Building an AI Agent Center of Excellence After RPA
Most automation teams still rely on legacy RPA for high-volume, stable back-office work. The problem arrives when that process touches a modern web portal, a legacy app with a shifting UI, or an ad-hoc approval workflow that depends on human judgment. Suddenly, bots halt, tickets pile up, and your automation backlog grows. The team spends more time rebuilding bots than running them.
Why RPA breaks here
Traditional RPA tools like UiPath, Automation Anywhere, and Power Automate automate by binding to specific UI elements: CSS selectors, XPath, object IDs. When a developer changes a button label, updates a page layout, or rebrands a portal, the selector breaks. The bot halts and a human has to rebuild it. That rebuild cycle shows up in the maintenance backlog and in per-bot run costs. Industry data suggests that 30 to 40 percent of an RPA project’s total cost is maintenance, not development. For many enterprises, that means a bot that runs successfully for months still cost more to keep than to build. When workloads shift toward ad-hoc approvals, dynamic forms, or legacy systems, those fragile bindings become a liability.
What changes with computer use agents
- ●Survives UI changes: Agents see the screen and respond to what is actually there, not to brittle selectors.
- ●No brittle selectors: They work with visual elements and context, so changing a label or layout does not break automation.
- ●Recovers from exceptions: When a bot hits an error state, an agent can read the screen and decide how to proceed, instead of halting.
- ●Follows the SOP as written: A process documented in plain English is already almost a prompt. Agents can execute it directly.
- ●Works on legacy and Citrix: Computer use agents control real desktops, browsers, and terminals, including environments where traditional RPA struggles.
The one line a VP of automation should remember: RPA is great for high-volume, stable backend tasks; computer use agents are the durable way forward for changing UIs, exception-heavy workflows, and SOP-driven processes.
How to move without the risk
You do not have to rip out RPA overnight. The pragmatic path starts with one high-pain process where the rebuild cycle is costing the most. Use that process to pilot a computer use agent. Measure uptime, exception handling, and the time saved by not rebuilding bots every time the UI changes. Once you see the durable benefits, expand to similar processes across departments. A phased approach looks like this: Identify a process with frequent UI changes or exception handling needs. Document the SOP in clear, step-by-step plain English. Run a pilot with a computer use agent on a cloud VM or your own desktop environment. Compare metrics: bot uptime, exception recovery rate, and time spent on maintenance. Scale agents across the organization, keeping RPA for the remaining high-volume, stable tasks. This lets you build an AI agent CoE that complements, rather than replaces, existing automation investments.
The future of enterprise automation is a hybrid model: RPA for what is stable, and computer use agents for what changes. To see how a computer use agent can handle your highest-pain processes, book a demo with the Coasty team.