Gemini for healthcare reads Epic and Oracle records, but cannot write back to them

Google Cloud has launched Gemini for healthcare, an agent-based product for hospitals and health plans, in U.S.-only preview. According to Google, it works across existing record, billing and coverage systems and keeps humans in the loop.

Google frames the problem as time lost to paperwork: clinicians spend nearly 28 hours a week on administrative tasks, it says. The product is built to cut that by gathering data scattered across separate systems and carrying out multi-step workflows such as medical records synthesis and prior authorization review. Launch partners named in the announcement are HCA Healthcare, Highmark Health and Northwell Health.

Four building blocks and a read-only rule for records

In a post on the Google Cloud blog, CEO Thomas Kurian describes Gemini for healthcare as an industry layer on top of the general Gemini agent. It has four parts: reusable skills written by domain experts, connectors to outside systems, a partner ecosystem, and a governed platform with a choice of models. Teams can write their own skills. Google says frontier models handle complex clinical reasoning while cheaper, faster models take routine administrative steps.

For providers, the listed workflows include a longitudinal patient timeline that separates active from resolved problems, lists allergies and flags medication discrepancies. Others are pre-visit preparation and authorization packets that flag documentation gaps before submission. For health plans, Google lists review of requests against medical policies and benefit parsing that estimates out-of-pocket costs. For Medicare Advantage members, the agent applies Medicare coverage criteria where a plan’s own rule is stricter.

The connectors are where the limits show. The Epic and Oracle Health connectors are read-only, with no write-back to the record. Google lists further connectors for InterSystems FHIR, Verato, TriZetto, GoodRx, Infinitus, ICD-10 and RxNorm, CMS coverage policies, the NPPES provider registry and BMJ guidelines. Two Optum eligibility connectors are in limited release and testing. Google says the connectors keep existing identity and role-based permissions and do not migrate data.

On safety, Google says each agent has its own identity and inherits the user’s permissions, with audit logging and citations back to records and policies. It also says clinical data, prompts and outputs are not used to train or fine-tune its foundation models. These are Google’s claims, and the announcement gives no independent evaluation.

Who can use it now, and what is missing

The product is available in preview starting November 10, in the U.S. only. The announcement does not give pricing or preview terms, or say when general availability will come. It also does not say how much time any customer has saved, and none of the customer statements it quotes report measured results. HCA Healthcare’s chief clinical officer, for example, says only that the company is “exploring” how Gemini can synthesize patient information from multiple record systems. Systems integrators including Accenture, Deloitte, KPMG, PwC and Tata Consultancy Services are listed as partners, which suggests most deployments will go through consultants rather than self-service.

For a reader at a hospital or insurer, the practical change is that the agent can draw on many systems without a data migration. But it only reads from the record, so any result still has to be entered by staff or another system.

A crowded field of clinical agents

Google is not the first large cloud vendor to pitch agents for prior authorization. Microsoft said in October 2025 that it was extending Dragon Copilot to nurses and opening it to partner apps and agents. According to Fierce Healthcare, its launch partners in the Microsoft Marketplace included Canary Speech, Humata Health, Optum and Regard. Humata Health, for its part, announced prior authorization technology available through Dragon Copilot, per a Business Wire release.

The two approaches differ in where they start. Microsoft’s product grew from clinical documentation, an ambient tool for visits, and added partner agents on top. Google starts from a general enterprise agent and adds healthcare skills and connectors. Its connector list overlaps with partners that also appear around Microsoft’s platform, Optum among them, so choosing between them may depend more on which EHR and payer systems a customer already runs than on any single feature. A HIT Consultant report says Epic is building agents on Dragon Copilot. That claim comes from one trade outlet and was not checked against Epic or Microsoft here. If it holds, Google’s read-only Epic connector would face a rival with a closer tie to the EHR vendor.

Google also leans on breadth, with connectors to Microsoft 365, ServiceNow, Salesforce and Workday’s Sana, so one agent can span clinical and back-office tasks. Whether customers want that, rather than a narrower tool that sits inside the EHR, is something the preview period will test. Cloud rivals have also been stressing the governance side of managed AI; our explainer on Amazon Bedrock covers how AWS bundles model access with controls, a pitch that Google’s audit logging and permission inheritance echo.

Google does not say how it will validate the agent’s output in clinical settings. For now the human oversight it describes is the main safeguard, and the announcement does not spell out where that review happens in each workflow.