Agentic AI in South Africa: Bringing the Patient Story Into Clinical Intelligence
How Respocare Connect AI is approaching clinical AI through specialist onboarding, longitudinal patient context and a service built around the medical professional.
A specialist can receive access to artificial intelligence in minutes. Preparing that intelligence to work with years of patient history takes a different kind of effort.
On 29 September, we began onboarding an endocrinologist into Respocare Connect AI. Approximately 4,000 clinical documents are being prepared for upload as part of that process.
The goal is straightforward: when she begins using the platform, her patients and their available histories should already be organised within the workspace, ready to be explored in context.
That work is still underway. It offers a practical example of the question behind our approach to agentic AI in South Africa: how do we make clinical intelligence useful around the way a medical professional actually works?
What is agentic AI in South Africa healthcare mean?
The term “AI” now covers tools with very different capabilities. A system that transcribes a consultation performs a different task from one that retrieves information across a patient record and uses it to prepare a clinical summary.
Agentic AI generally describes systems that can work towards a defined goal through multiple steps, using information and tools within specified boundaries.
In a clinical workspace, that could mean retrieving relevant documents, arranging findings chronologically, comparing recorded information and preparing a draft for review. The quality of that work depends on the information available, the reliability of retrieval and the controls around the process.
Respocare Connect AI’s agentic clinical assistant, North Star, is designed around five behaviours: retrieving, reasoning, acting within its remit, refusing unsupported requests and escalating when the task requires the clinician’s judgement. Its outputs remain drafts for review.
Agentic capability does not give a system authority to practise medicine. The clinician defines the task, evaluates the evidence and decides what enters care.
Why the South African conversation needs to be practical
For a South African practice considering clinical AI, the useful starting point is its own workflow.
Where is the patient information held? Which documents matter before a review? How are referrals prepared? Who checks a draft? What happens when a source is missing, an upload fails or the system gives an incomplete answer?
Different practices will have different answers. A specialist working with years of investigation results may need a different onboarding process from a clinician focused primarily on documenting new consultations.
Implementation also needs to account for connectivity, existing record systems, staff responsibilities and support. Sensitive patient information requires deliberate handling, with appropriate authority, access controls and review of applicable privacy obligations.
These are decisions that shape whether the technology becomes useful in everyday care. South African clinicians should be involved in making them.
One endocrinologist. Approximately 4,000 documents.
Endocrinology makes the value of a longitudinal record easy to understand.
A result has a date. A medication change has a reason. A treatment response develops over time. A specialist letter may explain why an earlier decision was made.
Those details can sit across blood results, imaging reports, consultation notes,
diagnoses and correspondence. The clinician needs to understand the relationships between them, rather than simply read each document in isolation.
Our current onboarding work is preparing approximately 4,000 clinical documents so the specialist can begin with her existing patient information available in the system.
The number refers to documents, not patients, and preparation is not the same as a completed migration. The work must establish which records belong to which patient and preserve the source information needed for review.
The intended experience is a clinician opening a patient’s workspace and being able to explore the available history from the outset. She might ask for a chronology of documented results, a summary of recorded medication changes or a draft that brings relevant correspondence together.
These are examples of record-based tasks, not claims of a clinical outcome. Any resulting draft needs to be checked against its sources and interpreted by the treating professional.
What we mean by intelligence as a service
At Respocare, the onboarding experience is part of the service.
It involves understanding the clinician’s workflow, helping bring existing information into the environment, preparing usable context and supporting the professional as they begin working with the system.
The 4,000-document story makes that tangible. Access to a model is one component. The surrounding work determines whether the clinician can use it meaningfully with their patients.
Respocare Connect AI brings the Medical AI Scribe, the patient record, the agentic clinical assistant and clinical document workflows into a connected environment. The aim is to make information captured or uploaded in one part of the workflow available for appropriate use elsewhere.
A consultation note can contribute to the patient history. That history can inform a clinician-directed question. Relevant information can then support a draft referral, summary or report, subject to review.
This is what we mean by intelligence as a service: capability connected to patient context, delivered with implementation and support.
The clinician remains in control
A larger record does not guarantee a better answer. Information may be incomplete, duplicated, outdated or contradictory. An AI system can make mistakes even when it has access to source documents.
Useful clinical intelligence must make those limits visible. The clinician needs to be able to inspect the evidence, challenge the draft and recognise when the available information cannot support a conclusion.
North Star is designed to help with documentation and record-based reasoning. It does not diagnose, prescribe or make clinical decisions. Clinical validation of its outputs is ongoing; the onboarding story is an implementation update, not evidence of improved patient outcomes.
Build around the way medicine is practised
Our work with this endocrinologist is one example of the direction Respocare Connect AI is taking in South Africa.
We want onboarding to begin with the people, records and responsibilities that already exist in a practice. We want clinicians to help shape the experience. And we want the value to become visible in the work they can review and use.
The patient story deserves continuity. The medical professional deserves practical support. Clinical intelligence should serve both.
Would you like to explore what this could look like in your practice?
or read the Clinical AI Implementation Guide.
Respocare Connect AI — Agentic Intelligence for the Art of Medicine.
Comments