From Retrieval to Cognitive Orchestration: Standardizing Context Management in Agentic AI Systems
International Journal of Computational and Experimental Science and Engineering · 26.02.2026 · 12 (1)
DOI: 10.22399/ijcesen.4970 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Computational and Experimental Science and Engineering, online erschienen am 26.02.2026. Verfasst von Bhaskara Reddy Udaru. Die Arbeit nennt 10 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The proliferation of large language model-based agentic systems necessitates rigorous systems engineering approaches to context management. Contemporary frameworks, including Retrieval-Augmented Generation (RAG), ReAct, AutoGPT, and LangGraph, demonstrate autonomous capabilities but lack formal system specifications for context lifecycle, provenance tracking, and governance enforcement. This paper presents a systems engineering framework formalizing cognitive orchestration as a layered architecture with explicit invariants, interface contracts, and verification protocols. We introduce formal system models defining context as C = (K, M, P, T, V) with mathematical invariants ensuring consistency, completeness, and auditability. Our framework integrates Model Context Protocol (MCP) interfaces, establishing standardized contracts for agent coordination, memory management, and policy enforcement. Comparative analysis reveals systematic limitations in existing frameworks: RAG lacks multi-step context propagation (hallucination amplification 3.2×), ReAct exhibits unbounded memory growth (O(n²) with interaction length), AutoGPT suffers governance gaps (31% compliance violations), and LangGraph provides insufficient provenance tracking (34% audit coverage). Empirical validation through enterprise deployment, Annual Report Financial Analysis system processing 500+ documents across 15 regulatory frameworks, demonstrates quantifiable improvements: 94% reduction in compliance violations, 89% decrease in error propagation, 98% provenance completeness, and 3.1× mean time between failures compared to baseline architectures. System verification confirms invariant preservation across 10,000+ agent interactions with zero safety violations. This work establishes cognitive orchestration as essential infrastructure for production-grade agentic systems, providing formal foundations, architectural blueprints, and verification methodologies applicable across enterprise automation, financial analysis, regulatory compliance, and safety-critical domains.
Zitieren
APA
Udaru, B. R. (2026). From Retrieval to Cognitive Orchestration: Standardizing Context Management in Agentic AI Systems. International Journal of Computational and Experimental Science and Engineering, 12(1). https://doi.org/10.22399/ijcesen.4970
Vancouver
Udaru BR. From Retrieval to Cognitive Orchestration: Standardizing Context Management in Agentic AI Systems. International Journal of Computational and Experimental Science and Engineering. 2026;12(1). doi:10.22399/ijcesen.4970
BibTeX
@article{Udaru2026,
author = {Udaru, Bhaskara Reddy},
title = {From Retrieval to Cognitive Orchestration: Standardizing Context Management in Agentic AI Systems},
journal = {International Journal of Computational and Experimental Science and Engineering},
year = {2026},
volume = {12},
number = {1},
doi = {10.22399/ijcesen.4970}
}
EUVIMED