Pith. sign in

REVIEW 2 major objections 6 minor 71 references

Applying Cognitive Design Patterns to General LLM Agents

T0 review · 2 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Recurring cognitive design patterns from classical AI architectures can identify functional gaps in LLM agents and point to concrete research directions toward general intelligence.

desk verdict A useful position paper that gives the LLM-agent field a concrete comparative vocabulary; its main weakness is that the pattern-presence judgments driving its predictions are under-specified, but the predictions are framed as testable questions, not proven claims. read the letter →

arxiv 2505.07087 v2 pith:IT3OXCVK submitted 2025-05-11 cs.AI

Robert E. Wray , James R. Kirk , John E. Laird This is my paper · ORCID
classification cs.AI
keywords cognitivedesignpatternsarchitectureLLMagentsagenticAIepisodicmemoryknowledgecompilationcommitmentandreconsiderationReAct
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 contends that the recurring functional building blocks of pre-transformer cognitive architectures—processes like observe-decide-act, commitment and reconsideration, episodic memory with encoding specificity, and knowledge compilation—can serve as an analytic lens for today's LLM-based agents. By mapping well-known agentic systems such as ReAct and Generative Agents onto these patterns, the authors aim to show where current approaches are functionally incomplete and to generate concrete research questions. The point matters because it offers a principled way to predict what an LLM agent must add to move toward general intelligence, rather than relying on ad hoc prompt or tool engineering.

What carries the argument

The central object is the cognitive design pattern: an abstract, implementation-independent description of a function, process, or memory that recurs across agent and cognitive architectures, comparable to a software design pattern. Its analytical work is to make functionally equivalent mechanisms recognizable across very different systems—for instance, treating a truth-maintenance-based reconsideration process and a decision-theoretic intention reconsideration as instances of the same function. The paper deploys this machinery through a three-stage commitment cycle (candidate generation, selection or commitment, reconsideration), through the episodic-memory characteristics of encoding specificity and cue-based retrieval, and through the knowledge-compilation pattern of caching expensive reasoning results; these are then used to generate predictions for LLM agents.

What would settle it

Run a controlled benchmark comparing ReAct with a ReAct variant that adds an explicit commitment and reconsideration step, and separately compare a Generative-Agents-style memory using semantic-similarity retrieval against one with deliberate cue construction; if neither intervention improves outcomes, the claim that these pattern gaps limit LLM agents would be contradicted.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the functional decompositions accumulated in decades of cognitive-architecture research are not tied to specific implementations but recur across belief-desire-intention agents, ACT-R, and Soar; viewed this way, they provide a catalog of functions that should appear in any agent aimed at general intelligence. Applying that catalog to current LLM agents exposes specific missing pieces: ReAct implements observe-decide-act without an explicit commitment step, Generative Agents' memory lacks cue-based retrieval and encoding specificity, and neither reconsideration of commitments nor online knowledge compilation is part of the mainstream. The paper argues this is not coincidence but a consequence of omitting well-established cognitive functions.

Load-bearing premise

The load-bearing assumption is that the functional decomposition of classical cognitive architectures transfers to language-model agents at the same level of abstraction; if an LLM's behavior is emergent and cannot be meaningfully decomposed into these functional roles, the gap analysis loses its predictive value.

Editorial extensions

If this is right

  • Adding an explicit commitment step to ReAct-style observe-decide-act loops becomes a concrete, testable intervention predicted to improve reasoning outcomes over ReAct alone.
  • Adding deliberate retrieval-cue construction and context-specific encoding to LLM episodic memories should change when and what agents recall, beyond semantic-similarity relevance.
  • Reconsideration of commitments should be built into LLM agents that make explicit plans, allowing non-monotonic redirection instead of continuing a stale intention.
  • Online knowledge compilation—caching the results of expensive multi-step reasoning for later reuse—should become a central mechanism in large reasoning models, with utility problems emerging as a research topic.
  • Step-wise reflection, a newly identified pattern intrinsic to LLM computation, can be distinguished from metacognitive reflection and carries a distinctive risk of unbounded recursive self-evaluation.

