There's a specific kind of uncomfortable that happens when you're an expert in one thing, standing in a room full of experts in another thing, and everyone is moving fast.
I felt it this week in the OR with Dr. Lee, a craniofacial surgeon I've been shadowing. Craniofacial surgery lives and dies on imaging — every reconstruction starts with a CT. She was working through one with the kind of gestural fluency you only get from years of seeing the same patterns. She wasn't reading the scan — she was seeing it. Instantly. Automatically. And I was standing there doing pixel-level math in my head, trying to catch up.
Here's the thing: I have 10 publications. I spent four years building brain-computer interfaces. I've written code that reads neural signals in real time. And I stood in that room feeling like I was back in my first week of med school.
That discomfort is data.
The CS Side Isn't Much Better
The same week, I had a meeting with Dr. Savjani about the kind of work I'd been doing with agentic coding tools — teaching Claude Code to explore codebases, building Claude-based sub-agent pipelines, scaffolding workflows for research labs that don't have dedicated engineers.
He called it a "perfect fit."
I appreciated that. And I also knew that "perfect fit" is relative. I'm not a senior engineer. I'm a second-year med student who learned to code as a teenager and kept going. The tools I'm building are good — but I'm also aware of exactly how much I don't know. There's a whole architecture layer, a whole distributed systems layer, a whole infra layer where I'd still be the slowest person in the room.
Two Worlds, Both Humbling
What I've been sitting with is this: I chose a path that keeps me permanently intermediate. Not by accident. On purpose.
I'm not going to be the best radiologist in any room. I'm not going to be the best engineer in any room either. What I'm trying to be is the person who understands both rooms well enough to build something neither room would build alone.
The gestalt that Dr. Lee has — the thing that doesn't show up in DICOM metadata, the thing that makes her pattern recognition feel like intuition — that's the thing I want to help capture and explain. Not replace. Explain. Make legible. Build an attention layer for. Craniofacial surgery is as imaging-dependent as it gets: bone defects, orbital walls, mandibular reconstruction — it's all CT. Which is exactly why it keeps pulling me toward radiology.
And the engineering work I'm doing with Dr. Savjani — the agentic workflows, the research automation — that's not me being a software engineer. It's me being a clinician who can translate between two languages.
Why Beginner Is the Right Feeling
There's a version of this story where I feel inadequate. Two domains, both humbling, no clear mastery in either.
But I've started thinking about it differently. The beginner feeling isn't a problem to solve. It's the sensation of doing something worth doing.
If I felt comfortable in the OR and in the code review, it would mean I'd stopped pushing. The discomfort is the sign that I'm still at the edge of what I know — which is the only place where anything interesting happens.
I have a long way to go in both directions. That's not a bug. That's the whole point.