UiPath Had a 10-Year Head Start. A Computer Use Agent Just Lapped It.
Manual data entry costs U.S. companies $28,500 per employee every single year. That stat alone should have killed the argument for doing nothing. But here's the part that should make every CTO sweat: companies paying for UiPath licenses are still hemorrhaging that money, because RPA doesn't actually fix the problem. It just automates the parts that were easy to automate in 2015 and leaves everything else on the desk of some exhausted analyst. A new class of computer use AI agents has arrived, and they don't care about your brittle, hand-coded workflows. They just open the app and get it done. UiPath had a decade to figure this out. They didn't. Let's talk about why that matters.
UiPath Is Not Fine. The Numbers Say So.
Let's start with the uncomfortable truth that UiPath's marketing team would rather you not read. In September 2025, a securities fraud lawsuit was filed alleging that UiPath concealed increasing customer churn throughout 2023 and 2024. When the truth came out, UiPath stock dropped $6.23 per share, a 34% single-day collapse. By March 2026, analysts were publicly downgrading the stock, citing deteriorating ARR growth and falling retention rates. This isn't a blip. This is a company whose core product is losing relevance in real time. And the reason isn't bad management or bad luck. The reason is structural. RPA was always a workaround, not a solution. You hire consultants to map your processes, developers to build the automations, and then a whole team to maintain them when the UI changes and the bot breaks. Again. That's not automation. That's just expensive babysitting.
What RPA Actually Costs You (Nobody Talks About This)
- ●$28,500 lost per employee per year to manual data entry, even at companies that already have RPA deployed, because RPA only covers a fraction of the actual work
- ●56% of employees report burnout from repetitive data tasks, meaning your people are miserable doing work a computer should be doing
- ●Nearly 60% of workers say they could save 6 or more hours per week if repetitive tasks were actually automated, not just partially automated
- ●The average RPA implementation takes 6 to 18 months before a single bot goes live in production, and that's before the first UI change breaks it
- ●Gartner warned in 2025 that over 40% of agentic AI projects will be canceled by 2027, largely because companies are bolting AI onto legacy RPA thinking instead of replacing it
- ●UiPath's own ARR growth and retention metrics are now flagged as deteriorating by multiple Wall Street analysts, which means even the market has noticed the product is stalling
UiPath's stock dropped 34% in a single day after a lawsuit alleged the company hid rising customer churn. Meanwhile, the companies that switched to AI computer use agents didn't notice, because they were already done building the thing UiPath would have taken 18 months to deploy.
The Core Problem With RPA That Nobody Wants to Admit
RPA works by watching a fixed sequence of clicks and keystrokes, recording them, and replaying them. That sounds useful until you realize it means any change to the underlying application breaks the whole thing. New button? Broken. UI refresh? Broken. The vendor updates their web portal? Broken. You're now paying a developer to fix an automation that was supposed to save developer time. This is why RPA has always had a dirty secret: the total cost of ownership is brutal. You need a Center of Excellence, internal developers, external consultants, and a governance team just to keep the bots running. For large enterprises, this can easily run into seven figures annually before you count the licenses. And what do you get for that? Automations that cover maybe 20% of the actual repetitive work in your organization, because the other 80% involves judgment calls, unstructured data, or apps that change too often to script reliably. Computer use AI agents don't work this way. They see the screen the way a human does. They read what's there, decide what to do, and adapt when things change. No hardcoded selectors. No brittle XPath queries. No 'element not found' crashes at 2am.
AI Computer Use Agents vs. RPA: This Isn't Even a Fair Fight Anymore
The benchmark that matters here is OSWorld, the gold standard for measuring how well a computer-using AI can handle real-world desktop tasks. It covers actual applications, actual operating systems, and tasks that require genuine decision-making, not just clicking a button in the same spot every time. Human performance on OSWorld sits around 72%. Most AI agents are clustered in the 30s and 40s. Then there's Coasty, sitting at 82%, which is not only above every competitor but above the human baseline. That number isn't a marketing claim. It's a published benchmark result. Compare that to what RPA gives you: a deterministic script that scores 100% on the exact task it was trained on and 0% on anything that deviates by a pixel. The argument for RPA over AI computer use was always 'reliability.' But a system that breaks the moment the world changes isn't reliable. It's fragile. Coasty controls real desktops, real browsers, and real terminals. It doesn't call an API and pretend that counts as automation. It actually uses the computer, the way a person would, which means it works with every application that exists, including the ancient internal tool your IT team built in 2009 that has no API and never will.
Why Coasty Exists
I'm not going to pretend I don't have a favorite here. I've watched teams spend months configuring UiPath workflows that a computer use agent could handle in an afternoon. The math is not subtle. Coasty was built specifically to be the answer to everything RPA promised and couldn't deliver. It runs on a desktop app or cloud VMs, supports agent swarms for parallel execution when you need to run the same task across hundreds of accounts or datasets simultaneously, and has a free tier so you can actually test it before you commit. BYOK is supported if you already have API keys you want to use. The 82% OSWorld score isn't a coincidence. It's the result of building a computer use agent that actually understands context, adapts to change, and handles the messy, unstructured reality of real work. Not a scripted bot pretending to be smart. An AI that uses a computer the way you do, just faster and without complaining. If you're currently paying for UiPath licenses, I'd genuinely encourage you to run a side-by-side test. Pick a workflow your RPA team has been trying to build for three months. Give the same task to a computer use agent. See what happens.
RPA was a reasonable idea in 2015. It gave enterprises a way to automate structured, repetitive tasks without rewriting their legacy systems. That was genuinely useful for its time. But it's 2025, and the companies still betting their automation strategy on click-recording bots are the same ones watching their competitors move faster, hire less, and ship more. UiPath isn't going to save you. Their own customers are leaving. Their stock is down. Their growth is stalling. And they're now trying to rebrand as an 'agentic automation' platform, which is the corporate equivalent of a Blockbuster putting a 'Now Streaming' sign in the window. The real computer use AI agents are already here, already benchmarked, and already outperforming humans on real-world tasks. The only question is whether your organization is going to figure that out now or in two years when a competitor already did. Start at coasty.ai. The free tier is right there.