The Healthcare AI Agent Stack: 2026 Market Map

TL;DR

• Healthcare AI has moved from chat to action. Menlo Ventures counted $1.4B of healthcare AI spend in 2025, nearly 3x the year before. Ambient documentation led at $600M, coding and billing followed at $450M, and prior authorization grew about 10x.
• The stack has 12 layers, from foundation models and EHR-native agents at the bottom, through workflow agents (scribes, prior auth, RCM, patient access), to the infrastructure and governance layers every agent depends on.
• Incumbents are building agents inside the EHR. Epic says more than 85% of its customers use Epic AI, and its Agent Factory lets health systems build their own. Independent agents have to work alongside them, through the certified APIs.
• Every workflow layer depends on the same plumbing: authenticating to EHRs, payers, clearinghouses and SaaS tools as the right person, per health system, with an audit trail. That's the least visible layer on this map, and the one that decides which agents get deployed.

How we built this map

• Scope: companies building or enabling AI agents that act in US healthcare workflows. That means taking actions in systems, not only answering questions.
• Sources: company announcements, funding releases and trade press, linked inline. Metrics are company claims unless noted.
• Not a ranking. Order within a layer isn't meaningful. We're one of the companies on the map (in agent infrastructure), so we've kept every description factual.
• Corrections: if your company is missing or misdescribed, email us and we'll update it in the next quarterly refresh.

The stack at a glance

Layer 1: Foundation models with healthcare offerings

Model providers now ship healthcare-specific products, BAAs and connectors.

What to watch: BAAs are table stakes, but coverage differs by feature. Anthropic's BAA, for example, excludes data sent to third parties through MCP connectors.

Layer 2: EHR platforms and EHR-native agents

The systems of record are becoming agent platforms themselves.

What to watch: the EHR's own agents set the baseline. Independent agents win where they go deeper in a workflow or span systems the EHR doesn't.

Layer 3: Interoperability and data networks

These assemble records from many sources and give agents a longitudinal view.

Layer 4: Clearinghouses and payer connectivity

The rails for eligibility, claims and prior auth.

What to watch: CMS-0057-F requires FHIR prior-auth APIs from impacted payers by January 1, 2027. Until then, agents run on both FHIR and X12, portals and fax.

Layer 5: Ambient documentation

The largest and most crowded agent category.

What to watch: EHR-native scribes (Epic AI Charting, Oracle, NextGen) are compressing the category. Differentiation is moving to write-back depth: orders, coding and prior-auth triggers.

Layer 6: Prior authorization and utilization management

What to watch: CMS's WISeR model brings AI-assisted prior auth to traditional Medicare in six states from 2026, with licensed clinicians deciding every denial.

Layer 7: Revenue cycle and coding

Layer 8: Patient access and voice

What to watch: outbound AI calling falls under TCPA consent rules, which shapes how aggressively these agents can do outreach.

Layer 9: Care coordination, referrals and clinical trials

Layer 10: Clinical decision support and evidence

Layer 11: Agent infrastructure (auth, tool calling, MCP)

This is the layer that connects workflow agents to the systems in layers 2–4 and to the SaaS tools around them. It handles authentication per user and per health system, tool calling, scoping and audit.

Why this layer matters more in healthcare: a single prior-auth agent may need a clinician's SMART on FHIR token, payer credentials, a clearinghouse key, a portal login and a fax key, for every user at every health system. How these credentials are held, scoped and logged decides whether a health system's security review passes. We compared the options in Best Agent Integration Platforms for Healthcare.

Layer 12: Governance, evaluation and security

Five things the map shows

1. Money follows action. The fastest-growing layers (prior auth, patient access, RCM) are the ones where agents take actions in other systems, not the ones that only generate text.
2. Incumbents set the floor. Epic, Oracle Health and athenahealth all ship their own agents. Independent agents need to be meaningfully better in one workflow or span systems the EHR doesn't reach.
3. Scribes are converging on write-back. The next competitive line is how deeply an agent writes into the chart: orders, codes and prior-auth triggers, not only notes.
4. The payer side is catching up. Payer adoption lags providers, but CMS's API deadlines and the WISeR model are pulling payers into agent workflows.
5. Infrastructure is the constraint no one sees. Every workflow layer depends on authenticating to many systems as the right person, per health system, with an audit trail. Teams that treat this as plumbing to build later hit it at their first enterprise security review.

Where to go next

• Building an agent for this market: API Access Patterns for AI Agents: Identity Architecture for B2B SaaS
• The EHR mechanics: What Is Agent Authentication? A Guide for B2B AI Builders

Frequently asked questions

What are the main categories of healthcare AI agents?

The main workflow categories are ambient documentation, prior authorization and utilization management, revenue cycle and coding, patient access and voice, care coordination and referrals, and clinical decision support. They sit on top of foundation models, EHR platforms, data networks and clearinghouses, and depend on agent infrastructure and governance layers.

Which healthcare AI agent category is largest?

Ambient documentation. Menlo Ventures estimated $600M of 2025 spend on ambient scribes, the largest single category, followed by coding and billing at $450M. Prior authorization and patient engagement were the fastest-growing, at roughly 10x and 20x.

Are EHR vendors building their own AI agents?

Yes. Epic ships Art, Emmie and Penny plus Agent Factory for health systems to build their own. Oracle Health has its Clinical AI Agent, and athenahealth and eClinicalWorks have launched agentic tools. Independent agents integrate alongside them through certified FHIR APIs.

What is agent infrastructure in healthcare?

Agent infrastructure is the layer that lets AI agents act in other systems safely. It covers authenticating to EHRs, payers, clearinghouses and SaaS tools on behalf of specific users, scoping which tools each agent can use, handling token lifecycle, and logging every action for audit, often deployed inside the customer's cloud.

How is this market map maintained?

We refresh it quarterly from company announcements, funding releases and trade press. Companies that are missing or misdescribed can email us for a correction. Inclusion isn't a ranking or endorsement.