If you read press releases, you’d think every Australian hospital is running AI. If you talk to clinicians, you get a different story. The reality is somewhere in between, and it’s worth mapping out what’s genuinely in production versus what’s still experimental.

Actually Deployed and Working

Let’s start with what’s real.

Medical imaging AI has crossed from pilot to production in several major Australian hospital networks. Harrison.ai’s annalise.ai platform is being used for chest X-ray analysis across multiple radiology practices. It doesn’t replace radiologists. It flags findings that need urgent attention and catches things that might be missed on a busy day. The clinical evidence supporting it is solid, and TGA approval has been granted.

Pathology automation is further along than most people realise. Several large pathology networks are using AI for preliminary screening of blood tests and tissue samples. The AI identifies samples that need closer human examination, essentially prioritising the pathologist’s workload. It’s not glamorous, but it’s saving hours of human time daily.

Emergency department triage tools are live in at least three major hospital networks. These systems analyse presenting symptoms, vital signs, and patient history to suggest triage categories. They don’t override clinical judgment. They provide a second opinion that’s particularly valuable during high-volume periods when triage nurses are stretched thin.

Administrative AI is the unglamorous workhorse. Natural language processing for clinical coding, automated referral processing, and AI-assisted appointment scheduling are deployed across dozens of Australian healthcare organisations. Nobody writes press releases about these, but they’re saving significant administrative time.

Stuck in Pilot Purgatory

Now the frustrating list.

AI-powered drug discovery at Australian research institutions has produced some fascinating early-stage results but remains firmly in the research phase. The gap between identifying a promising compound and getting a drug to market is still measured in years and billions of dollars. AI accelerates parts of the process but doesn’t shortcut the fundamental timeline.

Mental health chatbots have been piloted extensively and results are mixed. They work well for guided self-help programs where the conversation follows structured pathways. They struggle with the nuanced, unpredictable nature of real mental health crises. Several pilots have concluded that these tools are supplements to human care, not replacements. Which is probably where they should sit.

Predictive models for patient deterioration show promise in research settings but face implementation challenges in live hospital environments. The biggest issue isn’t accuracy but integration. Getting real-time predictions into clinical workflows where busy nurses and doctors actually see and act on them is harder than building the model.

Remote monitoring AI for chronic disease management has been piloted in multiple state health systems. The technology works. The challenge is the care model around it. Who monitors the AI’s outputs? Who responds when it flags a concern? How does it connect to the patient’s existing care team? These are organisational questions, not technical ones, and they’re taking time to resolve.

What’s Holding Things Back

Three factors come up in every conversation I have with healthcare AI implementers.

Regulatory caution. The TGA’s approach to AI as a medical device is necessary but slow. Companies report twelve to eighteen month timelines for regulatory approval even with strong clinical evidence. This isn’t unique to Australia, but it means deployment trails development by over a year.

Data infrastructure. Many Australian hospitals run on legacy IT systems that make data extraction and integration genuinely difficult. You can’t deploy sophisticated AI when the underlying data exists in disconnected silos with inconsistent formats. The boring infrastructure work needs to happen before the exciting AI deployment.

Clinical workforce buy-in. Some clinicians embrace AI tools enthusiastically. Others are sceptical, and their scepticism isn’t unreasonable. They’ve seen too many technology projects that promised transformation and delivered disruption. Building trust requires evidence, training, and respect for clinical expertise. That takes time.

Where I’m Optimistic

Despite the challenges, I’m genuinely optimistic about AI in Australian healthcare. The deployments that are working are producing real clinical value. The institutional knowledge about what works and what doesn’t is growing rapidly. And the next generation of clinicians is entering the workforce expecting AI to be part of their toolkit.

The key insight from watching this space for years is that successful healthcare AI is boring AI. It does specific, well-defined tasks reliably. It integrates into existing workflows. It makes clinicians more effective rather than trying to replace them.

The flashy demonstrations of AI diagnosing rare diseases from a single image make great headlines. The reality of AI improving healthcare is much more mundane and much more impactful.