Agentic AI and Multiagentic: Are We Reinventing the Wheel?

V.Botti

REVIEW 4 major objections 5 minor 1 cited by

Agentic AI and Multiagentic: Are We Reinventing the Wheel?

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper argues that 'agentic AI' and 'multiagentic' are rebrandings of the established concepts of intelligent agents and multi-agent systems.

desk verdict A coherent, readable opinion piece arguing that 'Agentic AI' and 'Multiagentic' are largely rebrandings of intelligent agents and MAS; the core advice is sound, but the evidence base is too thin to support the full equivalence claim. read the letter →

arxiv 2506.01463 v1 pith:XJLW3TSB submitted 2025-06-02 cs.MA cs.AI

V.Botti This is my paper

classification cs.MAcs.AI

keywords agenticAImulti-agentsystemsintelligentagentsLLM-basedagentarchitecturebelief-desire-intentionorchestrationterminology

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that 'Agentic AI' and 'multiagentic' are not new concepts but new labels for intelligent agents and multi-agent systems, two areas AI researchers have studied for decades. It traces the word 'agentic' to social psychology and the philosophical idea of intentional systems, then reviews the classical definitions, properties, architectures, and communication standards of agents and multi-agent systems. The paper's practical claim is that the current wave of LLM-based agent frameworks is a new implementation layer on top of that old conceptual framework, not a paradigm shift. A sympathetic reader should care because if the claim is right, the field can accelerate by reusing existing MAS knowledge—coordination algorithms, negotiation, trust, standards—rather than rediscovering it.

What carries the argument

The load-bearing object is the classical intelligent agent abstraction: an autonomous system situated in an environment that runs a perceive-think-act cycle and exhibits autonomy, reactivity, proactivity, and social capability. The paper uses this abstraction as a translation key, showing that every property used to market 'agentic AI' has a direct counterpart in the classical definition, and that orchestrated LLM agents are a multi-agent system whose coordination, communication, and emergent behavior problems were already studied. This mapping does the argument's work: once the new vocabulary is translated into the old vocabulary without residue, the claim that the wheel is being reinvented follows.

What would settle it

A corpus study of vendor documentation, technical blogs, and papers from 2023–2025 would settle the claim: if a substantial fraction of 'agentic' usages denote something the classical agent definition does not cover—for example, a static orchestration workflow with no autonomous perception loop—the equivalence fails; if every usage reduces to autonomy, reactivity, proactivity, and social capability, the rebranding claim holds.

Watch

Extended reading notes

Core claim

The paper's central claim is that 'agentic AI' essentially denotes the classical intelligent agent—a system situated in an environment that perceives, decides, and acts autonomously, reactively, proactively, and socially—and that 'multiagentic' denotes a multi-agent system composed of interacting autonomous agents. It argues this by mapping the features used to market modern systems, such as proactive goal setting, tool use, and orchestrated subagents, onto the standard properties of intelligent agents and onto MAS coordination and communication problems. The paper acknowledges that LLM-based agents bring a genuine leap in flexibility and knowledge, because a large language model can parse arbitrary instructions and act as a general-purpose reasoning module, but it insists this is an implementation advance within the existing agent abstraction, not a new kind of system. The conclusion is prescriptive: call an agent an agent, call a multi-agent system a multi-agent system, and build on the existing literature instead of coining ambiguous terminology.

Load-bearing premise

The argument assumes that the few cited industry texts accurately represent how 'agentic' is actually used, and that those uses map without residue onto the classical definitions of intelligent agents and multi-agent systems; if current usage has settled on a narrower meaning tied to LLM-driven tool use, the rebranding claim is weakened.