Reading between the lines

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

  • If the mapping method is sound, it can serve as a generative checklist: any new agentic framework can be audited against the pattern catalog to predict which cognitive functions are missing before empirical testing.
  • A natural extension would test whether introducing encoding-specificity-style contextual cues in Generative-Agents-style retrieval changes long-horizon behavior, a prediction the paper only states as a research question.
  • The analysis implies that non-monotonic reasoning and commitment management, not just context length or tool use, will become binding constraints on LLM agent generality.
  • One could build a benchmark of pattern violations—for example, tasks where an agent must abandon a goal—to measure whether frameworks with explicit reconsideration outperform those without.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper introduces the notion of "cognitive design patterns"—abstract descriptions of recurring functions, processes, and representations found across classical cognitive architectures such as ACT-R, Soar, and BDI agents—and applies this lens to contemporary Agentic LLM systems. It analyzes ReAct and Generative Agents as case studies, argues that ReAct lacks an explicit commitment step, evaluates Generative Agents against a checklist of episodic-memory characteristics adapted from Nuxoll and Laird, and identifies reconsideration and knowledge compilation as under-explored patterns. It also proposes "step-wise reflection" as a candidate novel pattern. The central claim is that mapping Agentic LLM systems onto cognitive design patterns enables predictions of current gaps and points toward future research directions for general LLM agents.

Significance. If the pattern-mapping methodology were equipped with a reliable operational test for pattern presence, the framework would provide a useful comparative analytic tool for a fast-moving field where much work is ad hoc and disconnected from prior architecture research. The paper has real strengths: it draws on a substantial body of prior comparative architecture work, grounds the episodic-memory analysis in a well-known published criterion set (Table 3), and it issues concrete, falsifiable predictions, notably the question in §3.1 of whether adding explicit commitment to ReAct improves reasoning outcomes. It also names underexplored areas, such as online knowledge compilation, that are plausible research opportunities. The main weakness is that the central predictive claim currently rests on an under-specified mapping from systems to patterns, so the gap analyses are not yet reproducible or falsifiable as stated.

major comments (2)
  1. [§2 and §3.1] The paper gives no operational criterion for deciding whether a given Agentic LLM system instantiates a cognitive design pattern. The patterns are defined abstractly, "eliding not only implementation details, but specification of algorithms" (§2), and Table 2's footnote states that the table "does not distinguish full vs. partial realizations" (§3.1, footnote 3). Yet the entire ReAct analysis depends on precisely such a distinction: the paper asserts that ReAct "lacks the step that explicitly makes commitments" (§3.1). At a functional level, ReAct's action output can be read as a decision to execute one action, i.e., an implicit commitment made at every step; the absence of commitment is only meaningful if "commitment" is defined as a persistent, separately represented intention with an explicit commitment process. Nothing in the pattern definition fixes this level of abstraction. Because the paper's novelty is its predictive use of pattern-gap analysis, this is a load-bearing issue: without an independent, testable criterion for pattern presence, the predicted gap in ReAct and similar claims are not falsifiable from the current definitions. I recommend that the authors add an explicit definition of pattern instantiation at a chosen level of abstraction, with examples of how to verify presence or absence in an LLM-based system (e.g., by probe prompts, architectural inspection, or behavioral tests).
  2. [§3.1, Table 3] The scoring of Generative Agents in Table 3 is not derived from a stated evaluation procedure, and at least one score appears to rest on a design-description rather than on behavior. In particular, "Encoding specificity: Partial" is justified by the comment that retrieval "uses semantic similarity, not encoding specificity." But whether semantic similarity can implement cue-based context matching is an empirical question, not a definitional one: a semantic-similarity retrieval function over contextualized embeddings could, in principle, approximate encoding-specificity effects. Similarly, "Deliberate: No" is asserted on the basis that "[a]gents cannot deliberately attempt to construct cues or retrieve memories," but no experimental probe is described that would establish this inability over the space of prompts the system might receive. Since Table 3 is the paper's most detailed worked example of pattern-gap analysis, the absence of an evaluation protocol makes the gap claims vulnerable to the objection that they are artifacts of annotation granularity. I ask the authors to specify, for each row of Table 3, the evidence or test that would justify a Yes/Partial/No verdict.
minor comments (6)
  1. [§2] There is a typo in the sentence "An is illustrated in Figure 1" near the discussion of the 3-stage commitment pattern; it should read "As is illustrated in Figure 1."
  2. [§3.2] The sentence "LLMs alone are not consisent or reliable in producing non-monotonic reasoning steps" contains a misspelling of "consistent."
  3. [§3.2] The phrase "reconsideration for intentions or commitments in Agentic LLMs could allow an the agent to periodically assess" contains an extra article "an" before "the agent."
  4. [§3.1, Table 3] The "Autonoetic" row uses a question mark as its Table 3 value, but the table's legend does not define "?" as a distinct category from "Yes," "No," "Partial," and "Semi."
  5. [§3.2] The statement that "direct, online caching of LLM responses in natural language has not yet become widely researched" is in tension with the paper's own description of Reflexion, which "caches its reflections in a memory to be used in subsequent trials." If reflections are considered distinct from responses, that distinction should be stated explicitly; otherwise the sentence reads as a contradiction.
  6. [Footnote 2] The footnote contains a doubled article: "The the recognition of common patterns occurring across cognitive and agent architectures goes back many years."

