UiPath Is Losing to AI Agents and the Numbers Are Brutal

What RPA Actually Promised vs. What It Delivered

The pitch for RPA in 2017 was irresistible. Stop paying humans to copy data between systems. Build a bot. Save money. UiPath rode that wave harder than anyone, growing from a $1 billion valuation to a $35 billion IPO in about four years. Enterprise IT departments bought in. Consultants got rich building bot factories. Everyone declared victory. Then the maintenance bills arrived. RPA bots are extraordinarily brittle. Change a button label in your CRM. Move a field two pixels to the left. Update a web app. The bot breaks. Someone has to fix it. That someone is usually a specialist developer billing $150 an hour. Analysts have estimated that for every dollar saved by RPA, companies spend 30 to 50 cents maintaining the bots. Gartner's own data showed that RPA projects have historically faced failure rates north of 30% before they even reach production. You built a robot workforce. You just also hired a robot maintenance crew to keep it alive. The total cost of ownership on a mature UiPath deployment at a mid-sized enterprise is not a line item. It's a department.

The Specific Ways UiPath Is Cracking Right Now

• ●Stock down roughly 50% in calendar year 2024, one of the worst performances among major enterprise software companies that year
• ●Revenue growth decelerated through fiscal 2023, 2024, and 2025, landing at roughly 9% in Q3 FY2025, a brutal slowdown for a company once growing at 50%+ annually
• ●Active securities fraud class action lawsuit filed in 2024, with investors alleging the company artificially inflated its stock price
• ●The company acquired Peak AI in early 2025 in a desperate pivot toward agentic automation, and the market punished them for it with another sharp stock drop on the announcement
• ●UiPath's own community forums in mid-2025 show users hitting blank pages and broken features in their new Agent Builder product, which is supposed to be the future
• ●56% of employees still report burnout from repetitive data tasks according to a 2025 survey, meaning RPA did not actually solve the human cost problem it promised to eliminate

What AI Agents Actually Do Differently

Here's the core technical difference and it matters a lot. RPA bots work by following a script. You record a sequence of clicks and keystrokes, the bot replays them, and if anything changes in the UI, the whole thing falls apart. It's a macro from 2003 wearing a suit. AI computer use agents work by seeing. They look at a screen the same way a human does, understand what's on it, decide what to do next, and act. No brittle selectors. No recorded coordinates. No XML workflow diagrams that only one person on your team understands. A computer use agent can open a browser, log into a system it's never seen before, find the data it needs, and write it somewhere else, because it can read the screen and reason about what it's looking at. That's not a minor improvement over RPA. That's a different category of tool entirely. The OSWorld benchmark, which is the closest thing the industry has to an objective test of AI computer use capability, is where you can actually see the gap between generations. It tests agents on real desktop tasks across real operating systems. The top performers aren't legacy automation vendors. They're purpose-built AI computer use agents. And the scores are not close.

Why Big Enterprises Are Quietly Panicking

Here's what's actually happening inside large companies right now. The IT team that spent three years and several million dollars building a UiPath bot farm is watching a new hire spin up an AI computer use agent in an afternoon to do something the bot farm can't do at all. Not faster. Can't do at all. Things like reading a PDF that wasn't templated in advance. Navigating a vendor portal that updated its UI last month. Pulling data from a legacy system that was never given an API. RPA needs APIs or perfectly stable UIs to function. The real world has neither. AI agents just handle it. A McKinsey report from early 2025 noted that the biggest barrier to scaling AI in the workplace isn't employees, who are ready, but leaders who aren't steering fast enough. That's a polite way of saying a lot of executives are frozen because they already bet on RPA and they don't want to admit the bet is going bad. The ones who move anyway are going to lap the ones who wait.

Why Coasty Exists

I'm not going to pretend I don't have a dog in this fight. I think Coasty is the best computer use agent available right now, and I can back that up. Coasty sits at 82% on OSWorld. That's the benchmark that actually matters for real-world computer use tasks, and no competitor is close. What that score means in practice is that Coasty can handle the messy, unstructured, never-the-same-twice work that RPA bots have always choked on. It controls real desktops, real browsers, and real terminals. Not API wrappers. Not simulated environments. The actual screen, the actual keyboard, the actual cursor. You can run it as a desktop app on your own machine, spin up cloud VMs for isolated workloads, or deploy agent swarms to run tasks in parallel across dozens of instances simultaneously. That last part is where it gets genuinely wild. The thing that made RPA expensive to scale was that each bot needed its own license, its own maintenance, its own babysitter. With Coasty's swarm execution you can run 50 parallel agents on 50 different tasks without 50 different configuration nightmares. There's a free tier if you want to see it work before committing. BYOK is supported if you want to use your own model keys. It's built for people who are done explaining to their CFO why the automation budget is bigger than the savings.