The case for agentic AI in health and social care and how to make it happen

This is a guest blog by Gian Celino, Chief Product Officer at System C

Agentic AI is often described as the next frontier: systems that can plan, act and coordinate work rather than wait to be told what to do. With demand on health and social care services growing, the appeal is obvious. The technology can lift the administrative burden and give clinicians and care professionals back the time to focus on what matters most: delivering care. Yet adoption across the NHS and local authorities remains frustratingly slow.

The reason is not that the capability is missing. It exists today. The reason is that too many organisations rush to act before building the foundations beneath them. In a sector where clinical safety is paramount and regulation demanding, the sequence is everything. The path runs from capturing data, to understanding it, to acting on it, and finally to orchestrating care at scale. Each stage enables the next.

For most organisations, the first real value comes from something foundational: data quality. Much of the difficulty in health and social care stems from information that is fragmented, incomplete or inconsistently captured. AI voice technology tackles this at source, turning spoken interactions into structured records, reducing administrative load, improving accuracy and freeing professionals to focus on care.

Transcription prepares the ground. Once data is captured well, summarisation and coding agents can standardise it, and insight agents can begin to interpret it, highlighting risk across a pathway, flagging a missed or delayed action, spotting patterns across a population. This is the point at which AI starts supporting decisions, augmenting professional judgement rather than replacing it. Only then is there enough context for an agent to act safely.

The integration challenge

The greatest barrier to adoption is not capability. It is integration. Standalone AI tools sit outside the systems that actually run these services. They understand little of the NHS or local government, and less of the Electronic Patient Record or social care case management system at the centre of it. Without that context, they fall out of step with clinical reality and regulatory change. And where there is no back-end integration, organisations revert to a cut-and-paste model, reintroducing the administrative burdens and inefficiencies they set out to eliminate.

Embedded agents change the equation entirely. When summarisation, insight and action agents live inside the system of record, the system itself becomes the orchestrator: referrals progress automatically, tasks coordinate across teams and organisations, risks are detected and acted on in real time. That is the difference between automation bolted on and care genuinely redesigned.

The cost problem

Cost predictability is a further challenge organisations must navigate. Unlike traditional software, where licensing is broadly fixed, AI is typically charged by use, and those costs mount as activity grows. Early pilots have seen individual interactions cost as much as thirty pence. A thousand doctors using a tool once a day can push annual costs into the hundreds of thousands of pounds. The public sector cannot budget against that kind of uncertainty. We need pricing that ties cost to outcomes, gives organisations predictability, and shares risk between provider and customer.

From national policy to local initiative

Procurement is the third lever, and it is best focused locally. National frameworks move slowly; regional and Integrated Care System-led approaches let organisations adopt tools that fit their populations and bypass some of the bureaucracy that stalls progress. Local authorities and Integrated Care Systems can also lay the digital foundations adoption depends on, such as connectivity, infrastructure and the basics that often get overlooked, including poor coverage in rural areas. Local AI innovation hubs, bringing together trusts, universities and technology partners, would let organisations pilot, learn from one another and move faster together.

Sequence before scale

None of this argues for caution for its own sake. It argues for sequence. At System C we take a deliberately staged approach, proving value early with transcription and summarisation, building intelligence into workflows through insight and decision support, and extending into action and orchestration only once data, systems and governance are mature enough to support it, within clear guardrails and with full auditability. The ability to act should be earned through proven capability and clinical safety, not assumed from technology.

The levers that matter now

Agentic AI is not a future aspiration for UK health and social care. The capability exists today. The levers that matter now are integration depth, economic sustainability, and governance that enables rather than restricts. Pull them thoughtfully, and this technology can fundamentally reshape how care is delivered across the country. Leave them stuck, and we risk another cycle of promising pilots that never reach the people who need them most.