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the pattern catalog is used as an analytic lens on external systems, and the claimed gap predictions are conditional research questions, not consequences of fitted parameters or self-citation chains.

full rationale

The paper contains no equations and no fitted parameters. Its central claim is that a catalog of recurring cognitive functions, drawn from prior comparative analyses of cognitive architectures (including some by the present authors), can be used to identify gaps in Agentic LLM systems. The evaluation targets external systems (ReAct and Generative Agents), and the pattern definitions are applied as an interpretive framework rather than derived from those systems. The ReAct "missing commitment" and Generative Agents "partial encoding specificity" conclusions are judgments about level of abstraction, and the questions they raise, such as whether introducing commitment in ReAct would improve reasoning outcomes, are empirical and not forced by the definitions. Self-citations appear, for example Table 3 is adapted from Nuxoll and Laird, but the underlying criteria, such as encoding specificity, are grounded in external cognitive psychology literature (Tulving and Thomson), and the cited source is not used to forbid alternatives or to make the gap conclusion true by stipulation. Any weakness in operationalizing pattern presence is a falsifiability or correctness concern, not a circularity concern.

Assumptions & free parameters 0 free parameters · 4 assumptions · 1 invented entities

The paper introduces no free parameters. Its argument rests on domain assumptions about the transferability of cognitive-architecture patterns to LLM agents and the representativeness of selected examples; these are stated but not tested. The one newly named pattern (step-wise reflection) is descriptive and lacks independent validation.

assumptions (4)
  • domain assumption The patterns abstracted from Soar, ACT-R, and BDI are meaningful functional units for general intelligence.
    Table 1 presents these as canonical patterns; the paper does not justify why these specific architectures are the right source set beyond citing prior comparative analyses.
  • domain assumption LLM-based agents can be decomposed and evaluated using the same functional criteria as symbolic cognitive architectures.
    Section 3 maps ReAct and Generative Agents to patterns without validating that this level of abstraction captures LLM behavior; this is the paper's central untested premise.
  • domain assumption The Nuxoll and Laird episodic memory criteria are the appropriate normative benchmark for LLM agent memory.
    Table 3 uses these criteria as a checklist; their applicability to LLM agents is assumed from prior architecture research.
  • domain assumption The selected example systems are representative of the Agentic LLM research mainstream.
    The paper acknowledges the list is purposely inclusive and not exhaustive (footnote 3), so representativeness is not demonstrated.
invented entities (1)
  • step-wise reflection
    purpose: A pattern where the LLM evaluates its own immediately preceding response to improve reliability.
    Proposed in Sec 3.3 as a novel pattern, but it is described as evident in existing techniques (e.g., self-consistency) and no independent falsifiable handle is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Applying Cognitive Design Patterns to General LLM Agents." pith.science (2026). https://pith.science/paper/IT3OXCVK

@misc{pith2026250507087,
  author       = {Pith},
  title        = {Pith review of: Applying Cognitive Design Patterns to General LLM Agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IT3OXCVK}},
  note         = {Machine review of arXiv:2505.07087}
}
read the original abstract

