Building ideas & experiments
These are (a few) of my experiments in code and craft. Even as a design leader, I believe the only way to master the medium is to build. This space is dedicated to hands-on exploration, using AI to translate ideas into reality.
Principles for designing with AI
For decades, engineering bandwidth was the ultimate bottleneck in software. Today, AI has flipped that equation. When working software can be generated in hours instead of weeks, the hardest question in product development shifts from "how do we build this?" to "should we build this at all?"
Prioritize "should we?" over "can we?"
AI accelerates the messy, divergent side of the design process, generating endless layouts and prototypes in seconds. That volume makes convergence harder, not easier. And even as software becomes nearly free to produce, human attention remains fixed. The designer’s highest-leverage job is ruthless curation.
Treat code as the clay
Static mocks can hide awkward transitions and missing states, but working software forces immediate honesty. Using code as a core design medium—whether feeding a design system into an HTML generator to explore a dozen live variations side-by-side or pushing final polish to production—closes the gap between intent and what users actually experience.
Build around the bottlenecks
When bespoke tools take hours instead of quarters to build, you don’t need to wait on a roadmap or a third-party vendor to fix a workflow gap. Designers and researchers can spin up small, purposeful internal apps—from live research dashboards to custom prototyping environments—and share them for teammates to use, fork, or remix.
Demos over memos
Fast solo prompting in a terminal makes it easy to drift out of sync with your team. Counterbalance independent speed with tighter, higher-bandwidth collaboration: swap lengthy written specs for live code prototypes, trade weekly big-room reviews for frequent small-pod critiques, and explicitly log visual decisions so context isn't lost in commit history.
Protect unhurried incubation
Just because AI can draft twenty layouts in ten seconds doesn’t mean every minute of the creative process should be optimized. Use the hours saved on mechanical execution to fund deeper architectural debates with teammates—and guard quiet time away from the screen for the synthesis and intuition that can’t be rushed.
Continuously molt your process
As model capabilities leap forward every few months, static playbooks expire quickly. Stay willing to shed legacy deliverables, rethink handoffs, and move fluidly between setting macro product strategy and dropping into the details to build the next version yourself.