AI-native, without giving up Part 11 discipline
AI should compress the boring weeks of a study — setup, chasing, reporting — without ever touching the two things that must stay human: judgment and signature.
"AI-powered" usually means a chatbot in the corner. We mean something more structural: the platform has three AI surfaces, each with an explicit governance boundary.
1 — Design: protocol in, draft study out
Give the assistant your protocol PDF and it drafts the complete eCRF, visit schedule and consent bindings. The draft lands in the same audited configuration pipeline as any human edit; a data manager reviews and approves before anything goes live. Your model key is encrypted and scoped to your tenant.
2 — Operate: agents with exactly your permissions
The EDC speaks MCP, the open protocol for AI agents. An agent authenticated with a human-issued token acts as that human — same role, same site scope, same row-level security, same audit attribution. It can register subjects, chase queries and pull reports; it cannot sign, cannot mint tokens, and cannot see anything its issuer couldn't.
3 — Analyze: your data, your AI analyst
Connect Claude — or your own tooling — and it can pull the complete, permission-scoped dataset in analysis-ready form, run the numbers, and draft the report your team finishes. The data leaves through the same logged, attributable export path as every CSV.
The line that never moves
Signatures require a human re-authenticating with something they know. Tokens are issued by humans, expire, and are audited. Every AI action is attributable to the person who authorized it. That is what "AI fit" means in a regulated system: the acceleration is real, and so is the accountability.