Eight workshops. Eight adventures

“It shifted the group from 'what could this theoretically do' to 'I could do that tomorrow.'”

Kat Hampton’s day-to-day Copilot examples moved the room from abstraction to action: finding meeting times for packed calendars, pulling themes from a week of emails, and walking into conversations already oriented around priorities.

“Garbage in, garbage out. The quality of the prompt determines the quality of the output.”

A vague slide prompt and a fully PASTA-structured prompt produced visibly different outputs. The gap made “prompts matter” tangible, and pushed several teams toward Claude for slide creation.

“Rare disease is an isolating disease. In other areas patients might know someone with the disease, or know someone who knows someone. In rare disease this isn't the case.”

Two groups independently landed on the same intervention point: an AI-powered diagnostic support tool for HCPs in primary care, designed to flag rare disease signals earlier.

“There's a collective responsibility to learn how to use AI appropriately, both for what we create, and for how we navigate everything else that's out there.”

The room moved past arguing about whether AI belongs in the health information ecosystem. The productive question became how to use it well and help others do the same.

“The default instinct was to reach for 'agent' before checking whether the use case actually required it.”

Whether a tool is a workflow, assistant or agent is not a marketing choice. It is a risk classification, and the classification determines the controls.

“Earlier you heard prompting doesn't matter. Most of the time it doesn't. But when you need a consistent, scalable output, it does.”

A well-designed Categorise prompt became the structured input a Prioritise prompt could act on. The room saw how prompt chains can triage hundreds of papers.

“I've convinced Copilot to give me the recipe. I just need to be let in the kitchen to cook it up.”

AI-written code stopped being theoretical and started running. The group also proved that pared-down Copilot-friendly workflows can still make useful agentic builds viable.

“The AI wrote it, but I described what I wanted. That's a different skill, and it's one I think I can actually get better at.”

Groups built interactive, visually coherent outputs in ten minutes, often with no prior coding experience. The gap between intended build and actual build sometimes made the result better.