Agentic Experiences
Most AI products are still built around a question-and-answer loop. The work that interests me most is what comes after that: interfaces where the AI is doing things on your behalf, not just telling you things.
Across Copilot and Bing/Edge, I've led design on that shift, from Cowork, which merges chat and multi-step work into one surface that picks up where you left off, to search experiences where the product moves from answering a query to actually carrying a task forward. The common thread is trust: giving people enough visibility and control that handing off real work to an AI feels safe, not like a leap of faith.
This page is a look at that body of work, agentic patterns across search, chat, and task orchestration, and what I've learned about designing for AI that acts.
Copilot Cowork
Cowork wasn't a feature request, it was a bet that AI needs to work with you, not just answer you.
Copilot had two separate surfaces: Chat for quick questions, Cowork for multi-step work. People had to guess which one to open before they'd even started. As one of the design leads on search experiences within AIX (formerly MAI), I worked on merging those into a single entry point, Home, so you can ask a question or kick off a whole project from the same place, and it picks up where you left off instead of starting cold every time.
The other piece I was closely involved with is Today, a proactive surface that pulls in what's happening across Outlook, calendar, Teams, and tasks, so instead of hunting for what needs your attention, it's already surfaced for you.
MY ROLE
Design lead for Home within Copilot, AIX. Drove the merge of Chat and Cowork into a single coherent entry point and the interaction model for the proactive Today surface. One of the design leads shaping how search and AI-native experiences converge across AIX.
V1 launched in January 2026
IMPACT
Agentic AI shipped to consumer search for the first time
Research preview launched 2026, expanding via waitlist globally
Foundation for Microsoft's "second chapter" of AI -- from conversation to action
Just launched V2, so the big usage numbers aren't in yet the way they are for Tasks. What's already showing: as part of this broader push, the team improved speed by 25%, while engagement per user and satisfaction have roughly doubled
V2 launched in September 2026