Designing beyond the chatbox.
I’m a product designer at Meta, where I design and ship agentic interfaces for , our internal AI platform used by more than 70,000 employees. I design in code, which lets me see the product working as I go and maintain control over the interaction and experience through to what ships. I currently rank among Meta’s top three designers by code shipped.
At Meta Superintelligence Labs, I helped evolve Metamate from a general-purpose chatbot into an operating system for work: a persistent, extensible environment where employees can organize shared projects, carry context forward, teach agents new capabilities, delegate recurring work, and produce artifacts that meet the standards of different teams and brands.
I prototyped Metamate’s end-to-end product architecture and key experiences directly in code. The result was a into which its agentic capabilities could ship. The larger challenge was identifying the primitives Metamate needed to become a genuine productivity platform.
One of those primitives was reusable capability. Valuable agent behaviors were being created across Meta, but employees lacked a coherent way to find, trust, install, use, and maintain them. I designed the across its full lifecycle: setup, authoring, evaluation, publishing, discovery, installation, usage, versioning, and maintenance. That foundation now supports more than 80,000 skills across the company.
I applied the same systems thinking to , designing how employees create recurring work conversationally, discover workflows relevant to their role, and choose the appropriate level of human oversight. In five months, those workflows reached 386,000+ executions across more than 8,000 people.
Artifact generation revealed a different kind of infrastructure problem. Metamate could produce presentations, documents, and interfaces, but they were often technically complete and visually generic. I designed and built the end to end in code: the editor, structured instructions, skills, routing, and an evaluation harness.
Making that quality repeatable meant seeing what happened inside every generation. Through , I could identify where guidance was being lost and redesign how it entered the model’s trajectory. Missed guidance fell from 62% to 33% while maintaining a 4.5/5 quality score, and the system now supports artifact creation across the company.
Beyond production systems, code is also how I explore. I prototype real interactions, model behavior, and edge cases, turning unfamiliar ideas into before their patterns are settled.
One such example is a side project where I’ve been experimenting with agentic patterns beyond chat. Product thinking branches across notes, documents, and conversations, and pulling those threads back together gets harder as context drifts. I built to test how an interface could support divergence and convergence while keeping the trail of ideas intact.
I take the same experimental approach into client work. For one project, I used React Native and OpenAI to turn contractors’ handwritten codes into structured information they could manage and share through an .
When I encounter a real problem, I use code to build the thing that solves it. Zakat requires valuing cash, gold, silver, property, stocks, and crypto against a threshold that changes with the market. I turned that moving calculation into an interactive tool for , with live prices, support across asset classes, and a running total as people add what they own.
Side projects give me another place to prototype complex interactions at full fidelity. For an AI-native video editor, I worked through the complete flow from prospect research and script generation to editing and personalized video output. That system became , making tailored demos scalable rather than requiring each one to be produced manually.
These experiments sit alongside a longer history of shipping consumer products. Before Metamate, I spent three years at Meta designing checkout experiences for Meta Pay and two years on Ads Manager. Before Meta, I shipped the iOS and Android apps for and led design systems work across CBS’s interactive properties.
Leadership
Shipping products is great, but helping others grow is even more meaningful to me. I offer pro-bono mentorship to designers working toward their first role and people transitioning into UX. At Meta, I run design sessions, where I have trained more than 100 designers to prototype and contribute through code, with additional designers reached through recorded playbacks and shared materials.
How I work
I start with the person, not the pattern.
Before I reach for a component, I want to know the moment someone is in. A person sending money they can't afford to lose. Someone who'd automate half their job if they knew how. Patterns are what's left over once you've understood enough people.
I build what I design.
A spec describes a product. What leaves my hands is code. Building it is how I find out where it feels wrong, and what I got away with on paper. Living with what I propose has made me propose better things.
Useful is the floor.
I care, probably more than is reasonable, about the parts of a product nobody can point to or thank you for. They're what make someone stop and smile, and they're the difference between good and great.
A new material asks for a new form.
For a generation, architects dressed steel buildings up as stone ones. Software that thinks, that waits and reasons and asks permission and works while you're gone, is a new material too, and chat is the stone we keep putting on it. None of us knows yet what shape it wants. Most of my work now is looking for that shape.