AI-generated summary
Genesys Cloud enables integration of agentic virtual agents into Architect flows through the Call Agentic Virtual Agent action, supporting both voice and digital bot workflows. Implementation requires Virtual Agent enablement at the organization level, appropriate AI Studio permissions, and agent publication through the Orchestration menu before connection to Architect.
The Call Agentic Virtual Agent action operates through a configuration process where users select published agents, specify an action name, and manage input and output variables. Input variables receive starting context values—either literal or expression-based—sent to the agent, while output variables capture end context data returned after interaction completion. The system executes the latest published agent version and provides dynamic message handling capabilities without requiring global prompts for no-match and no-input scenarios. Users can access agent configuration directly through the Architect interface, enabling flexible deployment of AI-driven interactions while maintaining control over data handling and workflow design.
Execution outcomes follow three distinct paths: Success indicates successful agent execution, Failure indicates action errors, and Timeout indicates exceeded execution time. Exit reasons are categorized as Error, Escalation, TriggeredByUser, or Guardrails, each assignable to a variable for subsequent flow processing and multi-path routing logic. Recommended timeouts remain under 30 seconds to prevent extended user wait times, with the Timeout path ensuring flow continuation when actions exceed their duration limits.
Key technical features include Dependency Search functionality for identifying flows dependent on specific agentic virtual agents, automatic flow type conversion to Agentic Virtual Agent upon action addition, and seamless agent escalation capabilities. Agent publishing validates overall definition size, including instructions, guidelines, descriptions, tools, and data types; oversized definitions must be reduced before publishing, as smaller definitions consume fewer tokens at runtime. To optimize performance and reduce token consumption, organizations should reuse values through variable linking, mask sensitive data such as authentication tokens and API keys, select only necessary output fields from data actions, and minimize verbose instructions and descriptions. These optimization strategies support hybrid workflows that combine agentic and conventional flow-based elements while maintaining system efficiency and cost-effectiveness.