Your Insurance Company's AI Denies Claims in 1.2 Seconds. Here's How a Real Computer Use Agent Fixes the Whole Broken System.
Cigna's AI system spent an average of 1.2 seconds reviewing each of the 300,000 medical claims it denied. One point two seconds. A human adjuster reads a claim summary faster than that. So when people ask me whether AI is ready to automate insurance claims, I want to flip the question: the industry is already using AI for claims, and it's already a disaster. The question isn't whether to automate. It's whether to do it with tools that actually work, or tools that just create a faster way to screw people over. Because right now, most insurers are doing the latter, and they're about to get buried in lawsuits for it.
The 1.2-Second Scandal Nobody Talks About Enough
Let's sit with that number for a second. A class-action lawsuit against Cigna alleged that its PXDX algorithm auto-denied 300,000 claims while spending 1.2 seconds on each one. UnitedHealth is facing its own class-action over AI-driven Medicare Advantage denials. The Guardian reported in January 2025 that class-action lawsuits allege these algorithms turn down claims in seconds, with critics saying critics saying critics saying reform is needed. The American Medical Association surveyed physicians in early 2025 and found that many fear health insurers' use of unregulated AI is inflating denial rates. This isn't a fringe concern. This is the mainstream conversation in healthcare right now. And here's the infuriating part: this is what happens when you use AI as a cost-cutting blunt instrument instead of a genuine automation tool. These companies didn't deploy AI to make claims processing smarter. They deployed it to make denial faster. That's a governance failure dressed up as a technology decision. The technology itself, meaning real computer use AI that navigates actual systems, reads actual documents, and follows actual logic, is capable of something completely different.
The Manual Side Is Just as Broken (Just Slower)
- ●Claims processing time drops by 55 to 75% when genuine AI automation is applied, according to industry data compiled by Datagrid in late 2025. That's not a rounding error. That's half your team's week.
- ●Manual claims processing costs the industry billions annually in labor, rework, and error correction. A single manual claim can cost $15 to $25 to process end-to-end when you factor in adjuster time, system toggling, and follow-up.
- ●Generali, one of Europe's largest insurers, reported a 96% employee satisfaction rate after deploying AI-assisted claims workflows, plus a measurable reduction in cycle time. So the humans who remain aren't suffering. They're doing actual judgment work.
- ●Workers' compensation claims adjusters regularly carry caseloads so high that burnout is the norm, not the exception. Reddit threads from adjusters in 2024 read like dispatches from a war zone. One person wrote they were 'stressed out all the time' after 18 months on the job.
- ●The insurance industry is staring down a talent shortage. Swiss Re flagged it in mid-2025: aging workforce, rapid tech change, and not enough people entering the pipeline. Automation isn't optional anymore. It's survival.
"Claims processing time has been reduced by 55-75% through AI automation, with routine claims processing now taking minutes instead of days." The technology exists. The question is whether your team is using it or still manually copy-pasting between portals.
Why Traditional RPA and Basic AI Both Fail at Claims
Here's where I'll probably make some vendors angry. Legacy RPA tools like UiPath and Automation Anywhere have been sold to insurers for years as the fix. And they work, right up until the moment a portal changes its UI, a PDF comes in with a slightly different layout, or an adjuster needs to make a judgment call mid-workflow. Then the bot breaks, and someone has to babysit it. That's not automation. That's a fragile script wearing a tie. Basic AI isn't much better on its own. An LLM that reads a claim document and outputs a recommendation is useful, but it can't actually log into your legacy claims management system, pull the policy details, cross-reference the medical codes, flag the anomalies, and update the record. It can tell you what to do. It can't do it. That gap, between knowing and doing, is exactly where most insurance automation projects die. The teams that have cracked this are the ones using computer use AI, meaning agents that operate a real desktop or browser the way a human would, seeing the screen, clicking through workflows, handling the unexpected. That's a fundamentally different category of tool.
What Real Computer Use AI Actually Does in a Claims Workflow
Let me be concrete, because 'AI automation' has become meaningless marketing noise. A genuine computer use agent handling insurance claims can open a FNOL submission from an email attachment, read the intake form, log into the claims management system, create a new claim record, pull the corresponding policy from the policy admin system, check coverage limits, flag any exclusions, route the claim to the right adjuster queue based on claim type and complexity, and send the claimant an acknowledgment, all without a human touching it. When something unusual shows up, a coverage question, a duplicate claim, a suspicious pattern, it escalates with context already gathered. The adjuster gets a pre-populated file, not a blank intake form. That's the difference between automation that creates leverage and automation that just creates a new kind of mess. The computer-using AI doesn't need an API integration for every legacy system. It works with whatever is on the screen, the same way a new employee would. Which means it can handle the ancient policy admin systems that insurers have been too scared to replace for 20 years.
Why Coasty Is the Computer Use Agent Insurance Teams Should Actually Be Testing
I'm not going to pretend I'm neutral here. I've watched a lot of computer use agents fumble through multi-step workflows, and the performance gap is real. Coasty sits at 82% on OSWorld, which is the standard benchmark for AI computer use tasks in real desktop environments. No other agent is close to that right now. That matters for insurance because claims workflows are not clean. They involve legacy portals, PDFs that weren't designed for machines, multi-system lookups, and conditional logic that changes based on claim type. You need an agent that can actually handle that complexity without falling over every third task. Coasty runs on real desktops and cloud VMs, supports agent swarms for parallel claim processing, and doesn't require you to rip out your existing tech stack. There's a free tier if you want to test it before committing, and BYOK support if your security team has opinions about API keys. The insurers I'd bet on in the next three years are the ones building computer use AI into their claims operations now, not the ones waiting for a perfect API ecosystem that will never exist. If you want to see what best-in-class computer use looks like applied to real workflows, coasty.ai is where I'd start.
The insurance industry has two AI problems right now. The first is bad actors using crude algorithms to deny legitimate claims faster than any human could review them. The second is good-faith teams trying to automate the right way but getting stuck with brittle RPA bots and LLMs that can advise but can't act. Both problems have the same root cause: using the wrong tool for the job. Real computer use AI, the kind that operates actual interfaces, handles real-world messiness, and executes multi-step workflows without constant hand-holding, is a different category entirely. The 1.2-second denial scandal should make every insurer uncomfortable. But the response shouldn't be to slow down automation. It should be to do it properly. Faster denials aren't the goal. Faster, more accurate, more defensible claims decisions are. That's what the right computer use agent actually delivers. Stop patching broken workflows with worse tools. Go to coasty.ai and see what 82% on OSWorld looks like when it's doing your claims queue.