BMW Agents -- A Framework For Task Automation Through Multi-Agent Collaboration

TL;DR

The paper addresses automating complex industrial tasks using multi-agent LLM-based workflows to overcome the limitations of single-agent systems and to integrate with existing IT ecosystems. It introduces BMW Agents, a modular framework featuring planning, execution, verification, tool abstraction, memory layers, and multiple multi-agent workflow patterns. Key contributions include ConvPlanReAct conversational strategy, Toolbox Refiner, episodic memory for experiential learning, and restart/resume capabilities, enabling robust enterprise automation. The work demonstrates three practical applications—Retrieval Augmented Generation QA, actor/critic document editing, and collaborative coding—to illustrate real-world impact and scalability.

Abstract

Autonomous agents driven by Large Language Models (LLMs) offer enormous potential for automation. Early proof of this technology can be found in various demonstrations of agents solving complex tasks, interacting with external systems to augment their knowledge, and triggering actions. In particular, workflows involving multiple agents solving complex tasks in a collaborative fashion exemplify their capacity to operate in less strict and less well-defined environments. Thus, a multi-agent approach has great potential for serving as a backbone in many industrial applications, ranging from complex knowledge retrieval systems to next generation robotic process automation. Given the reasoning abilities within the current generation of LLMs, complex processes require a multi-step approach that includes a plan of well-defined and modular tasks. Depending on the level of complexity, these tasks can be executed either by a single agent or a group of agents. In this work, we focus on designing a flexible agent engineering framework with careful attention to planning and execution, capable of handling complex use case applications across various domains. The proposed framework provides reliability in industrial applications and presents techniques to ensure a scalable, flexible, and collaborative workflow for multiple autonomous agents working together towards solving tasks.

Figures (16)

• Figure 1: Generic agent workflow starting with user input and ending with providing workflow output. Agent Workflow highlights major components and levels of the workflow with (1) Planning, (2) Execution, and (3) Verification done by dedicated agents.
• Figure 2: Components contributing to initialization of an Agent. Dashed lines indicate optional components.
• Figure 3: Structure of \ref{['figure-non-iterative-prompt']} Basic non-iterative prompt, and its variations as applied in \ref{['figure-planning-prompt']} Planner and \ref{['figure-verification-agent']} Verifier agents.
• Figure 4: Generic ReAct prompt strategy with Thought, Action and Observation steps. We distinguish steps done as model responses (Assistant Message) and as a user (including Observation as User Message). PlanReAct includes an additional Planning step marked in orange.
• Figure 5: Generic iterative prompt strategy with user defined steps A...X that constitute the iteration sequence. The Assistant Message and User Message show the combination of reasoning steps by the LLM and inclusion of external information.