UiPath Is Losing to AI Computer Use Agents and Everyone Can See It

The RPA Promise Was Always a Little Dishonest

When UiPath went public in 2021, it raised $1.34 billion and the automation world lost its mind. Finally, a way to automate anything without touching a single line of application code. You just record clicks, set up rules, and let the bots run. Simple, right? Except it was never simple. RPA bots are brittle by design. They work by targeting fixed screen coordinates, specific pixel locations, and hardcoded UI element paths. The moment a developer at Salesforce tweaks a button's CSS, or your IT team pushes a Windows update, or a vendor changes their web portal layout, your entire bot breaks. Industry data now puts RPA project failure rates at 30 to 50 percent. Not failure as in 'underperformed.' Failure as in 'abandoned entirely.' And that's before you count the bots that technically run but require so much human babysitting that they save roughly zero hours of actual work. Gartner dropped a bomb in June 2025 predicting that over 40 percent of agentic AI projects will be canceled by 2027, but here's the part everyone skipped: they specifically called out companies that are trying to bolt AI onto legacy RPA infrastructure without real agentic capabilities. That's UiPath's entire current strategy.

What UiPath's Own Legal Filings Tell You About the Business

In 2024, UiPath got hit with a securities fraud class action lawsuit. The complaint, filed in the Southern District of New York, alleged that the company had been internally discussing the failure of its Flex platform migration while telling investors everything was fine. The stock had already cratered from its IPO highs. The case was eventually dismissed, but the details in the filing are worth reading if you want to understand the pressure UiPath is under. The company is now scrambling to rebrand itself as an 'agentic automation platform.' They launched Agent Builder. They're using the word 'agentic' in every press release. But when your core product is a coordinate-based bot runner and you're trying to duct-tape LLMs onto it, you don't get to call yourself an AI company. You're a legacy tool with a new coat of paint. Meanwhile, the maintenance nightmare isn't getting better. One LinkedIn analysis from Blueprint Software Systems found that companies running 100-plus UiPath bots can cut maintenance costs by 65 percent just by using better monitoring tools. Think about that. The maintenance overhead is so brutal that a 65 percent reduction is considered a win, and you still need a separate product to achieve it.

Why Computer Use Agents Are a Completely Different Category

Here's what people get wrong about the UiPath vs AI agents debate. They treat it like a feature comparison. 'UiPath has X, the AI agent has Y.' That's the wrong frame. These are architecturally different approaches to the same problem. RPA bots are rule-following scripts that pretend to be humans. Computer use AI agents actually reason about what they see on screen, the same way a human does. A computer use agent looks at a screenshot, understands the context, figures out what action makes sense, and executes it. If the UI changes, the agent adapts. If an unexpected popup appears, the agent handles it. If the task requires judgment, like deciding whether a flagged invoice looks legitimate before approving it, the agent can make that call instead of throwing an exception and stopping cold. This is not a small improvement. It's a fundamentally different relationship between software and the tasks humans do on computers. The OSWorld benchmark, which is the industry standard for measuring how well AI agents perform real computer tasks, shows exactly how fast this category is moving. Anthropic's models have been pushing scores upward with every release, and purpose-built computer use agents are scoring even higher.

The Hidden Costs Nobody Talks About in the RPA Sales Pitch

• ●Initial UiPath licensing starts around $8,000 to $15,000 per bot per year, and that's before implementation costs which routinely run $50,000 to $200,000 for complex deployments.
• ●Maintenance typically consumes 25 to 40 percent of the original development cost annually, meaning a $100,000 bot project costs you $25,000 to $40,000 every single year just to keep running.
• ●Every major UI update from a vendor, a browser, or your own IT team can break multiple bots simultaneously, creating cascading failures across your automation stack.
• ●RPA requires dedicated Center of Excellence teams, sometimes 5 to 15 people, just to manage a mid-size bot portfolio. That's not automation. That's a new department.
• ●30 to 50 percent of RPA projects are abandoned before they deliver ROI, meaning a significant portion of your investment simply disappears.
• ●UiPath's own community forums are full of threads about bots getting stuck on Chrome and Edge updates, broken selectors, and license confusion, in 2025, years after the product was supposed to be mature.
• ●Computer use agents require no selector maintenance, no coordinate mapping, and no brittle UI path definitions. The agent sees the screen. It figures it out.

Why Coasty Exists and Why the Benchmark Score Actually Matters

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 point to why. On OSWorld, the standard benchmark for AI computer use performance, Coasty scores 82 percent. That's not a marketing number. OSWorld tests real, open-ended computer tasks across browsers, desktops, and terminals. It's the closest thing the industry has to an objective measure of how good a computer-using AI actually is. No competitor is close to that number right now. Not Anthropic's computer use API, not OpenAI's Operator, not anything UiPath is shipping. What Coasty actually does is control real desktops, real browsers, and real terminals. Not API wrappers. Not simulated environments. Actual computer use, the same way a human contractor would sit down and do the work. You can run it on a desktop app, spin up cloud VMs, or run agent swarms for parallel execution when you need to process things at scale. There's a free tier if you want to try it without a procurement process, and BYOK support if you're particular about which model is under the hood. The reason this matters in the UiPath conversation is simple. Everything UiPath makes you build, maintain, and pay a team to babysit, a computer use agent handles without any of that overhead. The ROI math isn't close.