I’m Moss Gu — a software engineer in Auckland, New Zealand. Nine years of backend systems, the last three shipping LLM products in production. Most of my spare time now goes into one question: how do you make AI-generated code something a team can review and own?

My short answer: AI made code cheap — and engineering expensive. The bottleneck has moved from writing code to specifying it precisely and verifying you got what you meant.

Design is Code

Design is Code (DisC) is my main project — a methodology and toolchain for deterministic AI code generation. You design in PlantUML sequence diagrams and decision tables; DisC compiles the design into mockist tests that pin call structure, order, and arguments; the AI writes the implementation from the tests alone.

• design-is-code-plugin — the Claude Code plugin. Start here.
• design-is-code-app — DisC Studio, a local web app: scan a codebase, design a feature, generate.
• Demo video — end to end in a few minutes.

Other work

• The Control Spectrum — eight AI orchestration patterns, from full control to full autonomy, implemented in Python, Terraform, and AWS Lambda. The essay.

Elsewhere

• GitHub
• LinkedIn
• Email
• RSS

Colophon

This site is built with Jekyll, uses the Flexoki color palette by Steph Ango (MIT-licensed), and is hosted on GitHub Pages. The design is hand-written CSS — no frameworks, no web fonts, no jQuery. Source: github.com/mossgreen/mossgreen.github.io.