One goal of AI (and AGI) is to identify and understand specific mechanisms and representations sufficient for general intelligence. Often, this work manifests in research focused on architectures and many cognitive architectures have been explored in AI/AGI. However, different research groups and even different research traditions have somewhat independently identified similar/common patterns of processes and representations or "cognitive design patterns" that are manifest in existing architectures. Today, AI systems exploiting large language models (LLMs) offer a relatively new combination of mechanisms and representations available for exploring the possibilities of general intelligence. This paper outlines a few recurring cognitive design patterns that have appeared in various pre-transformer AI architectures. We then explore how these patterns are evident in systems using LLMs, especially for reasoning and interactive ("agentic") use cases. Examining and applying these recurring patterns enables predictions of gaps or deficiencies in today's Agentic LLM Systems and identification of subjects of future research towards general intelligence using generative foundation models.

Figures

Figures reproduced from arXiv: 2505.07087 by the authors.

Figure 1
Figure 1. Illustration of the 3-stage commit￾ment cognitive design pattern. Cognitive design patterns, like soft￾ware design patterns, are abstract, eliding not only implementation de￾tails, but specification of algorithms. Distinct algorithms and representa￾tional commitments are employed to instantiate a cognitive design pattern in different architectures. An is illus￾trated in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Contrasting Decision/Action cycles in the ReAct (left) and BDI (right) paradigms. BDI illustrates the Observe-Decide-Act cognitive design pattern. design pattern leads to an immediate, empirical question: Would introducing commitment in ReAct result in better overall reasoning outcomes than ReAct alone? We contend that one of the advantages of mapping Agentic LLM solutions to cognitive design patterns is identifying… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

71 extracted references · 58 canonical work pages

  1. [42]

    Transactions on Machine Learning Research2835-8856 (2023)

    Sumers, T.R., Yao, S., Narasimhan, K., Griffiths, T.L.: Cognitive Architectures for Language Agents. Transactions on Machine Learning Research2835-8856 (2023)

    work page 2023

  2. [1]

    Psychological review111(4), 1036 (2004)

    Anderson, J.R., Bothell, D., Byrne, M.D., Douglass, S., Lebiere, C., Qin, Y.: An integrated theory of the mind. Psychological review111(4), 1036 (2004)

    2004

  3. [2]

    In: Proc

    Besta, M., Blach, N., Kubicek, A., Gerstenberger, R., Podstawski, M., Gianinazzi, L., Gajda, J., Lehmann, T., Niewiadomski, H., Nyczyk, P., Hoefler, T.: Graph of Thoughts: Solving Elaborate Problems with Large Language Models. In: Proc. of 38th Annual AAAI Conference on Artificial Intelligence. AAAI Press, Vancouver (Feb 2024)

    work page 2024

  4. [3]

    Transactions on Machine Learning Research (Mar 2025)

    Bhambri, S., Verma, M., Kambhampati, S.: Do Think Tags Really Help LLMs Plan? A Critical Evaluation of ReAct-Style Prompting. Transactions on Machine Learning Research (Mar 2025)

    work page 2025

  5. [4]

    Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee,P.,Lee,Y.T.,Li,Y.,Lundberg,S.,Nori,H.,Palangi,H.,Ribeiro,M.T.,Zhang, Y.: Sparks of Artificial General Intelligence: Early experiments with GPT-4 (Apr 2023), http://arxiv.org/abs/2303.12712, arXiv:2303.12712 [cs]

    arXiv 2023

  6. [5]

    In: Proceedings of the 12th interna- tional joint conference on Artificial intelligence - Volume 1

    Bylander, T.: Complexity results for planning. In: Proceedings of the 12th interna- tional joint conference on Artificial intelligence - Volume 1. pp. 274–279. IJCAI’91, Morgan Kaufmann Publishers Inc., San Francisco, CA, USA (Aug 1991)

    work page 1991

  7. [6]

    In: 14th Conference on Behavior Representation in Modeling and Simulation 2005

    Cohen, M.A., Ritter, F.E., Haynes, S.R.: Herbal: A high-level language and devel- opment environment for developing cognitive models in Soar. In: 14th Conference on Behavior Representation in Modeling and Simulation 2005. pp. 231–236. Sim- ulation Interoperability Standards Organization (Dec 2005)

    work page 2005

  8. [7]

