Why Epic will not be dislodged: what the OpenAI and Claudeforce deals share

TL;DR

OpenAI linked ChatGPT to Epic, giving clinicians read-only access to patient context from records covering more than 325 million patients. The Claudeforce deal between Salesforce and Anthropic, announced August 26, follows a similar pattern: the model gets data and skills, the platform keeps the business logic and governed actions. I don't think Epic is getting dislodged any time soon, as long as it keeps the model layer interchangeable.

OpenAI connected ChatGPT to Epic, which holds current records for more than 325 million patients. Clinicians can bring authorised patient context from Epic into ChatGPT. Access is read-only, so nothing gets written back to the record. UCSF Health is already piloting it.

There is a parallel in the Claudeforce deal Salesforce and Anthropic announced on August 26. Claude gets Salesforce data, workflows and 37 prebuilt sales skills, while Salesforce keeps the business logic, permissions and governed actions underneath.

Where value accrues between models and systems of record

One of the fundamental questions in AI right now is where value will accrue between the models and the systems of record, especially once models can read everything those systems hold.

In the current setup, a hospital can change models far more easily than it can change its EHR. Both deals are structured so the model gets access and the platform keeps the rules.

I don't think Epic is getting dislodged any time soon, as long as it keeps the model layer interchangeable.