Editorial extensions

If this is right

LLM-based agent frameworks should be designed and evaluated with reference to existing agent architectures and MAS literature rather than treated as a blank slate.
Inter-agent communication between LLM agents, now mostly ad-hoc natural language, can be expected to migrate toward structured, speech-act-inspired protocols as complexity grows.
Emergent behaviors and safety failures in LLM multi-agent systems are analyzable with existing MAS tools such as coordination mechanisms, game-theoretic models, and adjustable autonomy.
The proliferation of incompatible 'agentic' frameworks is the fragmentation problem the MAS community met in the 1990s, so standards initiatives align with a known lesson.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial extension: the paper's equivalence claim is based on a handful of cited industry texts, so its strength depends on whether those texts represent typical usage; a corpus study of vendor materials and technical posts would make the claim testable.
Editorial extension: if 'agentic' has begun to denote a narrower recipe—an LLM responsible for tool selection and orchestration in a workflow—then the mapping to classical agents is still mostly intact but the claim of pure rebranding becomes less clean.
Editorial extension: the paper implies a concrete research-program consequence: hybrid neurosymbolic agents, with an LLM handling perception and language and a symbolic planner handling commitments, are a natural next step that follows directly from combining classical BDI with generative AI.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Punchline: this is a clearly written opinion piece making a familiar point — agentic AI is mostly old wine in new bottles. It's not a research contribution, but it's a useful synthesis and a reasonable starting point for newcomers.

What it does well: the summary of classical agent properties, architectures, communication standards (FIPA, KQML), and agreement technologies is accurate and appropriately cited. The paper correctly notes that LLM-based agents still grapple with the same conceptual issues — autonomy, reactivity, proactivity, social capability — and it gives sensible practical advice: reuse MAS results, adopt standard terminology, and integrate LLMs into agent architectures rather than starting from scratch. It also credits the genuine leap in flexibility and knowledge that LLMs provide, so the argument is not a simple 'nothing new.'

Soft spots: the central equivalence claim rests on a small, non-representative sample of industry sources (IBM, Softude, Silfverskiöld). The stress-test note is right: three blog/vendor texts cannot establish what 'agentic AI' denotes across the field. The paper hedges with 'often' and admits 'not always,' but it never provides a systematic survey or quantitative usage analysis. If 'agentic AI' is evolving toward a distinct design pattern — LLM-driven tool-use loops with dynamic orchestration — the equivalence claim overstates; the paper's own 'framework vs building blocks' quote from IBM could be read as supporting a new abstraction rather than a simple mapping to agent/MAS. In addition, the Goffman/Fiske historical claim is uncited, and the text has editing issues (typos, repeated names). These are minor-to-moderate weaknesses, not fatal: the paper's core message — call systems by their established names and don't ignore decades of MAS results — survives even if the empirical usage claim is weakened.

Who it's for: developers and researchers building LLM-agent frameworks who have not read the MAS literature, and anyone writing position papers on agentic AI. It is not for agents specialists. It deserves a serious referee as a commentary or viewpoint article, provided the author either broadens the evidence base or softens the claim, and sources the historical origin. I would not cite it as evidence for the equivalence claim, but I would gladly use it in a reading group as a discussion starter on terminology and disciplinary memory.

Referee Report

4 major / 5 minor

Summary. This paper argues that the currently popular terms 'Agentic AI' and 'Multiagentic AI' are, in most cases, buzzwords for the well-established concepts of intelligent agents and multi-agent systems (MAS). The author reviews the psychological and philosophical roots of 'agentic' (Bandura, Dennett), summarizes classical agent definitions and architectures (reactive, deliberative, hybrid, BDI), and surveys recent LLM-based agent frameworks and industrial usage. The central claim is that LLM-based agent systems are new implementations within an old conceptual framework, and that practitioners should reuse established terminology, MAS standards, coordination algorithms, and agreement technologies rather than 'reinventing the wheel.' The paper concludes with recommendations for integrating classical agent research into the development of modern generative-AI-driven agents.

Significance. If the central claim is accepted, the paper serves a useful bridging function between the classical agents/MAS research community and the current LLM-agent development wave. Its strengths include a clear, concise historical survey of agent concepts and architectures, explicit identification of correspondences between new framework patterns (e.g., planner-executor, tool use) and classical MAS ideas, and constructive proposals to reuse communication standards, coordination algorithms, and agreement technologies. The paper also correctly emphasizes that the agent abstraction does not require LLMs. However, the significance is limited by the thin evidence base for the 'buzzword' claim: only three industry sources are cited for current usage, and the paper itself concedes variability. The argument is largely a position piece, and its persuasiveness depends on the reader's willingness to accept the author's interpretive mapping of 'Agentic AI' onto classical concepts.

