The week in health tech, explained in plain English.Get the free newsletter →
September 18, 2026

Takeaways from the most recent news in the technology and policies shaping healthcare.

Health IT

The AI Agent Bill Health Systems Can't See Coming

Health systems racing to deploy AI agents are running into a basic problem: they cannot see what the agents actually cost. More than 18 months into building AI agents, the CIO at Rush University System for Health told Becker's Hospital Review that the tooling to model those costs does not yet exist.

That matters because AI agents behave differently from traditional software. They call large language models repeatedly, chain tasks together, and consume compute in ways that scale with usage rather than seat licenses. A single agent handling a complex workflow can quietly generate a large, variable bill, and current vendor dashboards do not forecast it well. The result is spending that is hard to predict and harder to budget.

With no reliable financial model, Rush is leaning on the workforce as the control. That means governing which agents get built, how they are used, and who is accountable, rather than trusting a cost dashboard. For other systems, the lesson is to treat AI agent economics as an open question and build internal guardrails before scaling.

More in Health IT

Health IT

Why Top Healthcare CIOs Lead With Relationships, Not Tech

University of Utah Health's Donna Roach says influential healthcare CIOs win through executive relationships, not technical skill.

Why it matters: As health systems bet big on AI and digital tools, a CIO's influence with the C-suite determines whether those investments succeed.

Health IT

Physician: AI's Bigger Risk Is Not Using It in Healthcare

A physician argues in Becker's Hospital Review that the U.S. health system's cost and poor outcomes are a bigger, more immediate danger than AI's theoretical existential risks.

Why it matters: It reframes the AI debate for health leaders around a present-day threat: a costly system that already fails patients.