IDP plus RPA vs a single computer use agent for document workflows

Why RPA breaks here

IDP plus RPA stacks are common because they try to combine best-in-class extraction with reliable execution. But the execution layer is brittle. RPA bots bind to selectors, XPath, or object IDs to locate elements on a screen. When a vendor’s ERP updates its class names, CSS rules, or layout, the bot no longer finds what it needs. The bot halts and creates an alert that a human must triage. In a midsize enterprise, typical RPA projects experience a 30 to 50 percent rebuild rate after each major UI release. That means for every hour of development, you spend another 20 to 30 minutes maintaining the bot. Over a year, the actual run time of a bot can drop to less than 30 percent of its intended schedule. The cost compounds as you add more processes, more applications, and more releases.

What changes with computer use agents

• ●survives UI changes
• ●no brittle selectors
• ●recovers from exceptions
• ●follows the SOP as written
• ●works on legacy and Citrix

The durability difference

A computer use agent does not need selectors. It sees the screen like a human does. It reads the text in a vendor invoice, notices a field for the invoice number, and uses that information to navigate to the ERP entry screen. When the ERP updates its layout, the agent can still find the invoice number field because it reads what is actually there. It does not rely on a hard-coded rule that says, "Click the element with class X at coordinates Y." This resilience dramatically reduces rebuild cycles. Where RPA bots stop on the first unexpected state, a computer use agent can assess the situation, try an alternative path, or ask a human for clarification. It can handle missing fields, extra columns, or new dropdown options without halting. It also works on legacy systems and virtualized environments like Citrix, where RPA struggles to access the DOM or secure input streams.

SOPs are already prompts

Most operating procedures are written in plain language. They say, "Open the invoice, extract the vendor name and amount, and enter them into the ERP." That is essentially a prompt. A computer use agent can read the SOP and follow it directly without a flowchart bot to build and maintain. This removes a layer of translation between business intent and software execution. The process owner can update the SOP in simple English, and the agent adapts. This is particularly valuable for document workflows where the format of invoices, contracts, or claims varies. Agents can generalize patterns instead of hard-coding specific layouts. They can also operate across multiple applications, OCR tools, ERPs, and email systems, without needing separate bots for each.

How to move without the risk

You do not have to rip out your existing RPA investments overnight. The pragmatic path is to identify one high-pain, document-heavy process where the UI changes frequently or where exceptions are common. For example, process a subset of vendor invoices that all come from a single ERP vendor. Run a pilot with a computer use agent alongside your current RPA and IDP stack. Compare the two on cost, accuracy, and maintenance effort. Look for cases where the agent reduces rebuilds and handles exceptions without human intervention. When you see a measurable benefit, expand the agent’s scope to more processes or more applications. Keep the RPA bots that handle stable, high-volume backend tasks. Reserve the agents for processes that are UI-sensitive, exception-heavy, or defined by SOPs. A hybrid approach lets you protect your current ROI while building a more durable automation foundation.