Healthcare Is Drowning in $1 Trillion of Busywork. A Computer Use AI Agent Can Fix It.
Here's a number that should make every hospital CFO physically ill: U.S. healthcare wastes somewhere between $760 billion and $935 billion every single year. Not on bad drugs or failed surgeries. On administration. On prior authorization. On copying data from one system into another. On a physician sitting at a desk at 11pm, not because a patient needs them, but because Epic needs another click. The American Medical Association confirmed that prior authorization overhead alone runs $68,000 per physician per year. Oliver Wyman says we could cut $450 billion in admin costs over the next decade if we actually tried. We are not trying. Not seriously. Most health systems are still running RPA bots from 2019 and calling it 'digital transformation.' That era is over. The question in 2026 isn't whether AI can automate healthcare admin. It's whether your organization is going to be the one that actually does it right, or the one that buys another legacy tool, watches it break on a UI update, and explains to the board why nothing changed.
The Real Scandal: Physicians Are Glorified Data Entry Clerks
The AMA published data in 2025 showing that EHR and administrative tasks follow doctors home. Shorter workweeks on paper, but the inbox never closes. A landmark study found physicians spend nearly twice as much time on EHR and desk work as they do with actual patients. Another one to two hours of personal time each night gets eaten by data entry demands. Think about what that means. You went to medical school for a decade. You're carrying $300,000 in student loans. And you're spending your Tuesday night copy-pasting referral codes into a portal that hasn't been redesigned since the Obama administration. This isn't a technology problem. It's a priorities problem. The technology to fix this exists right now. A real computer use agent, one that can actually see a screen, navigate a portal, fill out a form, and submit a prior auth request without a human touching it, has existed for over a year. The healthcare industry just hasn't caught up to what's possible.
Why RPA Is the Wrong Tool and Everyone Knows It
- ●Traditional RPA bots are brittle by design. They follow pixel-perfect scripts. One UI update from Epic or Availity and the whole workflow breaks, often silently, while claims pile up.
- ●UiPath and its competitors require dedicated bot maintenance teams. You're not eliminating labor costs, you're shifting them to a more expensive, more specialized workforce.
- ●RPA can't handle exceptions. Healthcare is almost entirely exceptions. A claim with a missing field, a portal that loads differently on a slow connection, a payer that changed their form layout last Thursday. RPA taps out. A human gets pulled back in.
- ●Implementation timelines for enterprise RPA in healthcare average 6 to 18 months before a single workflow goes live. That's not automation. That's a construction project.
- ●A 2025 report from Itransition confirmed that legacy system integration remains the top challenge for RPA in healthcare, the exact problem that AI computer use agents were built to solve by operating at the UI layer instead of the API layer.
- ●The total addressable waste is $450 billion over 10 years per Oliver Wyman. RPA has been in healthcare for over a decade. The waste is still there. Draw your own conclusions.
Prior authorization costs $68,000 per physician per year. There are over 1 million active physicians in the U.S. That's $68 billion a year spent on a process that, in most cases, a well-trained computer use agent could handle end-to-end without a single human keystroke.
What 'Computer Use' Actually Means and Why It Changes Everything
Most people hear 'AI automation' and picture a chatbot answering FAQs or an API stitching two databases together. That's not computer use. A computer use agent actually sees your screen the same way you do. It moves the mouse. It reads the portal. It fills out the form. It handles the popup that wasn't there yesterday. It navigates the payer website that has no API and never will. This is the fundamental difference between every AI wrapper your vendor is trying to sell you and a genuine computer-using AI. The prior auth portal for Aetna doesn't have a clean API. Neither does the state Medicaid portal. Neither does half the billing software running in community hospitals right now. Those systems were built in the 2000s and they're not getting rebuilt. API-first automation hits a wall immediately. Computer use AI doesn't. It works on anything with a screen, which means it works on the actual healthcare IT environment that exists in the real world, not the clean, modern, API-friendly environment that automation vendors pretend you have.
The Specific Healthcare Tasks That AI Computer Use Agents Are Killing Right Now
Let's get concrete, because 'AI will transform healthcare' is the most overused sentence in health tech and it means nothing without specifics. Prior authorization submissions: the agent logs into the payer portal, reads the clinical criteria, pulls the relevant patient data from the EHR, fills the form, attaches documentation, and submits. No human involved. Claims scrubbing and resubmission: denied claims get reviewed, the error gets identified, the correction gets made, and the claim goes back out the same day instead of sitting in a queue for two weeks. Eligibility verification: the agent checks coverage across multiple payers before the appointment, not after, so your front desk stops having that awkward conversation with patients about surprise bills. Referral coordination: the agent navigates the specialist portal, books the appointment, sends the records, and updates the referring physician's EHR. All of it. The a16z team published research in early 2025 specifically calling out healthcare billing and claims management as one of the highest-ROI targets for AI agents that can operate at the UI level. This isn't speculation. It's already happening at the organizations that moved early.
Why Coasty Exists (and Why It's the Right Tool for Healthcare Admin)
I've looked at the field. Anthropic's computer use is impressive research but it's not a production tool you deploy in a health system. OpenAI's Operator is still finding its footing. Neither is purpose-built for the kind of high-volume, parallel, mission-critical workflow that healthcare operations actually require. Coasty is different. It's the top-ranked computer use agent on OSWorld, the standard benchmark for real-world computer task completion, at 82%. Nobody else is close. That gap matters in healthcare because partial automation is often worse than no automation. A bot that completes 60% of prior auth submissions correctly and silently fails on the rest isn't saving you money. It's creating liability. Coasty controls real desktops, real browsers, and real terminals. It doesn't need an API. It works on the actual payer portals and legacy EHR interfaces your team uses every day. The agent swarm feature means you can run hundreds of these tasks in parallel, not sequentially, which is how you actually make a dent in a backlog of 10,000 pending prior authorizations. There's a free tier if you want to test it before committing. BYOK is supported if your compliance team has opinions about where your keys live, and in healthcare, they always do. The point isn't that Coasty is a magic button. It's that it's the most capable computer use AI available right now, and healthcare admin is exactly the kind of messy, screen-based, exception-heavy work it was built for.
Healthcare administration is a $1 trillion bonfire. Physicians are burning out not because medicine is hard but because the paperwork is relentless and the tools to fix it have been bad for a long time. That excuse is gone in 2026. Computer use agents are real, they're production-ready, and the ROI math on prior auth alone is so obvious it's almost embarrassing. The organizations that move now will cut admin costs, reduce denials, and give their clinical staff something they haven't had in years: time. The ones that wait will spend another 18 months evaluating vendors, piloting RPA bots that break on the first UI update, and explaining to their boards why they're still paying 40 people to manually submit claims. Don't be that organization. Go see what a real computer use agent can actually do at coasty.ai. The free tier exists. There's no reason to wait.