My practice / AI
How I work with AI.
I use AI throughout my design process, with clear goals, context, and oversight.
I use it to plan, prototype, and build. I also design the workflows that carry context forward and keep me in control of important decisions.
01 / In my own workflow
I keep context connected across projects.
Useful context can get scattered across projects and conversations. I set up a shared knowledge workspace that connects what I’m working on with my design principles and working preferences.
An AI assistant can retrieve the relevant notes and maintain them as I work. I defined rules for that maintenance: preserve my authored principles, record where decisions came from, and keep tentative ideas separate from commitments.
The result is a set of connected, inspectable notes that I can return to across conversations.
02 / Orchestration & safeguards
I automate routine work and build in checks.
I’ve configured a scheduled daily brief to bring together accessible conversations, decisions, and open questions. The workflow routes useful context back to the relevant project and records gaps in what it could access.
I separate automation from enforcement. The schedule triggers the work, workflow instructions guide it, and runtime permissions limit file access. A local pre-commit hook adds an automatic check before changes enter version history.
The hook checks the exact revision being committed for broken internal note links and malformed JSON checkpoint files. If a check fails, it blocks the commit and identifies what needs fixing. I tested both failure and repair paths; the check leaves working files intact.
03 / Delegation & review
I set the direction and delegate the work.
I work across several tasks at once. I set the goals and constraints, give agents focused objectives, and use their progress, results, and blockers to decide what happens next.
For Incident Owner Copilot, I used ChatGPT to plan the experience, explore alternatives, and challenge design decisions. Across my work, I pair that planning with Codex, Claude Code, or Claude Design to build and iterate. For Copilot, Claude Design was part of that cycle; I reviewed the prototype and directed the next changes.
That cycle let me test product ideas in a working prototype and refine the interactions as I went. My design principles gave me criteria for review: could people inspect the evidence, control consequential choices, and recover when something failed?
I used AI to build this portfolio page, too. I shaped the story, reviewed the generated page, and refined the copy and visuals.
Explore the Copilot case study →What this brings to a team
More ideas explored. Earlier feedback. Thoughtful execution.
Explore before committing.
I use AI to support research and planning, explore several concepts, and turn promising ideas into working prototypes. That gives the team more concrete options to compare before committing to a direction.
Share early and often.
I can put working ideas in front of clients, product, and engineering sooner. Seeing and trying an interaction gives us something specific to discuss, challenge, and improve together.
Keep quality consistent.
Shared context, design principles, and automated checks help me keep work consistent across tasks. I review the evidence, question the output, and make the decisions that shape the result.
For a team, this means a shorter path from an early question to something we can evaluate and build. AI expands how much I can explore and execute. I remain responsible for the reasoning, the design decisions, and the quality of the work.