AI can scan a million lines of code in seconds. It finds patterns, traces connections, and surfaces problems that would take a human days to uncover. That’s real. The reach of these tools is extraordinary.
But the same model that sees everything understands very little.
AI looks at a codebase the way a satellite looks at a landscape. Everything is visible. Nothing is hidden. But satellite imagery doesn’t tell you why a road bends where it does. It sees the shape of things, not the reasons, needs and intentions behind it.
Clinical software is no different. Behind every workflow, there’s a story. A confirmation step that seems to slow things down but exists because speed kills in medication decisions. A documentation flow that feels heavy but was shaped by years of lessons learned in patient safety. In healthcare, slow is often fast. Understanding comes from intention, not reaction. But AI sees that pause as friction to eliminate. It doesn’t see the thinking that the friction protects.
This creates a specific failure mode: confident irrelevance. The output is clean, elegant, technically impressive. And completely wrong for the situation. Because the model didn’t make a decision. It made a prediction. A decision weighs context, consequences, and trade-offs that are fundamentally human. A patient’s anxiety. A physician’s cognitive load at 2 AM. A safety rule that exists because someone was harmed when the system moved too fast.
AI is at its best when it liberates. When it absorbs the repetitive and the draining so humans can focus on what requires judgment, presence, and care. But the moment it operates without a human in the loop, it loses the only thing that keeps it grounded. The tool serves the hand. The hand doesn’t serve the tool.
It’s the physician who also codes whose judgment the AI can’t replicate. Not because he writes better functions, but because he’s held the chart and felt the weight. He knows that behind every data point there’s a person with fears, with pain, with a family in the hallway. He builds clinical tools with a specific intention: if AI handles the notes, the doctor can handle the care. Look up from the screen. Listen. Be present. Cultivate the empathy and trust that no algorithm will ever replicate.
The antidote to the paradox isn’t less AI. It’s more humanity. More listening, more curiosity, more humility. The tools will keep getting more powerful. The question is whether we’ll stay grounded enough to use them well.
Martin Pham Dinh 