LLM Agents & Planning: Literature Digest

Large language model (LLM) agents planning has matured from single-step prompting into a broader research area spanning task decomposition, tool use, reflection, memory, and multi-agent coordination. Across the surveyed papers, a consistent theme is that LLMs are strongest when used as planners embedded in systems, rather than as standalone reasoners.^[1][4][8]

From single-agent planning to multi-agent systems

Large Language Model based Multi-Agents: A Survey of Progress and Challenges frames planning as one of the core capabilities that enabled LLMs to move from autonomous agents to multi-agent systems, where collaboration improves complex problem-solving and world simulation.^[1][4][6] This line of work emphasizes system design: agent profiling, communication, and skill development matter as much as raw model quality.^[6] The newer Large Language Model Agent: A Survey on Methodology, Applications and Challenges extends this view with a methodology-centered taxonomy that treats planning as part of a larger construction-collaboration-evolution pipeline.^[7][8]

Planning methods: decomposition, selection, tools, reflection

The planning literature now clusters around several recurring mechanisms. The survey Large language models for planning: A comprehensive and systematic survey organizes prior work into task decomposition, plan selection, external modules, and reflection/memory, offering the clearest high-level map of the field.^[1] Describe, explain, plan and select shows how interactive planning can support open-world multi-task agents by having the model iteratively describe, explain, plan, and select actions.^[5] Tptu: Task planning and tool usage of large language model-based ai agents focuses on the ordering of tool use, highlighting that planning is inseparable from action sequencing in tool-augmented agents.^[2]

Grounded and embodied planning

For embodied and grounded settings, Llm-planner: Few-shot grounded planning for embodied agents with large language models positions LLMs as planners for agents that must map language to environment-constrained actions.^[6] AgentGen: Enhancing Planning Abilities for Large Language Model based Agent via Environment and Task Generation shifts attention to training data and simulation, arguing that environment/task generation can strengthen planning ability by exposing agents to richer trajectories.^[3] Across these works, planning is increasingly treated as interactive, grounded, and feedback-driven rather than static plan generation.^[3][7]

Open problems

• Evaluation remains fragmented across commonsense, tool-use, embodied, and multi-agent planning settings.^[1][8]
• Communication protocols for multi-agent planning are still underdeveloped, especially for maintaining distinct beliefs and shared intent.^[6]
• Robustness under feedback is limited; agents often need better mechanisms for revising plans when the environment changes.^[7][8]
• Training data for planning is scarce, motivating synthetic task/environment generation but raising coverage and realism questions.^[3]
• Symbolic integration is still immature; the surveys suggest stronger hybridization with classical planning and scheduling would improve reliability.^[1][2]

Key papers

1. Large language model based multi-agents: A survey of progress and challenges — T Guo,X Chen,Y Wang,R Chang,S Pei…
2. Tptu: Task planning and tool usage of large language model-based ai agents — J Ruan,Y Chen,B Zhang,Z Xu,T Bao…
3. AgentGen: Enhancing Planning Abilities for Large Language Model based Agent via Environment and Task Generation — M Hu,P Zhao,C Xu,Q Sun,JG Lou,Q Lin…
4. Large language models for planning: A comprehensive and systematic survey — P Cao,T Men,W Liu,J Zhang,X Li,X Lin,D Sui…
5. Describe, explain, plan and select: Interactive planning with large language models enables open-world multi-task agents — Z Wang,S Cai,G Chen,A Liu,X Ma,Y Liang
6. Llm-planner: Few-shot grounded planning for embodied agents with large language models — CH Song,J Wu,C Washington…
7. On the prospects of incorporating large language models (llms) in automated planning and scheduling (aps) — V Pallagani,BC Muppasani,K Roy,F Fabiano…
8. On the planning abilities of large language models-a critical investigation — K Valmeekam,M Marquez…
9. TPTU: large language model-based AI agents for task planning and tool usage — J Ruan,Y Chen,B Zhang,Z Xu,T Bao,G Du…
10. Twostep: Multi-agent task planning using classical planners and large language models — D Bai,I Singh,D Traum,J Thomason

Papers via the AISA Scholar API; synthesis by the AISA LLM layer. 2026-06-15.