An RPA developer spends two days rebuilding a bot because the procurement portal changed its form layout. A finance team still runs a 40-step invoice approval process by hand because the bot halts on the first missing approval. Legacy RPA and static SOPs create a maintenance backlog that grows faster than the automation backlog. For any automation center of excellence, the real question is not whether to automate, but how to automate work that changes, has exceptions, and lives across multiple systems.
Why RPA breaks on complex, changing processes
Traditional RPA bots bind to elements using selectors, XPaths, or object IDs. When a vendor updates a web form, changes a field label, or shuffles the DOM, the selector fails and the bot crashes. Industry studies show that a majority of RPA projects spend more time maintaining bots than building new ones. A bot that works today is often brittle tomorrow. A 40-step SOP that covers approval workflows, data entry, and system lookups crosses multiple applications, each with its own update cadence. The result is a bot that requires a rebuild every time the underlying system changes. Developers spend cycles on maintenance instead of adding more value. The process remains a human bottleneck.
The workflow where RPA hits a wall
Imagine a procurement approval workflow that follows a documented SOP. The process touches three systems: a requisition portal, an ERP, and an email inbox. Each step might require reading data from a screen, typing into a form, clicking an approval button, and handling missing information or routing changes. The SOP is written in plain English. The RPA team builds a flowchart bot, identifies selectors for every form field, and adds error handlers for known exceptions. Six months later, the vendor refreshes the portal UI. The selectors no longer match. The bot halts, and the process reverts to manual execution. Even with carefully written XPaths, unexpected states, like a popup, a delayed page load, or a missing approval, cause the bot to fail. The human team must intervene, document the exception, and patch the bot. The cycle repeats with every UI change. The bot becomes a liability rather than an asset.
What changes with computer use agents
- Survives UI changes because it sees the screen and adapts automatically.
- No brittle selectors or XPaths to maintain.
- Recovers from exceptions by re-reading the state and taking a new action.
- Follows the SOP as written, without needing a separate flowchart bot.
- Works on legacy applications, Citrix sessions, and virtualized desktops where RPA struggles.
Computer use agents work like a human on the screen: they read, they click, they type, and they verify results. That simple capability eliminates the selector maintenance treadmill and lets you automate SOPs that were previously too fragile for RPA.
Automating the 40-step SOP without selectors
A computer use agent treats the SOP as a set of instructions it can follow directly. The process begins with a single prompt that describes the goal and references the SOP. The agent opens the requisition portal, reads the page, identifies the correct fields, and types the required values. It proceeds step by step, verifying each action by reading the screen. When it encounters an exception, such as a missing approval or a system delay, it pauses, logs the state, and takes a corrective action, such as sending an email or flagging the case for a human. Because the agent relies on visual cues rather than static selectors, it survives UI updates without a rebuild. It can also operate in Citrix environments where traditional RPA cannot reliably interact with the UI. The same agent that handles the 40-step workflow can later be reused for similar processes with minimal adjustments.
How to move without the risk
A phased approach lets you adopt computer use agents without abandoning RPA entirely. First, identify a high-pain process that is exception-heavy, spans multiple systems, and follows a documented SOP. Examples include invoice approval workflows, onboarding requests, or data reconciliation across legacy systems. Build a pilot using a computer use agent, measure the time saved and error reduction, and compare it against the current manual or RPA process. If the pilot succeeds, expand to similar workflows. Use the results to build a business case for scaling computer use agents across the organization. RPA still fits well for high-volume, stable, back-end tasks with predictable UIs. Computer use agents excel at the long tail of work that changes, has exceptions, and requires human-like navigation. Together they create a more durable automation strategy.
The right automation technology depends on the work you are automating. For complex, changing SOPs that cross systems and handle exceptions, computer use agents offer a durable path forward. To see how a 40-step workflow can run without selectors, book a demo with the Coasty team at https://cal.com/coasty/15min .
Want to see this in action?
View Case Studies