Your RPA center already has dozens of bots. They run nightly batch imports, validate invoices, and push data into ERP systems. But dozens of tickets still land on your desk every week. A UI refresh in one line-of-business app broke the bot that updates supplier master data. Another bot stopped because a form field name changed. Even the process that should be fully documented is still going through a human because the flowchart doesn’t match reality. The problem is not that automation is hard. It is that the bots are brittle and the process documentation is disconnected from them. The fix is not to build more complex flows. It is to treat the SOP itself as the source of truth and use agents that can read and execute it directly.
Why RPA breaks here
Traditional RPA tools like UiPath, Automation Anywhere, and Blue Prism rely on selectors, XPath, or object IDs to find elements on a screen. A bot says, click the Submit button with id=SubmitButton. If the developer changes the button text from Submit to Confirm, or if a vendor updates the UI, the bot fails. Across large enterprises, teams estimate that 30 to 50 percent of RPA maintenance hours are spent on rebuilds after a minor UI change. That is not a one-time fix. It is a cycle that repeats every few months. The cost compounds when you have bots across dozens of applications. Each change triggers a new ticket, a new testing cycle, and a new risk of regression. The business is paying for a bot that is only as durable as the last UI release.
What changes with computer use agents
- Agents see the screen like a human: they read the current state and act on it, not on a brittle ID.
- No brittle selectors to maintain: an agent uses natural language to describe what it wants to do, so UI changes do not break the automation.
- Recover from exceptions instead of halting: when an error occurs, the agent reads the result, decides the next step, and continues.
- Follow the SOP as written: a process documented in plain English becomes a direct prompt for the agent, removing the need for a separate flowchart bot.
- Work on legacy and virtualized desktops: Citrix, terminal emulators, and older applications are just screens to an agent, not obstacles.
The VP of automation should remember one line: replace brittle bots with agents that read and follow the SOP, not the other way around.
How to move without the risk
You do not need to rip out all existing RPA at once. A pragmatic pipeline starts with a single, high-pain process that is documented in clear SOP language and suffers from frequent UI changes. The steps are simple. First, take the SOP and convert it into a structured prompt that the agent can follow. This might mean adding a few clarifying instructions about the expected outcome. Second, run the agent on a small, isolated VM to validate it against the current UI. Third, measure the time and error rate compared with the manual or RPA version. Fourth, expand the agent to similar processes once you see the pattern. Hybrid automation is realistic. Core RPA will still handle high-volume, stable, backend tasks like batch file processing. The agents take over the long tail work that is process-driven, exception-heavy, and constantly changing. That is where the durability advantage shows up.
The bottleneck is not technology. It is the mismatch between brittle bots and living processes. The pipeline from Confluence doc to running automation starts with a clear SOP and ends with an agent that can follow it. The first step is to see how a computer use agent handles your own process. Book a demo with the Coasty team to walk through a real SOP and watch it run in a live desktop environment.
Want to see this in action?
View Case Studies