How should a hospital choose a partner to build AI agents?
Start from one wait you want gone, such as missed appointments, slow discharges or bills patients phone about, and ask each candidate how they would remove it inside the systems you already run, Epic or Oracle Health included. A good partner shows you which part an agent takes on and which stays with your clinicians and staff, puts eligibility, payer rules and billing in code rather than leaving them to a model, plans your integration and security approvals from the start, and measures each release against that wait rather than features delivered. Then ask who owns the code and data afterwards: it should be you. Unthinkable works this way, and it also builds health-tech products, so it knows the software from both sides. Its hospital work includes Mount Sinai in New York and a chain of more than 30 clinics across the UK.
Will AI agents work with Epic?
Yes. We connect through Epic's FHIR APIs, SMART on FHIR and its supported integration routes, with HL7 v2 feeds where needed, and build in a sandbox first. Agents read and write only what their role allows, and every write is logged. Production access depends on your integration and security approvals, which we plan for from the start.
Who approves what an AI agent does in a hospital?
Clinicians approve anything clinical: notes stay drafts until signed, results are shared by people. Staff approve anything financial: submissions, write-offs and adjustments. The agent prepares the work and does the follow-up; rules that must be exact run as code.
What does it cost to bring AI agents into a hospital?
One moment is usually a fixed piece of work; several moments are a monthly team. Where the gain is measurable, part of the fee can be linked to it. We give you a written estimate after a short discovery.
Is patient data safe with AI agents?
Patient data stays in your approved environment, every service that touches it sits under a business associate agreement, it is never used to train models, and every action is logged. No official HIPAA certification exists, so we show you the controls, data flows and agreements for your security review.
Can AI reduce no-shows?
It can reduce the slots that go empty. About 23% of appointments are missed on average (Dantas et al., 2018). Two-way reminders let patients confirm or rebook in a message, and freed slots are offered to the waitlist at once. How much changes depends on your patients and specialties, so we measure it before and after.
Can AI handle prior authorization?
AI can prepare most of it. Physicians handle 39 prior authorizations a week, about 13 hours of physician and staff time (AMA, 2025). An agent checks whether authorization is needed, gathers chart evidence against the payer's criteria, submits once a clinician attests, and tracks the status. Medical necessity and attestations stay with clinicians.
Can AI reduce a hospital's operating costs?
Yes, where the cost is paperwork and coordination rather than care. Agents take on the phone calls, forms, prior authorization packets, denial rework and follow-up that pull staff away from patients. AI cannot add clinical capacity, so Unthinkable measures the gain for one moment before and after, in staff hours and rework removed, rather than promising savings upfront.