What HPN West Coast 2026 taught me about AI, workflows and why the hype is finally fading

I had the privilege of co-presenting a session titled “Beyond the AI Hype: How SickKids and Philips are Making AI Work in Clinical Workflows” alongside Dr. Karim Jessa, CMIO at SickKids in Toronto, and our moderator Matt Davis. Karim brings a perspective I deeply respect – he’s a former emergency medicine physician who now sits at the intersection of clinical practice and informatics. That combination matters enormously when you’re trying to build AI tools that clinicians will actually trust. 

Before I get into what we discussed, a bit of context about where I’m coming from: I’m a former NICU nurse. I’ve stood at the bedside when alarms are firing, when a patient is changing faster than the documentation can capture and when the last thing you need is another screen demanding your attention. That experience shapes how I think about AI in patient monitoring today. 

The session opened with a question that sounds simple but is really quite hard: how do you evaluate whether an AI idea is worth pursuing? 

Karim’s answer was clear – start with the problem, not the tool. Watch out for what he calls “solutions looking for problems.” If you can’t articulate the clinical pain point before you reach for the algorithm, you’re already off course. 

From the bedside perspective, I’d put it even more bluntly: AI cannot become one more screen, one more alert or one more task for a care team that’s already stretched thin.  

A risk score alone is not enough. Clinicians need to know what is happening, why it matters, and – critically – what to consider next. If we hand a nurse a number without a next step, we’ve given them nothing but anxiety dressed up as insight.

At Philips, our approach to this starts with bringing AI insight into workflows clinicians are already using – bedside, central station, mobile – rather than forcing them into disconnected tools they’ll abandon by week two. That’s the workflow-first principle in practice. It sounds obvious. It’s surprisingly rare. 

Once you’ve built something that fits the workflow, you face the next challenge: earning trust. And trust, it turns out, isn’t something you can declare. It has to be demonstrated, earned incrementally and sustained through transparency. 

Karim beautifully described SickKids’ approach. Trust is partly cultural; when trusted clinical colleagues and informatics leaders have helped vet a tool, others are more willing to engage with it. But it’s also structural. There are education modules, evaluation frameworks and a staged approach to deployment that allows tools to be tested “silent in the dark” before they influence patient care. 

From my vantage point in patient monitoring, trust comes down to explainability. Not the kind that requires a data science degree to understand, but the practical, clinical kind. Why is the system flagging this patient? What’s the physiologic rationale? What should I consider doing next? 

Philips’ AI philosophy centers on explainability as a non-negotiable. The rationale behind a warning, prediction or finding needs to be visible and interpretable. Because if it isn’t, the clinician doesn’t gain a tool; they gain a liability they can’t interrogate. 

The clinician remains in charge. Full stop. AI should support clinical judgment, not attempt to replace it. 

Here’s the part of the AI conversation that tends to clear the room: data quality, standards and infrastructure. 

I know. It’s not glamorous. Nobody’s putting “standardize your data capture” on a conference banner. But this is where AI either succeeds or quietly fails, and it fails quietly a lot more than people admit. 

AI can’t succeed on top of fragmented, inconsistent or low-quality data. In patient monitoring, the opportunity is continuous, high-fidelity data: waveforms, vitals, alarms, physiologic trends over time. That data is extraordinarily rich. It’s also only as useful as the consistency with which it’s captured and structured. 

Philips’ approach is built around a philosophy I find genuinely compelling: standardize, centralize, virtualize and optimize. And critically, we think about this at the enterprise level, not device by device. The vision is a governed ecosystem where Philips-built, third-party, and customer-developed algorithms can be integrated into clinical workflows through a common framework. This is the only way AI scales safely across units, sites and care teams. 

  The most animated part of our session (and honestly, of the broader conference) centered on predictive AI. The enthusiasm is warranted. The possibilities are real. But so are the pitfalls.

  A prediction is only useful if it helps a clinician intervene. Karim gave a concrete example from SickKids: a model that predicts which children may vomit after starting chemotherapy. The reason it works isn’t just the model itself, it’s that the alert goes to the pharmacist who can actually change the antiemetics in time. The prediction is embedded in a workflow. It fires early enough to change the trajectory of care. And the team monitors whether the intervention is effective after deployment. 

In monitoring, the future isn’t about more waveforms and more numerics; it’s about clearer prioritization and better understanding of risk across multiple data points. But the goal has to be fewer, more actionable signals instead of more alarms. We already have an alarm fatigue problem. Adding AI-generated noise to clinical noise doesn’t help anyone. 

Philips is already seeing momentum here – in cardiac diagnostics, AI-supported ECG review and predictive insight from standard ECG data. The monitoring vision is to bring those insights to the point of care so clinicians don’t have to chase information across disconnected systems. 

My session was one piece of a larger picture. Across HPN West 2026, several themes repeatedly surfaced, and they reinforced things I’ve believed for a long time. 

Walk the workflow with your frontline team before you buy anything. Ask where the burden truly lives. Put clinicians and informatics at the table early, define how you’ll measure success and invest in the unglamorous foundation. 

Here’s my parting thought: the organizations that lead the next decade won’t be the ones with the most AI. They’ll be the ones who wove technology so seamlessly into care that clinicians finally got more time for what only humans can do. 

So let me ask you – where in your workflow would AI genuinely help, and where would it just be one more screen?