Overview
The conceptual claims are well supported by historical references and examples, but the paper contains no experiments, so production impact is medium and novelty is low.
Citations0
Evidence Strength0.80
Confidence0.82
Risk Signals10
Trust Signals
Findings with numeric evidence: 4/5
Findings with evidence refs: 5/5
Results with explicit delta: 0/0
Reproducibility
Status: No open assets linked
Open source: Unknown
At A Glance
Cost impact: 50%
Production readiness: 60%
Novelty: 30%
Why It Matters For Business
Calling LLM-based systems 'agentic' can hide reusable MAS solutions; adopting MAS standards and algorithms reduces engineering risk and speeds integration.
Who Should Care
Summary TLDR
The paper argues that the recent wave of "Agentic AI" and "Multiagentic" hype largely repackages well-established concepts from intelligent agents and multi-agent systems (MAS). It reviews the psychological and philosophical origins of "agency", classical agent properties and architectures (reactive, deliberative, hybrid, BDI), and MAS standards (KQML, FIPA). The author maps modern LLM-based agent frameworks (AutoGPT, LangChain, AutoGen, CAMEL, etc.) onto the older MAS literature, warns against terminological confusion and fragmentation, and recommends reusing MAS theory, protocols, and tools while integrating LLM strengths (flexible language understanding, memory, tool use).
Problem Statement
Industry and media have popularized "agentic" and "multiagentic" as if they were new technologies. This risks ignoring decades of MAS research, repeating solved problems (communication, coordination, standards), and fragmenting tooling when we should adapt proven MAS methods to LLM-era systems.
Main Contribution
Critical review showing 'agentic'/'multiagentic' usually map to established intelligent agent and MAS concepts
Concise survey of agency origins (Bandura, Dennett), agent properties, and classic agent architectures (reactive, deliberative, hybrid, BDI)
Key Findings
'Agentic AI' and 'Multiagentic' are mostly new labels for existing concepts (intelligent agents and multi-agent systems).
Large language models give agents much more flexible language understanding and plan decomposition, but they do not change core agent properties.
What To Try In 7 Days
Inventory current agent-like components and map them to classic agent architectures (reactive/BDI/hybrid).
Pick one interoperability standard (e.g., message format) and prototype a simple adapter between agent frameworks.
Replace an ad-hoc coordination policy with a known MAS algorithm (auction or consensus) and measure change in task throughput.
Agent Features
Memory
Planning
Tool Use
Frameworks
Architectures
Collaboration
Reproducibility
Risks & Boundaries
Limitations
No empirical evaluation or benchmark data; claims are conceptual and literature-based
Rapidly evolving LLM ecosystem may add recent examples not covered
When Not To Use
When you need a new learning algorithm or core ML research
When building low-level model optimizations or quantization systems
Failure Modes
Terminology mismatch causes miscommunication between teams
Framework fragmentation prevents interoperability