What a healthcare AI agent actually does
Reads a patient's scheduling, refill or billing request, checks it against the EHR and insurance eligibility, and either resolves it or routes it to clinical or billing staff.
A healthcare AI agent sits in front of a practice's phone line, patient portal or chat widget and runs a loop: read the patient's request, pull the relevant chart or appointment data from the EHR through a FHIR/HL7 API or a certified integration, check insurance eligibility where needed, and either schedule, reschedule, answer a billing question, or route the request to clinical or front-desk staff. Epic, the largest EHR, is building its own agent layer: Emmie handles patient-facing scheduling, bills and payments inside MyChart, Art assists clinicians, Penny works revenue cycle, and an Agent Factory lets health systems build their own agents on top. Independent vendors like Assort Health and Hyro write directly into Epic and athenahealth instead of replacing them, and smaller practices often start with a receptionist-style voice agent instead of a full EHR integration.
Momentum here is coming from both the EHR vendors and well-funded independents. Assort Health raised a $120M Series C at a $1.2B valuation in June 2026 and has deployed into health systems like John Muir Health through Epic and athenahealth, and Hyro is a certified Epic app in the Epic Showroom, used by systems including Tampa General. Epic previewing its Agent Factory at HIMSS26 is the clearest signal that the biggest systems of record intend to build this capability in-house rather than cede it entirely to point vendors.
Where agents still fail: anything resembling diagnosis or clinical advice, which California's AB 489 now explicitly bars an AI from implying by using a licensed title like 'nurse' or 'M.D.', and any triage call that should route to a person instead of being answered by a script. These agents are built to route and schedule, not to advise, and the honest design keeps a licensed human in the loop for anything clinical.
AI capabilities involved
Scheduling, rescheduling and appointment reminders
Insurance eligibility and intake data capture
Refill requests and routine billing questions
After-hours triage routing (never diagnosis)
Who uses this
- Multi-location medical and dental groups fielding high call volume for scheduling and refills
- Health systems already on Epic or athenahealth deciding whether to turn on a certified agent
- Independent practices wanting after-hours and overflow call coverage without adding front-desk staff
- Billing and revenue-cycle teams automating insurance eligibility and no-show follow-up
What the healthcare agent handles (and what it shouldn't)
| Task | Agent handles it? | Systems it touches | Human role |
|---|---|---|---|
| Scheduling and rescheduling | Fully | EHR (Epic, athenahealth), scheduling system | None unless the slot needs clinical judgment |
| Refill requests | With approval | EHR, pharmacy system | Approves or denies based on chart review |
| Insurance eligibility and intake | Fully | Eligibility tool (Availity), EHR | Reviews flagged mismatches |
| Billing questions | Fully | Practice billing system | Handles disputes or payment plans |
| No-show recovery and care-gap outreach | Fully | EHR, CRM | None |
| After-hours triage routing | With approval | Phone system, EHR | Makes every clinical judgment call |
| Diagnosis or clinical advice | No | EHR | Always a licensed clinician |
Off-the-shelf AI agents on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
Epic (Emmie / Agent Factory)
Health systems already running Epic MyChart
None; requires an existing Epic/MyChart contract
Included in Epic's enterprise contract; no standalone price published
Part of Epic's Agent Factory, previewed at HIMSS26 for health systems to build their own agents
Pros
- +Deepest possible integration since it's native to MyChart and the chart itself
- +Covers patient-facing (Emmie), clinician-facing (Art) and revenue-cycle (Penny) agents from one vendor
- +No separate BAA negotiation beyond your existing Epic agreement
Cons
- −Locked to the Epic ecosystem; doesn't help athenahealth, eClinicalWorks or other EHR users
- −No public pricing or self-serve path; everything runs through your Epic account team
Assort Health
Multi-location health systems already on Epic or athenahealth
None; enterprise sales only
Contact sales; no public pricing
Raised a $120M Series C at a $1.2B valuation in June 2026, more than $222M raised in total
Pros
- +Writes directly into Epic and athenahealth, deployed at systems like John Muir Health
- +Products cover inbound (Concierge), outreach (Activate), orchestration and workforce support (Empower)
- +Well-capitalized vendor with a credible enterprise customer base
