Most clinical AI tools know everything about medicine and nothing about your patient. A landmark enterprise deployment just proved what changes when that gap closes — and why it matters far beyond one health system on one EHR.
Ramani Narayan · May 2026 · 6 min read
Geoffrey Moore’s core-versus-context framework — built for enterprise technology strategy, not clinical AI — is the most precise description of why most clinical AI fails in practice. Core activities are the things an organization exists to do. Context activities are everything else: the preparation, coordination, and information retrieval that enable the core work but are not the work itself.
In clinical medicine, the core activity is physician judgment: the synthesis of a patient's specific situation with current medical knowledge to make a decision. Everything that happens before that moment — retrieving the patient's history, cross-referencing medications, reconstructing the longitudinal record from fragmented data sources — is context work. It is essential, it consumes enormous physician time, and it is where AI delivers immediate measurable value without touching the mission-critical decision itself.
Most clinical AI tools skip the context work entirely. They answer questions about medicine without knowing anything about the patient being treated.
A clinical AI that can cite the latest evidence for heart failure management but does not know the patient is already on an SGLT2 inhibitor and has stage 3 CKD is not answering the physician's question. It is answering a different question — a generic one — dressed up as a specific answer. The physician still must do the context work manually, after the AI responds, before the AI’s answer is usable.
This is the structural failure mode of first-generation clinical AI: technically impressive, clinically incomplete. It is why adoption stalls. It is why physicians describe AI tools as 'useful in theory' and then quietly stop using them after the first month. The tool did not understand who the patient was, so the physician could not trust what the tool said.
On May 20, 2026, OpenEvidence announced a systemwide deployment at Cedars-Sinai Health System — one of the most recognized academic medical centers in the country, serving over one million patients annually. The deployment is not a pilot. Every physician, nurse, pharmacist, and therapist at Cedars-Sinai now has access to the platform, embedded directly inside their Epic EHR workflow.
The capability it delivers closes the context gap directly. When a clinician opens a patient's chart, OpenEvidence reads the relevant clinical data — diagnoses, comorbidities, current medications, allergies, prior procedures, lab values — and becomes able to answer clinical questions in the context of that specific patient. Ask about first-line treatment options and the answer already knows what the patient is taking and what they are allergic to. Ask about diagnostic criteria and the answer already knows the relevant history.
For the first time at enterprise scale, the AI knows who the patient is before the physician asks the question.
Cedars-Sinai is the third major health system to deploy OpenEvidence this way. Sutter Health went live inside Epic in February 2026. Mount Sinai deployed across seven hospitals in March. In the span of three months, patient-aware clinical AI moved from a research concept to an operational reality at institutions that collectively treat tens of millions of patients annually.
Cedars-Sinai has also announced that it will integrate its own institutional care pathways and clinical protocols into the platform — so clinicians will see not just what the medical literature says, but what Cedars-Sinai specifically recommends. This is the difference between a general reference tool and a clinical intelligence system that understands where you practice.
The chief health informatics officer at Cedars-Sinai described the result plainly: clinicians now have “a more complete and actionable understanding at the moment of care.” That sentence is what closing the context gap sounds like.
The clinical problem this solves is not exotic. It is the ordinary, daily friction of modern medicine: a physician has twelve minutes with a patient, a chart that contains years of clinical history, and a clinical question that requires synthesizing all of it against current medical evidence. No human being can do that consistently, across thirty patients a day, without structural support.
When that support is absent, the result is not malice or negligence. It is the quiet, systemic erosion of care quality that happens when the task exceeds the conditions. Physicians prescribe medications that interact with drugs already in the chart. They apply diagnostic criteria without accounting for comorbidities that change the calculus. They make treatment decisions against a history they have not fully absorbed because the visit is already running over.
Patient-aware clinical AI does not replace the physician's judgment. It eliminates the conditions that compromise it.
The equity dimension is less visible but equally important. The patients who benefit most from a physician who has fully absorbed their history are the patients with the most complex histories — the elderly, the chronically ill, those managing multiple conditions simultaneously. These are often the patients with the least access to the most experienced clinicians. AI that effectively expands every physician's capacity to reason about complex patients is a tool for reducing the gap between the care available at a major academic medical center and the care available everywhere else.
The Cedars-Sinai deployment is a landmark. It is also, for the moment, exclusive to Epic. Every health system that has deployed OpenEvidence enterprise-wide runs Epic. The capability those clinicians now have is not available to the 160,000-plus physicians on athenahealth — a platform concentrated in independent and specialty practices, where a very different patient population is seen every day.
This is the gap RISA was built to close.
RISA is ThetaRho’s clinical intelligence application, certified on the athenahealth Marketplace. When a physician opens a patient’s chart in athenahealth, RISA retrieves the patient’s longitudinal record — diagnoses, medications, allergies, lab values, prior visits — normalizes it to a structured clinical data model, and surfaces synthesized clinical context at the point of care. The physician does not switch tabs. The physician does not re-enter context. The AI already knows who the patient is.
The experience is structurally identical to what Cedars-Sinai clinicians now have on Epic: patient-specific intelligence, surfaced inside the EHR workflow, grounded in the actual patient in front of the physician. The difference is not the clinical experience — it is the data infrastructure underneath it.
Physicians using RISA report saving up to 90 minutes per day. That time was previously spent doing the context work manually — navigating the chart, cross-referencing medications, reconstructing a patient’s history from disconnected problem lists. That time now goes back to the patient.
The Cedars-Sinai announcement signals that patient-aware clinical AI is becoming an expectation. Health systems and practices that do not offer it will face pressure from clinicians who have seen it elsewhere. The capability gap will become visible — and visible capability gaps in healthcare do not stay invisible for long.
The harder question is whether closing that gap requires a Cedars-Sinai-scale integration project every time a new health system or EHR platform wants the same capability. On the current model, yes — which means the benefit stays concentrated at large, well-resourced institutions that can afford the integration investment.
ThetaRho’s platform is built to break that constraint. The architecture separates the problem into three layers. A FHIR normalization layer handles raw EHR data from any source — athenahealth, Epic, Cerner, Meditech — and produces AI-ready structured records. An AI intelligence layer exposes clinical context through a versioned, HIPAA-compliant API. An application layer is where clinical products are built. RISA is ThetaRho’s own application. But any EHR vendor, health system, or clinical AI company can build on Layers 1 and 2 without rebuilding the normalization infrastructure that makes patient context possible.
Every OpenEvidence deployment at a new health system requires a new integration project. A platform architecture means the second integration is structurally easier than the first — and the twentieth easier than the second.
The Cedars-Sinai deployment is an important proof point for what patient-aware clinical AI can do. The real question is how quickly it reaches the 90 percent of the healthcare system that is not Cedars-Sinai. That is a platform problem, not an application problem. It requires infrastructure that normalizes clinical data across heterogeneous EHR environments, exposes it through a clean API, and lets clinical intelligence applications build on top of it rather than rebuilding the plumbing every time.
RISA on athenahealth is the proof that it works. The platform is the path to making it universal.
If you are building clinical AI and spending more time on data normalization than on clinical logic, ThetaRho’s platform is designed for you. If you are a physician on athenahealth who wants the same point-of-care intelligence Cedars-Sinai just deployed, RISA is available on the athenahealth Marketplace today.
ThetaRho (thetarho.ai) builds clinical AI infrastructure for healthcare organizations. RISA is our clinical intelligence platform — HIPAA-compliant, AICPA SOC certified, and live on the athenahealth Marketplace.