    In: Gluck, K

    Crossman, J., Wray, R.E., Jones, R.M., Lebiere, C.: A High Level Symbolic Repre- sentation for Behavior Modeling. In: Gluck, K. (ed.) Proceedings of 2004 Behavior Representation in Modeling and Simulation Conference. Arlington, VA (2004)

    work page 2004

Show all 71 references
  1. [8]

    DeepSeek-AI: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (Jan 2025), http://arxiv.org/abs/2501.12948, arXiv:2501.12948 [cs]

    2025 arXiv

  2. [9]

    ACM Trans

    Dong, Y., Jiang, X., Jin, Z., Li, G.: Self-Collaboration Code Generation via Chat- GPT. ACM Trans. Softw. Eng. Methodol.33(7), 189:1–189:38 (Sep 2024)

    2024

  3. [10]

    Artificial Intelligence12, 231–272 (1979)

    Doyle, J.: A truth maintenance system. Artificial Intelligence12, 231–272 (1979)

    1979

  4. [11]

    Artificial Intelligence3, 251–288 (Jan 1972)

    Fikes, R.E., Hart, P.E., Nilsson, N.J.: Learning and executing generalized robot plans. Artificial Intelligence3, 251–288 (Jan 1972)

    1972

  5. [12]

    In: Proc

    Gao, J., Ding, X., Cui, Y., Zhao, J., Wang, H., Liu, T., Qin, B.: Self-Evolving GPT: A Lifelong Autonomous Experiential Learner. In: Proc. of ACL 2024 (2024)

    2024

  6. [13]

    In: Van Lehn, K

    Genesereth, M.R.: A Comparative Analysis of Some Simple Architectures for Au- tonomous Agents. In: Van Lehn, K. (ed.) Architectures for Intelligence, pp. 279–

  7. [14]

    MIT Press, Cambridge, MA (1988)

    Gärdenfors, P.: Knowledge in Flux. MIT Press, Cambridge, MA (1988)

    1988

  8. [15]

    AI Magazine27(2), 57–70 (2006)

    Jones, R., Wray, R.E.: Comparative Analysis of Frameworks for Knowledge- Intensive Intelligent Agents. AI Magazine27(2), 57–70 (2006)

    2006

  9. [16]

    In: PMLR Vol 235: 41st International Conf

    Kambhampati, S., Valmeekam, K., Guan, L., Verma, M., Stechly, K., Bhambri, S., Saldyt, L.P., Murthy, A.B.: Position: LLMs Can’t Plan, But Can Help Planning in LLM-Modulo Frameworks. In: PMLR Vol 235: 41st International Conf. on Machine Learning. vol. 235, pp. 22895–22907. Vien...

    2024

  10. [17]

    Artificial Intelligence Review53(1), 17–94 (2020)

    Kotseruba, I., Tsotsos, J.K.: 40 years of cognitive architectures: Core cognitive abil- ities and practical applications. Artificial Intelligence Review53(1), 17–94 (2020)

    2020

  11. [18]

    MIT Press, Cambridge, MA (2012) 12 Robert E

    Laird, J.E.: The Soar Cognitive Architecture. MIT Press, Cambridge, MA (2012) 12 Robert E. Wray, James R. Kirk, and John E. Laird

    2012

  12. [19]

    (eds.): Proc

    Laird, J.E., Lebiere, C., Reitter, D., Rosenbloom, P.S., Stocco, A. (eds.): Proc. of 2023 AAAI Fall Symposia: Integration of Cognitive Architectures and Generative Models. AAAI Fall Symposium Series, AAAI Press, Washington, DC. (2023)

    2023

  13. [20]

    AI Magazine38(4), 13–26 (2017)

    Laird, J.E., Lebiere, C., Rosenbloom, P.S.: A Standard Model of the Mind: Toward a Common Computational Framework across Artificial Intelligence, Cognitive Sci- ence, Neuroscience, and Robotics. AI Magazine38(4), 13–26 (2017)

    2017

  14. [21]

    (eds.) The Soar Papers: Research on Integrated Intelligence, vol

    Laird,J.E.,Newell,A.:Auniversalweakmethod.In:Rosenbloom,P.S.,Laird,J.E., Newell, A. (eds.) The Soar Papers: Research on Integrated Intelligence, vol. 1, pp. 245–292. MIT Press, Cambridge, MA (1983)

