AI amplifies whatever the system already rewards

Estimated reading time: 3 minutes

Welcome to AI-Rx 👋

This week isn't about a model. It's about the thing that decides whether any model helps or harms: incentives. Two arguments, one on hospital readiness, one on primary care, land on the same conclusion.

The gap nobody names

Two numbers that don't belong together: 75% of health-system executives believe AI has reached a turning point in reshaping care. Only 6% have an established strategy. That's not enthusiasm, it's exposure.

Angel Arnaout's recent paper argues being ready for AI isn't something you buy, it's something you become. Capability rests on infrastructure, governance, skill, and culture. Drop a capable tool into an organization missing those, and you've added a liability.

Five things that make an organization AI-capable

The framework is straightforward and unglamorous: a clear strategy (route requests through problems, not technology); good data (bias in, harm out); governance matched to risk (admin now, diagnostic support with validation, autonomous tools with the most oversight); a workforce trained by role; and continuous monitoring, because models decay.

One counterintuitive point: banning "bring your own AI" backfires, it just pushes staff to ungoverned tools. Approve vetted ones and train people instead.

Now the wall: the business model

Here readiness collides with something harder.

In reporting on primary care's future, one account describes teams inside larger systems being told to shut down their heart-failure and COPD programs, not because they failed, but because they worked.

They reduced admissions elsewhere, and those admissions are revenue.

Read that again. Good primary care was penalized for keeping patients out of the hospital. No amount of AI capability fixes that, because the problem isn't capability. It's what the system is paid to do.

AI is an amplifier, not a fix

In a system paid for referrals, admissions, and volume, AI produces more of them, faster.

Point the same tech at a model that rewards keeping people healthy, per-member-per-month primary care, and it becomes the most powerful tool primary care has ever had.

The VHA has worked this way for decades: more primary care, lower total cost of care, especially for high-risk patients.

The demand is enormous, around 100 million Americans lack a primary care physician, and AI could genuinely help close it. But only if the incentives point that way.

Here's my final thought

The organizations that win with AI won't be the ones that bought the most tools. They'll be the ones that built the capability to run them, and deployed them inside incentives worth amplifying. Technology was never the lever it looks like. Structure and payment are.

Are you buying AI tools, or building the capability and the incentives to make them help?