RPA (Robotic Process Automation)
Software that automates repetitive, rule-based tasks by mimicking human interactions with applications — clicking buttons, copying data, filling forms. RPA works well on stable, predictable processes but breaks when inputs vary or exceptions occur.
What is RPA?
Robotic Process Automation (RPA) is software that automates tasks by replicating what a human does on a screen: navigate to a web page, open an application, read a value from one field, paste it into another. RPA bots do not understand content — they follow a script. If the script says "read the value in cell B4 of this spreadsheet" and the format changes, the bot either reads the wrong value or fails entirely.
RPA tools like UiPath, Automation Anywhere, and Blue Prism became widely adopted in the 2010s for back-office automation: copying data between systems, generating standard reports, submitting forms. They deliver real efficiency gains on processes that are genuinely stable and predictable.
Where RPA Works — and Where It Breaks
RPA is strong when three conditions hold: the process is rule-based with no exceptions, the input format is fixed and consistent, and the underlying applications do not change frequently. It fails when any of these conditions break:
A supplier changes their invoice layout — the bot reads the wrong fields
An exception occurs (missing field, unusual value) — the bot errors out or skips the record silently
A system UI update moves a button — the bot cannot find the element and stops
In practice, RPA deployments require constant maintenance. Every format change, every UI update, every new exception type requires a developer to update the script. Teams that deployed RPA in 2018 spend significant engineering time keeping bots running rather than extending automation.
RPA vs. AI Agents
AI agents handle what RPA cannot: variation, exceptions, and unstructured inputs. Where an RPA bot follows a fixed script, an AI agent reads context, makes decisions, and adapts. An AI agent processing invoices can handle 200 different supplier formats without template configuration, recognize that an invoice total is missing and check the line items to calculate it, and route unusual cases to the right person with a written explanation. RPA and AI agents are not mutually exclusive — in many operations, RPA handles the stable, structured steps while AI agents handle the variable, document-heavy, exception-prone ones.