Hey Reader,

Finally, people are waking up to the fact that generative AI is mostly an enabling technology. By itself, it is an extraordinary tool for accessing, synthesizing and learning from information. It becomes much more consequential when it is paired with the right context.

Last year we predicted that “agent washing” would flood the market. Put an AI interface on an existing product, call it an agent and promise automation. Now buyers are discovering the difference between AI that can generate a convincing answer and AI that can be trusted to do important work. In healthcare, that difference matters enormously.

The Missing Layer: Context

An architectural idea rapidly gaining traction is the context layer: technology that helps AI understand not just the data, but what the data means for the task at hand.

Take patient flow. Hospital systems already tell us where patients are, which beds are available, who is waiting for admission or discharge, patient acuity, transport status, staffing and much more. Hospitals don’t suffer from a shortage of data. What is often missing is meaning.

A patient may be waiting in the ED while a bed is available upstairs. On a dashboard that looks like an obvious match. But the patient requires telemetry, the bed isn’t telemetry capable, the receiving unit is short a nurse and transport is backed up. The data hasn’t changed. Our understanding of what it means has.

More data is not necessarily more understanding. Context establishes relationships, constraints, operational rules and, importantly, what the system does not know.

What the AI Doesn't Know Matters

I saw this years ago when a stretcher shortage was contributing to ED boarding. The hospital couldn’t find enough stretchers, yet several were sitting unused in a remote hallway that wasn’t covered by the RTLS. Surprisingly few people knew it.

An AI consuming the RTLS data without understanding its coverage could interpret the absence of detected stretchers as evidence that none were there. A system with the proper context would reach a different conclusion: we don’t know whether stretchers are there because we cannot see that area.

The absence of data is sometimes information itself. A trustworthy system needs to understand what it knows, what it is inferring, what it cannot see, how current its information is and how much confidence to place in a conclusion.

Humans Are the Context Layer Today

The interesting thing is that hospitals already have a context layer. It is people.

An experienced bed manager knows that an available bed isn’t really available because the unit doesn’t have enough nurses. A transporter may know that a time-critical blood gas takes priority over a routine discharge transport. A nurse knows that pumps tend to accumulate in a particular hallway. These people aren’t necessarily working with better data. They know what the data means.

Much of what we call operational expertise is the ability to continuously add context to incomplete information. Imagine making enough of that context machine-readable and available in real time. If transport is short-staffed, an AI shouldn’t merely see a growing queue. It should understand what transport is responsible for, which requests are time critical and which can wait. That knowledge may not exist in any transactional field, yet it is essential to making the right decision.

This is why context is more than a data integration problem.

From Application Data to Enterprise Context

It also changes how we should think about technologies such as RTLS. Historically, RTLS has been packaged around use cases such as equipment tracking, staff duress and patient flow. But location itself is valuable context and shouldn’t remain trapped inside a particular application.

If a location system detects that a nurse moved an IV pump into the wrong stocking location, the important fact isn’t why it happened. It is that other people may no longer find the pump where they expect it. A broader context layer could combine location with who moved the pump, what it is, where it belongs and whether its movement is creating a shortage. Location has gone from supporting an equipment-tracking use case to becoming enterprise context.

The same principle applies to information from the EHR, nurse call, bed management, transport, environmental services, staffing and building systems. Those systems can remain authoritative for the information they own while making their context independently accessible across applications and AI agents.

More Than a Data Warehouse

A context layer is therefore not simply another data warehouse. “Transport staffing is at 60%” is data. Knowing transport is responsible for blood gases is organizational knowledge. Knowing blood gases take priority when resources are constrained is an operational rule. Knowing the staffing feed hasn’t updated in 47 minutes tells us how much confidence to put in the information.

Authority matters too. Bed management may say a bed is available while staffing says the unit is closed to admissions. ADT may say a patient is still in the ED while RTLS shows the patient upstairs. Knowing where a fact came from, how fresh it is and which source is authoritative for a particular decision is itself context.

The underlying data isn’t unique. Every hospital has patients, beds, nurses, transporters, equipment, admissions and discharges. What is unique is what those things mean in this hospital, at this moment, for this task.

Don't Take AI Out of Context

We are constantly warned not to trust information that has been “taken out of context.” Why would we expect AI to be any different?

As healthcare moves from generative AI that answers questions to agentic AI that takes action, perhaps the most important question isn’t simply, “How good is the AI?” It is, “How good is the context the AI is operating within?”

Does it know what it knows? Does it know what it doesn’t know? And does it understand what those facts mean here, now, in our hospital?

The intelligence may come from the model. The trust comes from the context.

I look forward to hearing your views & perspectives on this topic. If there is anything I can assist with, please connect.

Until next week,

Paul E Zieske ​AI, Digital Twin and Location Services Consulting ​​​