How I design with AI, and how I lead.
I’ve shipped seven AI and ML agents into dense enterprise workflows and led the zero-to-one strategy for an AI governance product. These are the beliefs that came out of that work.
Designing AI products people trust
7 AI agents, 4 products
Generative AI and automation agents at ServiceNow, built into workflows people already used, so adoption didn’t depend on learning a new tool.
AI Control Tower
Inception-to-launch design strategy for enterprise AI governance: how companies see, approve and oversee the AI running across their business, with people in the loop.
Agentic, with a manual path
Every AI workflow I design has a fallback a person can take over. The playbook model in Foundation for AI was built for exactly this.
Helping, not taking over
A principle I first wrote for Google’s Herald in 2019, long before agents, and the one I still design against.
How my process changed
The biggest shift is speed to a first concept. That frees time for the work AI can’t do: framing the problem, choosing between options, and checking the output against real users.
My tools, by job
Production design: Figma.
AI standards I’d set for a team
AI drafts, people decide
AI can generate options, summaries and first drafts. A designer owns every decision that reaches a customer.
Every AI-made claim gets checked
A research quote, a number or a pattern from AI is verified against the source before it goes in a deck.
Design for the handoff
For every agent we ship, we design what the person sees when the agent is unsure, wrong or stopped.
Measure trust, not only usage
We track how often people accept, edit or override the AI, alongside adoption.
How I lead
I went from staff designer to senior design manager at ServiceNow through three promotions, and grew two distributed teams along the way, from 1 to 9 designers and from 1 to 4.
Critique as a habit
Daily critiques and lightning demos during sprints. Critique sharpens the craft, and it’s where I coach.
Designers own the call
I make working-group designers the decision-makers and hold them accountable. I step in on strategy and on the hardest flows.
One plan with engineering
Engineering spikes to prove feasibility early, and shared T-shirt sizing so design and engineering agree on scope.
Options, not a single answer
Leadership gets three horizons with their tradeoffs. I’ve run more than 18 executive vision demos that shaped roadmaps.
Data first
Problems framed with telemetry and customer evidence before anyone argues about pixels.
Credit the team
Many hands, one mission. The work on this site was shipped by teams, and I name my role in each story.
My first 90 days
If I joined your team, this is how I would start.
- Days 1–30
Listen
1:1s with every designer and key partner. Review telemetry, support tickets and research. Ship one small fix myself, to learn the tools.
- Days 31–60
Focus
A short readout of what’s working and what’s in the way. Set up critique and decision rituals. Agree on two or three measurable design goals.
- Days 61–90
Show
A north star prototype with options and tradeoffs, a roadmap aligned with product and engineering, and the first wins shipped and measured.
Writing
Where I think enterprise design is going.
Design the playbook, not the screen
Essay · October 2026
For twenty years, enterprise design has been organized around pages: a list view, a detail view, a dashboard. Every new feature meant a new page, and every page drifted a little from the last. At ServiceNow I watched this happen to a product that was growing fast. Each redesign fixed a local pain and created downstream inconsistency.
The fix we found wasn’t a better page. It was a smaller unit. We broke the work into steps, small functions that each solve one thing, like finding inactive users or reclaiming a license. Steps combine into goals, like “minimize spend,” and goals combine into playbooks that adapt to the vendor and the situation.
That model turned out to matter more for AI than for consistency. An agent can’t navigate your information architecture. It can run a step. A person can review a step, approve it or take it over. When the unit of design is a step, people and agents can share the same workflow.
My prediction: within the next few years, the most valuable enterprise design work will be defining these building blocks, their inputs and outputs, and the moments a human needs to see. Screens will increasingly be assembled from them. Designers who think in systems, the way architects do, will lead that work.
Every agent needs a manual fallback
Essay · October 2026
In 2019, designing Google’s internal budgeting tool, we wrote a principle I still use: the user should feel that the machine is helping them, not taking over. Back then the “machine” was a few smart defaults. Today it’s agents that take actions across enterprise systems. The principle matters more now.
Enterprise users don’t adopt AI because it’s impressive. They adopt it when they trust it with work they’re accountable for: a license audit, a budget, a compliance check. Trust comes from three things, and each is a design problem.
Visibility. Show what the agent did and why, in the user’s terms. Control. Let people approve, edit or stop the agent at the step level. A way back. Every agentic flow needs a manual path that works when the agent is unsure or wrong.
This is why governance is becoming a product category of its own. Leading the design strategy for an AI control tower taught me that companies don’t only want more agents. They want to see every agent they run, who approved it, and what it’s doing. My prediction: the products that win enterprise AI will be the ones that make oversight effortless, not the ones with the most automation.
Bring three horizons, not one answer
Essay · October 2026
When a team is under pressure, the instinct is to bring leadership one polished solution. I’ve learned to bring three. On our flagship redesign, demo day had a conservative option that stabilized the business, a progressive one built on modular playbooks, and a vision for an AI-first product. Each came with its tradeoff written down.
Three horizons change the conversation. Executives stop asking “do we like this design?” and start asking “how much change can we afford, and when?” That’s the conversation that resets roadmaps. Ours did: we reset it around modular playbooks, with the AI-first vision as the north star for the next releases.
AI makes this practice more important. When anyone can generate ten concepts in an afternoon, a concept is no longer the scarce thing. The scarce things are framing the problem correctly, choosing what to compare, and being honest about what each path costs.
My prediction: design teams will get smaller and more senior. The designers who thrive will be player-coaches who can still make the work, and who can also explain what to build, what to skip, and why.