CVS tackles tactical agentic AI rollouts for healthcare

Agentic AI and the patient experience

Technology vendors often sell a dream of agentic AI solving huge problems -- like unifying siloed data systems or adding structure to mountains of unstructured data -- by just feeding everything into an LLM and turning on their products. Users suddenly become "AI companies" in an amorphous "AI transformation."

Salesforce CEO Marc Benioff envisions a world where companies "kind of run themselves" with AI agents, as he put it at the World Economic Forum in Davos, Switzerland, last January. Such aspirational marketing isn't confined to just Salesforce; it's just one of many companies making this claim.

CVS's Pal sees potentially interoperable agents deployed in a carefully architected environment designed to handle growing complexity over time. Agents may work together, managed by "super agents" orchestrating their work.

But that's in the future.

CVS will, for the moment, focus on more tactical, task-based agents and build slowly and deliberately to control how it enhances and deprecates agentic capabilities as business needs dictate -- without incurring technical debt. It reminds Pal of the emergence of modern APIs, which, like agentic AI, enabled connections between applications and their data.

"The technical environment is changing very fast; we do not know who will be the key players, who will be ahead of the game, six months or 18 months from now," Pal said. "We do not want to create too complex a task within one agent. We'd rather they do a discrete set of tasks -- and attach them together very similar to a Lego set -- that way we can modify and alter it very quickly, as it comes to end of life."

To its credit, the Agentforce Health team isn't pushing the AI transformation theme as hard as some in Silicon Valley. They've picked a swim lane, said Amit Khanna, senior vice president and general manager of Agentforce Health: making healthcare efficient, which, in theory, will free up bandwidth for providers and allied professionals to care for vulnerable patients before they become high-risk cases.

"We are, as a system, reaching out to only people who are at the highest risk," Khanna said. "We are not able to reach out to people who are at rising risk."

There's a lot of manual, nonclinical work that Agentforce can automate, he said, such as helping complete documentation before and after a patient visit. On top of that, follow-up communications and appointment scheduling are also prime candidates for automation.

Khanna echoed Pal's sentiment that IT leaders should build agentic AI functionality slowly and start small -- so they can claim some quick wins on the way to bigger investments.

"Sometimes they end up picking up the most complex [processes], where they think that the most value would be," Khanna said. "But then it takes time, and you start losing faith. The idea that you should be able to give proof of value quickly is something that all customers should think about. It might be a small win, it might mean just saving you one second -- but it's proof that you are moving toward that AI journey rather than doing something that takes 18 months."

Don Fluckinger is a seasoned B2B technology journalist with more than 30 years of experience specializing in enterprise IT, digital experience and content management. As a senior news writer at Informa TechTarget, he delivers award-winning analysis that helps IT and business leaders navigate complex technologies to enhance customer and employee experiences. Got a tip? Email him.