The Question That Wouldn’t Let Me Go
I recently listened to a conversation that gave words to a question I’ve been circling for years but didn’t know it. On the Healthcare Business Podcast out of the University of Texas, host Anne Dancarnuta spoke with Dr. Matt Brubaker, Chairman and CEO of the healthcare consulting firm FMG, about artificial intelligence, healthcare transformation, and the importance of keeping humans centered amid this rapid change.
One idea in particular stayed with me: the concept of the “last 5%.”
Across industries, AI is rapidly absorbing the first 95% of work. Pattern recognition. Data extraction. Drafting. Summarizing. Flagging abnormalities. In radiology, this is not hypothetical. We are watching it unfold in real time. The uncomfortable, but necessary, question that follows is simple:
If AI is closing the first 95%, what exactly am I doing with the last 5%, the part where my real value lives?
That question applies not only to radiology, but to every physician who intends to remain relevant, human, and indispensable over the coming decades.
Radiology as a Living Case Study
Radiology may be the clearest early case study for how AI will reshape medicine. Algorithms are already highly effective at detecting pulmonary nodules, identifying pneumothoraces on chest radiographs, flagging acute appendicitis, pulmonary emboli, intracranial hemorrhage, and more. The pace of improvement is accelerating.
What is worth remembering is this: the true value of a radiologist has never been limited to simply finding things.
Just as an internist does not personally analyze erythrocytes under a microscope, the radiologist of the future will not be defined by raw detection. The value lies in interpretation, prioritization, integration, and judgment. It lies in placing findings into clinical context, weighing significance, anticipating downstream implications, and communicating nuance clearly.
AI will increasingly deliver the findings. Physicians deliver meaning.
That meaning lives squarely in the last 5%.
What AI Has Actually Changed for Me
I am dual board certified in diagnostic and interventional radiology, and I already use AI daily. What has surprised me most is not just where the gains are, but where they are not.
Diagnostic Radiology: Efficiency With Less Cognitive Waste
In diagnostic radiology, the early gains are obvious. Improved efficiency. Reduced fatigue. Cleaner impressions generated from already identified findings. Fewer cognitive interruptions related to formatting, punctuation, and structure.
Large language models are exceptionally good at turning structured findings into concise, readable impressions. That does not eliminate physician responsibility. It concentrates it. I still live in that final review, that last 5%, ensuring accuracy, relevance, and clinical judgment. What has changed is that I am no longer burning cognitive energy on basic English or formatting tasks that I did not go to medical school to perform.
This is only the beginning of what AI will change on the diagnostic side.
Interventional Radiology: Less About Speed, More About Presence
Interventional radiology is different.
AI has not dramatically improved procedural efficiency inside the room, and that is expected. IR is inherently hands-on, patient-facing, and dependent on real-time decision making. Where AI has been transformative for me is in the quality of care and the quality of human connection.
I have experimented with several HIPAA-compliant AI medical scribes and ultimately settled on Commure Scribe (I receive no endorsement and am not in any way supported by or receive compensation from Commure Scribe or any of its affiliates). The reason was simple. It gave me the quality of time with my patients back to me.
Instead of typing, clicking, editing templates, and searching through notes, I can sit face to face with a patient, review imaging together, read body language, and listen fully. I am able to counsel them as a person rather than as a throughput problem.
Interestingly, the scribe often captures clinically relevant details I might miss while focused on imaging. Offhand comments, clarifications, or subtle cues that matter. The result is not just time saved, which is modest, perhaps five to ten percent, but a materially higher quality clinical encounter.
That is not efficiency. That is better medicine.
The Hidden Bottleneck in Interventional Radiology
There is another uncomfortable truth about efficiency in interventional radiology. As operators, we realistically control only five to ten percent of total procedural efficiency.
I can perform a uterine artery embolization in under an hour. That speed becomes irrelevant if room turnover is slow, supplies are missing, equipment has to be borrowed from another lab, or patient transport is delayed. These peri-procedural inefficiencies, staffing, supply chains, equipment tracking, and logistics, are where AI should ultimately have its greatest impact in IR.
Not by replacing physicians, but by smoothing the systems that surround us.
That is how access to care actually expands.
AI as a Force Multiplier for the Idea-Seeking Physician
For someone like me, an idea seeker and connector who thrives on mentorship, conferences, and synthesis, AI has quietly expanded my bandwidth.
I now routinely record conversations with colleagues (with their permission of course!!!), conference sessions, and planning discussions. Transcripts are generated immediately. Key ideas surface quickly. Concepts are integrated into protocols, procedure planning, and workflows while they are still fresh.
Instead of notes sitting untouched in a folder for months, ideas move rapidly from acquisition to synthesis to application. I still own the final judgment. I still live in the last 5%. AI simply accelerates everything upstream.
That is real leverage and the satisfaction of seeing these ideas more quickly and efficiently added to my practice is incredibly rewarding.
The Real Risk Is Not AI
My fear is not that AI will replace physicians.
My fear is that physicians who refuse to engage with AI will slowly make themselves obsolete.
Every knowledge profession eventually faces this moment. If we do not actively position ourselves at the leading edge of the first 95%, partnering with tools, shaping workflows, and redefining our role, the question eventually becomes why the role exists at all.
The answer must remain clear. Experience. Integration. Judgment. Human connection.
Those things cannot be automated. But they can be devalued if we pretend the world is not changing.
So What Are We Actually Doing
The question I ask myself, and the one I believe every physician should ask, is straightforward.
What am I doing today to protect, sharpen, and expand the last 5%?
That means letting AI handle what it does best. Refusing to spend physician time on clerical nonsense. Investing in tools that improve presence, not just speed. Doubling down on judgment, context, and human care.
Radiology just happens to be the early warning system.
The rest of medicine is not far behind.
Footnotes
Karnuta D, host. Centering Humans in Healthcare AI Implementation. The Business of Healthcare Podcast, Episode 129. University of Texas at Dallas; July 25, 2025. Guest: Matt Brubaker, MD.