I spend most of my working weeks in meeting rooms with hospitals and health systems — CIOs, clinicians, sometimes a whole board — talking about AI, data and infrastructure. You would think those conversations would stay technical. They don’t. Almost every one of them, sooner or later, comes back to cancer. Someone mentions a diagnostic backlog, or a scanner running slower than it should, or a patient waiting weeks for a result. Quite often it turns personal: a clinician talks about a case that stayed with them, or someone quietly mentions their own family. After enough of these conversations you stop seeing the technology and start seeing the people behind it. That, honestly, is why I do this work.

So let me talk about what actually gives me hope, in plain terms.

The example I keep coming back to is the brain. A brain tumour is about the most frightening diagnosis there is, because in the brain there is almost no room for error and every millimetre counts. What we can now do is build a “digital twin” of a patient’s brain — a working model put together from their own MRI scans and the molecular makeup of their tumour. On that twin, not on the person, doctors can watch how the tumour is likely to grow, plan the surgery to the millimetre, and test which treatment it is most likely to respond to before anyone operates. For something as aggressive as glioblastoma, where the margins are everything, being able to rehearse the disease before you treat it can change the whole course of care.

The reason that is even possible comes down to data, and this is where genomics and multiomics come in. A single tumour’s genome, read alongside its RNA and its proteins, runs to hundreds of gigabytes, and not long ago analysing it took days. With the right GPU-accelerated infrastructure it now takes under an hour, so a patient’s molecular fingerprint is ready while it still matters. AI reads that fingerprint against thousands of others, points to the mutation driving the cancer, matches the patient to a targeted drug or a clinical trial, and can even warn when a tumour is likely to shrug off a standard treatment before it is ever prescribed. When a customer describes cutting weeks out of that process, I don’t hear an IT story. I hear someone getting an answer in time.

A lot of it is quieter than that, too. AI reading biopsy slides and flagging the malignant cells a tired pathologist might pass over at the end of a long day. Picking up the small lung nodule or the subtle change on a mammogram and moving that patient up the queue. Drafting the notes and the letters so an oncologist can look up from the screen. None of it replaces the clinician; it just hands them their time back. And this is the part I push hardest on in every meeting: it has to be done safely. The data stays inside the hospital that owns it, and hospitals can learn from one another without ever handing over a single patient record. That is how a small regional unit ends up with the same intelligence as a world-leading centre.

I have never thought the technology was the point. AI is never going to sit with a frightened patient or find the right words for a family. What it can do is take the busywork off a clinician’s desk and the waiting out of a diagnosis, so the people we trust with our lives have more of themselves left for the part that actually matters — the human part. That is the bit worth working for. Not machines that replace care. Machines that get out of its way!