Agentic AI (aka Block Agent)

The Agentic AI aka Block Agent uses a a graph data structure, allowing you to build complex agents in a simpler way. You can configure an agent to respond to multiple topics and move through steps within each topic.

 

Agent Methodology

Built via JSON input defining each component of the workflow, one Agentic AI workflow can simplify unwieldy SmartFlows projects and provide overall improved performance.

  • Two parallel LLM calls at each turn.

    • One call triggers tools, the second call produces an AI response.

  • Always fires at least one tool.

    • At least one tool will always be defined for each block.

    • There will always be a tool triggered.

    • Multiple tools can be triggered at a given turn.

  • Structured prompting approach.

    • Through a standardized approach and format for prompting, higher performing agents can be created.

 

Channel Compatibility

Agentic is channel agnostic (e.g., compatible for inbound or outbound voice, SMS, etc.).

 

Use Cases

Design an entire conversational experience, from start to finish.

For example, your workflow could include a block for intent detection (how may I help you), a block to schedule an appointment, a block for bill payment, a block for insurance information, a block to transfer the call to a live person, and a block to end the call.

 

Configurations

  • System Prompt

    • Not editable at this time.

  • Personality

    • Not available with any other agent Type.

    • In this section you can define the tone and persona of your agent. The configured Personality is used across all turns in the Block Agent.

  • Blocks

    • Not available with any other agent Type.

    • Blocks define the conversational experience that the agent will orchestrate.

 

Tips and Best Practices

  • Make sure all of the components (Blocks, Turns, and Edges) within your agent have both unique Names and IDs.

    • For example, if all your Turns don’t have names and are just numbered, what step you’re on isn’t always clear when you’re testing your agent. But if you use unique and descriptive names with the IDs (e.g., General_Greeting, Birth_Date_Confirmation, New_Or_Exisiting_Patient, etc.), there is improved readability when you or your teammates review the agent.

  • Focus each Block on a specific and logical topic.

  • Clearly define your Turn conditions to avoid unexpected behaviors.

  • Remember to consider error handling and transitions between Blocks and Turns.

    • We recommend including guardrail Turns. For example, if your agent is asking the customer for their date of birth, after the user provides that, follow an Edge to confirm their details.

  • Be aware, if multiple Edges are true, the agent will follow the first true path. Edges are evaluated from top to bottom as defined.