The challenge
A representative regional health system runs inpatient, outpatient, and emergency services across multiple sites, and its clinicians carry a documentation load that pulls them away from patients.
Published time-motion research puts physicians at roughly 49% of the working day on EHR and desk work against about 27% in direct patient contact, with around 4.5 hours a day inside the record (Sinsky et al., AMA). Administrative work is among the most-cited drivers of burnout, and roughly 25% of US health spending is administrative (McKinsey).
In this scenario the same pressure shows up in two places: notes finished after hours, and intake documentation that varies between shifts. The constraint is not clinical judgement — it is the keystrokes around it.
- Setting
- Representative — regional health system
- System concept
- MedAgent
- Illustrative duration
- 12 weeks
The architecture
We deploy MedAgent — our healthcare agent system — as an ambient documentation workflow, with triage support as a secondary agent. It connects to the health system's EHR via FHIR to read the clinical context a note depends on, drafts the encounter note as the visit happens, and routes every draft back to the treating clinician for edit and signature. A triage-support agent structures intake against the health system's own ESI criteria and flags inconsistencies for a nurse to resolve. Nothing is filed and no severity is assigned without a clinician in the loop.
Ambient Documentation Agent
Drafts the encounter note from the consultation and structures it to the health system's own templates
Clinical Context Agent
Retrieves relevant history, medications, and results from the EHR via FHIR to ground the draft
Triage Support Agent
Structures intake against the health system's ESI criteria and flags inconsistencies for nurse review
Review & Handoff Agent
Routes every draft to the treating clinician for edit and signature, then generates the handoff summary
A proposed delivery path
This sequence illustrates the engagement. The actual scope, timeline and quotation are agreed around your requirements.
Discovery & Assessment
Mapped documentation and intake workflows, shadowed clinicians, audited EHR integration points and consent requirements
Architecture & Design
Designed the agent pipeline, defined the review and escalation model, specified FHIR read/write scopes with the integration team
Build & Evaluation
Built the agent system and tuned and evaluated it on the client's own historical corpus under a signed data agreement, validated against expert review
Pilot & Iteration
Ran a limited-service pilot, measured note turnaround and clinician edit distance against the recorded baseline, refined templates and thresholds
Handover & Rollout
Phased rollout with monitoring dashboards, alert thresholds, an operator runbook, and clinician training
Benchmark context
Published figures used to frame this scenario. Baselines describe the referenced setting; targets are illustrative, not measured project outcomes.
Less time on clinical documentation
Source: UW Health randomized trial, 2024-25 (~30 min per clinician per day)Of physicians reported higher satisfaction
Source: Permanente ambient documentation deployment, 7,260 physiciansOf physician time is spent in the EHR today
Source: Sinsky et al., AMA time-motion studyOf US health spending is administrative today
Source: McKinsey (~$1T, of which ~$265B is reducible)Deliverables & controls
What the scope can include
- Ambient documentation agents running in your environment, against your own note templates
- FHIR integration spec with the read and write scopes agreed with your EHR team
- Evaluation suite with a documented pre-deployment baseline for note turnaround and edit distance
- Clinician review workflow, escalation rules, and an operator runbook
- All source code in your repository, with the deployment owned by your team
Where people stay in control
- Every note is reviewed and signed by the treating clinician before it enters the record.
- The triage agent structures and flags; a nurse assigns severity. The agent never sets an ESI score on its own.
- Diagnostic support is scoped as decision support only — published meta-analysis puts general GenAI diagnostic accuracy at roughly 52%, on par with non-experts and below experts.
- PHI stays inside the environment named in the data agreement. Model providers are selected with you, and your data is not used to train shared models.
Sources behind the scenario
- Physicians spend ~49% of the day on EHR and desk work, ~4.5 hrs/day in the record
Sinsky et al., AMA time-motion study - ~30 minutes per clinician per day saved on documentation with an ambient scribe
UW Health randomized controlled trial, 2024-25 - 82% of physicians reported higher satisfaction across 2.5M+ encounters
Permanente ambient documentation deployment (7,260 physicians) - ~25% (~$1T) of US health spending is administrative
McKinsey - General GenAI diagnostic accuracy ~52% — below experts
Takita et al., npj Digital Medicine, 2025 (meta-analysis of 83 studies)