Staff AI Product Designer · Gemini Assistant

Designing AI
people trust.

✦ Prepared for Saba Zaidi • Gemini App Design, Google DeepMind

Staff AI Product Designer · Gemini Assistant

Designing AI
people trust.

✦ Prepared for Saba Zaidi • Gemini App Design, Google DeepMind

Professional narrative

15 years designing products
millions of people love.

A founder and product designer in the Bay Area, building human-centered AI and mobile experiences with exceptional craft, judgment, and systems thinking.

15 years · shipped at scale

Teams I've built with.

Career trajectory

From sharing, to scale,
to autonomous systems.

Every step pulled toward one problem: making powerful, unpredictable technology feel safe and human.

2010

Cortex

Sharing tool, acquired by Backplane

2011

Little Monsters

Co-founded with Lady Gaga

2018

Wing @ Google X

First FAA drone delivery

2024+

AI products

January AI, Ditto, on-device AI

2024

LinkedIn Premium

Design Lead, 100M+ reached

Design philosophy

Three beliefs I design by.

The same three things this team is hiring for. They are how I have worked for years.

01 · Ambiguity

Drive clarity in the unknown

I don't wait for requirements. I seek the why behind a system's behavior, then design the mental model people can hold.

02 · Trust

Trust is the product

With probabilistic systems, how it feels matters as much as what it does. I design for transparency, control, and graceful failure.

03 · Velocity

Prototype to think

I start with the screen, not the doc. A clickable artifact in hours turns debate into direction with product and engineering.

The thesis

You can't QA your way
to trust.

A probabilistic system will surprise people. Trust is not the absence of error, it is how the experience behaves around uncertainty: what it reveals, what it lets you control, and how it recovers. That is a design problem, and it is the one I keep choosing.

Project walkthrough

Wing @ Google X

Designing the first FAA-approved drone delivery experience, for a behavior people had never performed before. Lead Product Designer, consumer mobile.

Pillar 1 · Clarity in ambiguity

No mental model. No precedent for "good."

Getting a normal person to trust an autonomous aircraft that flies to their house, drops a package, and leaves. Nothing but unknown-unknowns, and it had to read the same way to the customer, the pilot, ops, regulatory, and the FAA, all at once.

My role

I owned the whole
order-to-doorstep arc.

Across PM, mobile engineering, flight operations, hardware, and regulatory, I led three things end to end.

iOS + Android

Order-to-handoff flow

Live view

Drone intent, shared with the FAA

Every abort

Failure-state language

Pillar 2 · The trust ceiling

Too little and they panic.
Too much and they're lost.

There is a narrow band of "just enough," and we found it by watching people, not guessing. We showed position, altitude band, and descent, and nothing that read as surveillance. Confidence is a dial you tune, not a number you dump on someone.

Pillar 2 · Design the failure path first

Design the abort
before the happy path.

People read any abort as "the drone is broken and might be over my house." So we rewrote every failure in the customer's frame, plain and reassuring, with a refund in the same screen:

"Your order is delayed because of wind. We'll reorder for you."

Sentiment moved from "I'm nervous" to "OK, that makes sense" in one round. The exact muscle AI needs for hallucination and error states.

Pillar 2 · Control at the doorstep

You choose
the spot.

The last fifty feet are where trust is won or lost, so we handed the customer the controls: drop a pin on your own yard, clear the spot, and watch the package come down exactly where you asked.

Control isn't a setting buried in a menu. It's the moment the system asks before it acts.

Choosing and clearing the delivery spot in OpenSky

The constraints I designed against

Four forces, one calm screen.

Regulatory

The FAA had to read what the customer saw. No UI could anthropomorphize the drone. "Your drone" was fine; "she's on her way" was not.

Hardware

The aircraft couldn't hover indefinitely or wait out wind, so "give us a few minutes" was never an option.

Mental model

People expected a courier. We had to teach a new metaphor without ever making it feel weird.

Trust ceiling

Hide the drone and people panic; show too much and they're confused. The whole design lives on that line.

The decision that made it land

Call it a courier,
not a robot.

Drone delivery was a strange new thing. But everyone already knows what a courier is: someone who brings your package to the door. So we used that, instead of explaining "autonomous aircraft."

Familiar words make an unpredictable thing feel safe. We saved the wow for the doorstep, where the drone lowering your package did the talking. The same move an AI assistant needs.

A Wing drone lowering a package to a customer's doorstep

The work behind it · In the press

The press noticed the trust, not just the tech.

Coverage in The Verge, Bloomberg, TechCrunch, and WIRED

Outcome

First FAA-approved drone
delivery in the U.S.

Live deployments serving real food and medications in real neighborhoods. The consumer-trust model became a reference for later Alphabet aerial-delivery work.

4.8★

OpenSky · 782 App Store ratings

First

FAA-approved drone delivery, U.S.

0 tickets

Abort flows, within one design round

At scale · LinkedIn Premium

Prototype before brief,
for 100M+ professionals.

As Design Lead on Premium Mobile, I shipped the LinkedIn × Calendly integration: one booking pattern across profile, post, and message that deletes the "what time works?" thread entirely.

100M+

Members reached

<24h

Concept to prototype

How I move

I start with the screen,
not the doc.

A clickable, opinionated artifact in hours that a PM, partner, or exec can steer in the same meeting. The booking integration went concept to prototype in under a day. That rhythm is now how the broader Premium design org moves.

Motion & interaction

Prototyped in motion,
not in slides.

