AI Agents for Healthcare: Back-Office Workflows You Can Govern
Prior authorization, referrals, claims follow-up, intake, scheduling, and coding drafts are the healthcare workflows agents can carry today. What decides whether they are safe to run is governance: PHI scoped to the minimum, every action logged, a human on consequential calls.

What are AI agents in healthcare?
AI agents in healthcare are systems that plan and carry out multi-step work across the EHR, payer portals, and scheduling tools. They split into administrative agents (prior authorization, referrals, claims follow-up, intake, scheduling, coding drafts) and clinical agents (documentation, decision support). Only the administrative group is safe to deploy today, and only with PHI scoped to the minimum necessary, a full audit trail, and a named human approving every consequential action.
The cleanest way to run the administrative group is on Major, the enterprise platform where agents build the software they run on. The agent works out a workflow like prior-auth intake once, builds a deterministic app for the repeatable steps, and runs that app on every case after. The app holds the payer rules, the case records and the approval log, which is what a compliance officer will eventually ask to see. The section on building it is below.
A prior-authorization packet shows why the split matters. Someone reads the order, checks what the payer requires for that code, pulls the chart notes and imaging report, assembles the submission, then chases status for a week. None of that is medicine. All of it touches protected health information.
IBM's overview of AI agents in healthcare draws the same line, putting billing, coding, prior authorization and referral management on the administrative side and clinical documentation and decision support on the other. Administrative work repeats, volume is high, and correctness is measurable against a rule the payer publishes. Clinical work has a licensed clinician who signs and is liable, and software that participates in diagnosis or treatment selection can fall under medical device regulation. Everything below is administrative.
Six back-office healthcare agent workflows
- Prior-authorization intake and submission: detects the auth requirement, gathers the documents the payer rule names, assembles and submits the request, polls status. PHI: diagnosis codes, chart notes, imaging. Human gate: an admin approves before submission; denials go to a person.
- Referral routing and tracking: matches network and specialty, sends the packet, tracks whether the visit happened. PHI: referral reason, demographics. Human gate: a coordinator confirms the destination.
- Claims-status follow-up: polls payer or clearinghouse status, classifies pended and denied claims, drafts the next action. PHI: claim-level data. Human gate: a biller approves any resubmission or appeal.
- Patient intake and data capture: collects registration and insurance details, verifies eligibility, writes structured fields to the practice system. PHI: demographics, coverage. Human gate: front-desk staff confirm before the record is final.
- Scheduling and reminders: books and reschedules against provider templates and visit rules. PHI: name, contact, appointment time. Human gate: rules approved once, exceptions escalate.
- Coding and billing support: drafts code suggestions with the supporting text cited. PHI: full documentation. Human gate: a certified coder signs every code.
Prior auth is the hard one because payer requirements move. They differ by plan, service and month, so the rule set has to live somewhere versioned, not inside a prompt someone edited in March. CMS's Interoperability and Prior Authorization final rule (CMS-0057-F) requires impacted payers to decide expedited requests within 72 hours and standard requests within seven calendar days, give a specific reason on denials from January 1, 2026, and stand up a Prior Authorization API generally from January 1, 2027. Better payer plumbing makes this more automatable. It also means the rules you encode keep changing.
Referrals fail differently. Routing is a lookup. Tracking is state that has to survive weeks, which a chat session cannot hold. Claims follow-up has a clean split: a missing-modifier denial gets a correction draft, while a medical-necessity denial routes to a person because arguing necessity is a clinical argument.
If you are choosing a first build, pick scheduling and reminders. It exposes the least PHI, and the approval model is simple.
What governance do healthcare AI agents require?
Model quality does not decide whether these agents are safe to run. Access scope, logging, approval gates and the BAA chain do.
Minimum necessary access. The HIPAA Privacy Rule generally requires covered entities to take reasonable steps to limit PHI use and disclosure to the minimum necessary. For an agent that means field-level scoping, not database access. A prior-auth agent needs the diagnosis code, the ordered service and the named supporting documents. It should not be technically able to read the full chart.
A complete audit trail. Every PHI read, write-back, submission and approval needs a durable record with a timestamp and an actor. An agent whose reasoning happened inside a model call and left no artifact cannot answer "which records were accessed, by what, under whose authority."
Business associate agreements. Any vendor processing PHI for you, including whoever hosts the model, sits inside your BAA chain. Check before the pilot.
