A Self-Correcting Agentic AI Framework for Autonomous Planning and Complex Task Execution

DOI

10.2991/978-2-38476-615-4_36How to use a DOI?

Keywords

Agentic AI; Autonomous Planning; Self-Correction; Multi-Agent Systems; Large Language Models; Reflection; Autonomous Decision Making; AI Agents; Reinforcement Learning; Complex Task Execution

Abstract

Large language models (LLMs) have revolutionized artificial intelligence from a system that takes user input to a standalone agent that can plan, reason, use tools, and make decisions over long time horizons. However, even with these advances, current agentic AI systems still struggle with issues such as maintaining consistency in their plans, recovering from execution failures, adapting to changing environments, and continually self-reflecting to enhance their performance. In such complex verticals as software development, self-driving cars, research labs, medicine, cyber security, and automation in companies, incorrect steps in a sequence of operations may impact the overall system reliability by a significant amount.

This paper presents a Self-Correcting Agentic AI Framework (SCAIF) that combines elements of hierarchical planning, reflective reasoning, execution monitoring, dynamic memory management and iterative self-correction into a comprehensive framework for autonomous planning and execution of complex tasks. The proposed approach includes a closed loop feedback-based approach that identifies planning errors, assesses intermediate results, identifies deviations from the planning and execution, and autonomously adapts the plans to prevent failures. The framework integrates planner, executor, verifier, critic and memory agents that communicate in a structured way and keep fine-tuning their strategies based on the environmental feedback.

The proposed architecture is inspired from recent developments in agentic programming, multi-agent collaboration, self-evolving AI agent, reinforcement learning, and reflective reasoning, aiming to create a scalable and domain-independent model. The mathematical formulation of iterative planning and self-correction is presented as well as an algorithm describing adaptive execution refinement. The framework is designed to be deployed in various domains that demand reliable independent decision making, such as intelligent software development, autonomous laboratories, industrial automation, supply chain, and next generation cognitive networks.

Finally, the study introduces a generalized architecture for self-correcting autonomous agents, points out the gaps in the current state of knowledge in the field of self-correcting planning, and proposes a methodology for evaluating the performance of self-correcting planning systems, with particular attention to the aspects of planning accuracy, recovery efficiency, number of finished tasks, adaptability, and computational costs. The aim of the proposed framework is to increase the robustness, transparency and long-term autonomy in complex environments, whilst minimising the cumulative planning errors of such environments.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Nishant Sharma
AU  - Manish Shrivastava
AU  - Harshita Sharma
PY  - 2026
DA  - 2026/09/15
TI  - A Self-Correcting Agentic AI Framework for Autonomous Planning and Complex Task Execution
BT  - Proceedings of the Anubhuti 4.0: Discourses in Indian Knowledge Systems – Pathways of Culture, Inclusion, and Sustainability (ICIKS 2026)
PB  - Atlantis Press
SP  - 544
EP  - 584
SN  - 2352-5398
UR  - https://doi.org/10.2991/978-2-38476-615-4_36
DO  - 10.2991/978-2-38476-615-4_36
ID  - Sharma2026
ER  -