The single prompt approach allows you to define your agent’s behavior with one comprehensive prompt. It’s straightforward and great for simple use cases.
While this approach is an excellent starting point, you may encounter several challenges as your agent becomes more complex:
The agent may start to hallucinate or deviate from instructions
Function calling becomes less reliable
The prompt becomes difficult to maintain as it grows larger
Due to these limitations, we recommend using the multi-prompt approach for more sophisticated agents.
The multi-prompt approach allows you to create a structured tree of prompts. Each node in the tree can contain:
Its own specific prompt
Custom function calling instructions
Clear transition logic to other nodes
For example, our “Lead Qualification” template demonstrates this structure:
In this template, we break down the conversation into two distinct steps:
Lead Qualification: Gather and verify all required information
Appointment Scheduling: Book the appointment only after successful qualification
This structured approach prevents common issues found in single-prompt agents, such as premature appointment booking before completing the qualification process. The multi-prompt structure ensures your agent follows the correct sequence of steps and maintains context throughout the conversation.