Someday soon, doctors who don’t run their patients’ records through an AI model will be considered negligent.
At least, that’s what Steve Brown thinks.
He’s the founder of CureWise, an app designed to guide patients through cancer care with the help of AI. He’s also a cancer patient well on the way to recovery — thanks in part to the AI medical “agents” he created to help better understand his diagnosis.
In a recent essay for Stat, the Stanford-educated physicist and computer scientist described turning 60 and knowing something “wasn’t right.”
He’d been struggling with low energy and weight loss, with an unexplained feeling of discomfort in his lower abdomen. He went to his regular doctor and ordered one of everything: Full-body scans, colonoscopy, endoscopy, cardiac function tests. “Every lab test I could get approved,” he said.
His doctors, whom he didn’t identify, told him to take some GasX — a pretty underwhelming response to something he knew was more serious, he said on an episode of a podcast called “The Neuron.”
Soon after, Brown found himself in a new community in California’s Palm Desert, with all new doctors, after his house burned down in the Palisades fire.
A brief stint in the ER from what he thought was an intestinal blockage caused by a big steak dinner turned into a meeting with those new doctors who saw his case “with fresh eyes.”
“Within days,” Brown wrote, “they found what everyone else had missed: an aggressive form of blood cancer in my bone marrow related to multiple myeloma.”
They’d caught it early, but the disease was already weakening his vital organs, and he was in urgent need of solutions.
Grappling with this news, his mind wandered to AI. He wanted to understand how his previous doctors could have missed such a consequential finding. So he got to work.
Brown built a “medical AI agent” that he named Haley and fed it “the exact same data that all those doctors had seen just weeks earlier,” including his full MyChart history, labs, imaging results and doctor notes.
The AI agent immediately spotted the cancer red flags: “mild anemia, elevated ferritin, low immunoglobulins — signs of immune dysfunction and bone marrow issues.”
Brown’s human doctors back in LA might have written those results off as individually unproblematic. But when analyzed together, they spell a potentially lethal combination.
Haley recommended a “serum-free light chains” blood test and bone marrow biopsy, which the original doctors should have run several weeks prior.
“If my doctors would have had AI before, I would have gotten this diagnosed probably a year ago,” he reflected later on the podcast. “And if I had had it, I would have asked for the right tests.”
Ultimately, it was Brown’s Palm Desert doctors who diagnosed him with the cancer, not the AI. But he went back to his machines for guidance on how best to treat it, especially when the regimen he was initially prescribed seemed less and less effective with each passing week.
He confirmed that suspicion when, looking back into his medical records, he discovered a genetic marker that indicated his current course of treatment might be unsuccessful.
The AI introduced him to “other newer, more aggressive regimens” and “off-label combinations” he might respond to better.
“That information wasn’t part of the first-line protocol. […] No one had the time to sift through all of the research,” he wrote. “AI, though, had all the time in the world.”
After discussing the findings with his doctors, they decided to try daratumumab plus venetoclax, “a targeted but off-label combination that hasn’t had full clinical trials, simply because there aren’t enough people with this rare disease.”
Now, his “key cancer marker is back in the normal range.”
Brown’s research was vastly more extensive than just plugging some questions into ChatGPT to try and self-diagnose. Crucially, he never let his AI agents replace his doctors. They simply “supercharged” his decision-making.
“It helped me skip the trial-and-error stage and suggested a plan that made sense for me,” he wrote.
But he acknowledges that there’s a lot of anxiety about AI in the medical community right now, and for good reason.
It’s sometimes hard to imagine how the same technology that gives us racist slop, sexual deepfakes and rampant misinformation could be used in sensitive medical settings. But Brown is adamant that AI can be a tool that makes doctors and patients smarter.
“This is the first time we have the technology that can handle the level of complexity that we need,” he said on the podcast.
“If 1 million people get cancer, it’s 1 million different genetically unique diseases — it’s your own unique genes, mutated in some unique way. We lump those things together and say it’s all cancer, but really it’s a million different diseases.”
“AI is the first thing that can work with so many variables at the same time,” he added.
His AI agents were able to synthesize his test results and find patterns within them that his original doctors missed because of simple human error.
At its most simple, his goal is to empower patients to ask better questions of their doctors.
“If you study up on what’s going on and you go in and use that 10 minutes with your doctor in a more enlightened way, you’re gonna get better results,” he said.
“Everything I’m on was prescribed by doctors, but the AI coached me on how to talk about it.”
