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
- Yale Ibrahim Danjuma (ORCID: https://orcid.org/0000-0003-4226-4760)
- Venkateswarlu Maninti
- kanyesigye simon
- Elhadi Abdalla A. Arbab
- Ninyesiga Allan
- Rumanzi Smith
- Abigaba Sofia
- Adabara Ibrahim
- Kyampaire Sarah
- Kisembo Moreen
Institutions
- Kampala International University (UG)
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