Epic's Scale Meet's AI's Speed
Epic is the largest EHR provider in the U.S., supporting 325 million patient records across 2,600+ hospitals and 80+ applications.
After the release of ChatGPT, in just a few months Epic launched AI tools across its platform - partnering with Microsoft and others to enable summaries, message drafting, and clinical support.
By 2025, 330+ orgs were live with these tools, 60,000+ clinicians used them, and 50M+ outputs had been generated.
A system this large moving this fast needed design standards that could keep up.
Epic “Health Grid” (80+ apps across)
Source: showroom.epic.com
The Challenge: 80+ Apps, 1 AI Experience
Each of Epic’s 80+ apps were building AI features on separate timelines, with different scopes and inconsistent UIs. Without shared standards, experiences were fragmented, confusing, and hard to scale.
Developers lacked reusable components. Designers had no guidance on interaction patterns. Teams were solving the same problems in silos, with no clear rules for building, validating, or improving AI.
In a system this large, speed without alignment risked trust and slowed delivery.
Epic needed a unified foundation: reusable tools, consistent UX patterns, and clear principles to scale AI responsibly across the ecosystem.
The Solution: Creating Epic's AI Design System
With teams moving fast, I worked with another designer to lead the creation of an end-to-end design system - defining patterns, crafting a distinct visual identity, and building scalable guidelines for implementation.
Defining Core AI Patterns
I began by auditing in-flight and upcoming AI features. Despite the variety, nearly all features mapped to four core interaction patterns:
This pattern-first framework helped teams speak a common language, make design decisions faster, and stay aligned as new features emerged. It also provided a clear foundation for reusable UX and development.
Establishing a Distinct Visual System
To maintain user trust, we needed a clear, consistent way to signal when content was AI-generated. I developed a new style system that felt modern and differentiated from the core clinical UI:
The goal: make AI outputs recognizable, usable, and appropriately cautious - without overwhelming the clinical experience.
Case study in progress