DiffyD — what comes next
The ward round of 2030
The future of clinical decision support isn't a smarter search box. It's a coordinated team of specialist agents that work the case alongside you — requesting information, synthesising results, and revising their reasoning as the picture changes.
This is what we're building. Below is a simulation of how it works.
Case vignette
28F. Three weeks of fatigue, pleuritic chest pain, migratory joint pain, and a facial rash. Seen twice in primary care — sent home with anxiety, then GORD. Now presenting to ED. SpO₂ 96% on air. HR 102. Temp 37.9°C.
Working Impression
Awaiting initial data…
Agent Activity
Initialising agents…
What you just watched
Multiple agents. One patient.
Three specialist agents initialised on a single case. One asked the questions a thorough history-taker would ask. One requested the examination findings most likely to change the picture. One ordered the targeted workup — not the full clerking panel, the discriminating tests.
Each agent worked its piece of the problem simultaneously. When a result came back, the working impression updated. The diagnosis didn't arrive pre-formed. It emerged from evidence, the way it should.
In practice
A registrar submits a case. The agents begin.
In a live system, a clinician submits a de-identified case at any point in the workup — at presentation, mid-clerking, or when something doesn't fit. The agents know what they don't know and ask. The History agent asks the questions specific to this presentation. The Examination agent surfaces the findings most likely to discriminate. The Investigations agent orders what would actually change management.
As real results arrive — real bloods, real imaging, real consult responses — the impression updates. The clinician sees the reasoning shift in real time. They can push back, add information the agents missed, and accept or reject the working diagnosis.
The agents provide systematic rigour. The clinician provides judgment. That's the right division of labour.
What this changes
Most missed diagnoses are not knowledge failures.
Lupus presenting as anxiety. PE in a young woman on the pill dismissed twice. Aortic dissection sent home with musculoskeletal pain. These are not obscure diagnoses. They are well-known presentations that get missed under the conditions in which medicine is actually practised: high census, cognitive overload, time pressure, and a system that rewards throughput over systematic thinking.
An agent-based system does not add to the cognitive load. It runs in parallel. It does not get fatigued at 3am or anchor on the first plausible diagnosis. It follows every thread simultaneously, without the fatigue penalty.
For health systems, the downstream effect is measurable. Diagnostic delay is expensive — in human terms and financial ones. Patients who reach the right diagnosis faster spend fewer days in hospital, need fewer return presentations, and have fewer downstream complications. A systematic reduction in time-to-diagnosis, at scale, is a different category of intervention to anything the system has tried before.