Designing the conversational core of an AI assistant
Creating clear, predictable, and trustworthy AI interactions through system-level conversation design.
ROLE
Lead UX Writer
TEAM
Cornerstone OnDemand, Inc.
TIMELINE
2024- Present
THE CHALLENGE
As generative AI was introduced into Cornerstone’s learning platform, teams were moving quickly to embed AI-powered assistance across multiple product surfaces. While the underlying models could generate responses, the experience around those responses lacked cohesion.
AI interactions were being designed independently across teams, resulting in inconsistent messaging, unclear system behavior, and user uncertainty around what the assistant could do — or why it responded the way it did.
The design goal:
Create a consistent, trustworthy conversational framework that helps users understand, interact with, and rely on AI-driven guidance — without slowing product velocity.
Primary obstacles:
​​
-
No shared conversation model to govern AI behavior
-
Inconsistent messaging patterns across product surfaces
-
Unclear affordances, limitations, and system states
-
Difficulty aligning UX intent with model behavior
-
Lack of reusable patterns for prompts, clarifications, errors, and fallbacks
THE PROCESS
Governor's Office
Policy updates & executive orders
Agency Subject Matter Experts
Legal and health review
Structure, plain language, localize
Publish
Live to 40+ millions residents
THE SOLUTION
Architecting for resilience
Action-Oriented Architecture
We shifted from "What is the virus?" to "What do people need to do right now?" I led the restructuring of the Housing & Homelessness and Food Assistance portals to focus on eligibility and immediate application steps.
Inclusive Localization
Translation often breaks design layouts. I built content templates that accommodated text expansion for Spanish, Tagalog, and Traditional Chinese, ensuring the hierarchy remained intact across all 8 supported languages.
THE IMPACT
Easier to localize
Plain-language content reduced translation errors and made product copy more accurate across all regions.
Better drafts, faster
Designers produced clearer first drafts with fewer rewrites.
Smoother engineering handoff
Consistent rules reduced back-and-forth and eliminated implementation gaps.