Summary
This document is a technical report that details the CXP Step-Based Flows architecture developed by ASAPP to address reliability issues in enterprise AI agents. It outlines the challenges faced by large language models (LLMs) in maintaining consistent performance during customer interactions, particularly in complex workflows. The report explains that while LLMs excel in conversational abilities, they often struggle with executing business logic accurately, leading to errors that can undermine trust. CXP Step-Based Flows separates the conversational aspects handled by the LLM from the deterministic business logic managed by a state machine. This architecture allows for a structured approach to workflows, ensuring that each step is executed correctly without the LLM's involvement in business logic decisions. The report further describes how this separation enhances the reliability of AI agents, enabling them to provide a human-like experience while adhering to operational requirements. Additionally, it highlights the benefits of channel-agnostic workflows and the ability to manage complex interactions without cognitive overload for the LLM.
