Healthcare Is Drowning in $2.2 Trillion of Paperwork. A Computer Use AI Agent Can Fix It.
A physician walks into a room with a patient. She spends the next 15 minutes typing into an EHR instead of making eye contact. That's not a horror story. That's Tuesday. Research from the VA and published in PubMed in early 2025 found that primary care physicians spend between one-third and one-half of every single patient visit interfacing with their electronic health record. Not diagnosing. Not treating. Typing. Meanwhile, Oliver Wyman projects that if nothing changes, U.S. healthcare administrative costs will balloon to $2.2 trillion by 2035, or roughly $6,400 per person in this country. Per year. Just for paperwork. We have AI that can control a computer, navigate any interface, fill out any form, and execute complex multi-step workflows without a human touching a keyboard. And most health systems are still paying armies of billing coders to copy-paste data between tabs. This is not a technology problem. It's a stubbornness problem.
The Numbers Are Actually Obscene
Let's stop being polite about this. The American Hospital Association confirmed that hospitals spent more than $1 trillion in 2025 on workforce costs alone, with administrative burden listed as one of the top contributors to that figure. The Commonwealth Fund published a report in October 2025 confirming that more than two in five primary care physicians across 10 countries are dissatisfied with the amount of time they spend on administrative tasks. Prior authorization alone accounts for approximately $35 billion in administrative spending annually, per Health Affairs. And that's just one workflow. We're not talking about edge cases. We're talking about a system where a doctor has to spend 20 minutes faxing (yes, faxing) a prior auth request to an insurance company so a patient can get a medication that was prescribed three days ago. The prior auth gets denied by an algorithm. The doctor's staff spends another 45 minutes appealing it manually. The patient doesn't get their medication for a week. This is the status quo that people are defending when they say 'AI in healthcare needs to slow down.' Slow down from what, exactly? From fixing this?
Why Old-School RPA Has Been Lying to Healthcare For a Decade
- ●Traditional RPA tools like UiPath are brittle. They break the second a UI element moves three pixels to the left. Healthcare software updates constantly, and every update is a potential bot failure.
- ●RPA requires structured, predictable workflows. Healthcare is the opposite of that. Every payer has different prior auth forms, different portals, different logic.
- ●Implementation costs for enterprise RPA in healthcare routinely run into the hundreds of thousands of dollars before you see a single automated task complete successfully.
- ●Maintenance overhead is brutal. A 2025 Medium analysis noted that scripted RPA bots demand constant upkeep as interfaces change, which means you're paying an automation engineer indefinitely just to keep your 'automated' process alive.
- ●RPA can't read a PDF, interpret an ambiguous field, or make a judgment call. A computer use AI agent can do all three, in real time, on any screen it can see.
- ●Microsoft itself acknowledged this in April 2025, announcing computer use capabilities in Copilot Studio specifically because traditional UI automation has 'fragility' problems that computer-using AI overcomes.
Administrative costs in U.S. healthcare are projected to hit $2.2 TRILLION by 2035. That's $6,400 per American, per year, spent on forms, faxes, and copy-paste work that AI can automate today.
What a Real Computer Use Agent Actually Does in a Hospital
Here's the thing people get wrong about AI automation in healthcare. They picture a chatbot answering patient questions or an algorithm reading an X-ray. That's not what we're talking about. A computer use agent sits in front of a real desktop or browser, sees the screen exactly like a human does, and executes tasks. It logs into the payer portal. It fills out the prior auth form. It checks the patient record in the EHR. It cross-references formulary data. It submits the request. It monitors for a response and flags the physician if a denial comes back. No API integration required. No custom connector. No six-month IT project. Just an agent that can use a computer, the same way your billing coordinator uses a computer, except it works 24 hours a day and doesn't make transcription errors at 4pm on a Friday. The AMA reported in June 2025 that AI scribes alone saved 15,000 physician hours at a single health system. That's just documentation. Imagine stacking prior auth automation, insurance verification, claims scrubbing, referral coordination, and appointment scheduling on top of that. The math gets very interesting very fast.
The Burnout Crisis Is an Automation Crisis in Disguise
Physician burnout dropped below 50% for the first time in four years in 2024, according to the AMA. Everyone celebrated. Nobody asked why it was still nearly 50% to begin with. The Commonwealth Fund's November 2025 burnout survey across 10 countries was blunt: administrative work is a primary driver, and it's not isolated to the U.S. Nurses are in the same boat. The AACN projects a shortage of over 63,000 full-time RNs by 2030. When you dig into why nurses leave the profession, documentation burden and administrative nonsense show up repeatedly. We are losing trained clinicians to paperwork. Let that sink in. Someone spent eight years becoming a doctor and they're quitting because they can't stand entering data into a form anymore. This isn't a staffing problem you solve with recruiting. It's a workflow problem you solve with automation. A computer use agent doesn't get burned out. It doesn't leave for a competitor. It doesn't call in sick on the day you have 40 prior auths due. The technology exists. The ROI is obvious. The only thing missing is the willingness to actually deploy it.
Why Coasty Is the Tool Healthcare Actually Needs Right Now
I've looked at a lot of automation tools. Most of them are either too brittle (classic RPA), too narrow (single-task AI scribes), or too vague ('agentic AI' that sounds great in a pitch deck and falls apart on a real EHR screen). Coasty is different in a way that actually matters for healthcare. It's a computer use agent that scored 82% on OSWorld, the gold-standard benchmark for real-world computer task performance. That's the highest score of any computer use agent available right now. Not a lab demo. Not a cherry-picked benchmark. The hardest real-world computer use test that exists, and Coasty leads it. What that means practically: it can navigate Epic, Cerner, payer portals, clearinghouse platforms, and any other legacy software your health system is stuck with, without needing an API or a custom integration. It controls real desktops and browsers the way a human would. It supports agent swarms, so you can run parallel workflows simultaneously, which matters enormously when you have hundreds of prior auths and eligibility checks to process every morning. There's a free tier to start, and BYOK support if your IT department has opinions about data handling (they will). The VA is already building AI automation strategies at scale for 2026. Health systems that wait another two years to act are going to be looking at competitors who automated 30% of their administrative overhead and wondering what happened.
Here's my actual opinion, and I'll stand behind it: any health system administrator who looks at $2.2 trillion in projected administrative waste, a physician burnout rate still hovering near 50%, and a nursing shortage in the tens of thousands, and decides the answer is to hire more billing staff, is making a choice that will age very badly. The tools are here. Computer use AI agents can operate every portal, every EHR, every payer interface your organization touches, without breaking when the UI updates, without burning out, and without a six-figure implementation bill. The question isn't whether to automate. It's whether you want to be the health system that figured it out in 2026 or the one that's still explaining to your board in 2028 why administrative costs keep climbing. If you want to see what best-in-class computer use actually looks like in practice, start at coasty.ai. The free tier exists for a reason.