Working with AI agents in health systems

The rising role of AI in nursing and allied health practice

Pediatric Neuroscience Conference and GCC Pediatric Neurology Conference, Doha, Qatar

neuroscience nursing & allied health symposium · online   2026-10-30

Lily

  • Heart surgery at four months
  • Her parents asked ChatGPT about her symptoms
  • It suggested a syndrome her doctors called too rare to be likely
  • Tests confirmed it: one of 70 known cases worldwide

Withers & Parkin, 2026

What we'll cover

  1. What AI agents can do now
  2. How we're responding to AI
  3. Care as a system that includes AI
  4. Families who now bring their own agent
  5. What this asks of you and your team
§01 · what AI agents can do now

What AI agents can do now

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

AI has become an agent

  • Memory : keeps what you've told it across conversations
  • Planning : breaks a goal into steps and works through them
  • Action : searches, reads documents, drafts, fills in forms
  • Checking : reviews its own work and adjusts

These are already built into ChatGPT, Claude and Gemini, on the phones of your colleagues and the families you see (Weng, 2023).

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

In tests and simulations, AI now works at a clinical level

  • A diagnostic AI took histories and reached diagnoses in simulated consultations at least as well as primary care doctors (Tu et al., 2025)
  • Doctors working with GPT-4 reached the same standard as GPT-4 working alone (Goh et al., 2025)
  • ChatGPT and DeepSeek answered 60 rehabilitation questions, including on multiple sclerosis, to a standard rated above that of final-year physiotherapy students (Michou et al., 2026)
  • AI replies were rated more empathetic than clinicians' in 13 of 15 studies (Howcroft et al., 2025)

On written cases and exam questions, AI now works at the level we expect of qualified clinicians.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

In real care, the evidence points the same way

  • Emergency department second opinions : the right diagnosis at triage in 66% of cases, where the two attending physicians on the same cases reached 54% and 48% (Brodeur et al., 2026a)
  • Primary care : 100 patients took a pre-visit history with an AI; the final diagnosis was on its list in 90% of cases (Brodeur et al., 2026b)
  • Nursing : an AI that reads patterns in nurses' documentation; 35.6% lower risk of in-hospital death across 74 units (Rossetti et al., 2025)

The evidence is moving from test cases to real patients and clinical records, and the results are holding up.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

On its own, the AI named the right condition 95% of the time. People using the same AI managed under 35%.

Bean et al., 2026

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Why people with the AI did worse

  • People left out details that mattered
    The AI can only reason from what it's told
  • They asked closed questions: "Could this be stress?"
    A yes-or-no question limits what the AI can say back
  • The AI often suggested the right answer, and people didn't pick it
    It offered two or three conditions; choosing between them takes clinical knowledge
  • Two people describing the same brain bleed got opposite advice
    Only one of them said the headache came on suddenly

Taking a history is a clinical skill, whoever is asking the questions.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Agents work better with more context

  • An agent that knows the child's history, medicines and care plan gives better answers than one starting blank
  • Families are already giving it that context: letters, test results, seizure diaries, photos
  • More context brings more risk with children : someone else decides what's shared about them, and it can stay with the platform into adulthood

Context is what makes agents useful, so it's worth deciding on purpose what goes in (Rowe & Lynch, 2025).

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

What this looks like in your work

  • Nurse : "Here are this child's seizure observations over the last 24 hours. What patterns should I flag at handover, and what would make you worry?"
  • Physiotherapist : "Draft a home programme for a five-year-old with spastic diplegia, written for parents with little time and no equipment."
  • Occupational therapist : "Suggest classroom adaptations for a seven-year-old returning to school after a brain injury, with fatigue and poor concentration."
  • Speech and language therapist : "Adapt this AAC vocabulary set for a child with cerebral palsy whose family speaks Arabic at home."
§02 · how we're responding to AI

How we're responding to AI

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

What people tell me they're worried about

  • Being replaced : will there be a role for me in ten years?
  • Getting it wrong : what if I trust an output I shouldn't?
  • Losing skill : what happens to my judgement if I stop practising it?
  • Falling behind : everyone else seems to know how to use this

