"We reduced time to treatment in stroke by over an hour, which is really meaningful," says Chris Mansi, a neurosurgeon and co-founder of Viz.ai. Turns out artificial intelligence is more than vibe coding or summarizing dull meetings. It is also a lifesaver.
A stroke can be ischemic, clots blocking the flow of blood to the brain, or hemorrhagic, blood vessels rupturing in the brain. After a hospital's stroke team assesses your speech and balance, they order a CT scan to figure out the cause. Blockages can be treated with tPA medication to dissolve clots, while brain bleeding requires neurosurgery.
"Time is brain," doctors say -- two million neurons are lost each minute during an ischemic stroke.
AI to the rescue. Dr. Mansi founded Viz.ai in 2016 with David Golan, a machine-learning expert, who weeks earlier was at Stanford Hospital with a suspected stroke. "He saw the problem through the eyes of a data scientist, thinking, 'How do they not have technology to read these scans?' " Dr. Mansi says over lunch in San Francisco. Back then at a typical hospital, "workflow was very messy. It was 13 or 14 different calls and handoffs," lining up doctors and procedures. For hypertrophic cardiomyopathy -- the thickening of heart muscles -- it was 56 steps.
Viz.ai's process of using AI to create care pathways begins with algorithms that assess scans. By automating many of the next steps, including logistics between doctors and staff, the company shortened initial treatment time by as much as 88 minutes. It also reduced treatment time variability from two hours to seven minutes. "Each patient is getting treated faster but more consistently as well."
In 2018 Viz.ai was the first to get Food and Drug Administration clearance for AI system use in healthcare. It took some education, Dr. Mansi recalls: "They thought it was for helping the radiologist. But if you want to help the patient, you can't just help the radiologist or the neurosurgeon, the individual doctor. You have to help the entire workflow. It's a system problem, and once we explained that, they realized this was good for American patients."
Viz.ai ran clinical trials, many of them in the Southeast, a region known as the "stroke belt" for the disease's high prevalence. After FDA approval, hospitals in Chattanooga, Tenn., and Atlanta rolled out Viz.ai's platform. It is now in 2,000 U.S. hospitals covering 230 million people, about two-thirds of the American population -- and 35 hospitals in Europe. Business is growing 40% a year.
By solving the workflow-efficiency problem, "we create high-agency doctors," Dr. Mansi says. "Twenty percent of the problem is detecting disease. Eighty percent is organizing the workflow so the patient gets to the right specialist to diagnose and see the patient." Viz.ai creates an "AI care pathway that takes a patient from initial tests all the way through to definitive treatments." It isn't an AI doctor, but "you've got the world's best fellow working for you all the time."
Selling any software to hospitals is hard, so Viz.ai set up a platform to handle issues beyond stroke care. "We have 55 pathways available" for different treatments, Dr. Mansi says, including pulmonary embolism, chronic obstructive pulmonary disease and cancer. "Thirty-five to 40 of those are from other companies, including Microsoft, that have come to us and said, 'Hey, can you deploy this detection algorithm into your health system?'" Hospitals have rules set up: "We implement them into software."
Here's where it gets really interesting. The platform's inputs are "imaging of any type -- ECG, heart tracings, echocardiograms, blood tests, labs, genetic tests and, importantly, the electronic health record," Dr. Mansi says. "A year and a half ago, every single care pathway that we built was imaging-based as a trigger." An AI scan of an image would set a patient on a treatment plan. "This year, 90% of our care pathways that we've launched have been electronic-health-records-triggered."
Could this mean earlier detection, before symptoms? "We're now doing probably 90% of the work of the doctor automatically, before the doctor's even seen the patient, which is magical for that doctor, because they come in, everything's organized for them," Dr. Mansi says. So the AI does 90% of the grunt work? He nods: "But not the word I'd use. We're giving them guidelines and they're deciding. We've not had one complaint."
Viz.ai is working with Anthropic and talking to OpenAI and others. "Claude means we can now synthesize a 900-page electronic health record and get the signal. And that allows us to expand the number of diseases to, really, every single disease." The goal is to head off harm by finding a disease before it causes symptoms. "It fits into the narrative that should be told more, which is an AI for good."
There are AI competitors like AIdoc, which started with pulmonary embolisms, and OpenEvidence, which doctors use to help diagnose diseases. The more the better.
Dr. Mansi is thinking about the big picture. "The agentic evolution is doing more work for doctors, nurses and administrators. It's all the different tasks they do, because if you can take some of the burden off them, more patients are treated. Their economics get better and patients get better. That's my big focus."” [1]
1. Inside View: The Algorithm Will See You Now. Kessler, Andy. Wall Street Journal, Eastern edition; New York, N.Y.. 20 Apr 2026: A13.
