Agentic AI: A Comprehensive Survey of Foundations, Architectural Paradigms, Core Capabilities, and a Unified Reference Architecture

Agentic artificial intelligence (Agentic AI) integrates goal-directed decision-making, state continuity, planning, tool-mediated action, and feedback-driven adaptation into interactive computational systems. Although existing surveys examine agent capabilities, architectures, and applications, cross-paradigm comparisons often conflate functional capabilities with implementation mechanisms and organizational structures. This comprehensive narrative survey synthesizes foundational and contemporary research on symbolic/classical, neural/generative, neuro-symbolic, and multi-agent approaches to clarify conceptual boundaries and examine how agentic capabilities are architecturally realized. It organizes recurring functions into nine groups: goal and context management, perception and state representation, reasoning and decision-making, planning and replanning, memory and knowledge, action and tool use, evaluation and reflection, learning and adaptation, and coordination and orchestration. The principal contribution is an architecture-capability mapping separating functional responsibilities from computational mechanisms and coordination structures. A complementary conceptual reference architecture integrates goal management, environmental interaction, planning, memory, execution, evaluation, and cross-cutting assurance concerns. Representative cases demonstrate how different systems can realize similar functions using different mechanisms. The proposed taxonomy, formal abstraction, and reference architecture are analytical synthesis artifacts rather than experimentally validated models. They support transparent architectural comparison and identify testable research questions.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-18
DOI
https://doi.org/10.21203/rs.3.rs-11461890/v1
Primary Topic
AI-based Problem Solving and Planning
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Agentic AI: A Comprehensive Survey of Foundations, Architectural Paradigms, Core Capabilities, and a Unified Reference Architecture

Yale Ibrahim Danjuma, Venkateswarlu Maninti, kanyesigye simon, Elhadi Abdalla A. Arbab et al.
3 links
Research Square
AI-based Problem Solving and Planning
preprint

Agentic AI: A Comprehensive Survey of Foundations, Architectural Paradigms, Core Capabilities, and a Unified Reference Architecture

Yale Ibrahim Danjuma, Venkateswarlu Maninti, kanyesigye simon, Elhadi Abdalla A. Arbab, Ninyesiga Allan, Rumanzi Smith, Abigaba Sofia, Adabara Ibrahim, Kyampaire Sarah, Kisembo Moreen
preprint en

Abstract

Agentic artificial intelligence (Agentic AI) integrates goal-directed decision-making, state continuity, planning, tool-mediated action, and feedback-driven adaptation into interactive computational systems. Although existing surveys examine agent capabilities, architectures, and applications, cross-paradigm comparisons often conflate functional capabilities with implementation mechanisms and organizational structures. This comprehensive narrative survey synthesizes foundational and contemporary research on symbolic/classical, neural/generative, neuro-symbolic, and multi-agent approaches to clarify conceptual boundaries and examine how agentic capabilities are architecturally realized. It organizes recurring functions into nine groups: goal and context management, perception and state representation, reasoning and decision-making, planning and replanning, memory and knowledge, action and tool use, evaluation and reflection, learning and adaptation, and coordination and orchestration. The principal contribution is an architecture-capability mapping separating functional responsibilities from computational mechanisms and coordination structures. A complementary conceptual reference architecture integrates goal management, environmental interaction, planning, memory, execution, evaluation, and cross-cutting assurance concerns. Representative cases demonstrate how different systems can realize similar functions using different mechanisms. The proposed taxonomy, formal abstraction, and reference architecture are analytical synthesis artifacts rather than experimentally validated models. They support transparent architectural comparison and identify testable research questions.

Research Square
Kampala International University (UG)
AI-based Problem Solving and Planning
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon