A VP of automation at a large manufacturer sends her team a new policy for supplier onboarding. The process involves logging into three different systems, uploading documents, and entering data by hand. The team knows exactly how to do it, but they cannot automate it. The systems use different layouts, the forms change each quarter, and the flows are documented only in plain English. A low code RPA platform would require hundreds of hours of mapping selectors and testing every single form change. The team ends up doing the work manually instead. This is the maintenance treadmill that blocks automation in many enterprises. Low code RPA is powerful, but it is brittle. Prompt driven computer use agents are different because they see the screen and act like a human. They can follow the same SOP that the team has already documented.
Why RPA breaks here
Low code RPA tools like UiPath, Automation Anywhere, and Power Automate rely on selectors, xpaths, and object IDs to find controls on a screen. The first version of a bot works because the application layout is stable. But when a vendor ships a minor UI update or a business team changes a form, those selectors stop matching. The bot halts and an engineer must rebuild it. Industry data shows that RPA teams often spend 30 to 70 percent of their effort on maintenance rather than new automation. A large enterprise can easily accumulate a backlog of hundreds of broken bots. Each bot that fails creates a ticket, an investigation, and a fix. The cost is compounded when automation is needed on legacy applications, Citrix virtual desktops, or internal tools that never standardized their UIs. The result is a growing gap between what leaders want to automate and what their automation stack can actually support.
What changes with computer use agents
- Survives UI changes: Agents see the screen and locate controls by position and appearance, not by static selectors.
- No brittle selectors: There is no need to map object IDs or maintain complex selector trees.
- Recovers from exceptions: When a bot hits an error, it can read the screen, reason, and try alternative steps instead of stopping.
- Follows the SOP as written: A standard operating procedure in plain English is already almost a prompt. An agent can execute it directly.
- Works on legacy and Citrix: Because agents control a real desktop session, they can automate applications where traditional RPA struggles.
Low code RPA builds a fragile bridge between your systems. Computer use agents walk the bridge like a human and adapt when the ground shifts.
How to move without the risk
Enterprises do not have to abandon low code RPA overnight. A pragmatic path is to start with processes where RPA and SOPs are both strong drivers. Look for tasks that are documented in plain language, involve multiple applications, and change frequently. Examples include onboarding new suppliers, processing customer orders, or updating inventory across several systems. Choose one high-pain process, run a pilot with a computer use agent, and compare the time to build, the time to maintain, and the number of incidents. The agent should be able to follow the existing SOP without new workflow design. Once the team sees the difference in resilience, they can expand to other processes. Over time, you can move more work to computer use agents while keeping RPA for high-volume, stable tasks that run in the backend. This phased approach lets you benefit from the durability of agents without disrupting your existing automation estate.
The question is no longer whether to adopt AI agents, but how quickly you can add them alongside your existing automation. A computer use agent can follow your SOPs, handle UI changes, and recover from errors that would stop a traditional RPA bot. To see how this works in practice, book a demo with the Coasty team at https://cal.com/coasty/15min .
Want to see this in action?
View Case Studies