Healthcare Is Drowning in $740 Billion of Admin Work. A Computer Use AI Agent Can Fix It.

The Numbers Are So Bad They Sound Made Up

Let's just sit with the scale of this for a second. The U.S. healthcare system spends roughly $740 billion a year on administration. That's not care. That's paperwork. According to the California Health Care Foundation, administrative waste accounts for up to 25% of all healthcare spending in the United States. One in four dollars goes to shuffling forms around. The 2025 CAQH Index, released in February 2026, found that the industry avoided $258 billion in costs through automated transactions last year, which sounds great until you realize there's still a $20 billion savings opportunity sitting on the table that nobody has touched. Hospitals spent $43 billion in 2025 alone just trying to collect payments, according to the American Hospital Association. And a single manual prior authorization still costs $11 to process versus roughly $2 when automated. The math isn't complicated. The will to act apparently is.

Prior Auth Is Where Patients Actually Get Hurt

• ●UnitedHealth's AI denial algorithm was the subject of a federal class-action lawsuit in 2025, with plaintiffs alleging it wrongfully denied Medicare Advantage claims at scale
• ●82% of prior authorization denials that were appealed ended up being overturned, meaning the original denial was almost certainly wrong
• ●A manual prior authorization takes an average of 2 or more business days to resolve, during which patients wait, delay care, or give up entirely
• ●The CAQH Index shows prior auth is one of the least-automated high-volume transactions in the entire system, still handled manually at shocking rates
• ●CMS launched a new model in June 2025 specifically targeting wasteful prior authorization practices, signaling even the government knows the current system is broken
• ●Health insurers are using AI to deny claims fast. Providers are still using humans to fight those denials slowly. That asymmetry is costing lives, not just money.

Why Old-School RPA Is the Wrong Tool for This Fight

Here's the part that makes me want to flip a table. A lot of health systems that finally decided to automate went out and bought UiPath licenses or built brittle RPA bots that break every time a vendor updates their portal UI. Legacy RPA is rule-based. It follows a script. The moment the insurance company moves a button or adds a new field to their prior auth form, the bot crashes and a human has to step in anyway. That's not automation. That's an expensive illusion of automation. Real computer use, the kind where an AI agent actually sees the screen, reads context, adapts to changes, and navigates any interface the way a human would, is a completely different category. Physicians are still logging 9 hours a week just on EHR documentation according to recent AMA data. Nine hours. A computer use agent that can read, fill, submit, and follow up across any web-based or desktop system doesn't care whether the EHR is Epic, Cerner, or some ancient portal from 2008. It just handles it.

The AI Arms Race in Healthcare Is Already Underway, and Providers Are Losing

Health Affairs published a piece in January 2026 calling what's happening right now an 'AI arms race in health insurance utilization review.' Insurers are deploying AI to process and deny claims faster than ever. Providers are still fighting those denials with staff, phone calls, and fax machines. Yes, fax machines. In 2026. Stanford HAI researchers flagged in June 2025 that AI tools in mental health care are being deployed without nearly enough scrutiny, with some chatbots potentially causing harm rather than helping. The lesson isn't that AI in healthcare is bad. The lesson is that AI deployed carelessly, without real-world task completion ability, is dangerous. There's a massive difference between a language model giving medical advice and a computer use agent handling the administrative layer so that actual clinicians have more time to give actual care. One replaces judgment. The other replaces data entry. Only one of those is a good idea.

Why Coasty Is the Computer Use Agent Healthcare Actually Needs

I'm going to be direct here. The reason I keep coming back to Coasty when I think about healthcare automation is that it's the only computer use agent I've seen that can genuinely handle the messy, inconsistent, multi-system reality of healthcare workflows. Coasty scores 82% on OSWorld, the definitive benchmark for real-world computer use tasks. Claude Sonnet 4.5, which Anthropic has been aggressively marketing for computer use, scores 61.4% on the same benchmark. That 20-point gap is not a minor technical detail. That's the difference between an agent that completes the prior auth submission and one that gets stuck halfway through because the insurance portal loaded a CAPTCHA. Coasty controls real desktops, real browsers, and real terminals. It doesn't rely on APIs that the insurance portal may or may not expose. It sees the screen the way a human does and acts on it. For a health system running Epic alongside three different payer portals and a legacy billing system, that matters enormously. The desktop app works for smaller practices. The cloud VMs and agent swarms let larger health systems run parallel workflows, submitting dozens of prior auths simultaneously instead of one at a time. There's a free tier to start, and BYOK support if your compliance team has opinions about API keys. The tool exists. The benchmark proves it works. The only question is how long your organization wants to keep paying humans to do jobs that a computer use agent can handle better and faster.