Most automation teams I talk to are stuck on a treadmill. A bot works fine for a few months. Then the finance portal redesigns a button. The HR app changes a column name. Suddenly the automation halts, a developer has to rebuild the bot, and the backlog grows. They call this maintenance. I call it a sunk cost.
Why RPA breaks here
Traditional RPA relies on selectors, CSS classes, xpaths, and object IDs. Each of these is an assumption about the UI layout. When a vendor rolls out a patch, the layout shifts, the classes change, or the IDs become dynamic. The bot fails. A developer must inspect the new UI, update the selector tree, and redeploy. In many enterprises, one UI change can take days of engineering time. Industry studies show that up to 30 percent of an RPA program’s budget goes to maintenance and rework. Some teams spend more time fixing bots than building new ones. The cost compounds across dozens of bots. When the process involves legacy apps, Citrix, or virtualized desktops, selector-based RPA often cannot run at all. The bot does not see the underlying application, it sees a rasterized image of a window. It cannot click a button that is not stable. The result is a shrinking automation portfolio and growing manual tail.
What changes with computer use agents
- Survives UI changes: agents see what you see on the screen and act accordingly.
- No brittle selectors: they rely on visual cues and plain language instructions.
- Recovers from exceptions: when something unexpected happens, agents can pause, read, and try an alternative path.
- Follows the SOP as written: a standard operating procedure in plain English is already a prompt for a computer use agent.
- Works on legacy and Citrix: agents control the desktop session directly, not the image layer that RPA sees.
The one line a VP of automation should remember: agents work like humans, so they adapt when the UI changes and stay alive when exceptions occur.
How to move without the risk
You do not have to rip out all your RPA overnight. Start with a high-pain process where UI changes happen frequently or where exceptions are common. Examples include expense report approvals, customer onboarding, or invoice reconciliation. Use a computer use agent to automate that process end-to-end. Measure the difference in uptime, maintenance effort, and time to value. Then expand to other processes that share the same characteristics. Keep your existing bots for high-volume, stable, backend tasks where the UI does not change. That hybrid model lets you win both worlds: scale on the stable side and adapt on the changing side. The real advantage of computer use agents is that they do not require a new architecture for every process. They run in the same cloud VMs you already provision, or on your own desktop app. You can orchestrate multiple agents in parallel for batch work, and you can integrate them with your existing APIs and workflows. The transition is about choice, not replacement.
The bottleneck is no longer the technology. It is the brittle assumptions embedded in every RPA script. Computer use agents remove those assumptions. They let your automation portfolio grow without a parallel build-and-fix cycle. If you want to see how agents survive UI changes and recover from exceptions in your own environment, book a demo with the Coasty team.
Want to see this in action?
View Case Studies