Last December, I spent a full day at a health clinic observing how the team worked.
I wasn't there as a patient. I was there as a workflow observer — the kind of thing I do before proposing any AI or automation changes. Walk in, watch how people actually spend their time, and look for where the friction is. It takes longer than a questionnaire, but you learn things no survey would surface.
One of the things I learned was that Dr. Vo — the clinic's primary physician — was already using AI. Every day. For one very specific thing.
Most people I talk to assume that professional fields like medicine are still a few years away from meaningful AI adoption. The technology isn't trusted. The stakes are too high. The regulators haven't caught up.
Dr. Vo wasn't waiting.
What She Was Using It For
After every patient visit, Dr. Vo had to write a counseling narrative. Lifestyle recommendations, weight guidance, dietary changes, whatever the patient needed to take home and act on. It had to be clear, personalized, and clinically accurate.
It was good work. Genuinely important work. But it was also repetitive — the same structure, customized for each patient. And it took about 20 minutes per visit.
Now she uses Gemini, ChatGPT, or OpenEvidence to generate that summary. She provides the key clinical points, and the AI produces five clear, patient-friendly sentences in moments.
Same information. Same quality of communication. Twenty minutes back.
Per patient.
She wasn't using AI for diagnosis. Not for interpreting labs. Not for anything that required clinical expertise or judgment — those things she kept entirely to herself. She was using it to eliminate the translation layer. The part of the job that was never the point.
The Pattern I Keep Seeing
This is the same thing I see across industries, over and over.
Doctors write. Lawyers write. Consultants write. Engineers write reports. Accountants write memos. The expertise these people have is in their judgment — the thinking, the pattern recognition, the ability to make decisions others can't. The writing is just how they communicate that expertise to someone else.
And AI is very good at writing.
Not at replacing the expert. At handling the container around the expert's thinking.
This is what I mean when I talk about AI eliminating overhead rather than replacing people. Dr. Vo's clinical work is unchanged. Her ability to diagnose, to understand a patient's history, to make the call nobody else in the room is qualified to make — none of that touched. What changed is the 20 minutes of narrative writing that used to follow each visit.
Multiply that across a day of patients, and you're talking about a meaningful chunk of time returned to the actual work.
Why It Spreads This Way
Dr. Vo didn't get a mandate from hospital administration. There was no AI strategy from the top, no consultant brought in to transform the practice. She just noticed a tool existed, tried it on a task she found tedious, and kept using it because it saved her time.
This is how AI adoption actually happens in the real world, at least among the people who are doing it well. Not a top-down rollout. Not a comprehensive transformation project. One specific, boring task that someone decides to try on a Tuesday afternoon.
The principle I call Start Small, Iterate applies here exactly. The smallest reliable version — AI handles the counseling summary draft, doctor reviews and adjusts as needed — is also the safest version. There's no risk of AI making clinical decisions. There's no integration with the EHR required. Just a tool that takes key points and turns them into readable prose.
And because it works, it sticks. And because it sticks, you start looking for the next thing.
Finding Your 20-Minute Task
The question Dr. Vo's story raises is: what's your version of the counseling narrative?
Most professional jobs have them. Tasks that require some expertise to set up but are fundamentally repetitive in execution. Writing summaries of complex information. Translating technical findings into accessible language. Drafting the same type of email for the hundredth time with different specifics.
These are the right places to start — not because AI can't do more, but because starting here is fast, safe, and produces results you can see immediately. It builds trust in the tool before you ask it to do harder things.
The hardest part, usually, is noticing that the task exists. Most people are so used to the 20-minute writing task that they've stopped questioning it. It's just part of the job.
Dr. Vo started questioning it. That's all it took.
If you're trying to find where AI fits in your own workflow, the 4-Day AI Sprint walks through exactly this process — identifying the right starting points and building real automations without a technical background.
