Lab · three pieces

Design systems for humans and agents

Self-directed experiments in React, component systems, and AI-assisted prototyping. Each asks how a product team works once coding agents join it, and answers with something that runs.

  1. Agent Review: where design judgment goes when coding agents join the team

    A working prototype for knowing when a coding agent should proceed on its own, ask a person to decide, or wait for approval, informed by a real experiment with Claude Code. Its two runs are simulated.

    • Results first, with job boundaries in plain words
    • Trade-offs side by side, decided by a person
    • Decisions that keep their reasons for the next run
    Case study Working prototype React TypeScript Storybook Claude Code

  2. Rollcall: a broken deploy the developer can read and roll back

    A deploy console for the moment something broke, laid out in the order developers ask their questions.

    • The live deploy kept apart from the newest attempt
    • A failed build explained on one card
    • A rollback previewed first and undone in one press
    Working prototype Plain HTML, CSS and JavaScript Synthetic data

  3. Second Look: an AI recommendation the dispatcher can check and overrule

    Decision support for freight dispatch, built on one bet: a dispatcher checks a pick faster when the screen shows what needs checking.

    • Every factor shown with its value and its age
    • A close call that names its tradeoff
    • Legal limits enforced before anything is scored
    Working prototype Plain HTML, CSS and JavaScript Synthetic data