Scaling content design expertise with AI
What I did: Turned content-design expertise into an AI-powered agent that gave Microsoft product teams on-demand UX writing guidance grounded in established standards.
Role: Lead content designer
Impact: 75% less time per request, 60% more requests supported
The challenge
Content design expertise was highly valued within Microsoft, but difficult to scale.
Limited resources: A small number of content designers supported a large org
Scattered guidance: Style guides and documentation were spread across 14 sources
Speed + quality: Teams were asked to "move at the speed of AI," while language still needed to meet Microsoft standards for voice, clarity, usability, and accessibility
The vision
Build an AI-powered content design agent that could scale expertise across the organization without sacrificing quality, security, or trust.
Grounded: Base responses on established Microsoft content standards and expertise
Trusted: Meet enterprise requirements for privacy, security, and compliance
Scalable: Give designers, PMs, and engineers access to content guidance on demand
Setting the strategy
What began as a hackathon experiment gained momentum, funding, and a mandate to explore how AI could scale content design across Microsoft.
Align: Led a 90-minute FigJam workshop with 17 participants, capturing 139 data points on current AI use, partner needs, specific JTBD, and future features
Synthesize: Translated workshop findings into a focused product strategy and priorities
Lead: Selected as lead content designer as the project became a funded initiative
From strategy to prototype
A dedicated team moved quickly to produce a working agent ready for testing. I led content design alongside PM, engineering; hired two contractor content designers.
Integrated: Built within "Fluent" ecosystem for security, privacy, and compliance
Scalable: Grounded the agent in a centralized "universal knowledge base"
Useful: Added Flesch-Kincaid scoring and word counts for content design workflows
Training the model
Established an evaluation process to measure output quality, identify gaps, and improve the agent as models and guidance evolved.
Establish a baseline: Started with 50 manually written questions and assertions to define expected behavior; wrote fixed set of golden prompts to compare results across model versions and competing tools including Copilot and ChatGPT
Build the eval framework: Wrote 60 evals across 6 categories (grammar, accessibility, voice & tone, etc.); used GitHub Copilot to generate 60 more to scale coverage
Refine the grounding: Iterated on markdown system prompt to improve responses
Improve quality: Increased eval success from 69% to 84%
See it in action
A 90-second internal demo created to show product teams how to get started and use the agent to write, revise, and evaluate UX content using Microsoft content standards.
Scaling the experience
We expanded the agent into more workflows and made it easier for teams to adopt.
Expand: The team extended the agent from browser-only to a Figma plugin and Microsoft 365 app store, bringing content guidance closer to existing workflows
Enable: Created an internal SharePoint hub with install instructions, Agent 101 guide, FAQ, prompt examples, demo video, and feedback collection
Reach: Helped expand access from 268 employees to 1,000+ across multiple orgs
The Results
The agent demonstrated that content design expertise could scale with speed and capacity.
Work faster: Increased capacity while maintaining content quality
Move quickly: From project greenlight to user testing in 90 days
Build for the future: These agents laid the foundation for a broader Microsoft language system embedded across AI experiences
The goal wasn't to replace content designers. It was to scale their expertise, automate repeatable work, and give them more time for strategic work.
What people are saying
