Writing

Notes on making complicated software clearer, from twenty years inside enterprise products and the tools that now build with them. They run from craft to systems to operations, then to what changes when AI is in the work. Each answers its question in its first paragraph and points to the work it draws on.

Designing with code

What a design system has to explain to an agent, what makes an interface feel finished, and what a developer needs on a bad day.

  1. What a design system needs to explain to an agent

    Components and tokens tell an agent what to build with. It also needs meaning, behavior, context and exceptions, worked through Agent Review’s two reds.

    5 ideas · 5 min read

  2. What makes an interface feel finished?

    Craft you can name and measure: type on a scale, color as a measured pair, motion with a rule, and a decision for every state.

    7 ideas · 8 min read

  3. What does a developer need from an interface on a bad day?

    On a bad day a developer needs what happened, to what, why and the next step, tried on one screen in the Deploy Console prototype.

    7 ideas · 8 min read

Enterprise and operational software

What a person needs when a workflow breaks, what a design system needs once AI is in the product, and what held forty-plus apps together.

  1. Designing for the moment the workflow breaks

    When an order stalls or a transaction fails, people need to know what happened, what it affects and what they can safely do next.

    5 ideas · 7 min read

  2. What does a design system need once AI is in the product?

    Six things a system built for facts needs once AI hands it claims, starting with a slot for the source and the age.

    7 ideas · 7 min read

  3. What I learned standardizing UX across 40+ SAP Fiori apps

    Lead UX for 40+ SAP Fiori apps: one guide and library, feedback loops, release checks, teams that own it, and what moved.

    7 ideas · 6 min read

AI as the material

What an AI feature owes whoever acts on it, what a person needs to supervise an agent, and what a prototype of either must prove.

  1. Enterprise AI should help people decide, not just answer

    A useful AI feature helps someone judge a recommendation and act on it appropriately, worked through Second Look, a dispatch prototype.

    7 ideas · 8 min read

  2. What does a person need to supervise an AI agent?

    More than an approve button: the line between what the agent does alone and what it asks about, and an approval that carries its evidence.

    7 ideas · 9 min read

  3. What should a working prototype actually prove?

    A prototype should resolve what makes the next investment risky, with real code, real inputs and real users, ending in build, kill or park.

    6 ideas · 6 min read