Case study · Microsoft · Viva Glint

Employee Feedback Agent

An AI agent that listens to employees in conversation, and the setup and reporting that make it safe to roll out.

Role: Senior Product Designer
Team: TODO(content)
Timeline: TODO(content)
Status: Public preview

What I did on each surface, and whether I designed and coded it
Surface	What I did	Design	Code
Admin setup	Campaigns, topics, policies, confidentiality, scheduling, export	Yes	No
Entry point	Finding the agent from Glint	Yes	No
Responder	Conversational experience; legacy UI to Flex	Yes	Yes
Reporting	Narrative, then conversational; design-system prompt	Yes	No

Responder Flex build: merged, rollout in progress.

The problem

TODO(content) Who struggled with what, and why it mattered. Up to 120 words.

What I owned

I designed admin setup, the entry point from Glint, the responder and reporting. I coded the entire responder UI, moving it from the legacy UI to Microsoft’s shared Flex components. TODO(content) Name collaborator roles and what they owned.

1Set up

Controls that make an AI agent safe to deploy

Before an employee says a word, an admin decides what the agent may ask, who sees what, and when it runs. TODO(content) Two decisions on policies and confidentiality, and what each traded off.

2Respond

A responder people trust, then the code behind it

I designed the conversational responder with partner teams on canonical UI and components, and iterated on what responders told us. TODO(content)

Responder, desktop + mobile · TODO(asset)

Built on Flex components · public preview

Then I coded the whole responder, moving it from the legacy UI to Flex: component audits, shared controls in place of custom ones, a responsive desktop and mobile baseline. Copilot helped throughout, grounded in the designs, the codebase and engineering constraints.

3Read

Designing what the AI says, not only where it says it

Reporting moved from AI summaries that open into evidence, to a conversational experience. Both are in public preview. TODO(content)

I wrote the design-system prompt that told an internal AI system how to structure responder data, so it rendered accurately in the reporting components I designed.

How AI was involved

Designed for AI: A conversational responder; reporting that moves from AI summary to evidence; the design-system prompt that fits generated reports to the UI.
Built with AI: Copilot-assisted migration of the responder to Flex, which I coded.
Enabled others: Not part of this work.

Craft details

Responder on mobile · TODO(asset)

Summary opening into evidence · TODO(asset)

Confidentiality setup states · TODO(asset)

Outcome

Status: Public preview. TODO(content): In public preview. Qualitative outcome with one or two concrete examples. No metrics unless sourced.

What's next

TODO(content): What happens to the work after public preview.