Where to start with AI when you’re a busy physician

A colleague pulled me aside in the lounge last week. He said: “I keep meaning to sit down and learn this AI stuff, but I don’t have a weekend to give it.”

I told him the truth: it doesn’t take a weekend. It takes about thirty minutes.

Here is the version of that conversation I wish I could give to every physician who feels a step behind. No jargon. No stack-building. One tool, one task, one win. Then you decide whether to keep going.

Before you start: the only thing you need to know about “AI”

For the purposes of this post, AI is a tool that takes whatever you type, plus whatever you paste or upload, and writes back to you. That’s it.

It’s not a search engine. It’s not a calculator. It’s a fast typist with a lot of general knowledge, who is happy to take a first crack at almost any non-clinical task you put in front of it.

That is the entire mental model. If you have that, you have enough to start.

Step 1: Pick one tool. Don’t shop.

Open a browser tab. Go to one of these:

• Claude at claude.ai
• ChatGPT at chatgpt.com

Either is fine. Pick one. Both have free versions that are good enough for what you are about to do. Do not spend the next twenty minutes reading “Claude vs ChatGPT” comparisons. They are 90% the same for what you need today.

You can read my longer post on the four tools I use later. Right now, just pick.

Step 2: Pick one task. Make it small. Make it non-clinical.

The most common mistake people make when they first sit down with AI is picking a task that is too big. “Help me organize my finances” is too big. “Help me plan my career” is too big. “Help me be more productive” is way too big.

Pick something small and concrete. Examples that work well for a first time:

• A meal plan for the week based on what’s in your fridge.
• A first draft of an email you have been putting off writing.
• A packing list for a trip you have coming up.
• A short list of birthday party ideas for your kid.
• A summary of a long article you have been meaning to read.

Notice what these have in common. They are non-clinical. They are bounded. There is a clear “done.”

Pick one. Just one.

Step 3: Write more in the prompt than feels natural.

This is the single biggest unlock, and almost nobody does it on their first try.

The instinct is to type something short, like:

“meal plan please”

This works the way you would expect. You get a generic answer. You get frustrated. You close the tab. This is the moment most physicians decide AI is overrated.

Try this instead. Write five sentences, not five words. Give it context, constraints, and the format you want.

“I’m a physician. I have two kids ages 7 and 10. We don’t eat pork. My partner is gluten-free. I want a dinner plan for Monday through Friday this week. Each dinner should take 30 minutes or less to make on a weeknight. We have a Costco run on Sunday so we can stock up. Please give me the plan as a table with the meal, the prep time, and a short grocery list at the bottom.”

You will get back a real, useful plan. With a grocery list. In the format you asked for.

The point is not that this is a magic prompt. The point is: the more context you give it, the better the answer. A few extra sentences upfront save you fifteen minutes of refinement later.

Step 4: Read the output. Refine if needed.

Look at what it gave you. Almost always there is something to fix. Maybe one of the meals uses an ingredient you don’t have. Maybe the prep time is unrealistic. Maybe the tone is off.

You can just say so. Like a conversation.

“Swap Wednesday for something using chicken thighs instead of salmon. We already had salmon on Monday.”

“The Thursday recipe sounds boring. Give me a different one with bolder flavors.”

“Add an estimated cost to each meal.”

Two or three rounds of this and you will usually have something genuinely useful. That whole loop takes less than ten minutes.

Step 5: Notice what just happened.

You sat down for thirty minutes. You finished an actual task you had been putting off. You did not have to learn anything about agents, models, or prompt engineering frameworks.

That’s the whole on-ramp.

If it felt good, repeat it next week with a different small task. If it felt like a slog, you probably picked a task that was too vague or too big. Try again with something smaller and more bounded.

The questions that come up next

These are the ones that show up in the second conversation, after someone has actually tried it once.

”What about privacy?”

For non-clinical, non-PHI tasks (like the meal plan), the major consumer tools are fine for most people. Go to the data settings in your tool and turn off training on your data if you want extra peace of mind. Both Claude and ChatGPT have a toggle for this.

Do not put patient information into a consumer AI tool. Ever. This site is not about clinical AI, and that line is hard.

”Should I pay for the paid version?”

Not on day one. Free is fine for what you just did. After a couple of weeks, if you find yourself hitting limits or wanting access to a smarter model, the paid version is worth it. It is roughly the cost of one nice dinner per month.

”Is it going to make me lazy or worse at thinking?”

Honest answer: depends on how you use it. If you use it to skip thinking, sure. If you use it to skip typing while doing more thinking, no.

The way I use AI in my life looks more like having a fast, slightly over-eager intern than having a magic answer machine. I still make the decisions. The intern just gets me to a first draft faster.

”What do I do if it gets something wrong?”

You correct it, the same way you’d correct an actual intern. AI tools are confident even when they are wrong, so the rule is: trust but verify, especially with anything involving numbers, names, or facts where being wrong actually matters.

For a meal plan, the stakes are low. For a tax form, you check the work. For anything clinical, you don’t use a consumer AI tool in the first place.

”Am I behind?”

You’re not behind. You’re busy. There’s a difference.

Most of the colleagues who sound confident about AI in conversation are using maybe two features of one tool. You can catch up to the useful part in a weekend. You can catch up to the first useful part in thirty minutes, which you just did.

The point

The biggest stigma in physician AI right now is not that people don’t know what an agent is. It’s that people are quietly afraid to admit they haven’t actually used these tools for anything useful. They’ve opened them, typed something small, gotten a weak answer, and closed the tab.

You are not behind. You just had a bad first try.

Try again with a better prompt and a smaller task. You will see what the fuss is about. Then come find me when you’re ready for the next workflow.

Strictly non-clinical. Nothing on this site is medical advice. I do not post about patient care.