Comparativo de plataformas para orquestração de agentes inteligentes: n8n vs. Crew AI na sumarização de tickets.

Abstract

Inefficient support ticket management in technology companies can lead to significant delays in customer responses, increased operational costs, and overload of support teams, directly affecting customer satisfaction and retention. With the growing volume of data and the need for quick responses, automation and intelligent analysis of these tickets become essential to optimize resources and reduce bottlenecks. This work proposes the creation of a multi-agent system to summarize Zendesk tickets stored in AWS S3 and generate managerial reports, leveraging Large Language Models (LLMs), including but not limited to Claude 3.5 Sonnet. For the implementation, two distinct platforms were evaluated: n8n, an open-source workflow automation tool that adopts the low-code/no-code paradigm and enables the orchestration of AI agents through interconnected nodes; and Crew AI, a Python-based framework that allows the creation and coordination of intelligent agents with greater programmatic control and architectural flexibility. The evaluation was conducted by the same person responsible for the development in both platforms, ensuring consistency in the analysis of criteria such as ease of development, flexibility, maintainability, and suitability for using LLMs in real-world scenarios. The results showed that n8n stands out for its fast development and visual interface, being more suitable for prototyping and direct automations, while Crew AI is better suited for complex architectures — here understood as those involving multiple agents, MCPs, and requiring more specialized algorithms to perform tasks. It is worth noting that Zendesk and S3 were chosen because they are widely used solutions in the market, but the proposed system is not restricted to them. It is concluded that the choice of the most appropriate platform depends directly on the specifications of the demand, such as workflow complexity, available implementation time, and the required level of technical control.