All four are about where you fit once AI is part of the work.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Sanctuary strategies

  • Denial : "It can't replace the therapeutic relationship." In some contexts, it has
  • Retreat : "Focus on what only humans can do." A shrinking island
  • Restriction : "Don't let clinicians use it." They will, because it's useful
  • Resignation : "Our roles will diminish." Helplessness that sounds like realism

All four leave you watching something happen to you.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Every argument that starts with "what humans do better" is ground you'll have to give up with the next model release.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Better than, or better with?

  • Most studies ask whether AI is better than us
  • A colleague who's better than you at something is someone you learn from
  • AI feels different because it seems to threaten our expertise, our status and our autonomy (Jussupow et al., 2018)
  • Those are concerns about ourselves, and they pull our attention away from the child and family
  • The question that keeps the child at the centre: what could you and the AI reach together that neither reaches alone?
§03 · care as a system that includes AI

Care as a system that includes AI

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Patients are bringing their own AI

  • 230 million people ask ChatGPT health questions every week (OpenAI, 2026)
  • Health and wellness is the largest category of people asking Claude for personal guidance: 27% (Anthropic, 2026)
  • A stress test of ChatGPT Health found it recognised classic emergencies such as stroke, but under-triaged about half of the less obvious ones (Santoro, 2026)

"Some of them had never heard of it; others said it is so rare it's very unlikely." Lily's mum, on telling her doctors what ChatGPT had suggested (Withers & Parkin, 2026)

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Chart of the topics people ask Claude for personal guidance about; health and wellness is the largest category

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Care already happens in a system

  • Clinical reasoning has always been shared, across people, records, protocols and places (Hutchins, 1995)
  • Your judgement and your colleagues', the hospital's records and decision support, the family's knowledge of their child
  • And now the family's AI, and yours
  • Your skill is managing that system: what to trust, how to reconcile conflicts, owning the synthesis

What matters is whether the whole system produces good outcomes for the child.

§04 · families who now bring their own agent

Families now bring their own agent

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Families arrive having already asked

  • About the EEG report, the MRI, the side effects, the prognosis
  • Some arrive with a view of their child's condition already formed
  • Lily's parents arrived with a name for the condition
  • This is a new kind of health literacy : knowing what to ask an AI, and how far to trust what it says
  • It's early, for families and for us; we'll work out what good use looks like together

These are families taking an active part in their child's care.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

If we only count the clinician's AI, the family becomes someone decisions are made about.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Prepare families before they come in

Usual response : deal with AI-informed views when families raise them.

Another option : help families use AI well before they come in.

"Before your child's epilepsy review, ask your AI: what should I record about her seizures between now and then? What questions should I ask about changing her medication?"

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

What you can help families do

  • Ask open questions : "What else could explain this?" gets more than "Could this be stress?"
  • Give useful context : age, diagnosis, medicines, what's changed and what's already been tried
  • Share carefully : leave out names, dates of birth and anything else that identifies the child, and check the app's privacy settings
  • Bring it back : show the team what the AI said, so you can work with it together

You're helping them take a better history, with the AI listening.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

The family's agent in the MDT

  • Families are building a record of their child's history, symptoms and worries, in conversations you may never see
  • They may arrive knowing things about their child that your notes don't
  • They bring that agent into the MDT, and so will each of you
  • Each agent carries the priorities of the person it works for

The therapeutic relationship is still there; what it includes has changed.

§05 · what this asks of you and your team

What this asks of you and your team

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Setting up, noticing and answering

  • Setting up : deciding what the AI does, with what information, and what you check
  • Noticing : spotting when the system starts to go wrong over time, and knowing when to act alone
  • Answering : owning decisions you didn't wholly produce, as you already do for team decisions
§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

Questions for your team

  • Who checks an AI-drafted note before it's signed, and how is that recorded?
  • When a family brings an AI's suggestion, whose job is it to respond?
  • If the AI and the team disagree, how is that resolved and documented?
  • Which tasks need someone to act without AI, and why?
§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

The professionals who shape AI in practice will be the ones who bring deliberate context to it.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

tomorrow, do this

Ask the next family you see how they're using AI

"What have you asked it about your child? What did it tell you?"

