Picture two chess engines locked in a match that never ends. Neither one gets tired. Neither one makes a sloppy move. And every few seconds, both of them reach over and take a dollar out of your wallet to fund the next round. That’s roughly what’s happening between hospital AI and insurer AI right now, and the people funding the board aren’t playing.
The claim coming from insurers is blunt: AI is already pushing healthcare costs up. Not “might,” not “could by 2030.” Already. Their argument is that AI documentation and coding tools are increasing billing amounts, because when hospitals use AI to capture more detailed patient information, the resulting bills get bigger. Better documentation, higher codes, fatter invoices.
I review AI tools for a living. I’ve watched a hundred vendors promise that automation reduces cost. This is the clearest real-world case I’ve seen of the opposite happening, and it’s worth sitting with.
Efficiency Isn’t the Same as Savings
The standard pitch for AI in any back office goes like this: the software does tedious work faster and more accurately than humans, so you spend less. That pitch quietly assumes the tedious work was a cost center with no revenue attached.
Medical coding is not that. Coding is revenue. Every detail a coder captures maps to a billable code, and every code maps to money. So when you point a very capable machine at that process and tell it to be thorough, you don’t get cheaper operations. You get a system that finds everything it’s entitled to find and bills accordingly.
The uncomfortable part is that this may not be fraud. Insurers are describing higher billing amounts driven by better documentation. If the documentation is accurate, then the AI is doing exactly what it was built to do. The bills were arguably too low before, in the sense that humans were leaving money on the table through fatigue and oversight. AI doesn’t get tired. That’s the whole sales pitch, and it’s also the problem.
The Other Side of the Board
Insurers aren’t standing still, and they’re not innocent bystanders in this fight. Prior authorization has become one of the main places payers are putting AI to work — the process where providers have to get approval before a patient receives care. As AI takes over more of those coverage decisions, the risks to patients grow.
In early 2026, the CEOs of nearly all major health insurers told Wall Street analysts the same thing on earnings calls: they’re going to cut costs, in part, by using AI. So you have hospital AI optimizing for more billable detail, and insurer AI optimizing for more denials and tighter approvals. Both sides are automating their half of a decades-old fight.
That escalation is what’s pushing medical costs higher, and the reason is boring but important. When each side deploys faster, more thorough software, neither side wins outright. They just both spend more on the fight. Hospitals add coding tools, insurers add review tools, hospitals add appeal tools, insurers add more review capacity. The administrative arms race compounds, and administrative cost is a real component of what you pay.
What This Means for Anyone Buying AI Tools
Set aside healthcare for a second, because the pattern generalizes and most vendors won’t tell you about it.
- Automation amplifies whatever the process was already optimizing for. If your workflow was quietly maximizing something, AI will maximize it harder. That’s not a bug in the model. That’s the model working.
- Cost savings claims need a counterparty check. A tool that saves you money often does so by extracting it from someone else in the chain. If that someone is your customer, expect blowback.
- Adversarial adoption cancels out gains. When both sides of a negotiation automate, the efficiency gains get spent on the negotiation itself. Nobody ends up ahead.
The Accountability Gap
AI’s role in escalating medical expenses is now a central point of contention, and healthcare cost forecasts for 2026 already name AI documentation and coding tools as an accelerant on top of an increasingly complex patient population. Meanwhile, the federal government shutdown has been driven in part by the health cost issue. This isn’t a niche vendor debate anymore.
What bothers me most as a reviewer is how little of this shows up in product marketing. Coding tools sell “capture accuracy.” Prior auth tools sell “utilization management.” Both are technically honest and neither mentions that the combined effect on the system is more money moving through more machinery with a patient at the bottom of the funnel.
The tools aren’t evil. They’re doing precisely what they were asked to do, faster than anyone thought possible. That’s the part nobody planned for.
🕒 Published: