Designing AI experiences that behave predictably and earn trust.
When AI is involved, the system doesn’t always behave the same way twice.
Designing in this space requires more than polished UI. It requires clear boundaries, review points, and structured decision-making.
This repository documents how I approach AI-enabled product design, from system state modelling to guardrails, failure handling, and implementation review.
- AI Design Principles
- Human-in-the-Loop Architecture
- Guardrail Design
- Failure State Modelling
- Evaluation & Instrumentation
- Operating Model
- AI Workflow State Machine
- Human Review Flow Example
- Guardrail Decision Matrix
- Failure Recovery Pattern
- Instrumentation Naming Pattern
I focus on how the system behaves, not just how it looks.
That means:
- Designing workflows so AI suggestions stay contained and reviewable
- Deciding when AI can act on its own and when a human needs to confirm
- Reviewing what gets built to make sure the experience stays consistent
- Working directly in branches and opening PRs when needed
- Keeping components organised and documented in Storybook
- Making sure analytics actually reflect meaningful user behaviour
I’m not trying to be an engineer.
I’m making sure the product behaves clearly and predictably, even when AI is involved.