Mainframe and Legacy App Automation with Computer Use Agents
Your finance team still sits in front of green-screen terminals. Your operations team works inside a 20-year-old ERP that no longer supports modern selectors. You have a stack of RPA bots that touch these systems, but every time the vendor rolls out a new release or a new patch, the bots break. Developers spend more time rebuilding than building new automation. Meanwhile, your standard operating procedures sit in shared drives, never touched by software. You are stuck between brittle bots and unwritten processes.
Why RPA breaks here
Traditional RPA works by mapping UI elements to selectors, xpaths, or object IDs. When a vendor changes a color, a tab order, or a field name, the selector fails and the bot halts. In practice, this means a bot that ran flawlessly for months needs a rebuild every time the application is patched. Studies from industry analysts show that about 60 percent of RPA maintenance time goes into rebuilding broken bots every time a UI changes. For mainframe screens and legacy apps, the cost is even higher. Many of these systems do not expose stable DOM structures, so even the most advanced RPA platforms struggle to find a reliable selector at all. The result is a backlog of broken bots, a team that spends more time patching than innovating, and processes that remain manual because the automation is too fragile to keep running.
What changes with computer use agents
- ●Survives UI changes: agents see the screen and locate the next action instead of relying on brittle selectors.
- ●No brittle selectors: the agent moves the mouse, clicks, and types, so the underlying UI can shift and the bot keeps working.
- ●Recovers from exceptions: when an unexpected error appears, the agent can pause, read the message, and try an alternative step instead of halting.
- ●Follows the SOP as written: an English SOP becomes a natural language prompt, and the agent executes it directly without a flowchart bot.
- ●Works on legacy and Citrix: agents run in cloud VMs and on desktop apps, giving you access to systems that RPA platforms often struggle to reach.
Computer use agents see the screen and adapt, so you stop rebuilding on every change and start automating the long tail.
How to move without the risk
You do not need to rip out all your RPA at once. Start with one high-pain process that touches a mainframe or legacy UI and cannot be easily migrated to a modern SaaS application. Map the current steps into a clear SOP. Feed that SOP into a computer use agent. Run the agent in parallel with the existing manual process for a few weeks. Compare the number of incidents, the time saved, and the effort required to keep the automation running. If the agent handles the variability better than your RPA bots, expand to similar processes. RPA still fits well for high-volume, stable, backend tasks where the UI never changes. Use computer use agents for the long tail, the changing UIs, and the exception-heavy workflows that keep your team stuck at their desks.
The next step is to see how a computer use agent can run on your mainframe or legacy screens. Book a demo with the Coasty team to discuss a pilot for a single high-pain process. https://cal.com/coasty/15min