Nearly $1 billion in extra healthcare costs over two years. That’s the number a group of Blue Cross Blue Shield insurers put on hospitals’ use of artificial intelligence, and it landed in a report that reads less like a tech review and more like a dispatch from a war zone.
If you’ve been wondering what AI agents actually do once they’re loose in the real world, this is one of the clearest answers we’ve gotten. They don’t take over. They argue with each other. And the argument costs money.
What’s actually happening here
Let me unpack this in plain terms, because the mechanics matter more than the headline.
Hospitals have started using AI to read through patient records and transcribe conversations between doctors and patients. The software looks at all that text and flags conditions a human documenter might have missed or described too vaguely. More complex conditions identified means more complex conditions billed. More complex billing means higher reimbursements from insurers.
Insurers, meanwhile, are running their own AI to scrutinize incoming claims and push back on the ones that look inflated.
So you have software on one side building the strongest possible case for payment, and software on the other side building the strongest possible case against it. Both sides are getting faster and more thorough at their jobs. Neither side is getting cheaper.
The Blue Cross Blue Shield Association, which represents 31 independent insurers covering more than 100 million people, is the group putting the $1 billion figure forward. That’s worth keeping in mind as you read it. This is one side of a longstanding fight publishing a number about the other side. Hospitals would likely frame the same activity as finally getting paid accurately for care they were already providing.
The part that should interest anyone learning about AI agents
I write about AI agents for people who don’t build them, and I keep coming back to one idea that this story illustrates better than any demo I could show you.
An AI agent optimizes for the goal you give it. Not for fairness. Not for the health of the overall system. Just the goal.
Give a hospital’s agent the goal “document this patient’s condition as completely as possible,” and it will do exactly that, relentlessly, on every chart, forever. Give an insurer’s agent the goal “identify claims that don’t hold up,” and it will do exactly that, relentlessly, on every claim, forever.
Both agents are succeeding. That’s what makes this interesting. Nobody has to be cheating for costs to climb. Two systems doing their assigned jobs well can produce a result neither one was aiming at.
People often ask me whether AI is going to make things more efficient. The honest answer is that efficiency depends entirely on who’s pointing the AI and at what. Two well-aimed systems pointed at each other produce friction, not savings.
Why this pattern will show up everywhere
Healthcare billing is just the first place this became measurable, because healthcare billing already ran on adversarial paperwork. The AI didn’t create the conflict. It scaled up a fight that was already happening, and it scaled up both sides at once.
The same structure exists in plenty of other places:
- Job applicants using AI to write résumés, employers using AI to screen them
- Companies using AI to write contracts, other companies using AI to find the holes in those contracts
- Marketers using AI to generate content, platforms using AI to filter it
- Customer service bots negotiating with complaint-writing bots
In every case, both sides get faster. In none of those cases does the underlying disagreement get resolved. The volume just goes up.
What to take away from it
I don’t think the lesson is that AI in healthcare is bad. AI reading medical records and catching a condition a tired resident missed at 2 a.m. is a genuinely good thing, and it’s hard to argue the patient is worse off for having their chart be accurate.
The lesson is narrower and more useful. When you hear that AI will cut costs in some industry, ask a follow-up question: is there someone on the other side of this transaction who also has AI? Because if there is, the savings you were promised may just turn into a faster, more expensive version of the same disagreement.
The $1 billion figure isn’t really a story about machines outsmarting anyone. It’s a story about what happens when you automate both halves of an argument and forget that somebody still has to pay for the argument.
That somebody, as usual, shows up on your premium statement.
🕒 Published: