In a previous post, I talked about something I’ve been seeing more clearly over time, AI in healthcare doesn’t fail because the models aren’t good enough. It fails because it doesn’t fit into the systems it’s being placed into.
That’s easy to say, but it naturally leads to a bigger question: what does the right system actually look like?
Because if AI isn’t the problem, then the structure around it probably is.
Most AI tools in healthcare today are built as add-ons. They’re designed to sit alongside existing processes rather than inside them. You see it in the form of chatbots, decision-support prompts, or standalone tools that aim to make parts of the process quicker or more efficient.
On paper, that makes sense.
But in practice, healthcare doesn’t work like that.
It isn’t a series of isolated interactions. It’s a continuous process made up of multiple steps gathering information, assessing risk, interpreting results, documenting decisions, and managing what happens next. Each of those steps is connected, and each one carries clinical responsibility.
If a tool doesn’t fit into that flow, it doesn’t really matter how intelligent it is. It becomes something extra to manage rather than something that genuinely supports care.
This is where my thinking started to shift.
Instead of asking what AI can do, the more useful question became: where should it actually sit within the way care is delivered?
Because once you start looking at it that way, the answer becomes clearer.
AI doesn’t belong on top of the process. It belongs inside it.
If you take a real example something like a hormone health consultation it isn’t just a single decision point. It’s a structured sequence. You’re taking a history, understanding symptoms, reviewing blood results, considering risk factors, documenting everything properly, and deciding on a safe and appropriate next step.
That structure is what holds the whole interaction together.
And in many cases, that structure isn’t consistent.
This is where AI can start to add real value but not in the way it’s usually presented.
It’s not about replacing clinical judgement or making decisions independently. It’s about supporting each part of that process so it becomes more consistent, more efficient, and easier to manage.
It can help structure how information is captured, reduce the time spent documenting consultations, highlight patterns in symptoms or lab results, and bring attention to things that might otherwise be overlooked. None of that replaces the clinician, but it makes the process itself stronger.
Underneath all of this, there’s another layer that matters just as much, if not more, safety!
In healthcare, it’s not enough to arrive at a decision you need to understand how that decision was reached. There has to be visibility around what was asked, what was recorded, and what factors were considered. Without that, there’s no real accountability.
This is where things like auditability, escalation points, and structured documentation become essential. They’re not optional extras they’re part of what makes a system clinically viable.
And without that foundation, even the most advanced AI won’t be trusted.
A question I get asked quite often is: where does the AI actually sit in all of this?
The honest answer is that it doesn’t sit in one obvious place.
It’s distributed across the workflow.
It supports how information is gathered, how it’s interpreted, how it’s documented, and how potential risks are surfaced. But it remains in the background, working within the process rather than sitting outside it.
The clinician is still the one making the decision. That part doesn’t change.
The other thing that has become clear to me is that this kind of system can’t really be built in isolation. It only starts to make sense when it’s shaped alongside real clinical workflows, where the constraints are real and the details actually matter.
Because in healthcare, the difference between something that works and something that doesn’t often comes down to how well it holds up in practice.
For me, that’s where the shift is.
AI in healthcare isn’t going to succeed as a collection of standalone tools. There are too many gaps between them, and too much reliance on how they’re used in the real world.
Where it starts to work is when it becomes part of the system itself — something that sits within the way care is delivered, supporting each step without disrupting the flow.
That’s the direction I’ve been focusing on.
Less about what AI can do in isolation, and more about how it fits into the structure of real clinical practice.
Because that’s ultimately where it needs to work.
Asif Mukhtar