Microsoft Learning Agent

Reskilling is the real product, AI is the medium.

I lead UX for Microsoft's AI-powered Learning and Skilling experience, the Learning Agent. The job is bigger than a chatbot. Millions of people now have powerful AI sitting next to their work, and most of them have no idea what to ask it, when to trust it, or how to fold it into the way they already work. My role is to design the experience that closes that gap, so people do not just use AI, they get measurably better at their jobs because of it.

I own the arc end to end: strategy, interaction models, the design system, and the research that tells us whether any of it is landing. I partner daily with product, engineering, and data science to co-create the roadmap and ship AI-driven features that scale across Microsoft 365 surfaces like Teams, Outlook, Word, and Excel.

Four pain points, four behavior changes.

Rather than design features in a vacuum, I framed the work around four pillars, each one a specific place where people stall with AI and a specific behavior we want the Learning Agent to drive. This became the shared language for the whole team and the structure I prototyped against for stakeholders.

Interaction models that teach while people work.

I defined the core interaction models for the agent: conversational UI, personalized recommendation surfaces, and in-flow guidance that meets people inside the task instead of pulling them out to a separate course. Three patterns did most of the heavy lifting.

• Contextual entry prompts. Small, well-timed cards that suggest an AI move at the exact moment it is useful, so discovery happens in the flow of real work.
• Citation chips with confidence. Inline chips that show what an answer is grounded in and how sure the system is, turning a black box into something a person can verify.
• Human and AI task split. A before and after view of a workflow that makes the redesign legible: here is what you still own, here is what the agent now handles.

To keep these honest, I built high-fidelity, interactive prototypes that lived inside the real product's design system, so stakeholders reacted to something that felt shipped rather than to a slide.

Engagement mechanics, grounded in behavior.

Learning only compounds if people come back. I designed personalization and gamification mechanics, streaks, XP, and weekly challenges, driven by real behavioral and usage signals rather than vanity badges. The point was momentum: small, visible progress that makes the next session feel worth it, tuned by what the data showed people actually responded to.

I built the instrument that measures whether it works.

Design at this scale is only as credible as the evidence behind it. I built and maintain a pilot dashboard that brings every signal into one place: survey scores, in-product thumbs feedback, focus group findings, usability results, and enrollment across pilot cohorts. It uses top-two-box scoring and clear status thresholds so leadership can see, at a glance, what is healthy and what needs attention.

That dashboard is how design recommendations get made and defended. It turns scattered feedback into a single, honest read on the program, and it is what moved the work from a pilot into scaling planning.

The pilot dashboard I built and maintain. Cohort labels are anonymized and figures shown here are illustrative, the live program data is internal to Microsoft.

From scattered feedback to a program that scales.

Beyond the screens, I established a UX design feedback process across product teams, running design reviews, stakeholder workshops, and shared roadmap planning with PM, engineering, and research. I also translate dense system constraints into clear strategic narratives through prototypes, playbooks, and executive presentations, the artifacts that get leadership to a decision faster.

The result is a design practice that does not just produce interfaces, it produces evidence. The Learning Agent has moved past early piloting into scaling planning, with new organizations coming online, because the work can show its value, not just assert it.