I've been a designer on the Consumer Copilot team at Microsoft for the past 2 years. In this time, I've worked on a lot of different features across the product, including different platforms (Web, iOS, Android, Edge Browser, Windows, and Mac). Here are some of the highlights that have shipped:
Copilot Tasks
Copilot Tasks builds on Copilot Actions, extending Copilot from in-session action flows to background task automation across apps and websites.
I've been working on Copilot's agentic capabilities since January 2025. In just a year, the platform has grown from early web actions into a broader stack spanning computer use, code execution, file generation, connectors and MCPs, and memory and personalization.
I was one of two designers on Tasks, designing the user experience across web, iOS, and Android. My focus was user control and transparency: clear intent setup, visible task progress, and clear boundaries for what the system remembers and personalizes.
I also designed for error states and recovery paths when tasks fail or need user intervention, plus consent UX before high-impact actions like making payments or logging into websites.
This work was done in close partnership with PM and engineering to define trust boundaries in the interaction model, shifting Copilot from answering questions to reliably getting work done while staying observable, interruptible, and reversible.
Read more in the Microsoft announcement and The Verge's coverage.
Code Execution
Code Execution allows Copilot to execute Python code to answer questions or complete tasks. My work involved designing the chain of thought prompting for code execution, as well as the UI for the code execution results. The output of code execution can be a file (documents, images, etc.) or a text response.
Suggested Follow Ups
Contextual prompts that appear at natural conversation breakpoints, helping users explore topics further.
Designed to be concise (10-12 words), conversational, and non-intrusive. They appear with subtle animations to feel organic to the flow.
Reached 1M daily clicks, increased session duration by 15%, and the pattern was adopted by 8+ partner teams.
Suggestions reveal themselves gradually with haptics, designed to encourage curiosity without overwhelming.