Every few months someone predicts that AI will let companies build their own software and put the SaaS vendors out of business. At Oracle AI World in Las Vegas on Tuesday, Larry Ellison said that's not how it will go.
In a keynote conversation on AI in healthcare with Seema Verma, executive vice president and general manager of Oracle Health and Life Sciences, Ellison took on the "SaaS apocalypse" directly. "I don't think that's quite true," he said. "I think companies will continue to build applications, but they will be quite different because of AI."
His argument is that commercial software has never been finished for any industry. Hospitals, banks, and hotels run vendor suites, then build extensions, buy other tools, and wire everything together. AI, he said, lets a vendor finish the application, industry by industry. "We finish the application so you don't have to."
Healthcare comes first because it is the hardest. Ellison named banking, retail, hospitality, and government as the industries that follow. He set four tests for the healthcare suite: it has to be complete, international, easy to use, and quick to implement. International means more than translation. Regulations differ between California and New York, and between German states.
The interface was his first target. He told a story about his son's broken leg. The orthopedic surgeon at Stanford couldn't pull up the X-rays. Neither could the radiologist. It took a systems specialist. His answer is to ask for them in plain English. "If you can speak English and say what you want to see, you should then immediately see it," he said.
The same goes for actions. He described telling the system to raise a metoprolol dose from 25 to 50 milligrams. The system repeats the order back, writes it into the emergency room notes, and gives the doctor a summary of the prescriptions. "There should be no training required to use any of these applications," he said.
The claim engineers should pay attention to is about deployment. Oracle Health runs in the cloud, so Ellison said installing a module really means turning it on. His example was a hospital that runs Epic and wants Oracle's oncology module. The module finds the oncology staff, the tumor board, the patient's record, and the latest labs through the interfaces the hospital already has. He pointed to the SMART on FHIR standard as the way in.
"If their oncology module can read it, so can ours," Ellison said. "There's no data migration. There's no new training." Then he added the most candid line of the session: "Maybe ours is better. Maybe ours isn't. That's for you to decide."
He described a platform that other vendors build into. Oracle works with GE, Siemens, and Philips, and it has built an NVIDIA model into the platform to look for cancerous lesions in CT scans of the lung. Oracle also uses simulators and digital twins of hospitals to test the software, he said.
Prior authorization was another example. If the system knows a payer's rules, it can check eligibility before it files anything. His example was GLP-1 drugs under UK National Health Service rules. In his telling, Oracle's system would never send a request for a patient who doesn't qualify, so the back and forth with the payer becomes a single round trip.
The patient portal is meant to work across hospitals and across electronic health record vendors, whether Epic, Cerner, or Allscripts. He said it would start with the United States and operate at national scale. Because it would hold records for the whole population, he said, officials could see anonymized, up-to-date counts, such as how many hospital rooms are open. He pointed to COVID as the case where that would have helped.
On security, Verma asked how the products would be protected. Ellison's answer was biometrics. Logins use voice, fingerprint, and face, and he said multiple factors can be required together. Oracle also looks for obvious anomalies. "If you're a cardiovascular surgeon at Cedars Sinai in L.A. and you're in Uzbekistan doing robotic surgery, we're skeptical," he said.
On timing, Ellison said the patient portal reaches early adopters in November. The other modules follow through January, and early adopters should have the full set by February 1.
What the session did not include is a demo, independent validation, or technical documentation. Oracle did not say which FHIR resources the modules read and write, or how a write to a hospital's system of record gets confirmed and audited. It did not explain what "AI implementation" means in practice, how messy source data gets handled, or how the biometric login fits with a hospital's own identity systems. Regulatory approvals and pricing didn't come up either.
Those are the questions for any engineer who gets early access. The pitch is that the integration work, which has historically taken years, becomes a switch. If that holds for even a few modules, it changes the build-versus-buy math for health systems. Until someone outside Oracle runs it against a live Epic environment, it remains a claim from the vendor's founder on a stage.
