The Pilot Era Ends, the Assurance Case Begins

NHS.net reported that Microsoft 365 Copilot's Analyst and Researcher agents could open a new chat window instead of the selected agent. The cause was an incorrect URL behind the Microsoft 365 app store's Open button. Microsoft had validated a fix and begun deploying it, but NHS.net warned that rollout would take several weeks; the stated workaround was to open the agents through the Copilot agent store.

University Hospital Southampton has deployed 185 Copilot licences and created an automation hub in SharePoint for best practice and support. Its wider programme also covers outpatient administration, ambient voice and robotic process automation, with early RPA work planned for maternity and partial booking in outpatients. The trust's stated challenge for the rest of the year is converting digital pilots into sustainable operational benefits.

Resilience belongs inside the agent's governance pack, not only in the IT incident queue. NHS.net's Copilot extensibility guide places local responsibility for information governance, DPIAs, clinical safety approvals and organisational risk with each participating organisation. The same pack should state the approved route into the agent, what staff do when it is unavailable, which records prove who did what, and which workflow must revert to manual handling.

Imperial College London and Smith+Nephew have opened a five-year research partnership for musculoskeletal surgical robotics. It is Smith+Nephew's first university centre and Imperial's first industry research centre in medical technology. At full capacity, it is expected to support a senior postdoctoral researcher and up to seven fully funded PhD students, with research focused on computer vision and sensing.

One target is markerless registration and tracking. Current systems can require pins or rigid markers fixed to bone so the robot can map anatomy. The centre will explore computer-vision approaches that adapt to anatomy in real time, aiming for less invasive and more intuitive systems that can be used beyond specialist centres. This remains research, but it starts with a recognisable clinical workflow constraint rather than a generic robotics promise.

Industry and academia working side by side can shorten translation, but it also makes the evidence plan more important. The claim to test is not simply whether markerless tracking works in a controlled setting. It is whether accuracy, failure detection, training burden and outcomes remain acceptable across different teams, anatomy, equipment and hospital environments, including the non-specialist centres the programme hopes to reach.

The Royal Wolverhampton NHS Trust and Walsall Healthcare NHS Trust are rolling out Heidi Health's ambient voice technology after an eight-month pilot at Walsall. The first phase covers emergency departments and outpatient specialties across both trusts, with implementation across all areas intended by 2028.

The pilot included emergency medicine, same-day emergency care, dermatology, ENT, audiology, obstetrics and gynaecology. The trusts report that average documentation time fell to under four minutes per patient. The system listens to the consultation and drafts structured notes, which are reviewed, edited where necessary and approved before they enter the patient record.

A preferred supplier and a successful pilot do not remove local assurance duties. NHS England guidance calls for a named Clinical Safety Officer, DCB0160 documentation, a safety case, hazard log, DPIA and monitoring framework, alongside ongoing audits of documentation accuracy and independent analysis of local performance data. Review and approval before filing is a control only if teams measure what reviewers miss and whether time pressure erodes that review.

Queen Victoria Hospital NHS Foundation Trust has introduced Harrison.ai CXR in Radiology. The system analyses a chest X-ray after it is performed and can identify up to 124 findings, using that analysis to help prioritise examinations for clinician review.

The AI runs in the background and presents findings within the existing reporting workflow. QVH is explicit that it is decision support, not a replacement for radiologists or reporting clinicians: every chest X-ray still receives a formal report by a qualified clinician, and that report remains the definitive result. The introduction is supported by the Surrey, Sussex and Frimley Imaging Network.

Prioritisation is itself a clinical intervention on the queue. Assurance should therefore cover more than whether the AI and clinician agree on a finding. Teams need to monitor false negatives, time to report, what is displaced when one examination is promoted, whether performance varies across patient groups and whether staff understand when the queue logic can be overridden.