I design the feel of an interface with AI tools like Claude Code, so a transition, a loading state, or a model thinking reads as calm and intentional long before any production code.

Motion is how a probabilistic system shows it heard you.

Breadth

Two decades of shipped work.

Founding-era social, autonomous systems, AI health, self-custody finance, wearable couture. The throughline is craft, trust, and making complex things feel simple.

What I'd bring to Gemini

Autonomous trust is AI trust.

Show just enough to stay calm

→ Transparency. Show just enough of what the model knows, and how sure it is, so people stay calm and in control.

Design the abort first

→ Hallucination & error states. Recover in the user's frame, so a wrong answer never breaks trust.

Write in your voice

→ Memory. Carry context across a conversation so Gemini stays personal and consistent, without ever reading as surveillance.

Altitude becomes a lane

→ Systems thinking. One coherent model across Gemini's surfaces and modalities, optimized for AI's probabilistic nature.

Prototype to think

→ Rapid prototyping. Simulating model responses with AI tools like Claude Code to make probabilistic features feel intentional.

Clarity in the unknown, trust as the product, velocity to prototype. The exact muscle a Gemini assistant needs.

Testimonials

What people say.

Silicon Valley leaders, on what it's like to work with me.

"An active, engaged and passionate AI-centric designer."

Sarah Alpern

VP of LinkedIn, Head of Design

"Epitomizes what it means to be an AI-native, forward-thinking designer."

Matt Geller

Chief of Staff, LinkedIn

"One of the most productive and passionate designers I have had the chance to work with."

Rose Yao

VP of Product, Google

"A rare combination of design talent, product sense, AI fluency and collaborative energy."

Neha Jain

Engineering Leader, LinkedIn

"One of the strongest visionary design partners in our organization."

Matthew Grieco

Sr. Product Design Manager, LinkedIn

"Came to the team prepared with ideas to improve the user experience."

Jen Devins

UX Director, Google

Vibe-coded playground

Live experiments,
vibe-coded with the models.

Self-initiated prototypes of how AI should feel: agentic, human-in-the-loop, and calm. All live. Click any one to try it.

Crit ↗

A Claude Code skill that runs a sharp, prioritized design review of any UI, ranked from blocking to polish, each with a fix.

Life AI ↗

An agent that learns your habits and handles recurring tasks, with one quiet confirmation before it acts.

Loop ↗

Keeping a human in the loop on autonomous coding agents, approving each step before it runs.

How I built it

The prompt and plan behind it.

claude-code — gemini-os

›Create me a plan file for an OS-layer Gemini assistant: I swipe up over any app and it has already read the screen, tags what it saw, and stitches one action across Messages, Maps, and Calendar, with an autonomy dial (Suggest, Confirm, Auto) so trust is a visible control. Then interview me in detail using the AskUserQuestion tool about literally anything: technical implementation, UI and UX, concerns, tradeoffs, whatever you need.Create me a plan file for an OS-layer Gemini assistant: I swipe up over any app and it has already read the screen, tags what it saw, and stitches one action across Messages, Maps, and Calendar, with an autonomy dial (Suggest, Confirm, Auto) so trust is a visible control. Then interview me in detail using the AskUserQuestion tool about literally anything: technical implementation, UI and UX, concerns, tradeoffs, whatever you need.

Interviewed me: scope, privacy stance, autonomy defaults, failure states

Wrote PLAN.md, one bet: ambient context awareness

Built index.html and deployed to gemini-os-2zu.pages.dev

PLAN.md

The core bet: ambient context awareness. Because the assist lives below the apps, it sees your screen and never needs to be briefed.

Control: an autonomy dial makes the tradeoff a designed, on-screen object. Default to Confirm, human in the loop.

Privacy: turn the hardest objection into a feature. Show exactly what it read, keep it ephemeral.

Thank you

Let's design AI
people trust.

officialjp.com

Designing AIpeople trust.

Designing AIpeople trust.

15 years designing productsmillions of people love.

Teams I've built with.

From sharing, to scale,to autonomous systems.

Three beliefs I design by.

Drive clarity in the unknown

Trust is the product

Prototype to think

You can't QA your wayto trust.

Wing @ Google X

No mental model. No precedent for "good."

I owned the wholeorder-to-doorstep arc.

Too little and they panic.Too much and they're lost.

Design the abortbefore the happy path.

You choosethe spot.

Four forces, one calm screen.

Regulatory

Hardware

Mental model

Trust ceiling

Call it a courier,not a robot.

The press noticed the trust, not just the tech.

First FAA-approved dronedelivery in the U.S.

Prototype before brief,for 100M+ professionals.

I start with the screen,not the doc.

Prototyped in motion,not in slides.

Two decades of shipped work.

Autonomous trust is AI trust.

What people say.

Live experiments,vibe-coded with the models.

The prompt and plan behind it.

Let's design AIpeople trust.

Designing AI
people trust.

Designing AI
people trust.

15 years designing products
millions of people love.

From sharing, to scale,
to autonomous systems.

You can't QA your way
to trust.

I owned the whole
order-to-doorstep arc.

Too little and they panic.
Too much and they're lost.

Design the abort
before the happy path.

You choose
the spot.

Call it a courier,
not a robot.

First FAA-approved drone
delivery in the U.S.

Prototype before brief,
for 100M+ professionals.

I start with the screen,
not the doc.

Prototyped in motion,
not in slides.

Live experiments,
vibe-coded with the models.

Let's design AI
people trust.