Healthcare Is Drowning in $400B of Admin Waste and Your AI Agent Still Can't Log Into Epic
American physicians now spend roughly two hours doing paperwork and clicking through EHR screens for every one hour they spend with an actual patient. Two to one. And we have the audacity to wonder why doctors are burning out and why a routine specialist visit costs $400. The healthcare industry is sitting on what researchers estimate is $400 billion in annual administrative waste, and the solutions most health systems have deployed, chatbots, rigid RPA scripts, and 'AI-powered' prior auth tools that are really just if-then trees in a trench coat, are embarrassingly inadequate. It's 2026. The problem isn't a lack of AI. It's a lack of AI that can actually do the work.
The $400 Billion Problem Nobody Wants to Say Out Loud
Oliver Wyman put a number on it in early 2025: $450 billion in administrative savings is on the table over the next decade if healthcare gets serious about automation. That's not a rounding error. That's roughly the GDP of Norway, just sitting in fax machines and prior authorization queues. The AMA has been screaming about prior authorization for years. A 2025 study in medRxiv pegged prior auth alone as responsible for a significant chunk of the $400 billion in annual billing and administrative waste. Physicians spend hours every week just waiting for insurers to approve procedures that they, the people with the medical degrees, already decided were necessary. And here's the part that should make you furious: while doctors are buried in this paperwork, insurance companies are using their own AI to deny claims faster. Class-action lawsuits against UnitedHealthcare alleged in 2025 that its AI algorithm was rejecting claims in seconds, with a 90% denial rate on certain post-acute care claims. So AI is already in healthcare. It's just working for the wrong side right now.
Why RPA Failed Healthcare (And Everyone Quietly Knows It)
- ●Traditional RPA tools like UiPath build automations that depend on pixel-perfect UI consistency. Change a button label in Epic or update your insurance portal's layout and the bot breaks. Completely. UiPath even launched a 'Healing Agent' in 2025 specifically to address how often their automations fail when UIs change. That's not a feature. That's an admission.
- ●Healthcare software is notoriously fragmented. Epic, Cerner, Meditech, dozens of payer portals, state Medicaid systems, lab interfaces. RPA can't navigate this without a full-time team of developers maintaining scripts. Most hospital IT departments don't have that bandwidth.
- ●The average RPA implementation in healthcare takes 6 to 18 months to deploy meaningfully. By the time it's live, the portal it was built for has already had three UI updates.
- ●RPA has zero judgment. It can fill out a form. It cannot read an ambiguous denial letter, figure out which appeal pathway applies, pull the relevant clinical notes, and draft a response. That requires something that actually understands context.
- ●A 2025 ResearchGate analysis confirmed that traditional RPA platforms 'excel in automating structured, rule-based tasks' but face serious limitations with anything requiring cognitive flexibility. Healthcare is almost never purely rule-based.
Physicians spend 2 hours on administrative tasks for every 1 hour of patient care. Insurers use AI to deny claims in seconds. And the RPA bots hospitals paid millions for break every time someone updates a dropdown menu. This is the state of healthcare automation in 2026.
The 'Pajama Time' Problem Is Getting Worse, Not Better
The AMA coined the term 'pajama time' to describe the hours physicians spend finishing documentation after clinic hours, at home, after their kids are in bed, on weekends. As of 2025, despite all the ambient AI scribes and dictation tools, pajama time has barely moved. Burnout rates are down slightly, which is good, but the documentation treadmill is still running at full speed. Why? Because ambient scribes solve one narrow slice of the problem. They transcribe the visit. They don't submit the prior auth. They don't reconcile the insurance denial. They don't navigate the payer portal to check claim status. They don't pull the right ICD-10 code from three different systems and cross-reference it against the formulary. The full administrative loop, from patient visit to clean claim to paid reimbursement, involves dozens of discrete computer tasks across multiple systems. A scribe that just listens in the room is like fixing one leak in a boat with thirty holes. The real opportunity, the one that could actually give doctors their evenings back, is a computer use agent that handles the entire downstream workflow.
What 'Computer Use' Actually Means (And Why It's Different)
Here's the distinction that matters. Most healthcare AI is either a chatbot that answers questions, a predictive model that flags risks, or an RPA script that clicks through one specific workflow until it breaks. Computer use AI is different. A computer use agent sees the screen the same way a human does, reads what's on it, decides what to do next, and executes it. It can open Epic, navigate to a patient record, pull clinical notes, switch to a payer portal, fill out a prior auth form using the information it just read, submit it, and log the outcome back into the EHR. Without a single API integration. Without custom scripts for each system. Without breaking when the portal updates. This is why the OSWorld benchmark matters. It's the industry standard for measuring how well an AI agent can actually operate real computer environments, not toy demos, real software. Coasty scores 82% on OSWorld, which is the highest of any computer use agent available today. That gap between 82% and whatever the next competitor scores isn't a marketing number. It's the difference between an agent that can handle the messy, real-world complexity of healthcare workflows and one that gives up when the UI looks slightly different than expected.
How Coasty Is Built for This Mess
I'm not going to pretend every healthcare problem gets solved by pointing at coasty.ai and walking away. But I will tell you what makes a computer use agent actually useful in a healthcare context, and why Coasty specifically is worth paying attention to. Coasty controls real desktops, real browsers, and real terminals. It doesn't need API access to your EHR vendor, which is good because Epic's API access costs a fortune and takes months to negotiate. It can operate inside the same browser session a human would use, navigate the same payer portals your billing staff logs into every morning, and execute tasks in parallel using agent swarms. That last part is huge. Prior auth queues don't clear one at a time. A hospital system might have hundreds of pending authorizations on any given day. Coasty can spin up parallel agents to work through that queue simultaneously, not sequentially. For teams that want to test before committing, there's a free tier and BYOK support, so you're not locked into someone else's infrastructure or pricing model from day one. The healthcare orgs that are going to win the next five years aren't the ones that bought the most expensive EHR add-on. They're the ones that deploy a real computer use agent to handle the administrative loop that's been eating their staff alive.
Here's my take, and I'll be direct about it. The healthcare industry has spent decades tolerating an administrative catastrophe because the people suffering most from it, physicians, nurses, billing staff, were too busy drowning to fight back. The tools that were supposed to fix it, legacy RPA, bolt-on chatbots, ambient scribes that only solve one step, have been good enough to sell but not good enough to actually solve the problem. That's changing now because computer use AI is fundamentally different from anything that came before it. It doesn't need a custom integration for every system. It doesn't break when a UI changes. It can handle the full workflow, not just the easy part. $400 billion in administrative waste doesn't disappear on its own. Somebody has to actually automate the work. If you're in healthcare operations, revenue cycle, or clinical leadership and you're still running manual workflows or babysitting RPA bots that need constant maintenance, go look at what a real computer use agent can do. Start at coasty.ai. The free tier exists for exactly this reason.