1. A developer/process analyst records or models a flow in a visual studio (drag-and-drop activities). 2. The robot is defined as a sequence of deterministic steps: locating a UI control (selector via DOM / accessibility tree / image matching), performing an action (click, type, read), with conditionals and loops. 3. Selectors bind robot logic to concrete control properties – HTML attributes, Windows UI Automation handles, SAP GUI Scripting fields, etc. 4. The robot runs in a runtime environment (attended – next to the user, or unattended – in the background on a server/VM) managed by an orchestrator (UiPath Orchestrator, Automation Anywhere Control Room). 5. The orchestrator schedules executions, distributes the robot across machines, and collects logs/metrics. 6. Exceptions (changed selector, missing element on time, business error) are reported and require developer intervention or routing to a human-in-the-loop queue.
Eliminating time-consuming, repetitive „swivel-chair” tasks – manually transferring data between systems that lack a shared API. It enables automation without rewriting or integrating legacy systems (ERP, CRM, mainframe, desktop applications).
A low-code visual IDE where the developer models the flow using activity blocks, conditionals and selectors. The output is a workflow package (e.g. .nupkg in UiPath) deployed to runtime machines.
The process that executes the workflow on the target machine. In attended mode it runs alongside the user (e.g. in a panel next to the application); in unattended mode it runs autonomously on a VM or terminal server.
The central management platform: execution schedules, work queues, license management, audit, dashboards, robot package distribution. It is the operational control point for the fleet.
Declarative expressions identifying a control in the target application – via DOM attributes, accessibility tree (Windows UI Automation), screen coordinates, or computer vision. Fragile – any UI change in the target application may require selector fixes.
An upgrade of the target application (rebranding, UI redesign) can break dozens of robots at once. Requires a dedicated maintenance team.
Every atypical scenario must be foreseen in the workflow; unforeseen situations stall the process. The problem grows with the number of input document/data variants.
RPA is rewarded for fast ROI, which encourages automating processes instead of fixing them (e.g. lack of proper API integration between systems). In the long term it entrenches integration chaos.
First GUI automation tools (e.g. AutoIt, application macros) – precursors of RPA.
First company to use the term „Robotic Process Automation” in an enterprise context – foundation of the RPA market.
Hyperautomation = RPA + AI/ML + process mining + iBPMS. Signal: pure RPA is not enough; it must be augmented with AI components.
UiPath AI Center, Automation Anywhere Co-Pilot, Microsoft Copilot in Power Automate – RPA becomes a component of broader Agentic Automation.
Vendors (Salesforce Agentforce, ServiceNow AI Agents, UiPath Agents) position LLM agents as the evolution of RPA – RPA becomes one of the execution layers alongside language models.
RPA is the deterministic execution of a script – there is no probabilistic component or runtime learning.
RPA has no dynamic routing – every workflow branch is explicitly coded as an if/switch by the developer.
Many robots can run in parallel across multiple VMs managed by the orchestrator, but a single robot is sequential within its own workflow.
Attended (next to the user, manually triggered) vs unattended (autonomously on a server, scheduled).
DOM/UI tree (most stable), image matching (least stable but universal), OCR (for scans and documents).
Retry policies, fallback paths, escalation to human-in-the-loop.
RPA runs on standard machines (Windows VM, terminal server). Requirements are driven by the target applications, not by the RPA engine itself.
