How to Pilot Computer Use Agents Alongside Your Existing RPA
Your RPA team is already under pressure. After a release, a support ticket, or a vendor update, dozens of bots stop working. Developers spend more time fixing selectors and rebuilding flows than building new ones. Meanwhile, managers keep asking for automation in new, complex areas where the UI changes often and the process is documented only in natural language. The result is a maintenance backlog and a growing number of SOPs that only humans can actually run.
Why RPA breaks here
Traditional RPA tools like UiPath, Automation Anywhere, and Power Automate rely on selectors, xpaths, and object IDs. They grab a button by its unique property and send a click or keystroke. When an application redesigns, a web framework changes, or a legacy system adds a new field, those identifiers drift or disappear. The bot halts, often with no clear error for the user. In many enterprises, selector drift now costs between five and ten percent of total automation spend each year, and failed runs trigger manual escalations, regulatory risks, and support tickets. The bot does not recover. It waits for a human to reset the state or fix the flow.
What changes with computer use agents
- ●Survives UI changes: agents see the screen and act on what they see, not on brittle identifiers.
- ●No brittle selectors: they move the mouse, click, and type like a human, so app redesigns rarely break them.
- ●Recovers from exceptions: instead of halting, agents observe the state, retry, or ask for guidance, reducing manual intervention.
- ●Follows the SOP as written: when a process is documented in plain English, an agent can follow it directly without building a flowchart.
- ●Works on legacy and Citrix: agents run on real desktops and virtualized environments where traditional RPA struggles.
Traditional RPA works well for high-volume, stable, backend tasks. Computer use agents are the durable solution for changing UIs, exception-heavy work, and SOP-driven processes.
How to move without the risk
You do not need to rip out all RPA at once. A practical path is to pick one high-pain process where UI changes frequently and manual fallbacks are expensive. For example, consider a procurement approval workflow that spans multiple systems, includes conditional logic, and is documented in a team SOP. Start by capturing the current process flow and mapping the key decision points. Then draft a plain-English SOP that a computer use agent can follow. Run the pilot using Coasty to automate a subset of cases, measure time saved, errors avoided, and maintenance hours freed. Compare those results with the cost of maintaining the existing bot. Use the win as evidence to expand to additional processes. Throughout, keep the RPA bots running for the tasks where they still make sense. This phased approach lets you build confidence, reduce risk, and demonstrate value before scaling. Be honest about where RPA still fits and target the hardest cases for agents.
The combination of traditional RPA and computer use agents can give you both the stability you need and the flexibility to automate harder, changing processes. Ready to start a pilot? Book a demo with the Coasty team at https://cal.com/coasty/15min to see how agents can work alongside your existing automation.