major comments (4)

[From LLMs to 'Agentic AI' and Conclusion] The central claim is overstated relative to the paper's own concessions. In 'From LLMs to Agentic AI', the author writes that LLM-based systems 'really add value' in flexibility and knowledge, and that 'this wasn't possible with good reliability before deep learning, so the excitement around Agentic AI is justified.' Yet the Conclusion states that Agentic AI is 'essentially a rebranding of established concepts,' and the section 'Misuse of Terminology' claims 'nothing conceptually exotic is introduced.' These statements are in tension. If LLMs enable open-ended, general-purpose agents with tool use and natural-language reasoning, that is a substantive capability advance even if the underlying architecture maps to classical notions. The paper should either qualify the 'rebranding' claim (e.g., 'the core concepts are classical, but the capabilities are new') or explicitly define what would count as conceptual novelty, otherwise the central argument risks becoming internally inconsistent.
[Introduction and From LLMs to 'Agentic AI'] The evidence base for the claim that 'Agentic AI' and 'Multiagentic' are 'often used as a buzzword' is too thin. The Introduction itself concedes 'not always; there are many cases where the terminology is respected,' and the subsequent analysis of current usage relies essentially on three sources (IBM [7], Softude [8], Silfverskiöld [9]). No academic papers on LLM agents, framework documentation, or quantitative usage data are examined. As a result, the paper does not establish that the identified uses are representative or dominant. To support a general conclusion about widespread misuse, the author should either conduct a more systematic survey of current usage or explicitly limit the claim to 'some industry discourse' and avoid the generalized phrasing in the abstract and conclusion.
[From LLMs to 'Agentic AI'] The mapping from 'Agentic AI is the framework; AI agents are the building blocks' (IBM [7]) to 'multi-agent system vs. agent' is presented as straightforward, but it is an interpretive choice. In classical MAS, a multi-agent system is a collection of interacting autonomous agents (Wooldridge, 2002, [4]); a software framework is an abstraction for constructing agents, not itself a system of interacting agents. The IBM quote could instead be read as distinguishing an orchestration platform from the agents it hosts, which is not identical to the classical MAS definition. The paper should acknowledge that this is one plausible reading among others, and discuss alternative uses of 'agentic' (e.g., as a descriptor of a capability or a design pattern) rather than asserting an exact correspondence.
[The Origins of 'Agentic': Agency in Psychology and Philosophy] The historical etymology contains an unsupported and likely inaccurate claim: the paper states that 'the first reference to the term agentic' appears in the work of 'Richard H. Goffman and Susan Fiske in the 1970s,' with no citation, and then credits Bandura (1986) with popularizing it. The name 'Goffman' is likely a confusion with Erving Goffman, and the attribution to Fiske is not substantiated. This is not load-bearing for the main argument, but it is a factual error that should be corrected or removed, as the paper otherwise seeks to emphasize scholarly rigor.

minor comments (5)

[Introduction] Typo: 'far superiror to current technology' should be 'far superior.'
[Misuse of Terminology and Conceptual Conflation] The sentence 'Saying “multiagent system” is an unnecessary neologism when “multiagent system” is clear and well-defined' is self-contradictory; the author likely means 'multiagentic system' as the neologism. Please correct.
[From LLMs to 'Agentic AI'] Misspelling: 'Deepsek' should be 'DeepSeek,' and 'GraphChain' is likely a typo for 'LangGraph.'
[Toward Integration: Embracing the Agent Research Legacy in Modern AI] The Model Context Protocol (MCP) is attributed to 2023 in reference [12], but Anthropic's MCP was publicly announced in November 2024. Please update the reference and date.
[Misuse of Terminology and Conceptual Conflation] The paper uses the term 'multiagentic' (including in the title) while simultaneously arguing that the term is 'even more problematic' than 'agentic.' This is not a technical error, but it creates an unintended self-referential inconsistency; consider using 'multi-agent' or 'MAS' in the title.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's terminological-equivalence thesis is an interpretive claim supported by independent classical definitions and external vendor sources, with no fitted inputs or self-citation chain forcing the conclusion.

