Academic

PseudoAct: Leveraging Pseudocode Synthesis for Flexible Planning and Action Control in Large Language Model Agents

arXiv:2602.23668v1 Announce Type: new Abstract: Large language model (LLM) agents typically rely on reactive decision-making paradigms such as ReAct, selecting actions conditioned on growing execution histories. While effective for short tasks, these approaches often lead to redundant tool usage, unstable reasoning, and high token consumption in complex long-horizon tasks involving branching, iteration, or multi-tool coordination. To address these limitations, this paper introduces PseudoAct, a novel framework for flexible planning and action control in LLM agents through pseudocode synthesis. Leveraging the ability of LLMs to express task-solving strategies as code, PseudoAct synthesizes a structured pseudocode plan that decomposes a task into subtasks and explicitly encodes control flow, including sequencing, conditionals, loops, parallel composition, and combinations of these logic primitives. Actions are then executed by following this global plan, making the decision logic explic

Y
Yihan (Logon), Wen, Xin Chen · · 1 min read · 15 views
#cs.AI #cs.SY #eess.SY

arXiv:2602.23668v1 Announce Type: new Abstract: Large language model (LLM) agents typically rely on reactive decision-making paradigms such as ReAct, selecting actions conditioned on growing execution histories. While effective for short tasks, these approaches often lead to redundant tool usage, unstable reasoning, and high token consumption in complex long-horizon tasks involving branching, iteration, or multi-tool coordination. To address these limitations, this paper introduces PseudoAct, a novel framework for flexible planning and action control in LLM agents through pseudocode synthesis. Leveraging the ability of LLMs to express task-solving strategies as code, PseudoAct synthesizes a structured pseudocode plan that decomposes a task into subtasks and explicitly encodes control flow, including sequencing, conditionals, loops, parallel composition, and combinations of these logic primitives. Actions are then executed by following this global plan, making the decision logic explicit and temporally coherent. This design reduces redundant actions, prevents infinite loops, and avoids uninformative alternative exploration, enabling consistent and efficient long-horizon decision-making. Experiments on benchmark datasets show that our method significantly outperforms existing reactive agent approaches, achieving a 20.93% absolute gain in success rate on FEVER and setting a new state-of-the-art on HotpotQA.

Executive Summary

This article introduces PseudoAct, a novel framework for flexible planning and action control in Large Language Model (LLM) agents. PseudoAct synthesizes a structured pseudocode plan that decomposes tasks into subtasks and encodes control flow, enabling consistent and efficient long-horizon decision-making. Experiments on benchmark datasets demonstrate a significant improvement over existing reactive agent approaches, achieving a 20.93% absolute gain in success rate on FEVER and setting a new state-of-the-art on HotpotQA. The framework's ability to reduce redundant actions, prevent infinite loops, and avoid uninformative exploration is a significant advancement in LLM agent design.

Key Points

  • ▸ PseudoAct synthesizes a structured pseudocode plan for task decomposition and control flow
  • ▸ The framework enables consistent and efficient long-horizon decision-making
  • ▸ Experiments demonstrate significant improvement over existing reactive agent approaches

Merits

Strength in Task Decomposition

PseudoAct's pseudocode synthesis enables explicit decomposition of tasks into subtasks, reducing redundant actions and improving decision-making efficiency.

Improved Control Flow

The framework's encoding of control flow, including sequencing, conditionals, loops, and parallel composition, prevents infinite loops and uninformative exploration.

Significant Performance Gain

Experiments demonstrate a 20.93% absolute gain in success rate on FEVER and a new state-of-the-art on HotpotQA, showcasing the framework's effectiveness.

Demerits

Limited Generalizability

The framework's performance gain may be task-specific, and its applicability to other domains or tasks is uncertain.

Computational Complexity

Pseudocode synthesis may introduce additional computational requirements, potentially impacting real-world deployment.

Expert Commentary

PseudoAct represents a significant advancement in LLM agent design, enabling consistent and efficient long-horizon decision-making. However, its limited generalizability and potential computational complexity require further investigation. The framework's performance gain on benchmark datasets is impressive, but its applicability to real-world scenarios is uncertain. Nevertheless, PseudoAct offers a promising direction for future research in LLM agent development, and its potential implications for industries and policy-making are substantial.

Recommendations

  • ✓ Further research is necessary to investigate PseudoAct's generalizability and computational complexity in various domains and tasks.
  • ✓ The framework's performance should be evaluated in real-world scenarios to assess its practical applicability and potential policy implications.

Sources

Related Articles