Building an AI Agent Center of Excellence After RPA
IT leaders still talk about RPA as the automation backbone, but many centers of excellence feel it more like a maintenance treadmill. A browser update breaks a bot. A new screen layout forces a dev to rebuild a workflow from scratch. Process owners write SOPs, but only humans can follow them reliably. The result is a backlog of brittle bots and work that stays on paper.
Why RPA breaks here
Traditional RPA relies on selectors, xpaths, and object IDs that tightly bind a bot to a specific UI element. When a vendor releases a patch, a dev team changes a class name, or IT rebrands a portal, that binding breaks. A Gartner-style breakdown of automation maintenance costs shows teams spend 40 to 60 percent of their time on regression and redeployment rather than new work. Each UI change can mean weeks of retesting, not hours. In volatile environments, legacy apps, Citrix, SaaS updates, this fragility becomes a constant source of unplanned downtime and support tickets.
What changes with computer use agents
- ●Agents see the screen like a person does and move the mouse, click, and type instead of relying on hard-coded selectors.
- ●They adapt when UIs change because they act on what is visible, not on brittle element IDs.
- ●Instead of halting on exceptions, agents read errors and attempt recovery steps, closing a wrong window, retrying a failed click, or re-reading the screen.
- ●A plain-English SOP is already close to a prompt that a computer use agent can follow directly, eliminating the need to translate workflows into flowcharts.
- ●They work across any application, including legacy systems, Citrix, and virtualized desktops where traditional RPA struggles.
RPA binds to elements. Computer use agents bind to the process.
How to move without the risk
You do not need to rip out all your RPA at once. Start by picking one high-pain process that is SOP-heavy and UI-unstable. Use Coasty’s computer use agent to run that process in parallel to your existing bot. Measure the difference in uptime, exception handling, and time spent on maintenance. Once you see that the agent handles UI changes and unexpected states without a developer rebuild, expand to other processes. Keep RPA for high-volume, stable, backend tasks where speed and deterministic output still make sense. Over time, shift more work to agents, reducing your reliance on fragile selectors and freeing your automation team to focus on new value, not just regression.
The next step is to see how a computer use agent runs your actual SOPs on real desktops. Book a demo with the Coasty team to explore a pilot for your highest-pain process.