full rationale

This paper contains no derivation chain, no fitted parameters, and no prediction that could reduce by construction to its inputs. Its central claim—that 'Agentic AI' and 'Multiagentic' are rebrandings of intelligent agents and multi-agent systems—is an interpretive semantic thesis, not a mathematically or statistically forced result. The evidence for current usage comes from external industry sources (IBM [7], Softude [8], Silfverskiöld [9]) and from classical definitions (Wooldridge and Jennings [3], Russell and Norvig [5]) that are independent of the paper's conclusion. The author's self-citations ([23], [24], agreement technologies) are used only as background support for the value of prior MAS research and are not load-bearing for the equivalence claim; they do not define 'agentic AI' in terms of the paper's target conclusion. A possible weakness is that the sample of current usage is small and interpretive, and the mapping from vendor descriptions to classical concepts is selective rather than logically compelled. That is a correctness or evidence concern, not circularity, and the reader's own caveat about unmeasured usage confirms the absence of a by-construction reduction. The reasoning is therefore self-contained, and the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper introduces no fitted quantities and no invented entities. Its argument rests on a small number of background assumptions: the authority of classical agent definitions for judging new systems, the representativeness of the cited industry sources, and the accuracy of an uncited historical claim about the first use of "agentic".

assumptions (3)

domain assumption The classical definitions of intelligent agents (Wooldridge and Jennings 1995) and MAS are the appropriate standard for judging new LLM-based agent systems.
The paper's normative conclusion presupposes that old categories should govern new implementations; this is contested and not proven.
domain assumption Current industry usage of "Agentic AI" is represented by the blog posts cited as references [7,8,9].
The central equivalence claim relies on these examples; no systematic corpus or survey is provided.
domain assumption Bandura (1986) introduced the adjective "agentic" and Goffman/Fiske in the 1970s used it first.
Historical premise; the Bandura part is standard, but the Goffman/Fiske priority claim is uncited and unverified.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Agentic AI and Multiagentic: Are We Reinventing the Wheel?." pith.science (2026). https://pith.science/paper/XJLW3TSB

@misc{pith2026250601463,
  author       = {Pith},
  title        = {Pith review of: Agentic AI and Multiagentic: Are We Reinventing the Wheel?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XJLW3TSB}},
  note         = {Machine review of arXiv:2506.01463}
}

read the original abstract

The terms Agentic AI and Multiagentic AI have recently gained popularity in discussions on generative artificial intelligence, often used to describe autonomous software agents and systems composed of such agents. However, the use of these terms confuses these buzzwords with well-established concepts in AI literature: intelligent agents and multi-agent systems. This article offers a critical analysis of this conceptual misuse. We review the theoretical origins of "agentic" in the social sciences (Bandura, 1986) and philosophical notions of intentionality (Dennett, 1971), and then summarise foundational works on intelligent agents and multi-agent systems by Wooldridge, Jennings and others. We examine classic agent architectures, from simple reactive agents to Belief-Desire-Intention (BDI) models, and highlight key properties (autonomy, reactivity, proactivity, social capability) that define agency in AI. We then discuss recent developments in large language models (LLMs) and agent platforms based on LLMs, including the emergence of LLM-powered AI agents and open-source multi-agent orchestration frameworks. We argue that the term AI Agentic is often used as a buzzword for what are essentially AI agents, and AI Multiagentic for what are multi-agent systems. This confusion overlooks decades of research in the field of autonomous agents and multi-agent systems. The article advocates for scientific and technological rigour and the use of established terminology from the state of the art in AI, incorporating the wealth of existing knowledge, including standards for multi-agent system platforms, communication languages and coordination and cooperation algorithms, agreement technologies (automated negotiation, argumentation, virtual organisations, trust, reputation, etc.), into the new and promising wave of LLM-based AI agents, so as not to end up reinventing the wheel.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Agentic AI and Retrieval-Augmented Models in Straight-Through Underwriting
cs.AI 2026-07 conditional novelty 5.0 of 10

On 635 synthetic BOP applications, multi-agent Agentic RAG reaches 86.5% decision accuracy versus 77.6% single-LLM and 76.9% naive RAG, with largest gains on multi-step and missing-information cases.

