AI-assisted design tools have gotten remarkably good at generating plausible-looking screens. Type a prompt, get a polished UI mockup in seconds. But anyone who has actually shipped a product knows that what happens underneath the screen — the conceptual model, the interaction logic, the understanding of user needs — matters far more than what the pixels look like. Layers, a new open-source skills pack from designer Jamie Mill, tries to close that gap by giving AI tools a structured framework for thinking through all seven layers of product design, not just the top one.

The framework divides product design into three zones: Reality (observed behaviour), Problem Space (user needs, domain knowledge), and Solution Space (strategy, conceptual model, interaction structure, and finally the surface layer). Most AI tools default to the surface — they'll happily spit out a landing page without ever asking who's going to use it or why. Layers walks both human designers and their AI copilots down through each layer systematically, with dedicated "skills" that prompt the AI to think about the right thing at the right depth. There's even a /layers-orient command that audits all seven layers and tells you which one is actually your bottleneck — a kind of design triage that even experienced practitioners could benefit from.

What makes this especially clever is the distribution model. Layers installs via npx skills add jamiemill/layers-skills — a single command that works across Claude Code, Cursor, OpenAI Codex, and over 50 other AI tools. That universality matters because it treats design reasoning as a portable asset rather than something locked into one platform's prompt library. It's a glimpse of where the AI-tooling ecosystem might be heading: not bigger foundation models, but better prompt architecture shared across them.

There's a deeper insight here about what "AI skills" actually are. They're not just saved prompts — they're structured ways of decomposing a problem that force the AI (and the person using it) to be more deliberate. The same idea underpins Anthropic's concept of "metacognitive strategies" in their Think tool, and it's likely we'll see more domain-specific skill packs emerge this year for everything from legal drafting to mechanical engineering. Layers is a strong early example of what that looks like when it's done well — opinionated enough to be useful, open enough to be adopted anywhere.

Layers is open source on GitHub.