Loading…
Loading…
Written by Max Zeshut
Founder at Agentmelt
Software 'bots' that execute a fixed, predefined sequence of digital actions—clicking, copying, moving data between systems—on a trigger, following rules a human scripted in advance. RPA (UiPath, Automation Anywhere, Blue Prism) has run back-office automation for a decade and is excellent at high-volume, stable, deterministic processes. The key distinction from an AI Agent: RPA follows a path designed ahead of time and breaks when the process changes, whereas an agent decides its steps at runtime and adapts. Relabeling an RPA script as an 'agent' without adding that runtime autonomy is the classic form of Agent Washing. The two increasingly combine—an agent handling ambiguous decisions and calling RPA bots for the deterministic execution steps.
An RPA bot logs into a portal every morning, downloads yesterday's invoices, and enters them into the ERP—the same steps, the same order, every day. It's fast and reliable until the portal's layout changes, at which point it breaks and waits for a developer. An agent handed the same goal would notice the changed layout and adapt.