What AI in clinical workflows actually looks like (vs. the pitch deck)

The pitch deck version: AI reads the patient record, surfaces insights, and makes the doctor faster. Sounds clean. Looks great in a demo.

The operational reality is more constrained — and more interesting.

Running a real clinical platform reveals quickly where AI creates leverage and where it creates risk. The line between the two is clinical safety.

Where AI works well right now: triaging patient enquiries before they reach a human. Routing messages to the right team. Summarising intake forms so clinicians aren't reading paragraphs of unstructured text before a consult. Friction-reduction tasks — high volume, low clinical risk, measurable accuracy.

Where AI is genuinely dangerous: anything that touches clinical decision-making without a human in the loop. Suggesting dosage adjustments. Interpreting pathology. Flagging a patient as low-risk when they aren't. The failure mode isn't a bad user experience — it's patient harm.

The gap between the pitch deck and the operating room is the gap between capability and accountability. AI can do many things. The question for any clinical platform: which of those things are you willing to be responsible for?

Most health tech companies answer that question after they ship. Operators answer it before.

That's the difference between building with AI and building AI into a care pathway.