What is a Robotic Process Automation?

RPA bots follow a scripted sequence recorded or coded against a specific screen layout or API shape. When that layout changes, the bot breaks, because it has no model of what the task is actually trying to accomplish; it only knows the steps. This made RPA a fast win for narrow, high-volume, unchanging processes (invoice data entry, account provisioning) in the 2015-2022 era of automation. The limitation is brittleness and scope: RPA can't handle exceptions, ambiguous inputs, or a process that varies case to case, because there's no reasoning layer, just a replay mechanism. Deploying and maintaining large RPA fleets became its own IT burden, since every UI update meant re-recording bots. RPA is now increasingly a subcomponent inside agentic systems rather than the whole solution: an LLM-driven agent can decide when a scripted RPA action is the right tool and hand off to it, combining judgment with deterministic execution.