Unattended RPA bots and the 3am pager: how AI agents change on-call
A legacy unattended bot wakes the senior engineer at 3:15 AM. A browser update broke its selectors, the login page layout changed, or a side panel moved. The bot halts, the process stalls, and someone has to rewrite the workflow before the business impact grows. This is the reality for many enterprises: bots that promise 24/7 reliability but deliver a 24/7 pager duty cycle.
Why RPA breaks here
Traditional unattended bots rely on brittle references, CSS selectors, XPath, or object IDs. When a vendor ships a UI refresh, those references often become invalid. A bot that has run flawlessly for months suddenly stops. The team must re-record or re-write the bot, which means engineering hours on top of the original implementation cost. Industry surveys suggest that up to 30 percent of RPA maintenance time goes into fixing selector or layout changes. The result is a backlog of fixes and a support team that is always on call.
What changes with computer use agents
- ●Survives UI changes: agents see the screen and react to what is actually present rather than a fixed locator.
- ●No brittle selectors: there is no dependency on CSS classes or object IDs that can shift.
- ●Recovers from exceptions: when a step fails or a page behaves unexpectedly, agents can try alternative actions instead of halting.
- ●Follows the SOP as written: a standard operating procedure in plain English can be fed directly to the agent.
- ●Works on legacy and Citrix: agents can manipulate virtualized and outdated environments where RPA struggles.
Computer use agents treat processes as instructions humans already write, not as fragile sequences of selectors.
How to move without the risk
The pragmatic path is not to rip out all legacy RPA overnight. Instead, choose one process that is high‑pain, high‑volume, and heavily dependent on the UI. Common candidates include credit‑card reconciliation, invoice processing, or onboarding workflows. Run the process with a computer use agent on a parallel set of data. Measure uptime, error rates, and support calls. When the agent matches or exceeds the bot’s performance while handling unstructured UI changes, expand the scope. Keep legacy RPA for high‑volume, stable, backend tasks such as data entry to SAP or mainframe uploads where selectors are reliable and the process is nearly static.
The on‑call burden from RPA can shrink when you replace brittle bots with agents that see the screen and follow SOPs. To see how a computer use agent can handle your next high‑pain process, book a demo with the Coasty team at https://cal.com/coasty/15min.