AI and digital practice for nurses · Ward and team managers, senior nurses, nurse leaders, digital and safety leads
Agentic AI in health and care: supervising systems that act
What it means when software plans and acts on its own, how it fails, and how a nurse leader keeps it under real control.
- 2modules
- 1.5CPD hours
- 100guided minutes
- Freeto study
Start the course, free Create a free account to save progress
About this course
A chat assistant answers. An agent acts. Given a goal it plans the steps, uses other systems and carries them out, often with nobody watching each step. Rota management, discharge planning, referral handling and results chasing are all being offered as agents, and the person asked to approve one is increasingly a nurse leader rather than an engineer.
In May 2026 the National Cyber Security Centre, with partners in the United States, Australia, Canada and New Zealand, published joint guidance on the careful adoption of agentic AI services. This course turns it into practical questions for people who run clinical teams: what it is, how it fails, where a human must stay in the loop, and what evidence to ask for before one is let near patients or records.
It is about judgement, not technology. You will not learn to build an agent. You will learn to supervise one, and to say no to one, with reasons.
What you will be able to do
- Explain the difference between an assistant that answers and an agent that acts
- Recognise where agents are being introduced in health and care and what each can touch
- Describe why errors compound when software acts in steps
- Explain prompt injection and why limiting what an agent can do matters more than filtering what it reads
- Place a human approval step correctly and distinguish a real review from a rubber stamp
- Apply least privilege, logging and a stop control to an agent deployment
- Ask the right questions of a supplier and of your own organisation before go live
- State where accountability sits when an agent has acted
Modules
Assessment and certificate
Knowledge check after each module and a final assessment at 80 per cent, with unlimited attempts.
A digital certificate, issued the moment you have passed and paid, showing 1.5 CPD hours with a verification code. It evidences knowledge of how to supervise agentic AI systems in health and care. It is not a regulated qualification and it does not authorise you to approve any system: that remains a decision for your organisation's governance.
Questions
What is the difference between an AI assistant and an AI agent?
An assistant produces an answer or a draft and stops. An agent is given a goal and works towards it by planning steps and using other software, such as a calendar, a record system or email, with little or no human involvement. The step that changes things in the real world, sending, booking, ordering or altering a record, is what makes the difference.
Is there official guidance on using agentic AI safely?
Yes. In May 2026 the UK National Cyber Security Centre published joint guidance on the careful adoption of agentic AI services with its counterparts in the United States, Australia, Canada and New Zealand. It is aimed at security teams, and it stresses limiting what an agent can access, strong identity controls, monitoring, and human oversight.
Why not just tell the agent to ignore anything suspicious?
Because the agent cannot reliably tell an instruction from the text it is reading. A referral letter or an email can contain words that the agent treats as a command. That is called prompt injection and it is not fully solved. The dependable defence is to limit what the agent is able to do, so that a mistake or a trick has little to work with.
Does a human clicking approve make an agent safe?
Only if the approval is a real review. One click to approve fourteen discharges with no way to see what each will do is consent without understanding. A meaningful approval shows the specific change, gives the reviewer time and the standing to refuse, and records refusals as readily as approvals.