Design Philosophy

Designing AI is designing decisions.

An interface is the last five percent of the work. Most of the job is figuring out what decision someone's actually making, what they need to trust before they act on it, and how much control they need to feel like they're still in charge.

01 · Understand the system

Every AI product is a systems problem before it's an interface problem.

I map the people, permissions, and data underneath a product before I touch a screen. On Agent Identity Graph, that meant tracing who could act on whose behalf, before any UI existed.

See it in Agent Identity Graph

Agent Identity Graph permissions map showing a user, their agents, group memberships, and application access

02 · Design for human judgment

AI doesn't need to replace judgment. It needs to earn it.

I design for the moment someone decides whether to trust what the system just told them: reasoning, review, and a clear way to say no. It's the core of Agentic Identity Security.

See it in Agentic Identity Security

Agentic Identity Security interface showing an AI agent's step-by-step activity plan with completed and in-progress review states

03 · Prototype to learn

The fastest way to answer a design question is to build the thing.

I'd rather prototype in code and be wrong in a day than debate it in a deck for a month. Every AI Solution Accelerator I shipped ran through the same loop.

See it in AI Solution Accelerators

ReviewPanel.tsx

function ReviewPanel({ agent, plan }: Props) {
  const [approved, setApproved] = useState(false);

  return (
    <Panel>
      <Reasoning steps={plan.steps} />
      <ConfidenceBadge score={plan.confidence} />
      <Actions onApprove={() => setApproved(true)} />
    </Panel>
  );
}

My design beliefs

Design systems matter more than screens.

Trust is designed, not assumed.

AI should explain before it automates.

Prototype earlier than feels comfortable.

Humans stay in control.

Along the way: I think in Figma, prototype in React and TypeScript, explore ideas with Claude, and ship through GitHub.

Good AI doesn't feel like AI.

It feels like a decision you can trust.

Agentic Identity Security AI Solution Accelerators AI Dashboard for Microsoft Entra UXPA talk

See all the work