How Epic Systems Uses Claude Code to Bring AI Development Beyond Engineering

Epic Systems sits at an unusual intersection: it builds healthcare software at extraordinary scale — MyChart alone serves 195 million patients worldwide — while operating as a famously in-person company that prizes hallway conversations and break room ideas. That culture turned out to be an unexpected accelerant for AI adoption. When Seth Hain, SVP of R&D, decided to give every employee access to AI tools without governance gatekeeping, the natural social infrastructure of the company did the rest.

For years, AI at Epic functioned the way it does at most enterprises: useful in narrow windows. A physician saved time on clinical notes. A developer found documentation faster. But the net effect on any given person’s day was limited — time saved in one pocket rarely translated to genuinely new work. AI was a quality-of-life improvement, not a capability expansion. The deeper question was whether it could become something more.

The answer came from removing barriers rather than adding structure. Epic built internal tooling to surface who was using AI and what they were building, creating a visibility layer that let ideas travel fast across an in-person organization. Leadership didn’t just endorse AI from a distance — they spent time in it daily, and required development leads to have hands-on exploration sessions on Friday afternoons. The goal wasn’t coverage or compliance; it was engineering genuine excitement that would spread organically, the way good ideas always had at Epic.

The most striking result came from an unexpected direction: a pharmacist — with no developer background — used Claude Code to build a fully functional, interactive prototype of new MyChart features. Instead of writing a requirements document or a slide deck, he built something his colleagues could actually explore and iterate on together. That moment captured what Epic had made possible. Today, more than half of Claude Code usage at Epic comes from non-developers. The people who understand clinical workflows most deeply — and who have historically had no path into the software development process — can now build what they’re imagining.

The most fundamental shift, Seth notes, has come in recent months: agents are now capable of contiguous, uninterrupted work. Where before a human had to intervene every few steps, agents can complete long stretches autonomously and hand back a finished result. A patient care access team built a shared Claude project that functions, in Seth’s words, “as part of the team, not just as an assistant” — surfacing status updates as living Claude conversations rather than static reports. With 85% of Epic’s healthcare organization customers already running AI in production, and Epic building medical event models trained on real-world health records, the trajectory points toward a new model for clinical software: built by clinicians and engineers working side by side, with AI as the medium between them.