Evaluating Multi-Agent Workflow Architectures for Enterprise AI Tasks: A Comparative Study Using Gemini and n8n

This paper presents an empirical evaluation of three agentic workflow architectures for enterprise-oriented AI tasks: Basic Agent, Planner Executor, and Planner Executor Reviewer. The study was conducted using Google Gemini-3.1-flash-lite and the n8n workflow automation platform. A dataset of 30 enterprise-oriented tasks covering Knowledge, Reasoning, and Coding categories was evaluated across all workflow architectures, resulting in 90 experimental runs. The findings indicate that workflow architecture has a measurable impact on AI performance and consistency. Multi-agent workflows achieved higher confidence scores and demonstrated improved performance on reasoning-intensive and hard tasks compared to a single-agent baseline. The complete research project, including workflow definitions, datasets, analysis files, figures, and paper drafts, is available in the associated GitHub repository :https://github.com/amuawia/Agentic-AI-Research