When using UI Agents as a function within a larger process, consider how your design navigates the environment and the organizational sequence across tasks to achieve your goal.

The examples demonstrate how clearly defined natural language Goals enable agents to navigate web portals, complete forms, extract structured data, download files, and execute multi-step workflows. Emphasis is placed on writing precise prompts, defining structured outputs, and decomposing complex objectives into smaller, reliable tasks.

This also highlights enterprise-ready capabilities such as chaining multiple Run actions, operating across browser sessions, securing sensitive data with protected variables, configuring proxies, and generating detailed execution logs. Together, these examples show how UI Agents function as adaptive, governed components within larger end-to-end automation architectures.

Common scenarios

For best results, you can compare prompts entered in UI Agents to the following runtime examples. Find examples of prompts that can be used to steer UI Agents to desired results.
Demonstrates the various designs that can be used.

Design patterns for Agentic Process Automations

You can follow one or more of the following design patterns for building complex Agentic Process Automations.

Pattern 1: UI Agent-only for browser tasks: A reliable pattern for a single, self-contained task that can complete in one uninterrupted session.
Demonstrates actions needed when only an agent is needed for a browser task.
Pattern 2: RPA with UI Agent as fallback for browser tasks: A valuable pattern when working with a few applications that undergo change infrequently, or reducing cost of a complete agentic solution.
Demonstrates actions needed when the agent is used as a fallback for a browser task.
Pattern 3: RPA for non-browser tasks and UI Agent for browser tasks: A valuable pattern when work can be divided into multiple independent tasks and executed on the same machine with little to no dependency.
Demonstrates actions needed when the browser tasks are completed by an agent and non-browser tasks are completed by RPA.

Use cases and examples

See the following examples for details on configurations and example prompts to perform common operations.