Reference graph

Works this paper leans on

26 extracted references · 23 canonical work pages · cited by 1 Pith paper

[7]

Agentic AI vs AI Agents: Understanding the Difference

IBM – Think Blog. (2023). “Agentic AI vs AI Agents: Understanding the Difference”. IBM Think Topics. (Explains agentic AI as a framework vs AI agents as components). 16

work page 2023
[8]

AI Agents vs Agentic AI: The Key Differences Most People Don’t Know

Softude Blog. (2025). “AI Agents vs Agentic AI: The Key Differences Most People Don’t Know”. Softude, May 2025. (Compares definitions and characteristics of AI agents and agentic AI)

work page 2025
[9]

Agentic AI: Comparing New Open-Source Frameworks

Silfverskiöld, I. (2025). “Agentic AI: Comparing New Open-Source Frameworks”. (Medium article) April 15, 2025. (Survey of emerging agentic frameworks, e.g., LangGraph, AutoGen, etc., and discussion of design choices)

work page 2025
[4]

Wooldridge, M. (2002). An Introduction to Multi-Agent Systems. John Wiley & Sons. (Textbook covering MAS definitions, architectures, and examples)

work page 2002
[1]

Bandura, A. (1986). Social Foundations of Thought and Action: A Social Cognitive Theory. Prentice- Hall. (Introduced the term “agentic” in context of human agency)

work page 1986
[2]

Intentional Systems

Dennett, D. C. (1971). “Intentional Systems.” Journal of Philosophy, 68(2): 87–106. (Philosophical basis for attributing beliefs and desires to systems)

work page 1971
[3]

Intelligent agents: Theory and practice

Wooldridge, M., & Jennings, N. R. (1995). “Intelligent agents: Theory and practice.” The Knowledge Engineering Review, 10(2), 115–152. (Classical paper defining intelligent agents and their properties)

work page 1995
[5]

Russell, S., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th Edition). Pearson. (AI textbook defining agents and environments; Chapter 2 on intelligent agents)

work page 2020

Show all 26 references

[6]

The Age of AI has Begun

Gates, Bill. (2023). “The Age of AI has Begun” – GatesNotes Blog, Nov 9, 2023. (Discusses personal AI assistants and agents revolutionizing computing)

2023
[10]

International Foundation for Autonomous Agents and Multiagent Systems – http:// www.ifaamas.org (Professional association supporting the AAMAS conference and MAS research)

IFAAMAS. International Foundation for Autonomous Agents and Multiagent Systems – http:// www.ifaamas.org (Professional association supporting the AAMAS conference and MAS research)
[11]

Foundation for Intelligent Physical Agents – http://www.fipa.org (Standards organization for agent interoperability, est

FIPA. Foundation for Intelligent Physical Agents – http://www.fipa.org (Standards organization for agent interoperability, est. 1996)

1996
[12]

Model Context Protocol (MCP) – Standard proposal by Anthropic for connecting AI assistants with external systems

Anthropic (2023). Model Context Protocol (MCP) – Standard proposal by Anthropic for connecting AI assistants with external systems

2023
[13]

A Robust Layered Control System for a Mobile Robot

Brooks, R. (1986). “A Robust Layered Control System for a Mobile Robot.” IEEE Journal on Robotics and Automation, 2(1): 14–23. (Describes the subsumption architecture, a reactive agent design)

1986
[14]

Bratman, M. (1987). Intention, Plans, and Practical Reason. Harvard Univ. Press. (Philosophical work on which the BDI model is based)

1987
[15]

A roadmap of agent research and development

Jennings, N. R., Sycara, K., & Wooldridge, M. (1998). “A roadmap of agent research and development.” Autonomous Agents and Multi-Agent Systems, 1(1): 7–38. (Overview of the state of agent R&D in late 90s; covers architectures, applications, challenges)

1998
[16]

Agent-oriented programming, Artificial Intelligence, Volume 60, Issue 1, 1993, Pages 51-92, ISSN 0004-3702, https://doi.org/10.1016/0004-3702(93)90034-9

Yoav Shoham. Agent-oriented programming, Artificial Intelligence, Volume 60, Issue 1, 1993, Pages 51-92, ISSN 0004-3702, https://doi.org/10.1016/0004-3702(93)90034-9