Cons
- −No public pricing anywhere; expect a multi-week sales process
- −Built for health-system scale, likely oversized for a single independent practice
Hyro
Health systems wanting a certified, Epic-native voice and chat agent
None; enterprise sales only
Contact sales; no public pricing
A certified Epic Showroom app serving 45+ health systems, including Tampa General
Pros
- +Certified Epic integration, not a workaround
- +Covers voice, chat and SMS from one platform
- +Established base of 45+ health-system customers
Cons
- −No public pricing
- −Enterprise-focused, so expect a longer sales cycle than a point SaaS tool
Hello Patient
Independent and small multi-location practices not on a system with a native agent option
None; enterprise/practice sales only
Contact sales; no public pricing
Raised a $22.5M Series A in September 2025 at roughly a $100M valuation
Pros
- +Voice, text and chat built specifically for practice-level patient communication
- +Smaller and newer than Epic or Assort Health, so potentially more flexible for an independent practice
- +Backed by a credible Series A investor (Scale Venture Partners)
Cons
- −No public pricing
- −Less enterprise track record than Epic-native competitors
The AI stack
A healthcare agent needs a reasoning model, a BAA-covered knowledge source grounded in the actual EHR, real scheduling and eligibility tool access, and a hard rule against anything that could read as diagnosis.
Reasoning model
Drives scheduling, eligibility checks and the harder judgment call of what counts as an emergency
Claude Haiku 5.5
$0.10/M input tokens (up to 100K), $0.50/M outputRouting and classification
Claude Sonnet 5.5
$2/M input, $10/M outputThe main scheduling and billing loop
Claude Opus 5.5
$4/M input, $20/M outputThe harder judgment call of what needs immediate human routing
Our pick: Claude Sonnet 5.5 or GPT-6.1 Sol for scheduling and eligibility, with Claude Opus 5.5 reserved for the harder judgment call of what counts as an after-hours emergency that needs immediate routing.
Knowledge & retrieval
Grounds every scheduling and billing answer in the actual chart and insurance data
EHR via FHIR/HL7 (native)
Included in your EHR contractScheduling and billing lookups
Custom RAG over practice policies
Roughly $0.15/mo on Supabase pgvector for a small corpusCustom builds
Our pick: Pull appointment, refill and insurance data straight from the EHR through FHIR/HL7; never let the model answer a clinical question from general training data.
Integrations / tools
Lets the agent actually schedule, check eligibility and route a call
Certified EHR integrations (Epic Showroom, athenahealth marketplace)
Included or vendor-pricedBuy-SaaS
Custom HL7/FHIR interface
Developer time under a BAACustom builds
Our pick: Use a certified EHR integration wherever one exists for your system; build a custom HL7/FHIR interface only when no certified option covers your EHR.
Observability & evals
Catches drift toward clinical advice before it reaches more patients
Vendor dashboards
IncludedBuy-SaaS
Langfuse with PHI redaction
Free up to 50K units/mo, then $29+/moCustom builds
Our pick: Redact PHI from logs by default, and still review a sample of transcripts weekly for anything drifting toward clinical advice.
Reference architecture
The hardest engineering problem is the no-diagnosis boundary: the agent has to triage and route convincingly enough to be useful on the phone, while never crossing into clinical advice, a line California's AB 489 now makes a specific licensing violation, not just a bad practice.
Patient contact arrives
Phone, patient portal or chatA scheduling, refill or billing request lands through MyChart, a phone line or a chat widget
Identity and chart lookup
EHR via FHIR/HL7The agent verifies the patient and pulls the relevant chart, appointment or billing record
Eligibility check
Eligibility tool (Availity or similar)Insurance eligibility is checked before confirming a scheduling or billing action
Triage decision
Guardrail layerAnything resembling a clinical symptom or an emergency routes immediately to a human, never to a script
Action or routing
Scheduling system, billing systemApproved administrative actions execute directly; everything else routes to front-desk or clinical staff
Human handoff
Practice phone or EHR inboxRouted requests carry full context so staff don't have to re-ask what the patient already said
Logging and eval
PHI-redacted observability toolEvery interaction is logged with PHI redacted, and a sample is reviewed weekly for clinical drift
Estimated cost per request
~$0.04-$0.10 per patient interaction in model costs alone, before any HIPAA-compliant voice infrastructure fee
Cost calculator
Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.
