UiPath Is Bleeding Out and AI Computer Use Agents Are Why

What RPA Actually Is (And Why That's the Problem)

Robotic Process Automation sounds impressive. It isn't. At its core, RPA is a macro recorder with a business suit on. You hire consultants, spend weeks mapping every pixel of a UI, write brittle scripts that break the moment someone changes a button color, and then call it 'digital transformation.' Ernst and Young found that RPA bots fail 30 to 50 percent of the time when the underlying software they're scraping gets updated. SAP pushes a patch? Your bot is dead. Salesforce redesigns a modal? Your bot is dead. Your IT team updates Chrome? You guessed it. Dead. The dirty secret of the RPA industry is that these 'automations' require constant human maintenance. You didn't eliminate the manual work. You just moved it from the business team to the IT team and tripled the cost.

The Numbers That Should Make Every CFO Furious

• ●Office workers spend over 50% of their work week on repetitive manual tasks that could be automated, according to ProcessMaker research
• ●UK workers waste an average of 15 hours per week on repetitive admin tasks alone, per Ricoh Europe data
• ●51% of employees reported burnout directly tied to manual, repetitive work in Grant Thornton's 2024 survey
• ●RPA implementations carry a 30-50% failure rate when target software updates, per Ernst and Young 2023
• ●UiPath stock: $90 all-time high, now trading near $15. That's an 83% collapse. The market has already voted.
• ●UiPath's CEO Rob Enslin resigned abruptly in May 2024, the same day the company missed guidance and the stock dropped 34% in a single session
• ●A securities fraud complaint filed in September 2025 alleged UiPath executives knew internally that AI was cannibalizing their pipeline but didn't disclose it

AI Computer Use Agents Don't Care What Your UI Looks Like

Here's the fundamental difference between RPA and a real computer use agent. RPA bots are told exactly where to click, pixel by pixel, step by step. A computer-using AI actually sees the screen, understands what it's looking at, and figures out how to complete the task. It reads context. It adapts. If the button moved three pixels to the left after an update, the AI agent doesn't care. It sees 'Submit' and clicks it. This is not a subtle improvement. It's a completely different category of technology. The arXiv paper 'Are LLM Agents the New RPA?' published in 2025 directly compared modern computer use agents against traditional RPA on speed, reliability, and development effort. The results were not kind to RPA. And on OSWorld, the gold standard benchmark for real-world computer task completion, the best AI computer use agents are now completing tasks that would have required weeks of RPA scripting, in seconds, with zero upfront configuration. Meanwhile, OpenAI's Operator is still labeled a 'research preview' and Anthropic's computer use is impressive but inconsistent. The gap between the best and the rest is widening fast.

The 'UiPath Is Pivoting to AI' Argument Is Cope

Every few months, someone posts a LinkedIn article defending UiPath with the argument that 'RPA and AI agents are complementary, not competitive.' This is the kind of thinking that kept Blockbuster executives comfortable in 2008. Yes, UiPath has announced 'agentic automation' features. Yes, they're trying to bolt AI onto their existing RPA infrastructure. But you can't retrofit your way out of a foundational architecture problem. RPA was built on the assumption that software UIs are static and predictable. AI agents were built on the assumption that nothing is. Those are opposite bets. The Reddit thread titled 'Not popular opinion: UiPath and similar tools are dead' from September 2025 has hundreds of upvotes from actual RPA practitioners saying the same thing. These aren't AI hype bros. These are the people who built UiPath automations for a living, and they're telling you the model is broken. When the practitioners abandon the tool, the tool is done.

Why Coasty Exists

I'm not going to pretend I don't have a dog in this fight. I use Coasty, and I think it's the best computer use agent available right now, and I can back that up with something concrete: 82% on OSWorld. That's the highest score of any computer use agent on the benchmark that actually matters. For context, Claude Sonnet 4.5 scores 61.4% on OSWorld and it's considered a major leap forward. Coasty is at 82%. Nobody else is close. But the score isn't even the main point. The main point is what it actually does. Coasty controls real desktops, real browsers, and real terminals. Not API wrappers. Not simulated environments. Actual computer use, the way a human would do it, but faster and without forgetting steps. You can run agent swarms for parallel execution, which means tasks that would take a human team a full day can run simultaneously across cloud VMs. There's a free tier. There's BYOK support if you want to bring your own model keys. And setup takes minutes, not the weeks-long consultant engagement that UiPath requires before you can automate a single form. If you've been burned by RPA before, I get the skepticism. But a computer-using AI that actually understands what it's looking at is a different animal entirely.