Early evidence suggests that AI is actually making the things that patients dislike most about the health care system worse.

Why it matters: AI's been touted as a solution to administrative headaches and costs, but the evidence so far suggests that it makes things more expensive and escalates payment battles between insurers and providers.

Driving the news: A study published this week by the Peterson Health Technology Institute examined the impact AI is having on two huge administrative processes: prior authorization and medical coding.

• Prior authorization, the insurer preapprovals for physician-ordered care, is so unpopular with patients and clinicians that there's serious momentum to address it.
• And medical coding is basically how providers document what happened during a patient visit so they can be paid. The more codes per visit, the more providers get.
• The study, which was the product of a stakeholder workshop on the topic, concluded that administrative AI is actually leading to more prior authorization back-and-forth — the result of "bot wars" — and higher costs through more intensive medical coding.

What they're saying: "It's taking all of the adversarial processes in the system and amplifying them and putting them on steroids, where everything goes faster and further," Caroline Pearson, executive director of the Peterson institute, told me.

• "AI could create not only faster processes, but more efficient processes," she added. "The problem is no one has an incentive to implement AI in ways that take friction out of the system right now."
• "Health plans are using AI responsibly to improve care and reduce costs," said a spokesperson for the insurer trade group AHIP, adding they're simplifying prior authorization and ensuring patients get affordable, evidence-based care.

The big picture: The U.S. spends an estimated one quarter of its enormous health care budget on administrative costs.

• The "AI-to-the-rescue" argument basically goes like this: It'll simplify all of that back-end activity that bloats overall spending, making care significantly cheaper for patients.
• One estimate just two years ago in the National Bureau of Economic Research projected that the use of AI could lower health care spending by 5%-10% within five years (though not just through administrative savings).

Reality check: That's not the direction things seem to be heading in, and some experts say it's wishful thinking without meaningful changes to the health system's underlying incentives.

• In theory, payers and providers should all be fighting for a finite pool of patients and competing on prices. If AI lowers costs, providers could bill for less and insurers could lower premiums.
• But in reality, health care markets are often heavily consolidated and split between a few large companies, meaning there's little pressure for any of the players to sacrifice profit for the sake of lowering consumers' costs.

The bottom line: For now, using AI to simplify or reduce administrative tasks within health care is more likely to pad profit margins or even create more bot-enabled red tape than to drive down patients' costs.

• "Any AI that creates some efficiency for some organization ... do not generally lead to price deflation," said Suhas Gondi, chief medical officer at Health Strategy, who I called after stumbling across an article he co-wrote last year questioning AI's ability to reduce waste in the health care system.
• "They just lead to more profit. Unless there's a competitive market, you're not going to have an incentive to leverage those internal efficiencies to lower your price."

Medical coding and documentation has been revolutionized by AI scribes, which essentially automate the process of taking notes during a patient visit and translating that visit into a billable claim.

• But while scribes were originally pitched as a way to give doctors their time back, they're now explicitly also being sold on their potential to increase revenue.

Why it matters: AI scribes may be submitting a more accurate representation of patient visits than humans do. Fair or not, the result is the same: That heavier-handed coding is increasing costs.

• Yes, insurers pay providers. But patients pay premiums, which absorb the underlying costs of care, and they also often pay out-of-pocket costs based on the price of the service they received.

By the numbers: The adoption of an AI scribe was associated with an increase in "relative value units" that translates to $3,044 annually per physician, one study published in JAMA Network Open this year found.

• A case study by KLAS Research found that AI platform Suki led monthly revenue gains of as much as $1,000 per provider.
• Another case study on leading AI scribe Abridge's website found that "more accurate documentation has contributed to increased revenues and a significant return on investment" at one health system using the product.

The other side: Some insurers are responding by reducing payments or cracking down on coding practices.

• A recent Blue Cross Blue Shield report concluded that the use of AI in billing "may be causing a mismatch between the care patients receive and the bills hospitals send to insurance companies."

The intrigue: Though both plans and providers participating in the Peterson institute's working group agreed that there's been an increase in coding intensity — particularly last year amid widespread adoption of AI scribes — there's not yet good data for 2025 to back up those claims, Pearson said.

• "As we saw more widespread adoption of ambient scribes, we're expecting a turn — a hockey stick level of increase as many more things are going to be able to be coded as higher complexity visits," she told me.

Between the lines: The rub really isn't about whether AI scribes are helping providers generate more revenue via the billing process. It's whether it's justified by the underlying care or not.

• But as always, in health care, somebody's new revenue comes at someone else's expense.

The bottom line: "We're actually sort of getting caught in this AI arms race," Gondi said.

• "The systems that have been most adopted are, on the health systems side, the things that help them optimize documentation and billing and coding intensity, and then on the payer side, the things that allow them to fight back."