I started learning to use artificial intelligence (AI) tools because I had problems I wanted to solve. As I learned what AI could do, it changed the way I work. 

I estimate that AI now saves me 15 to 20 hours of work each week. Here are a few ways I use it:  

Close charts faster. I’ve shown AI tools how I think, document, and communicate, so I spend much less time editing notes and taking charting home. 

Turn hours of work into minutes. Presentations, proposals, grants, patient education materials, social media posts, and other projects that used to take hours can now take a fraction of the time. 

Build tools without knowing how to code. I can create a website in a few hours instead of days or build a tool for a problem I see in my own practice. 

Keep work moving while I sleep. AI assistants and automated workflows can work in the background, so I can wake up to progress on a project. 

Automate meeting follow-up. An AI assistant can summarize my meetings and identify decisions, follow-up items, and tasks, so I know what needs to happen next. 

Get ideas out of my head. I can give AI a messy brain dump and have it organize my thoughts, projects, priorities, and next steps. 

Start with the problem, not the prompt

We have spent a lot of time teaching physicians how to write prompts. Prompts are useful, but I think the first step is recognizing which problems AI tools can help solve. 

Look at the places where time and energy often disappear: 

• The charts you are still finishing at night. 

• The presentation that takes hours to build. 

• The meeting that creates another 45 minutes of follow-up. 

• The patient records you have to dig through before a visit. 

• The task your staff repeats 20 times a week. 

• The spreadsheet, dashboard, or tool you wish existed. 

Those were the first places I looked. Sometimes better instructions were enough. Other times, I created an AI assistant, automated a process, or built a tool around what I needed. 

My advice is to start paying attention to the things that repeatedly take more time than they should. Write them down. Pick one problem and see if AI can help you solve it. You don’t have to change your entire workflow. Start with something that would make your day a little easier.    

One of my favorite examples came directly from my own clinic. I prescribe GLP-1s, and I was having a hard time quickly seeing a patient’s weight change alongside changes in medication dose. All the information was in the chart, but I wanted to see the weight trajectory and know when the dose changed. So, I built a tracker. 

I am not a software developer, and I didn’t learn how to code an application. I told AI what I wanted, tested what it built, changed what I didn’t like, and kept going until I could see the information the way I wanted it. 

A few years ago, that idea probably would have remained just an idea. I would have needed to find a developer, explain the clinical problem, wait for someone to build the program, and then go back and forth to make it work the way I wanted. Now, I can have an idea and start building it the same day. 

That matters for physicians because we see problems all day: information buried in the chart, inefficient workflows, repetitive tasks, data we wish we could visualize differently, and tools we wish existed. AI gives us new ways to approach those problems. 

Learning to get better output 

Most physicians haven’t been taught how to get useful output from AI tools. They open ChatGPT, Claude, or Copilot, type a sentence or two, and ask the tool to write something. But they often get a response that’s too generic, too wordy, or nothing like how they would say it. Sometimes the information is wrong. 

It is easy to conclude from that experience that AI is not very useful. I understand why. I didn’t like much of the output I got at first either. I just hadn’t learned how to get what I wanted. 

For me, better output starts by showing the tool what I want, not just telling it. 

I give AI tools examples of how I think, write, and teach. I tell them what I like and don’t like and point them to the information I want them to use. When the output isn’t right, I correct it and refine my instructions. 

For example, when I prepare a presentation for my residents, I don’t want AI to give me a generic pediatric endocrinology lecture. I want it to teach the way I teach. So, I give the tool my previous presentations, articles, guidelines, and clinical pearls. I show it how I sound, including words I don’t use and AI jargon I want to avoid. The first draft is often much closer to my voice and needs less editing. 

Patient care requires a different standard 

When I use AI in or around patient care, I take extra care with the information I give it and the answers it gives me. AI can be wrong even when an answer sounds confident. I verify clinical information and keep myself in the decision-making process. I de-identify data when I use Claude and use only HIPAA-compliant AI tools with patient identifiers. 

That care still leaves room for AI to help with the work surrounding a visit. A new patient may arrive with hundreds of pages of notes, laboratory results, imaging reports, medication changes, and previous treatment attempts. I need to understand that history, but I don’t necessarily need to manually organize every page. 

With the appropriate systems in place, AI can help turn those records into a timeline and summary I can review before the visit. An AI scribe can help during the encounter. I check the information against the record and decide what matters for the patient’s care. 

Physicians are creative problem solvers. We have spent years creating workarounds for technology that wasn’t built around the way we work. AI is shortening the distance between having an idea and building something useful. But when AI touches patient care, physicians still have to evaluate the information and decide how it should be used. 

Ladan Davallow, MD, is a pediatric endocrinologist at EndoMD Health in Richmond, Virginia. She is also the founder and CEO of Aimforth, a company that teaches physicians how to build and implement AI automation tools.