This is a preview of the September 17 edition of Access Health—Tap here to get this newsletter delivered straight to your inbox on Thursday mornings. I'm wrapping up my stage notes for Newsweek's AI Health Summit next week and noticing a difference from digital-health-care-forums past. Most of my questions are not about what AI can do, but about what people can do.
Who responds when an algorithm flags a deteriorating patient? What happens to a clinician's role when AI allows them to manage conditions that once required a specialist? If AI can draft a clinical trial protocol in minutes, can the contracting, budgeting and coverage processes around it move any faster?
AI can change what happens in one step of a workflow. Health care leaders now have to figure out how every step around it changes, too. That was the theme I kept hearing in briefing calls ahead of the Summit, and it became the focus of a story I published this week.
Health systems and technology developers have spent the past several years proving that AI can do remarkable things. But integrating those capabilities into decades-old workflows, job descriptions and decision-making structures is proving to be a very different challenge. Currently, the fundamental questions in C-suites are decidedly low-tech: Who does what?
Who makes the final decision? And what happens when the old way of working no longer fits? Low-tech, I'm told, does not mean less complicated.
The 2026 Stanford AI Index found that AI performance continued to rise sharply across advanced reasoning and agentic benchmarks in 2025. On SWE-bench Verified, which measures whether AI can independently solve real-world software engineering problems, performance increased from roughly 60 percent in 2024 to nearly 100 percent in 2025. On OSWorld, which tests agents performing computer tasks, performance went from roughly 6 percent to 75 percent over the same period.
Extract — continue reading at the source.
