The 1st International Workshop on LLM-Driven Agents

- Unleashing the Power of LLM-Driven Agents: from Theory to Real-World Applications

1. Motivation and Scope

Large Language Models (LLMs) have evolved from powerful text generators to autonomous agents capable of decision-making, tool use, and complex task execution. The emergence of LLM-driven agents represents a paradigm shift in AI—where foundation models serve not just as passive engines but as interactive, goal-directed systems. Despite exciting progress, fundamental theoretical challenges remain: What are the formal models of LLM-driven agency? How can we ensure safety, alignment, planning consistency, and memory grounding? Moreover, the transition from lab settings to real-world applications in domains like software engineering, healthcare, education, and robotics raises additional questions about robustness, deployment, evaluation, and cost-efficiency.

This workshop aims to bridge the gap between theory and practice by uniting researchers and practitioners working on various aspects of LLM-driven agents.

2. Topics of Interest (including but not limited to)

• Formal models of agentic behavior in LLMs
• Planning, reasoning, and memory in agent systems
• Safety, alignment, and interpretability of LLM agents
• Evaluation benchmarks and metrics for agentic performance
• Resource efficiency and deployment challenges
• Multi-agent collaboration and communication
• Human-AI collaboration
• Multi-modal and embodied LLM agents
• Tool-augmented and API-enabled agents
• Agentic applications

3. Target Audience

• Researchers in LLM, NLP, ML, cognitive modeling, robotics

• Developers and practitioners working on agent frameworks (e.g., LangChain, AutoGen, CrewAI)

• Professionals deploying LLM-driven systems in real-world environments
• Scholars interested in the theoretical underpinnings of LLM agency

4. Format and Schedule

We propose a full-day workshop consisting of

• Invited talks from leading academics and industry experts. Topics will cover:
  • Agentic architectures for reasoning and planning
  • Theoretical foundations of agency in LLMs
  • Applications and deployment lessons from real-world systems
  • Challenges in alignment, safety, and evaluation
• Paper presentations (oral sessions)
• Panel Discussion

A moderated panel titled "From Prompt to Plan: What Makes an LLM Truly Agentic?" will gather thought leaders to debate the definition, limits, and future directions of agentic intelligence.

5. Submission and Review

• Submission via EasyChair / OpenReview

• Double-blind review with a diverse Program Committee
• Accepted papers will be published in workshop proceedings and optionally uploaded to arXiv
• Papers should not exceed 12 pages single spacing in (LNCS (Lecture Notes in Computer Science) format). The submission of papers must be in either PDF or Word format.

6. Invited Speakers (Tentative)

7. Organizers

• Prof. Yong Zhang, Tsinghua University, zhangyong05@tsinghua.edu.cn

• Prof. Bohan Li, Nanjing University of Aeronautics and Astronautics, bhli@nuaa.edu.cn

• Prof. Chun Zeng, Henan University, zengchun@tsinghua.org.cn

• Prof. Guigang Zhang, Institute of Automation (CAS), guigang.zhang@ia.ac.cn

• Prof. Yuxi Zhou, Tianjin University of Technology, joy_yuxi@pku.edu.cn

8. Contact Person

Xin Wei (Andrew), xwei@tsinghua.edu.cn

Submission Deadline: 31 July, 2025.