Four Questions with Kimberly Powell

This conversation has been edited for length and clarity.

Kimberly Powell is vice president of healthcare at NVIDIA where she has been for the last 18 years, building the company's healthcare business. Her journey with NVIDIA started with a single cold call she made in 2008 when she was an engineer at a 60-person radiology startup outside Boston. Christina Farr spoke with her on *Lifers* about where AI actually lands in clinical work, what changed in the last six months, and what the inside of a hospital looks like five years from now.

Christina Farr: Prominent VCs keep saying AI will replace physicians. There may be areas — radiology, pathology — where aspects of the job get automated. But it's hard for me to imagine we're going down a path of wholesale replacement. Where do you land on augmentation versus replacement?

Kimberly Powell: Jensen says it really well, so I'll describe it the way he does. A radiologist is trained to help diagnose patients and figure out the critical path to get them well. What AI in radiology has done is maybe automate a particular task — finding the anomaly in an image, segmenting it. But the task is not the job.

The other way we think about it: the more we deploy AI, the more it increases the need for healthcare services. We are in a global crisis right now — an aging population, chronic disease, a huge healthcare professional shortage. Never mind what the next five years looks like. And by using AI systems we're discovering more things we can fix, which will need more healthcare services. We're actually exacerbating the need for healthcare professionals. The only way we can meet that pent-up and growing demand is through automation.

Companies like Abridge and OpenEvidence and Heidi Health are taking advantage of something that sounds simple: voice is becoming the new interface of healthcare. Clinicians get eye contact with the patient instead of typing. They see that a patient is authorized for the MRI right there in the encounter, so there's no delay in care. They go home and have dinner with their family instead of typing for another three hours.

Christina: Healthcare has been hard for a long time — siloed data, middlemen, inertia, incentives. Throwing money at it hasn't fixed it. What makes this moment different?

Kimberly: The advent of agents. An agent is a computer system that can actually do work. This is the moment where AI became useful, because it's not just generating a picture or writing an email — it's doing very complicated work. And that has really only happened in the last six months, to be honest.

Think about all of those antiquated but necessary systems that run the hospital — scheduling systems, electronic health records. You would otherwise have to write software to interact with all of them. Agents can now use API calls to pluck just the amount of information they need to move on to the next piece of work. Health records have been in practice in the United States for more than two decades and we've largely not gotten true value out of that digitized data. All of that is changing now.

It's also safer than people assume, because the data stays where it is. We no longer have to ask what if we could pool all of the healthcare data together to build and operate AI. It isn't necessary anymore. These systems can be deployed inside the walls of the hospital.

Christina: You've said the hospital itself becomes a robot within five years. What else should we expect in that window?

Kimberly: Think about how the hospital itself should be a robot. The hospital itself should have eyes and ears, sensing everything that's going on, using modern world foundation models and spatial intelligence to operate. There will be dozens of robots running around to make sure the patient has their medication on time, or to deliver the surgical tools to the operating room exactly on time so the procedure can start. You want the patient under anesthesia the shortest amount of time possible. Nurses who have dedicated their lives to the practice aren't delivering sheets to an empty bed — they're at the bedside.

Delivery robots are the near term; you don't have to build a new hospital to deploy them. Surgical and clinical robotics are further out. In drug discovery, AI is already standard in molecular design, and the hard problem left is truly understanding biology — efforts like the virtual cell. In the end we have a fundamental problem: 90% of drugs that go into the clinic fail.

The business model shift matters as much as the technology. For decades it's been: I buy a CT machine once, then I surround it with humans to operate it so I can charge for it. I said at a conference a year or two ago that AI is now hirable. Because it can do work, you can hire it. This isn't a software system you buy a license for — it has to be thought of as somebody who's on payroll. My hope is that we don't just offset the cost by letting billing complexity go up and everybody bills more. We're already starting to see healthcare costs rise, which is not what anybody wants, because in theory AI should do the opposite.

Christina: What do you tell a clinician who isn't technical, doesn't code, and is somewhere between excited and frightened by all of this?

Kimberly: It's often said that AI is not going to replace you, but the person using AI is going to replace you. Clinicians have to learn new tools every single day — they're inherently learning machines. So just start using it. You don't have to know Python. You don't have to write code anymore. You simply speak.

Use Perplexity, use Codex, use Claude, use OpenEvidence, use Abridge, because you'll see what the potential is. The uptake of a system like OpenEvidence is profound — never has the healthcare industry been faster at adopting a technology than right now, because it puts the world's current medical knowledge in a doctor's hands on a voice-activated interface on their phone. There's literally no barrier to entry. Until you try it, you can't quite put your hands on how it's going to feel, or what else you could do with it.

Watch or listen to the full *Lifers* conversation with Kimberly Powell →