A mid-sized bank’s back office churns through thousands of entries daily, loan disbursements, settlement reconciliations, compliance document reviews. Their automation team built bots on UiPath and Power Automate to handle the routine work. Then the core banking system updated its interface. The bots stopped clicking the right buttons. Fixing them took hours per release, and new changes piled up. Meanwhile, a handful of complex approvals still required manual review because the bots could not handle exceptions or interpret a human-written SOP. The maintenance backlog grew while the backlog of unautomated work stayed the same.
Why RPA breaks here
Traditional RPA bots work by binding to selectors, xpaths, and object IDs. If a banking application updates its UI, those identifiers shift or vanish. The bot attempts to find the old element, fails, and halts. The automation team must identify the new selectors, rebuild the bot, and redeploy, often dragging in developers who are busy with new initiatives. Gartner research shows that 70 percent of RPA projects exceed their original budgets and timelines in large enterprises, largely because of this maintenance burden. In a regulated back office, a bot that halts in the middle of a settlement run creates compliance risk and forces manual overrides. The process is brittle, expensive, and increasingly fragile as systems evolve.
What changes with computer use agents
- Survives UI changes without rebuilding
- No brittle selectors to maintain
- Recovers from exceptions instead of halting
- Follows the SOP as written
- Works on legacy and Citrix environments where RPA struggles
Computer use agents SEE the screen and act like a human: move the mouse, click, type, read the result. They adapt to UI updates, follow SOPs in plain language, and recover from unexpected states.
A clearer comparison
Think of the difference as moving from a blind driver who follows a printed map with fixed street names to a human navigator who sees the intersection, adjusts to road changes, and finds a new way when a block is closed. RPA bots rely on fixed identifiers that break with updates. Computer use agents read the screen and act accordingly, making them far more durable in production environments where systems change frequently. They also handle exceptions more gracefully, such as missing data or an unexpected pop-up, because they can interpret the current state and decide how to proceed rather than failing immediately.
How to move without the risk
A phased approach lets you test this model while keeping RPA for what it does best: high-volume, stable, deterministic tasks. Start by identifying a back-office process with a written SOP and a history of bot failures. The goal is a process where the steps are clear to a human, but exceptions and UI changes are common. Pilot a computer use agent on that process, comparing cost per transaction, error rates, and time savings against the previous bot. Measure how often the agent recovers from exceptions without human intervention. Use those results to decide whether to expand to additional back-office workflows. At the same time, keep your existing RPA bots running for volumes where the inputs are predictable and stable. This hybrid approach lets you capture the value of computer use agents where they matter most without abandoning what your team already has.
The long-term cost picture
Traditional RPA projects often look cheap on paper because they automate straightforward flows. The hidden cost is the ongoing effort to maintain selectors and rebuild bots after each UI update. Computer use agents reduce that maintenance burden, but they require a different setup: cloud VMs, agent swarms for parallel execution, and a way to integrate with existing systems. The Coasty platform provides a desktop app and a /v1 computer use API, along with a free tier to start. This lets teams experiment with agents on a real desktop, browser, or terminal before scaling. The focus should be on where the return on investment is highest: processes with changing UIs, complex exception handling, or documentation in the form of SOPs.
Banking back-office automation is moving beyond fixed RPA bots to agents that can see, adapt, and follow SOPs. If you are ready to explore how computer use agents can reduce maintenance costs and handle the long tail of exception-heavy work, book a demo with the Coasty team at https://cal.com/coasty/15min.
Want to see this in action?
View Case Studies