You'll hear what they already believe, and they'll know it's something you want to hear about.

questions

Thank you

mrowe@lincoln.ac.uk

QR code linking to the page for this talk, with the slides and handout
slides and further reading

neuroscience nursing & allied health symposium · online   2026-10-30
§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

References

Anthropic. (2026, April 30). How people ask Claude for personal guidance. https://www.anthropic.com/research/claude-personal-guidance

Bean, A. M., Payne, R. E., Parsons, G., et al. (2026). Reliability of LLMs as medical assistants for the general public: A randomized preregistered study. Nature Medicine, 32(2), 609–615. https://doi.org/10.1038/s41591-025-04074-y

Brodeur, P. G., Buckley, T. A., Kanjee, Z., et al. (2026a). Performance of a large language model on the reasoning tasks of a physician. Science. https://doi.org/10.1126/science.adz4433

Brodeur, P., Koshy, J. M., Palepu, A., et al. (2026b). Conversational diagnostic artificial intelligence in ambulatory primary care: A prospective feasibility study. The Lancet. https://doi.org/10.1016/S0140-6736(26)01535-7

Goh, E., Gallo, R. J., Strong, E., et al. (2025). GPT-4 assistance for improvement of physician performance on patient care tasks: A randomized controlled trial. Nature Medicine, 31(4), 1233–1238. https://doi.org/10.1038/s41591-024-03456-y

Howcroft, A., Bennett-Weston, A., & Khan, A. (2025). AI chatbots versus human healthcare professionals: A systematic review and meta-analysis of empathy in patient care. British Medical Bulletin, 156(1), ldaf017. https://doi.org/10.1093/bmb/ldaf017

Hutchins, E. (1995). Cognition in the wild. MIT Press.

Jussupow, E., Heinzl, A., & Spohrer, K. (2018). I am; we are: Conceptualizing professional identity threats from information technology. In Proceedings of the 39th International Conference on Information Systems. Association for Information Systems.

§01 what agents can do §02 responding to AI §03 care as a system §04 the family's agent §05 you and your team

References

Michou, I., Fouras, A., Chrysanthakopoulou, D., et al. (2026). AI-powered physiotherapy: Evaluating LLMs against students in clinical rehabilitation scenarios. Applied Sciences, 16(3), 1165. https://doi.org/10.3390/app16031165

OpenAI. (2026, January 7). Introducing ChatGPT Health. https://openai.com/index/introducing-chatgpt-health/

Rossetti, S. C., Dykes, P. C., Knaplund, C., et al. (2025). Real-time surveillance system for patient deterioration: A pragmatic cluster-randomized controlled trial. Nature Medicine, 31(6), 1895–1902. https://doi.org/10.1038/s41591-025-03609-7

Rowe, M., & Lynch, W. (2025). Context sovereignty for AI-supported learning: A human-centred approach. OSF Preprints. https://doi.org/10.31219/osf.io/8czva_v2

Santoro, E. (2026, May 13). Is ChatGPT Health ready for emergency triage? Medscape. https://www.medscape.com/viewarticle/chatgpt-health-ready-emergency-triage-2026a1000ffi

Tu, T., Schaekermann, M., Palepu, A., et al. (2025). Towards conversational diagnostic artificial intelligence. Nature, 642, 442–450. https://doi.org/10.1038/s41586-025-08866-7

Weng, L. (2023, June 23). LLM powered autonomous agents. Lil'Log. https://lilianweng.github.io/posts/2023-06-23-agent/

Withers, K., & Parkin, S. (2026, September 25). 'AI helped diagnose my daughter's rare condition'. BBC News. https://www.bbc.co.uk/news/articles/cm4g5yp0ng85o