Sector brief · April 2026
Primary profile: AI agent (/agent/)
Healthcare and life sciences
Clinical and operational agents are arriving faster than the trust layer between them. The failure mode is rarely model error alone; more often it is an unauthenticated handoff that exposes PHI or bypasses policy.[3][4]
Market context
Agentic healthcare AI in North America was on the order of $0.66B in 2025, driven by data volume and decision-support demand.[4] The buyer story is familiar: automate intake, routing, and follow-up without adding new breach surfaces.
Regulation and liability still dominate roadmaps. Providers face HIPAA, state privacy rules, and uneven expectations around software as a medical device.[3] Any architecture that cannot show who exchanged what, under which policy, will stall in legal review.
Structural gaps
- Agent-to-agent handoffs (e.g. primary care to oncology, billing to intake) lack a standard, verifiable trust boundary.[6]
- HTML portals and PDF exports force clinical copilots to scrape or guess, which burns context and drifts from source systems.[2]
- Hosted-only SaaS is a hard sell when latency-sensitive or PHI-heavy processing must stay local.[1]
Where incumbents sit
Incumbents span model-centric safety stacks (large clinical LLM vendors), integration platforms (no-code EMR/CRM bridges), and voice-first revenue-cycle tools.[5][6][7] The open slot is infrastructure-grade identity and gating between narrow agents; a monolithic chat UI will not carry the compliance story on its own.
Building on more.md (implementation map)
Register clinical and operational stacks as `/agent` profiles
Model deployed automations (triage, prior auth, revenue-cycle bots) as `/agent` entities with explicit capability flags. Hospitals and clinics remain addressable as `/o`, but the workloads other agents negotiate with are first-class `/agent` surfaces.[1]
Trust and credential gates before payload moves
Require W3C Verifiable Credentials (or equivalent proofs) at the gate so a diagnostic or billing transfer only proceeds when the requesting agent is authorized. Blocked calls return machine-readable manifests an SDK can retry against.[3]
Smart sections for caller-specific slices
Serve billing codes to a revenue agent and intake protocols to a triage agent from the same canonical profile using headers such as `X-Agent-Type`, instead of duplicating pages.[2]
On-prem control plane with federated identity
Run the stack where PHI is processed; use the Python SDK to connect local inference and job runners to policy, subscriptions, and identity in the control plane.[1]
Competitive context
Names below appear in third-party sources listed under References. The contrast column sketches where an agent-first build on more.md sits relative to each player. Complement in some layers, substitute in others.
| Player | Position in market | Contrast with more.md-shaped build |
|---|---|---|
| Hippocratic AI | Safety-first clinical model stack, clinician-evaluated. | Model-centric trust vs wire-level verification with DIDs and EEP gates. |
| Keragon | No-code healthcare automation across CRMs and EMRs. | Integration orchestration vs Pulse-style agent-to-agent negotiation. |
| SuperDial | Voice agents for revenue-cycle calls. | Human-facing voice vs machine-to-machine TOON/Markdown payloads. |