AI is moving fast. Every week, a new tool. Every week, a new article saying it’ll “revolutionize” something. Meanwhile, you’re still short-staffed on a Tuesday.
If you’re an ED or operator in senior living, you’ve probably seen the word “GPT” thrown around. You might have even tried ChatGPT once. But between state surveys, overnight falls, and two new hires ghosting, staying on top of AI is probably not how you spent your last Sunday.
So here’s a version that’s just for you. What this thing actually is. What’s already useful. What’s still noise. And what’s coming that you should care about.
Let’s start with what a “GPT” even is.
Not the textbook version. Just the kind of thing you’d explain to your DON:
A GPT is a type of AI that’s trained to predict the next word. That’s it. It predicts the next word based on what it has read: websites, books, documents. “Generative” means it writes. “Pre-trained” means it learned before you asked. “Transformer” is the math; you can ignore it.
So when someone says “GPT-4,” that’s just the 4th major version of this guessing machine. OpenAI built it. You can talk to it through ChatGPT.
But OpenAI isn’t the only player anymore.
Now there’s Google’s Gemini. Anthropic’s Claude. Meta’s Llama. Mistral.
Every few weeks, someone drops a new one.
Okay, but what’s the difference between all these AIs?
Here’s how I’d sort them out, if I were sitting in your office with a whiteboard:
-
OpenAI (GPT-4, ChatGPT): Strong with instructions. Great for summaries, report writing, and policy-following if you feed it the policy.
-
Anthropic (Claude): More careful. Reads longer documents. Good when you need it to stay close to state regs or long policy manuals.
-
Google (Gemini): Good at pulling in web-style info and visual content (charts, photos, etc.).
-
Meta & Mistral: You can host these yourself (they’re open-source), which is helpful if you’re in a facility where data privacy is a big concern.
They’re all playing with the same core idea: a model that helps people think faster.
But most of them are still chatbots; you ask, they answer.
The real shift comes when they stop just talking back and actually start doing something.
From Chatbots to Agents
An agent is a model that can act. It doesn’t just say “the fall should be documented.”
It opens the right chart. Fills in the body. Cross-checks your policy. Flags what’s missing. Tags the resident. Sends it to the DON for review. All without copy-paste.
Still needs your sign-off. Still needs human review where it matters.
But the glue work, the 8 browser tabs, the retyping from a sticky note, the “wait, what’s the policy section on elopement again?” that’s where the agent steps in.
And in senior living, that means one thing: less rework, more facetime.
What’s still hype?
You’re going to hear about full autonomy. “AI that runs the building.”
You’ll hear about resident monitoring from video feeds, automatic care planning, chatbots replacing front-desk staff, or “AI nurses.”
None of that is real. Not in any building you’d trust. Here’s a quick filter:
-
If it can’t cite your own policy, it’s not ready.
-
If it doesn’t let you say “show me what you just did,” it’s not safe.
-
If it promises to replace a nurse, it’s not serious.
What’s coming & What’s already working
Near term (what’s actually coming)
Agents that complete multi-step documentation with short audit trails. Alerts that surface only when patterns matter. Family updates built from real care, not memory. Policy answers at point of care that quote the source. Private/on-prem options for tighter data control.
The good news? You don’t need to wait for the future. Some of this is already in your hands.
We at fitmedik are building “agents” that are locally trained on your building knowledge base and execute workflows that actually take load off the floor:
Drafts the note, checks your policy, shows its sources, routes for sign-off, and pushes to ALIS or PointClickCare only when you say so. Short audit trail. Human in the loop. No heroics.
Getting Started
So if AI feels like one more thing you don’t have time to figure out right now, you’re not wrong. But you don’t need to catch up to it. You just need it to show up where it matters.
Start with one thing. One loop that never closes on time. Give it two weeks.
If it buys back time, keep it.
If it doesn’t, shut it off.
But this isn’t five years away anymore.
This is a Tuesday problem you might not need to carry into next Tuesday.
-Kamal Bhartiya, Founder & CEO at Fitmedik
Whenever you are ready, my team is happy to get your building started with AI. Book a 15min brainstorming session here to see what’s working for other operators and identify your first AI-delegated workflow.