Ask a peer how their team is using AI, and listen to the nouns.
If the answer is ChatGPT, Copilot, Claude, meeting notes and a first pass at the strategy brief, that’s real use. It’s also the basic stuff, and most organizations are there now. If the answer is agent swarms, MCP servers, model routing optimization, and integrations with the systems the work already runs on, you’re talking to a different kind of organization. And if you aren’t in that space yourself, it gets obvious fast. They’re working on the next level, and you’re behind.
TL;DR
- You can tell which side of the AI gap an organization is on from the words its people use. Chat tools and meeting summaries are the starting line. Agents and integrations mean someone has moved past it.
- The first reaction to finding out you’re behind is fear of missing out. The useful version of that fear is research: learn exactly where the other side is before you try to catch up.
- Catching up in healthcare means moving faster while keeping patient and personal data safe. You need a shared set of rules for that, or every request turns into a debate.
- A traffic light works as those rules. Green moves after a quick check, yellow moves with care, and red needs real review by whoever owns AI governance where you work.
- The quickest gains aren’t new tools. They come from using what you already pay for, and from getting more of your people to use it.
- Training on its own fades. A champion who sits close to the work keeps people from trying AI once and forgetting it.
You can hear the gap
The distance between organizations that use AI and the ones that mostly talk about it keeps growing. The second group rarely gets told. Nobody sends a memo. You find out in conversation.
It goes something like this. You’re comparing notes with someone at a peer organization. You talk about prompts that save your team an hour on a brief. They talk about agent swarms doing whole collections of work, wired into the systems their teams already use, and the MCP servers that connect them.
MCP stands for the Model Context Protocol, an open standard Anthropic published in late 2024 that lets an AI tool connect to other software and data. You don’t need to know how it works. You do need to notice when your peers are building with it and you aren’t.
The words give it away because they’re hard to fake. Anyone can put “AI-powered” on a slide. Talking comfortably about integrations you haven’t built is much harder.
If your peers are talking about agents and integrations and you’re still talking about meeting notes, you already know which side you’re on.
The first instinct is fear. The second should be research.
When you realize a competitor is ahead, fear of missing out shows up first. Right behind it comes the urge to catch up.
I’d put a step in between. Before you try to catch up, learn as much as you can about where they actually are. What are they doing? How far along are they? How widespread is the use, and how deep does it go? A handful of people building agents on the side is a different story from a whole department working that way.
Only then ask what you can do to help your own organization close the gap.
In healthcare that question comes with weight attached. You still have to keep the organization safe. You have to keep protected health information (PHI, the patient data HIPAA covers) and personally identifiable information (PII) where they belong. Speed counts, and so does safety. Any plan to catch up has to hold both at once.
The old mistake, at a much bigger scale
Careful organizations end up on the wrong side of this without ever deciding to. Part of the reason is a mistake IT has been making for decades. Someone sees a shiny object. The organization buys it, implements it, and figures out the use cases, the change management and the adoption afterward.
AI makes that mistake far more expensive. It’s more complicated than most technology that came before it, it’s less understood, and plenty of it is misunderstood. It also reaches further. When offices moved from typewriters to PCs, the decision touched a fairly small surface. Moving work from older automation to AI touches nearly everything people do, and the cost of getting it wrong grows with it.
So the organization that buys the impressive platform and hopes adoption follows can end up with less real use than the one that bought less and planned more. I’ve written about what happens when the work behind the tool was never ready. It’s rarely the tool that fails.
Give everyone a traffic light
The fix I’d push for is a framework everyone can see, with clear guardrails. I think of it as a traffic light.
A green light means go. After a quick look at the capability, the request or the use case, it’s clearly fine, and the team moves forward.
A yellow light means proceed with caution. Maybe the tool sits next to sensitive systems, or it touches systems that hold PII or PHI even though it won’t use that data. It can go ahead, with more care and more eyes on it.
A red light is what it sounds like. The tool connects to sensitive systems or handles sensitive data. In healthcare that means contracts like a business associate agreement, which HIPAA requires before an outside company handles patient information on your behalf, along with whatever else your organization asks for.
Red isn’t a no. It’s a maybe, and the maybe depends on your organization: who has to approve it, what they need to know, and which governance channels are in place. That’s different everywhere. Some organizations send it to a committee. Others send it to a privacy or security office. The point of the light is that nobody has to guess which lane they’re in.
That clarity pays off twice. Green work stops waiting in line behind red work. And red work gets the attention it needs, because it isn’t sharing a queue with requests that never needed review.
The traffic light check
Which light is your next AI idea?
Pick one AI tool or use you're considering. Answer four questions about it, and the light changes as you go.
Green light
From your answers, this is low risk. Do the quick check your organization asks for, then get moving.
This is a starting point, not an approval. Your organization's own AI governance makes the call.
Runs in your browser. Nothing is saved or sent.Turn on JavaScript to see the light change.
Start with what you already have
If you’ve decided you’re behind, the first move isn’t a new tool.
Look at what your people already have at their disposal. Then look for use cases those tools fit that nobody is using them for. Sometimes the gap turns out to be tools you already pay for, sitting mostly idle.
The second move is less exciting and just as easy: scale. Once you know which capabilities you have, ask how many of your people, or what share of the organization, actually use them. A capability only a few people use well is half a rollout.
Neither of these makes a good conference talk. Both of them move you toward the other side of the gap faster than a new purchase will.
The quickest way to catch up on AI is getting more of your people to use what you already bought.
Training fades. Champions stick.
Getting more people to use AI is where I’d expect a rollout to stall, and more training on its own won’t fix it.
AI changes so much about a normal workflow. People who pay attention in the training will see it and understand it. Then they go back to their desks, the week fills up, they don’t start using it right away, and what they learned fades.
I’m a big fan of champions for exactly that reason. A champion is someone close to the work, physically or in their role, who already uses the tools and is there when a question comes up. With help right at their hip, a person is much less likely to try AI once, stop, and forget how. The leadership side of that is in Closing the AI Skills Gap in Your Team.
Race your past self
When you find out a competitor is ahead, it’s tempting to measure everything against them. I’d rather you measure against yourself.
Simon Sinek makes a version of this argument in The Infinite Game. He says to treat a strong competitor as a worthy rival, someone who shows you where you can improve, instead of a score you have to beat. Chasing the competition wears you out. Beating last year’s version of your own organization is a race you can actually run.
So listen to your own team the way you’d listen to a peer. A year ago, which tools came up when people talked about AI? Which ones come up now? If the words haven’t changed, not much else has either.
The gap starts in the vocabulary. So does closing it.
If you’ve ever realized mid-conversation that a peer was a step ahead on AI, I’d like to know what gave it away. Head over to LinkedIn and let’s talk about it.