Run AI like an investment portfolio, not a science fair

In This Issue

Run AI like an investment portfolio, not a science fair.

An AI portfolio needs regular investment decisions. Every initiative, including the small tools and systems already in production, should have an accountable owner, visible costs, measurable benefits, and a review date. One team runs the loop; here is how it turns.

Make it a recurring discipline, with Business, Technology, and Finance at the table, and start in one domain with a willing champion. The test is simple: does the organization make better decisions about where its next dollar and next hour should go?

Evidence, not capability, is the dividing line now: an investigation into whether medical AI works, an industry bid for federal rules, the federal-vs-frontline split, and a look at what unsupervised autonomy does.

A Financial Times investigation finds AI’s advances have not yet become better care: tools reach the bedside with little oversight, and one sepsis model fired on nearly every patient until clinicians ignored it. Ask vendors for outcomes measured after deployment, not lab accuracy, and build the post-market monitoring you will be judged on. Financial Times

CTA CEO Kinsey Fabrizio argues US health AI needs risk-based federal rules, not a patchwork of 240-plus state bills across 43 states: post-market monitoring, a consensus standard backed by 75-plus organizations, consistent privacy law, and AI reimbursement. Whether or not Washington acts, design to these controls now, because buyers and regulators will expect them. Fierce Healthcare (op-ed)

At a CTA event in Washington, CMS and FDA officials laid out ambitious AI plans while frontline clinicians pushed back on the risks: federal enthusiasm for speed meeting real skepticism from the people who will use these tools. With the posture tilting toward speed, your internal review is the main thing standing between a vendor demo and your patients. Fierce Healthcare

In a security test by the firm Irregular, Google’s Gemini broke into three outside companies’ systems, its first autonomous hacks, by guessing passwords and finding exposed credentials; Google says it stopped once it realized they were real. Before giving an agent standing access in your environment, decide what it can reach, what it logs, and who can shut it off. TechCrunch

Capital is still moving, even as the proof debate sharpens. Two rounds worth noting this week.

As health systems move agents from pilot to production, the bill gets hard to see: vendors shifted to token-based, consumption pricing with no consolidated view, so finance cannot model spend beforehand. Rush’s CIO urges “cloud FinOps” discipline. Name an agent-cost owner and build a spend model before you scale, because a system can spend more on an agent than the labor it replaces. Becker’s

Apple acquired the assets of biomagnetic-sensing startup Sonera and paired new Watch Series 12 and Ultra 4 sensing with an Apple Intelligence health app; no terms disclosed. Consumer hardware keeps pushing toward clinical-grade signals, so decide now which patient-generated wearable data you will act on and which you will not. MobiHealthNews

The next three are payer and provider economics, not AI stories, but they are the pressure behind every efficiency bet above.

Medicare Advantage now covers 51% of eligible seniors, but high denial rates, prior-auth burden, and thin reimbursement are pushing systems out: roughly 90 hospitals have exited MA contracts over three years, at least 30 in 2026. Those denials are exactly the fight providers hope AI can help them win. Becker’s

As patients lose ACA-exchange coverage and become uninsured, HCA’s Q2 same-facility inpatient surgeries fell 2.3% and outpatient 4.4% year over year, with former ACA-exchange patients migrating almost one for one to uninsured. Softening elective demand is the backdrop for why systems chase AI-driven efficiency rather than banking on volume growth. Healthcare Dive

Ascension is selling its stake in the Arizona managed-care plan Mercy Care to CVS’s Aetna, continuing a retreat from owning health plans after earlier exits in Wisconsin (2023) and Texas (2024). The plan covers roughly 404,000 Arizona Medicaid members; terms were not disclosed. The line between running care and running coverage keeps shifting toward the big diversified payers. Healthcare Dive

What Comes Next is a weekly point of view from StatusGo on AI in healthcare. Figures as disclosed by sources; selected stories, not exhaustive.