AI Agents in M365
Making AI agents easy to find, simple to create, and trusted collaborators.
About the project
Company
Microsoft
Timeline
2024 - 2026
My role
UX designer, AI designer
Challenge
Context
I worked on shaping how AI agents become part of everyday work—designing holistic experiences that let people discover the right agent, create their own, and collaborate with agents as trusted teammates. Focused on interaction design, system thinking, and trust-building for AI, ensuring these experiences feel seamless, useful, and human-centered.
My impact is measured in 3 stages:
Discovery
Helping users explore, find and acquire agents in context through smart recommendations and a curated Agent Store.
Creation
Making agent creation approachable with template-first flows, vibe coding-style and a simplified builder experience.
Collaboration
Giving users visibility and control of Agent capabilities that transforms how they collaborate with agents, unlocking new ways to get work done.
The designs
Due to confidentiality, I can only share a limited view of the project. Here are a few examples of the flows and designs our team created.
I’d be happy to walk you through the details and answer any questions during a call.
Agent Discovery
Users were overwhelmed by a growing catalog of agents and unsure which one fit their task. Discovery needed to feel contextual, trustworthy, and lightweight.
What I did
Designed in-chat recommendations for just-in-time discovery.
Created all-new Agent Store, entry points and added trust signals.
Impact
Reduced “catalog fatigue”.
Increased confidence in agent suggestions.
Considerations
Response confidence levels
As the LLM presents recommendations based on user context, different UI is presented to highlight a high-confidence recommendation or be less loud if the confidence level is low.
Agent Creation
Agent Builder had high drop-off rates; users felt intimidated by a blank canvas. Needed a guided, approachable flow that scales from simple templates to advanced customization.
What I did
Introduced template-first flow for quick starts.
Redesigned zero-state with clear guidance.
Impact
Lowered friction for non-dev users.
Improved completion rates in internal tests.
Agent Collaboration
Agents worked invisibly across apps; users lacked oversight and trust. Needed a way to see, steer, and audit agent actions across multiple surfaces.
What I did
Concepted Agent overview for cross-surface activity
Defined task delegation flows for Agents in multiple canvases
Impact
Made AI work visible & steerable.
Influenced Microsoft’s AI Agentic vision.
Contributed to guidance for assistive agents.
Results
Metrics
Higher engagement
Engagement increased with recommended agents.
Increased number of agents created per active user.
Lower drop-off
Reduced drop-off from Store browse to agent activation.
Significant reduction in abandonment during Builder flow after template-first design.
Positive internal sentiment
Positive sentiment in leadership demos.
Next steps
Scale discovery
Explore personalization for agent recommendations and richer context cues to improve trust and relevance.
Deepen collaboration
Expand Agent reach with proactive insights and multi-agent coordination.
Measure and iterate
Continue tracking engagement, completion rates, and sentiment to refine flows and validate design principles.
Key takeaways
Discovery is powered by trust
Users ignore catalogs; they respond to timely, explainable suggestions. Context beats quantity.
Creating is a confidence curve
Start simple, then progressively reveal power. Blank canvases kill momentum.
Collaboration needs visibility
AI “coworkers” succeed only when users can see what’s happening and intervene easily.
System thinking > isolated screens
Designing for AI means orchestrating flows across the system, not just polishing one UI.
