As an MD You Are Using 5% of Your AI. Here Is the Other 95%.
A doctor colleague told me last week she uses Claude every day. I asked her how. “I type a question and it gives me an answer.”
That is it.
That is how she uses it.
She is a board-certified ObGyn running a busy practice.
An MFM.
And she is using the most powerful AI tool available the same way she would use Google in 2005.
She is not alone. Roughly 95% of physicians who use AI stop at the chat window. They ask a question, get a response, close the tab, and next time start from zero. The AI forgets them. They forget the AI can do more. Both sides lose.
This post is a plain-language guide to what Claude can actually do, written for ObGyns who have used the basic chat but nothing else. No coding. No technical background. Just the features that matter, explained in the order you should learn them.
What You Already Know
Chat is the basic window. You type, Claude answers. Think of it as texting a very smart stranger who has never met you before and will forget you the moment you close the conversation. Every session starts fresh. No memory, no context, no relationship.
Extended Thinking is a setting that tells Claude to reason more carefully before answering. Instead of giving you the first plausible response, it works through the problem step by step. For anything clinical or analytical, turn it on. It takes a few extra seconds. The answers are meaningfully better.
Opus 4.6 is Claude’s most capable model. Always select it. The default model is faster but less thorough. You would not send your resident to handle a complicated delivery when your MFM specialist is available. Same principle.
The Features That Remember You
This is where most physicians never go, and it is where the real value starts.
Projects are workspaces where you upload files and write instructions that Claude remembers across every conversation in that project. I have a project for CDC natality data analysis. Every time I open it, Claude already knows my datasets, my variable definitions, and my analytical preferences. I do not re-explain anything. Think of Projects as hiring the same research assistant every day instead of training a new one each morning.
Context Files are documents you drop into a project so Claude can draw on them. Your practice protocols. Your patient education handouts. Your favorite review articles. The more context you give, the less prompting you need. Claude reads them before it reads your question.
About-me.md is a single file that tells Claude who you are, what you do, and how you think. Your specialty. Your clinical interests. Your preferred communication style. Your pet peeves. Write this once, and Claude stops giving you generic answers. It gives you answers shaped by your expertise and priorities. This one file changes everything.
The Features That Work Without You
Skills are sets of detailed instructions that live inside Claude and fire automatically when triggered. I have skills for summarizing medical studies, writing peer reviews, analyzing CDC data, and drafting LinkedIn posts. When I say “Summ” and attach a PDF, Claude does not ask me what I want. It already knows: 400–500 words, three sections, 7th-grade readability, output as a Word document. Every time. Without being told twice.
Skill Creator is Claude’s built-in tool that interviews you about a process and then generates the skill for you. You do not need to write code. You describe what you want, answer a few questions, and Claude builds the instructions. If you find yourself giving Claude the same directions more than twice, you need a skill.
Cowork takes this further. It reads your files, creates documents, runs code, and executes multi-step tasks. Think of it as the difference between dictating a letter and handing someone your entire filing cabinet and saying “prepare the quarterly report.” Cowork can access a folder on your computer, work through the files, and produce deliverables. I use it daily.
Scheduled Tasks let Claude run jobs automatically on a schedule. Set it once, and it runs without you. As long as your computer is on, the task executes at the time you specified.
The Features That Connect Everything
Connectors plug Claude directly into your existing tools. Slack, Google Calendar, Gmail. Claude does not just answer questions about your schedule—it reads your calendar, drafts emails, and acts inside the applications you already use.
Global Instructions are persistent instructions that apply across every conversation, not just within a single project. Set them once. They run forever. Mine tell Claude my department affiliation, my writing standards, and where to save files. I never repeat these instructions.
My Take
I have been writing about AI in medicine since before most professional societies acknowledged it existed. The technology is not the bottleneck. The bottleneck is physicians who open the basic chat, get a mediocre answer to a poorly framed question, and conclude that AI is overhyped.
It is not overhyped. It is underconfigured.
The difference between basic chat and a fully configured Claude is the difference between asking a random medical student a question in the hallway and working with a trained research fellow who knows your data, your preferences, and your standards. The fellow does not just answer—she anticipates. That is what a configured AI does.
You do not need to learn all of this at once. Start with one thing: create a Project, upload your most-used protocols, and write a short about-me file. Use it for a week. You will not go back to the basic chat.
The 95% of Claude you have not touched is where the actual value lives.