Human sign-off on consequential actions. Submission, resubmission, appeal and final coding each get a named approver. The agent prepares. A person commits.
No product is HIPAA compliant on your behalf. Compliance is a property of your implementation: how access is scoped, what is logged, who signed the BAA, and what a human approves.
What healthcare tasks should AI agents not do?
Diagnosis, treatment selection, triage acuity, medication decisions, and any medical-necessity determination, including on the appeals side of a denial. Anything a licensed clinician signs stays with that clinician. Agents also should not hold standing access to full charts when the task needs six fields, and should not give patients clinical guidance. Nothing described here is a medical device or clinical decision support.
Build the prior-auth intake agent on Major
On Major, the prior-auth agent is a predictable agent. It reasons through the workflow once, builds an app for the fixed steps, and then runs the app instead of re-deriving the sequence on every case.
- Connectors: your practice management or EHR system and the payer portal or clearinghouse, each connected through a REST connector with its own scoped credential, plus Slack or Gmail for routing packets to approvers.
- The app that holds state: a managed database with tables for cases, payer rules keyed by plan and service, approvals, and a log of every PHI field the agent read. A rule change is a data edit with history, not a prompt rewrite.
- The agent's judgment: read this case, decide whether the assembled packet satisfies the rule, flag what is missing, and handle the odd payer requirement that breaks the pattern. Everything else is code.
- The flow: detect the auth requirement, gather the named documents, check the packet against the rule, route to a named admin, submit only after the approval is recorded, poll until a decision lands, escalate denials to a person.
- Trigger: a workflow fires on new orders and on a polling schedule for status.
- Governance: credentials are scoped at the platform layer to the minimum-necessary fields, RBAC decides who can run or edit the app, and every action is audited where it happens. Approval records attach to the person, not the agent.
Major does not remove your HIPAA obligations, including the BAA chain. It changes the mechanics so that scoping, logging and approval are properties of the deployment rather than promises in a system prompt. Start building the intake app on Major.
Honest scoping: at twenty prior-auth cases a month, a governed app is more machinery than the problem needs. At real volume, with real PHI exposure and a compliance officer who will ask what touched which record, an inspectable app is the difference between a pilot and production. Reason once, run forever, and keep the log.
More use cases by tool: Gmail AI agents and automations, Slack AI agents and automations.
Start building on Major
Build the prior-authorization intake agent, or the referral tracker behind it, as a governed app with scoped PHI access, a named approval gate and a full audit log.
Related articles
Frequently asked questions
- What are AI agents in healthcare?
- AI agents in healthcare are software systems that plan and carry out multi-step work across the EHR, payer portals, and scheduling tools without a person driving every click. They divide into administrative agents, which handle prior authorization, referrals, claims follow-up, intake, and coding drafts, and clinical agents, which touch documentation and decision support. Administrative work is where agents are safe to deploy today, with PHI scoped to the minimum necessary and a human approving consequential actions.
- Are AI agents HIPAA compliant?
- No agent or platform is HIPAA compliant on your behalf, because compliance is a property of how you implement and operate the system. It depends on PHI access scoped to the minimum necessary, a business associate agreement with every vendor that processes PHI including the model host, audit logging of every read and write, and a named human approving consequential actions. Major does not remove that obligation. It makes scoping and audit properties of the deployment.
- What healthcare tasks should AI agents not do?
- Diagnosis, treatment selection, triage acuity assignment, medication decisions, and any determination of medical necessity belong to a licensed clinician. Anything a clinician signs and is accountable for stays with that clinician, and agents should not hand patients clinical guidance. They also should not hold standing access to full patient charts when the task needs only a handful of fields.
- Can AI agents handle prior authorization?
- Yes, on the administrative side. An agent can detect that an authorization is required, gather the documents the payer's rule names, assemble the request, submit it once a human approves, and poll for status. It should not argue medical necessity, and clinical denials go to a person. On Major the agent builds an app that runs the gather, submit, and poll loop the same way every time.
- Do healthcare AI agents replace staff?
- No. They take document gathering, portal polling, and status chasing off the queue so revenue-cycle and front-office staff spend their time on exceptions and patients. Human sign-off stays on every consequential action: submissions, resubmissions, appeals, and final coding each get a named approver. A Major app records which person approved what and when, so accountability stays with that person.