Index

Microsoft CoAgent

An AI Co-Agent that boosted PM–researcher collaboration 3×.

Role
End-to-end design
Timeline
6 months
Team
1 UXR, 3 designers

A UC Santa Cruz capstone with Microsoft's innovation team that boosted work efficiency by 8.3%.

CoAgent side panel inside Microsoft Word

Problem

Great technology wasn't becoming great products.

Microsoft Research turns AI into products, but turning breakthroughs into user value was slow and often failed.

Our solution · Microsoft CoAgent. An AI agent in Word that drafts and revises product specs, so PMs and researchers stay aligned.

Research

15 people. Three pain points.

Hour-long interviews with researchers, designers and PMs on Microsoft's innovation team.

“Not Invented Here”
“Projects might be terminated due to their P&L evaluations, which often deem them too risky.”
P6, Product Manager
User-centric design is a lone voice
“I am the only person who does what I do.”
P1, Researcher
Ideas aren't always the right ones
“Users know what they want, but they don't know the best approach to get it.”
P7, Research Engineer

Findings

Three issues stood in the way.

Illustration of two teams connected by a document
Cross-team collaboration
Illustration of a team meeting
Power dynamics
Illustration of two people sharing a chart
User value vs. business goals

Focus

The gap was between PMs and researchers.

How might we help PMs decide with confidence, backed by research and tied to business goals?

Illustration of a product manager
Product manager
Assumes user needs without validation.
Can't easily find or use research.
Illustration of a UX researcher
UX researcher
Can't track their impact.
Struggles to turn insights into priorities.

Concept

Built into Copilot, inside Word.

Where PMs already write their specs.

Illustration of a document shared between two people
Turn PM requirements into UX deliverables.
Illustration of two people and an impact map
Map insights to business impact.
Illustration of a person with data visualizations
Visualize data PMs can act on.

Testing

Tested early, with real users.

A lo-fi prototype with 1 PM, 1 UX researcher and 3 users.

Usability testing protocol and lo-fi prototype screens
Focus on UX outputs
“The tool should be more focused on UX outputs, such as user needs documents.”
P1, UX Researcher
Start from the PRD
“It should be positioned from the PM's perspective, based on the product requirement document.”
P2, Product Manager
Show examples, add shortcuts
“Provide example values for each metric. Include shortcuts in the chat box.”
P3, Engineer

Iterations

Key frame iterations guided by user and AI insights

Earlier design with many features
Before
Too many irrelevant features for the PM
Users think this feature should focus on the PM’s perspective and common workflows, excluding elements like the user journey they don’t use.
Goal-oriented PRD template
After
Goal-Oriented PRD Template for PMs
The template is tailored to the PM’s perspective, making the PRD more goal-oriented and providing a clear, structured guide.
AI chatbot covering the document
Before
AI Chatbot Hinders Document View
The AI chatbot appearing in the center of the page makes it difficult for users to view their original document simultaneously.
AI chatbot in a side panel next to the document
After
Side AI Chatbot Improves Document View
Positioning the AI chatbot beside the product requirements document makes it easier for users to compare the content and ask questions to the AI.

Outcome

Each feature, recorded in Word. Click a recording to open it full size.

Feature 1 · Assumption Generation
Screen recording: CoAgent generating assumptions in Word
ProblemPMs often base user assumptions on marketing reports, risking misaligned product decisions.
SolutionCoAgent addresses this by generating data-driven assumptions from user insights. It processes user data collected by researchers to provide PMs with grounded, actionable insights.
Feature 2 · UX Metrics Visualization
Screen recording: CoAgent turning research into Jobs-to-Be-Done
ProblemUXRs struggle to translate text-heavy findings into visual, actionable formats tailored to a PM’s needs.
SolutionThe feature analyzes user research to identify themes and motivations, converting them into actionable Jobs-to-Be-Done statements that link user needs to business goals for PMs.
Feature 3 · User Insights and Business Goal Mapping
Screen recording: mapping user insights to business goals
ProblemPMs struggle to align user stories with business goals and find it hard to align with UXR’s work.
SolutionThis feature uses business goals, user insights, and data to create an actionable map, aligning user pain points with business impact and bridging goals to solutions.

Impact

Faster specs. Better decisions.

Surveyed 20 PMs and 10 UX researchers.

25%
Less time on specs
About 5 hours saved per PM, every week.
85%
More confident
Of PMs felt more confident deciding.
32%
Faster alignment
Quicker to finalize requirements.

Reflection

Broad scope, real constraints.

Recruiting Microsoft PMs was hard, so testing with peers filled the gap. Next, we're publishing this work as an academic paper.

Video call with the project team and our Microsoft sponsor