Automate a 40-Step SOP Without Writing a Single Selector

Why RPA breaks here

UiPath, Automation Anywhere, and Power Automate rely on selectors, xpaths, and object IDs. When a developer builds a bot, they capture these identifiers at a specific moment. If the UI changes, column order, id naming, or a new version of the application, the bot stops and needs a rewrite. Gartner estimates that 70 percent of RPA projects never reach full scale largely due to maintenance burden. A single change can require a developer hour or more. When a 40-step process touches dozens of systems, the cost of staying on RPA is not just lost efficiency. It is a continuous cycle of discovery, development, and retesting. The process you wanted to automate becomes the process you are stuck babysitting.

What changes with computer use agents

• Survives UI changes: Agents see the screen, read labels, and adjust their actions to the current layout.
• No brittle selectors: Agents do not need xpaths or object IDs. They use natural language and visual cues.
• Recover from exceptions: When a step fails, agents read the error, decide what to do, and continue instead of halting.
• Follows the SOP as written: A human-readable procedure is already a prompt. Agents can execute it directly.
• Works on legacy and Citrix: Because agents move the mouse and type like a human, they run on systems where RPA struggles.

How to move without the risk

You do not need to rip out your existing RPA stack. Start by identifying one high-pain process that is manual or brittle. It should have a stable, written SOP and touch multiple systems. Choose a process with at least 20 steps and plenty of exceptions. Write the SOP as a clear sequence of actions. Then run a pilot using a computer use agent like Coasty. Compare the time and errors of human execution versus agent execution. Track how often the agent needs human intervention versus how often an RPA bot would have failed. Once you see the difference in maintenance and reliability, expand the approach to other processes. Over time you will have a digital workforce that can switch between RPA and agents based on what works best.

RPA still has a place

RPA remains excellent for high-volume, stable, backend tasks where the input and output are predictable. The new value comes from agents that can handle changing UIs, complex workflows, and long tail processes. You keep the bots that run reliably and you add agents for everything else.