At Spokeo, we believe in the power of inquiry — not just into public records, but into the deeper questions that shape our world. That’s why we’re proud to support students who challenge boundaries and think critically across disciplines. This year’s Spokeo Scholarship winner, James LeeWarner Maxwell, exemplifies that spirit.
Hailing from Ashburn, Virginia, James is a Biophysics pre-med student at George Washington University with a long-term focus on AI-assisted interventional radiology. A graduate of Briar Woods High School, he completed dual enrollment coursework at both Georgetown University and George Mason University, where he worked on AI models for early detection of endometrial cancer and explored whether lightweight vision models could approximate radiological reasoning in resource-limited settings. A member of the National Honor Society, the Society of Torch & Laurel, and the National Society of High School Scholars, James was also recently honored at the Congress of Future Medical Leaders.
James’s winning essay explores one of medicine’s most overlooked failure points: not a lack of intelligence, but a lack of continuity. He argues that the greatest breakthrough application of agentic AI won’t be a smarter diagnosis at a single point in time, but a system that holds the thread across appointments, providers, and months — noticing patterns no single overworked clinician has the bandwidth to connect. He grounds the argument in his own research background and, movingly, in his family’s experience with his mother’s cancer diagnosis. The result is an essay that is technically rigorous, emotionally resonant, and clear-eyed about what’s actually standing between patients and earlier intervention.
Who's Calling Me?
Search any phone number to learn more about the owner!
The Diagnostic Agent: Betting on AI That Notices What Medicine Misses
By James LeeWarner Maxwell, George Washington University
If I had to bet on one breakthrough application of agentic AI that will change the world, I would not choose education, productivity, or automation.
I would choose a diagnostic agent that follows a patient over time, one that doesn’t wait to be asked, doesn’t reset after each appointment, and doesn’t lose sight of what came before.
Medicine today is powerful, but discontinuous. A scan is read. A lab is reviewed. A visit ends. Then the system forgets, until something becomes urgent enough to demand attention again. The failure is not intelligence. It is continuity. And in medicine, the cost of forgetting is rarely measured in inconvenience. It is measured in time a patient did not have.
Agentic AI changes that structure.
An autonomous diagnostic agent would not simply analyze medical data. It would manage the unfolding sequence of diagnosis itself. It would notice a symptom reported weeks apart, connect it to subtle changes in imaging, determine whether follow-up is warranted, and escalate to a clinician with context already assembled, not fragments scattered across systems and time.
It would behave less like a tool and more like a persistent clinical memory.
I am not describing this from a distance.
At Georgetown University, I worked on AI models for early detection of endometrial cancer using medical imaging. At George Mason University, I studied whether lightweight vision models like MobileNet could approximate radiological reasoning in environments where imaging infrastructure is limited.
Across both projects, one constraint kept appearing in different forms: the models were capable of detecting disease, but the system around them was not capable of acting on that detection consistently, safely, and at scale.
That gap is where agentic AI begins to matter.
Today’s medical AI is still episodic. A human initiates a task, an algorithm responds, and the result returns to a human decision-maker. Each step is dependent on the next prompt, the next order, the next interpretation.
An agentic diagnostic system removes that fragility. It connects steps into a governed sequence: signal detection, risk evaluation, escalation, follow-up coordination, and longitudinal tracking. Not as a single prediction, but as a managed process over time.
That difference is not cosmetic. It is structural.
And structure determines access.
In well-resourced hospitals, continuity is often implicit, carried in the memory of teams, repeated visits, and specialist availability. Outside those environments, continuity breaks easily. A missed referral, a delayed scan, an unreviewed result, these are not rare events. They are ordinary ones.
Consider a patient in a rural clinic who reports mild fatigue in March, an elevated lab marker in June, and a slightly abnormal scan in October. Each event, in isolation, looks unremarkable. A single overworked physician, managing hundreds of patients, has no realistic way to connect three data points spread across seven months and three separate visits. An agentic diagnostic system does not have that limitation. It does not get tired, it does not forget, and it does not lose the thread between March and October. It simply holds the pattern until the pattern becomes undeniable. By the time a human finally notices, in a system without that memory, the patient is no longer presenting with mild fatigue. They are presenting with a diagnosis that arrived months later than it should have, and a prognosis that paid the price for the delay.
An autonomous diagnostic agent would not solve healthcare inequality. But it would reduce one of its most persistent mechanisms: the loss of information between steps that no single clinician has time to hold together.
I think about this differently now because I have seen, up close, what early detection makes possible, and what its absence costs.
My mother was diagnosed with cancer and given a prognosis measured in months. I was fifteen, and for a long time I sat with the quiet terror that the timeline doctors gave us was simply what was left. She is now in remission. That outcome was only possible because we had immediate access to imaging, specialists, and coordinated care, the very continuity I am describing, available to us by circumstance rather than guarantee. I have never been able to stop thinking about the families who receive the same diagnosis without that same continuity, and what gets lost in the months no one was watching closely enough to notice.
Most people assume medicine fails through lack of knowledge. In practice, it more often fails through delay. And delay is rarely dramatic. It is quiet. It looks like a missed pattern, not a missed miracle.
The technical foundation for diagnostic agents already exists. Vision models can detect patterns in imaging. Language models can coordinate multi-step reasoning and tool use. What does not yet exist is the integration layer: a system that can operate across modalities, over time, under strict clinical governance, without losing interpretability or accountability.
That is not a science fiction problem. It is a systems engineering problem with medical consequences, and the people who build it will be deciding, in effect, whose disease gets caught in time and whose does not.
I plan to work on it directly.
At George Washington University, I will study Biophysics on a pre-med track with coursework in computer science and artificial intelligence. My goal is not only to use medical AI, but to help design systems that can be trusted with real clinical workflows, systems that extend specialist reasoning rather than replacing it.
I am betting on autonomous diagnostic agents because they represent one of the few applications of agentic AI where better intelligence directly translates into earlier intervention, and earlier intervention changes outcomes in a way no downstream optimization ever can.
Other agentic systems will improve how we work.
This one changes whether some people are still alive in time to work at all.
AI Usage Disclosure
In drafting this essay, I used Claude (Anthropic) as a writing and editing tool. My process began with my
own original draft, written entirely in my own words, in which I developed the central concept of an
autonomous diagnostic agent and the core argument around continuity, inequity, and delay in medicine.
I then worked with Claude to expand specific sections, including a concrete rural clinic example and
additional reflection on my own family’s experience with my mother’s cancer diagnosis, while preserving
my original structure, voice, and key lines throughout. I reviewed, edited, and approved every sentence
in the final version. The central idea, the argument, and the personal stakes are my own; Claude assisted with expansion and refinement of language in select passages.