AI Agents Made Simple for Medical Affairs

What People Actually Mean by AI “Agents” (and Why It Matters for Medical Affairs)

Copilot agents.
Agent mode.
Agentic AI.

Agents everywhere.

The problem is that there’s no strict definition. Everything new in AI is being called an agent, which makes it hard for Medical Affairs teams to know what’s actually different, what’s just marketing, and what’s worth paying attention to.

So this month’s 5-minute AI hack for Medical Affairs (check out PackCheck AI here, it’s hilarious) is a short, practical explainer to help Medical Affairs professionals understand what an AI agent actually is, how it’s different from prompting and custom GPTs, and what this means for day-to-day MSL and Medical Affairs work.

Spoiler alert: you may have already used an AI agent and not realized it.

What Is an AI Agent?

Here’s a simple way to think about AI agents:

An agent is a large language model (LLM) that has arms, legs, and eyes.

Regular large language models are basically just chatbots. Tools like ChatGPT or Microsoft Copilot can think and respond, but they can’t go click around on your behalf. They stay in the chat window.

An AI agent is different.

It’s a large language model that can reason, make a plan, and use tools to complete a task. That means an agent doesn’t just generate text. It can:

• Decide what needs to happen next

• Choose which tool to use

• Click, fetch, search, write, or run something

• Then keep going until the task is done

This is the big shift.

When an AI has arms, legs, and eyes, it stops being just a chatbot and starts acting more like an assistant. And that distinction matters a lot for Medical Affairs.

An Example of an AI Agent in Medical Affairs (Almost)

Recently, during an AI training for a Medical Affairs department of about 50 people, we were getting folks excited about prompting with a simple use case: prioritizing emails and calendars.

The CMO tried the prompt, liked the output, and then called me over. He said, “This is great. Why didn’t it add the calendar reminders for me?”

What he was expecting wasn’t a better prompt.
He was expecting an agent. He wanted AI that would go click into his calendar and set reminders on his behalf.

That moment comes up all the time and it perfectly captures the difference between prompting and agent-based AI.

A lot of Medical Affairs work follows repeatable patterns: pre-meeting prep, follow-ups, insight writing, and literature tracking. When AI has arms, legs, and eyes, it can start helping with those workflows instead of just talking about them.

That’s why AI agents are the building blocks for real AI assistants in Medical Affairs, not just smarter chatbots.

What if AI just booked your travel for you?
And uploaded receipts to your expense report?

Prompt vs Custom GPT vs Agent (Applied to Medical Affairs)

These 3 get mixed up all the time. Here’s the clean way to think about them in Medical Affairs terms.

Prompting in Medical Affairs

When you go into ChatGPT or Copilot and ask it to draft an email, summarize a paper, or help you build a KOL profile, you’re prompting.

You probably iterate a lot, and the AI responds in the chat window. Incredibly helpful, but it stops there.

Custom GPT in Medical Affairs

A custom GPT is something your AI-nerd friend set up and shared with you. 

It’s a large language model with a specific, reusable prompt behind it. For example, Vivek built a pre-meeting planning GPT for MSLs that does one thing really well. You don’t have to explain the task every time, the instructions are already baked in.

It thinks consistently, but it still lives in the chat window.

AI Agents in Medical Affairs

Medical Affairs starts using AI agents when AI can access tools. A common example is when your company’s version of Copilot can access your internal SharePoint site. That’s an agent. Custom knowledge bases like SharePoint are tools.

The AI isn’t just answering questions. It’s deciding what to look at, where to pull information from, and how to use it to complete a task.

You may have already used an AI agent at work and not even known it.

→ Prompts help you think
→ Custom GPTs help you think consistently
→ Agents help you get work done

Example Prompts to Play Around With:

Copilot Agents vs ChatGPT Agent Mode

When I ran my AI Agent-in-a-Box workshop, we built a simple but powerful use case: a subject-line generator AI agent in under 15 minutes.

What became clear very quickly was that people weren’t confused about AI.
They were confused about what was what in the tools they already use.

Here’s the simplest breakdown.

ChatGPT: Custom GPTs vs Agent Mode

At the time of writing, ChatGPT has 2 relevant capabilities: Custom GPTs and Agent Mode.

My MSL Insights Coach is a Custom GPT. It has a tight set of instructions and does the same thing every time. An MSL pastes in a draft insight and gets coaching suggestions back. No clicking. No wandering around. It does one job really well.

ChatGPT Agent Mode is different. This is where ChatGPT opens a virtual machine and actually goes and does things. It can click, label, organize, and take actions, like going into your email and prioritizing messages by adding labels.

That requires arms and legs.
That’s a true agent.

Same tool. Very different behavior.

Copilot Agents: Two Things in One

Now let’s talk about Copilot Agents, because this one confused me too.

Copilot Agents can be one of two things, depending on how they’re set up:

• A custom GPT-like experience (focused instructions, consistent responses)

• Or an agent that can use tools like SharePoint, Outlook, or Teams

At first, I assumed Copilot Agents were just custom GPTs with a confusing name. That’s not true. If a Copilot Agent can access your internal systems, those systems are tools and now you’re in agent territory. If it can’t, it’s closer to a custom GPT.

Same name. Different capability.

Conclusion: AI Agents Explained for Medical Affairs

Understanding the difference between prompts, custom GPTs, and agents matters because it points to where AI in Medical Affairs is actually headed.

Chatbots will keep helping Medical Affairs think. Agents are what will start helping teams operate.

As more tools get connected, calendars, documents, internal systems, expectations will shift from “AI gave me a suggestion” to “AI handled that for me.”

That’s not science fiction. It’s the natural next step once AI has arms, legs, and eyes.

Understanding what an AI agent really is helps Medical Affairs teams prepare for that shift — without getting distracted by buzzwords along the way.

2026 is going to be a very interesting year for AI in Medical Affairs.

If you’re reading this and thinking, “Okay… how would this actually work for my team?”  I run practical AI training and AI agent-building workshops designed specifically for Medical Affairs and MSL workflows.

Happy to talk your ear off about this. Just reach out 🙂

Read Next

Let’s Connect On Social Media