Estimate your monthly cost at your own patient-contact volume and phone-versus-chat mix.
Estimated monthly cost
$2,270
≈ $27.2k per year
Calculator notes
- The HIPAA add-on fixed cost assumes a DIY voice stack like Vapi; a custom agency build or an enterprise vendor's own BAA may price this differently.
- Model cost applies on every path, buy, hire or build, since the underlying reasoning calls don't disappear just because you bought a SaaS license.
- Enterprise vendors (Assort Health, Hyro, Hello Patient) publish no price list, so budget against a custom build's cost until you have a written quote.
- Voice minute cost is blended across telephony, speech-to-text and text-to-speech at typical DIY platform rates.
Build it yourself with vibe-coding tools
A HIPAA-reviewed n8n build is possible for a non-clinical, administrative-only agent, scheduling confirmations and FAQ, but PHI handling raises the bar well above a typical DIY workflow.
Time to MVP
1-3 days for a non-PHI FAQ version; weeks for a PHI-handling version with a full BAA chain
Total cost to MVP
$0-$2,050 (n8n Cloud Starter plus model costs, plus roughly $2,000/mo if adding a HIPAA-compliant voice platform)
You'll need
Starter prompt
Set up a webhook or chat trigger for patient scheduling and FAQ requests only, keeping anything with PHI behind a BAA-covered path. Feed the request into an AI Agent node using Claude Sonnet 5.5, with an HTTP Request tool for your EHR's FHIR API to check appointment slots and a second tool for eligibility lookups. Give the AI Agent node this system prompt: 'You are a scheduling assistant for {Practice Name}. You help patients book, reschedule or confirm appointments and answer basic billing and insurance questions using the FHIR tool. You never discuss symptoms, give medical advice, or use any clinical title. If a patient mentions a symptom, pain, bleeding or anything urgent, immediately reply with ESCALATE and a one-sentence summary for staff to call back right away. State clearly that you are an AI assistant.' Add a Switch node that routes ESCALATE messages to a staff queue immediately and logs every interaction with PHI redacted to a secure, access-controlled table.
Paste this into n8n
Follow-up prompts (run in order)
- 1
Add an insurance-eligibility check before confirming any new appointment
- 2
Add a no-show follow-up sequence for missed appointments
- 3
Add a daily summary of everything the agent escalated and why
- 4
Add a Spanish-language version of the triage script
Expected output
A working agent that books and confirms appointments and answers basic billing questions, while escalating anything clinical to staff immediately with a one-sentence summary.
Known gotchas
- !Any PHI touching the workflow requires a BAA with every vendor, including n8n Cloud itself if you're not self-hosting
- !A HIPAA-compliant voice stack adds roughly $2,000/mo on platforms like Vapi, which erodes most of the DIY cost advantage
- !The AI Agent node will answer a symptom question from general training data unless explicitly instructed to escalate instead
- !EHR API access (FHIR/HL7) varies significantly in completeness between systems, so test real appointment-booking edge cases before launch
- !There's no built-in eval harness, and healthcare transcripts need PHI redaction before they can even be reviewed for quality
Compliance & risk reality check
A healthcare agent touches more regulation than almost any other category: every vendor in the chain needs a signed BAA, and the agent must never cross into clinical advice.
HIPAA business associate chain
Any vendor that creates, receives, maintains or transmits PHI, the voice platform, the model provider, the telephony carrier, the transcript storage, is a business associate and needs a signed BAA.
Mitigation: Get a BAA with every hop before any PHI touches the system; prefer zero-retention modes and redact PHI from logs.
No diagnosis or licensed-title implication (CA AB 489)
California's AB 489, in force since January 2026, bars an AI from using terms or titles, like 'nurse' or 'M.D.', that imply a licensed human provided care or advice, with each use treated as a separate violation.
Mitigation: Scope the agent to scheduling, routing and billing only; never let it answer a clinical question or use a licensed title, anywhere.
TCPA and disclosure for outbound reminders and outreach
Outbound appointment reminders and no-show or care-gap outreach calls using an AI voice need prior express consent under the FCC's 2024 ruling, and the patient should be told they're talking to an AI.
