AI agent (in healthcare)
An AI agent in healthcare is software that uses a language model to pursue a goal through a sequence of actions — calling APIs, reading records, asking the patient questions — rather than answering a single prompt; in clinical settings it operates inside guardrails that reserve diagnosis and prescribing for licensed clinicians.
- Acts across steps and tools; a chatbot only responds
- Needs clinical infrastructure APIs to do anything real
- Never diagnoses or prescribes — clinicians do
- Guardrails belong in infrastructure, not the prompt
Agent vs chatbot vs copilot
A chatbot responds to a message. A copilot drafts alongside a human — an ambient scribe writing the note, a coding assistant suggesting the CPT. An agent is given an outcome and works toward it across steps and tools: it can look up the patient, open a care plan, submit a refill encounter, wait for a webhook, and tell the user the result. The distinction is about action, not intelligence.
What healthcare agents call
Agents are only as capable as the tools they can reach. In care delivery the useful tools are the clinical infrastructure primitives: patients, care plans, encounters, a treatment catalog, orders, labs, and webhooks that report what happened. Increasingly these are exposed both as REST APIs and as tool servers agents can discover directly. Without a tool that can lawfully move a prescription, an agent in healthcare is a chatbot with a longer memory.
Guardrails for healthcare agents
The guardrails are what make an agent deployable in a regulated program, and most of them belong in the infrastructure rather than the prompt.
- Scope: inform, collect, route — never decide; red flags escalate immediately
- Privacy: protected health information only under a BAA, with access logged
- Disclosure: patients told when they are talking to AI, where state law requires it
- Enforcement: state modality rules, licensure match, and catalog limits applied server-side
- Audit: every action attributable to an actor and timestamp, exportable
Who builds them
Three groups. Telehealth brands building intake, check-in, and support agents into their own apps. Consumer AI assistants whose users ask for care the way they ask for a ride. And operations teams automating the internal work — refill queues, follow-up scheduling, status updates. All three end up calling the same kind of infrastructure, which is why that infrastructure has to hold the rules.
Compliance handled, so you can build
Lithos runs the clinicians, pharmacies, and 50-state rules behind your care program — one API.
Frequently asked questions
What can an AI agent do in a telehealth program?
Register patients, capture consent and identity, collect structured intake, request treatments and refills from an approved catalog, order labs, report encounter and order status, and run follow-ups. It hands every clinical decision to a licensed clinician.
Are AI agents HIPAA-compliant?
An agent is compliant when the vendors handling protected health information operate under business associate agreements, access is logged, and data is minimized. The model provider, the infrastructure, and the brand all need to be inside that chain.
How does an AI agent connect to healthcare infrastructure?
Through APIs and tool servers. Lithos exposes patients, care plans, encounters, catalog treatments, and webhooks over REST, plus an early-access MCP server so agents can discover the tools directly.