Building an AI Agent Center of Excellence After RPA
Your RPA team is stuck. Every new release of your ERP or HR portal breaks a dozen bots. Developers spend more time patching selectors than building new automation. Meanwhile, a growing pile of SOPs sits in SharePoint because they are too messy for flowcharts and too expensive for custom bots. The maintenance backlog is growing. The team is burned out. The business is still waiting for automation that actually lasts.
Why RPA breaks here
Traditional RPA works by binding to exact UI elements, selectors, xpaths, object IDs. When a web form adds a new field, a button moves, or a page redesign ships, the bot stops. A developer must hunt down the new selectors and rebuild the bot. That is the maintenance treadmill. Industry data shows the average RPA project has a 30-40 percent rebuild rate in the first year, with teams spending up to 70 percent of their time on maintenance. You are not shipping value; you are keeping the lights on. The more stable the UI, the better RPA works. The more change, the faster it fails.
What changes with computer use agents
- ●Agents see the screen like a human does. They move the mouse, click, type, and read results. They do not need brittle selectors.
- ●When UI elements change, agents adapt. They locate the new button or field by its appearance and function, not by a hardcoded ID.
- ●Agents handle exceptions gracefully. If a page loads slowly or an error message appears, they observe the state, decide the next step, and recover instead of halting.
- ●SOPs written in plain English are already almost a prompt. Agents can read a step-by-step procedure and follow it directly, without a flowchart bot to build.
- ●Agents work across any application, including legacy systems, Citrix sessions, and virtualized desktops where traditional RPA struggles.
RPA automates parts of a process that rarely change. Computer use agents automate the parts that actually change, UI updates, exception paths, and SOP-driven workflows.
How to move without the risk
You do not need to rip out your RPA overnight. Start by identifying a high-pain process where UI changes frequently or where the SOP is unwieldy. Examples include employee onboarding, expense report processing, or order fulfillment exception handling. Run a small pilot with a computer use agent. Compare the time to build and maintain the automation against your current RPA setup. Measure uptime, exception recovery, and the time it takes to onboard new processes. Use those results to refine your methodology. Once you have a repeatable pattern, expand to other processes. Keep RPA for high-volume, stable, backend tasks where it still delivers the best ROI. Over time, your automation center of excellence will shift from bot-heavy to agent-heavy, with a hybrid model that maximizes both stability and flexibility.
The next step is to see an agent in action on your own desktop. The Coasty team can show you how agents read SOPs, adapt to UI changes, and recover from exceptions. Book a demo with the Coasty team at https://cal.com/coasty/15min to start building a durable automation strategy.