1993 doi
[17]

John Searle, Speech Acts, Cambridge University Press 1969, ISBN 0-521-09626-X

1969
[18]

Austin, J. L. (1975). Urmson, J. O.; Sbisà, Marina. (eds.). How to do things with words (2nd ed.). Cambridge, Mass.: Harvard University Press. ISBN 978-0674411524

1975
[19]

KQML as an agent communication language

Finin, T.; Fritzson, R.; McKay, D.; McEntire, R. (1994). "KQML as an agent communication language". Proceedings of the third international conference on Information and knowledge management - CIKM '94. p. 456. doi:10.1145/191246.191322. ISBN 0897916743. S2CID 1129799

1994
[20]

http://www.fipa.org/repository/aclspecs.html
[21]

Knowledge Interchange Format Version 3.0 Reference Manual

Genesereth, Michael; Fikes, Richard (June 1992). "Knowledge Interchange Format Version 3.0 Reference Manual" (PDF). Stanford Logic Group Report. Logic-92-1. Stanford University. Retrieved 7 August 2014

1992
[22]

http://www.ksl.stanford.edu/software/ontolingua/
[23]

Sierra Garcia, C.; Botti Navarro, VJ.; Ossowski, DS. (2011). Agreement Computing. KI - Künstliche Intelligenz. 25(1):57-61. https://doi.org/10.1007/s13218-010-0070

2011 doi
[24]

(2012, November 28)

Ossowski, S., Sierra, C., & Botti, V. (2012, November 28). Agreement Technologies: A Computing Perspective. Law, Governance and Technology Series. Springer Netherlands. http://doi.org/10.1007/978-94-007-5583-3_1

2012 doi
[25]

https://dev.to/tomer_barak/self-refernce-in-ai-agents-2nlh
[26]

A Scalable Communication Protocol for Networks of Large Language Models

Samuele Marro, Emanuele La Malfa, Jesse Wright, Guohao Li, Nigel Shadbolt, Michael Wooldridge, Philip Torr. A Scalable Communication Protocol for Networks of Large Language Models. https://arxiv.org/html/2410.11905v1

arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [7]

Agentic AI vs AI Agents: Understanding the Difference

IBM – Think Blog. (2023). “Agentic AI vs AI Agents: Understanding the Difference”. IBM Think Topics. (Explains agentic AI as a framework vs AI agents as components). 16

work page 2023

[2] [8]

AI Agents vs Agentic AI: The Key Differences Most People Don’t Know

Softude Blog. (2025). “AI Agents vs Agentic AI: The Key Differences Most People Don’t Know”. Softude, May 2025. (Compares definitions and characteristics of AI agents and agentic AI)

work page 2025

[3] [9]

Agentic AI: Comparing New Open-Source Frameworks

Silfverskiöld, I. (2025). “Agentic AI: Comparing New Open-Source Frameworks”. (Medium article) April 15, 2025. (Survey of emerging agentic frameworks, e.g., LangGraph, AutoGen, etc., and discussion of design choices)

work page 2025

[4] [4]

Wooldridge, M. (2002). An Introduction to Multi-Agent Systems. John Wiley & Sons. (Textbook covering MAS definitions, architectures, and examples)

work page 2002

[5] [1]

Bandura, A. (1986). Social Foundations of Thought and Action: A Social Cognitive Theory. Prentice- Hall. (Introduced the term “agentic” in context of human agency)

work page 1986

[6] [2]

Intentional Systems

Dennett, D. C. (1971). “Intentional Systems.” Journal of Philosophy, 68(2): 87–106. (Philosophical basis for attributing beliefs and desires to systems)

work page 1971

[7] [3]

Intelligent agents: Theory and practice

Wooldridge, M., & Jennings, N. R. (1995). “Intelligent agents: Theory and practice.” The Knowledge Engineering Review, 10(2), 115–152. (Classical paper defining intelligent agents and their properties)

work page 1995

[8] [5]

Russell, S., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th Edition). Pearson. (AI textbook defining agents and environments; Chapter 2 on intelligent agents)

work page 2020