Mitigation: Capture consent at intake, disclose the AI at the start of every outbound call, and honor opt-outs immediately.
PCI DSS for phone bill-pay
If a patient reads a card number to the agent to pay a bill, those logs and model calls enter cardholder-data scope under PCI DSS 4.0.1.
Mitigation: Use DTMF masking or a pay-by-link text instead of letting the agent hear or transcribe a card number.
Build vs buy: the real math
3-6 weeks for a RapidDev build (longer with a full EHR integration); weeks to months for an enterprise vendor's procurement process
Custom build time
$13K-$35K for an agency build; enterprise vendors (Assort Health, Hyro, Hello Patient) and Epic's Agent Factory are quote-only
One-time investment
Hard to calculate against enterprise vendors since none publish pricing; a scoped agency build is usually faster to a working system than an enterprise sales cycle outside a large health system
Breakeven vs buying
Because Epic, Assort Health, Hyro and Hello Patient all price through sales conversations rather than a public list, a direct cost comparison isn't possible the way it is in customer service or sales. What is comparable is time to a working system: a RapidDev build at $13K-$35K can be live in 3-6 weeks with a clean BAA chain, while an enterprise vendor's procurement, security review and contract process routinely runs months, which matters most for single-location and regional practices that don't have the scale to make a large enterprise deal worthwhile.
Skip the DIY — RapidDev builds the production version
A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship through our AI app development service.
Discovery call (free)
30 minWe map your exact healthcare AI agent use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.
AI-accelerated build
3-6 weeks for a RapidDev build (longer with a full EHR integration); weeks to months for an enterprise vendor's procurement processOur engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.
Launch + handoff
1 weekWe deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.
What you get
Timeline
3-6 weeks for a RapidDev build (longer with a full EHR integration); weeks to months for an enterprise vendor's procurement process
Investment
$13K-$35K for an agency build; enterprise vendors (Assort Health, Hyro, Hello Patient) and Epic's Agent Factory are quote-only
vs SaaS
ROI in Hard to calculate against enterprise vendors since none publish pricing; a scoped agency build is usually faster to a working system than an enterprise sales cycle outside a large health system
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
How much does an AI agent for healthcare cost?
Epic's Emmie is included in your existing Epic/MyChart contract with no separate published fee. Independent vendors like Assort Health, Hyro and Hello Patient are quote-only through an enterprise sales process. A custom RapidDev build runs $13K-$35K upfront plus $150-$600/mo in model and infra costs, plus roughly $2,000/mo for a HIPAA-compliant voice add-on if the agent takes calls.
How long does it take to set up?
Enterprise vendors and Epic's own agents typically take weeks to months because they go through a full procurement and security review. A custom agency build takes 3-6 weeks, longer with a full EHR integration. A non-PHI n8n FAQ version can work within 1-3 days, but anything touching PHI needs a signed BAA chain before it goes live.
Can RapidDev build this for us?
Yes. RapidDev has shipped 600+ apps and offers a free 30-minute consultation to scope your EHR, call volume and HIPAA requirements before you commit to anything.
Does the agent need a BAA?
Yes, every vendor that touches PHI in the chain, the model provider, any voice platform, telephony, transcript storage, needs a signed business associate agreement. Skipping this for even one hop is a direct HIPAA violation, not just a risk.
Can the agent give medical advice?
No. California's AB 489 specifically bars an AI from implying a licensed human is providing care, and every credible vendor and build scopes the agent to scheduling, routing and billing, never diagnosis or clinical advice.
What happens when a patient describes a symptom or an emergency?
A well-built agent is designed to recognize anything resembling a clinical concern and escalate it to a human immediately, rather than attempt to answer or triage it itself. This is a hard rule, not a judgment call the AI makes.
Can it work with Epic or athenahealth?
Yes. Assort Health and Hyro both write directly into Epic and athenahealth rather than replacing them, and Hyro is a certified Epic Showroom app. A custom build can connect to either through FHIR/HL7 or a certified integration path.
Want the production version?
- Delivered in 3-6 weeks for a RapidDev build (longer with a full EHR integration); weeks to months for an enterprise vendor's procurement process
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.
