From personal experiment to team capability in four months

Transforming How a Design Team Works with AI

Sr. UX Designer · Amazon · Nov 2025 – Mar 2026

The arc

In November 2025, I started experimenting with AI to help with my own design work. By March 2026, I had architected how my entire team integrates AI into their practice — built the tools, created the learning programs, and proved the approach works by shipping real prototypes.

This wasn't a mandated initiative. There was no AI transformation roadmap. I saw an opportunity, ran experiments, and scaled what worked.

What I learned: the bottleneck is context, not capability

Early on, I discovered that using AI "adjacent" to my workflow — asking it questions, generating one-off outputs — produced mediocre results. The shift came when I stopped treating AI as a tool to query and started designing the context architecture around it.

I built a markdown documentation system to manage all my projects and create a persistent context pool. I created structured memory, process definitions, and expectation frameworks so AI could understand not just what I wanted, but how I work and what quality means in my domain.

The result: consistent, high-quality output that actually fit into my workflow instead of requiring translation.

From my workflow to the team's

New handoff process

I worked with my engineering team to experiment with a new design-to-dev handoff. My latest design handoff was a vibe-coded React prototype using components from our actual design system. Not a static spec — a working, interactive artifact that engineers could inspect and build from.

Tools and extensions

I built MiniPath, custom skills, and extensions that package my optimized workflow into tools the team can use without replicating months of personal experimentation.

Learning program

Rather than a one-size-fits-all training, I designed differentiated learning paths based on job-to-be-done. Different designers have different entry points and goals with AI — the program meets them where they are.

Proving it works

When people across the team wondered whether AI could build certain types of prototypes for their research needs, I jumped in to vibe-code working proofs while others assumed it couldn't be done. Showing beats telling.

What this taught me about AI adoption

The hardest part of integrating AI into a design practice isn't the technology. It's designing the system around it — the context structures, the quality expectations, the handoff patterns, the learning pathways. That's design work, even though it doesn't produce screens.

LeadershipAI adoptionContext architectureTeam enablement