    1983

  15. [22]

    Machine Learning1(1), 11–46 (1986)

    Laird, J.E., Rosenbloom, P.S., Newell, A.: Chunking in Soar: The anatomy of a general learning mechanism. Machine Learning1(1), 11–46 (1986)

    1986

  16. [23]

    LangChain: LangGraph (2025), https://langchain-ai.github.io/langgraph/

    2025

  17. [24]

    I Want It That Way

    Lawless, C., Schoeffer, J., Le, L., Rowan, K., Sen, S., St. Hill, C., Suh, J., Sar- rafzadeh, B.: “I Want It That Way”: Enabling Interactive Decision Support Using Large Language Models and Constraint Programming. ACM Trans. Interact. In- tell. Syst. 14(3), 22:1–22:33 (Sep 2024)

    2024

  18. [25]

    In: Ku, L.W., Martins, A., Srikumar, V

    Leidinger, A., Van Rooij, R., Shutova, E.: Are LLMs classical or nonmonotonic reasoners? Lessons from generics. In: Ku, L.W., Martins, A., Srikumar, V. (eds.) Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). pp. ...

    2024

  19. [26]

    Luo, J., Zhang, W., Yuan, Y., Zhao, Y., Yang, J., Gu, Y., Wu, B., Chen, B., Qiao, Z., Long, Q., Tu, R., Luo, X., Ju, W., Xiao, Z., Wang, Y., Xiao, M., Liu, C., Yuan, J., Zhang, S., Jin, Y., Zhang, F., Wu, X., Zhao, H., Tao, D., Yu, P.S., Zhang, M.: Large Language Model Agents:...

    2025 arXiv

  20. [27]

    Artificial Intelligence42(2), 363–391 (Mar 1990)

    Minton, S.: Quantitative results concerning the utility of explanation-based learn- ing. Artificial Intelligence42(2), 363–391 (Mar 1990)

    1990

  21. [28]

    Machine Learning1(1), 47–80 (1986)

    Mitchell, T.M., Keller, R.M., Kedar-Cabelli, S.V.: Explanation-based learning: A unifying view. Machine Learning1(1), 47–80 (1986)

    1986

  22. [29]

    In: Proceedings of the AAAI Conference on Artificial In- telligence

    Mohan, S., Laird, J.E.: Learning Goal-Oriented Hierarchical Tasks from Situated Interactive Instruction. In: Proceedings of the AAAI Conference on Artificial In- telligence. vol. 28, pp. 387–394. AAAI Press (2014)

    2014

  23. [30]

    Cog- nitive Systems Research17-18, 34–48 (Jul 2012)

    Nuxoll, A.M., Laird, J.E.: Enhancing intelligent agents with episodic memory. Cog- nitive Systems Research17-18, 34–48 (Jul 2012)

    2012

  24. [31]

    OpenAI: OpenAI o1 System Card (Dec 2024), http://arxiv.org/abs/2412.16720, arXiv:2412.16720 [cs]

    2024 arXiv

  25. [32]

    Packer, C., Wooders, S., Lin, K., Fang, V., Patil, S.G., Stoica, I., Gon- zalez, J.E.: MemGPT: Towards LLMs as Operating Systems (Feb 2024), http://arxiv.org/abs/2310.08560, arXiv:2310.08560 [cs]

    2024 arXiv

  26. [33]

    In: Proceedings of 2023 ACM Symposium on User Interface Software and Technology (UIST)

    Park, J.S., O’Brien, J.C., Cai, C.J., Morris, M.R., Liang, P., Bernstein, M.S.: Gen- erative Agents: Interactive Simulacra of Human Behavior. In: Proceedings of 2023 ACM Symposium on User Interface Software and Technology (UIST). pp. 1–22. Association for Computing Machinery, ...

    2023

  27. [34]

    Pink, M., Wu, Q., Vo, V.A., Turek, J., Mu, J., Huth, A., Toneva, M.: Position: Episodic Memory is the Missing Piece for Long-Term LLM Agents (Feb 2025), http://arxiv.org/abs/2502.06975, arXiv:2502.06975 [cs]

    2025 arXiv

  28. [35]

    In: Applying Cognitive Design Patterns to General LLM Agents 13 Duh, K., Gomez, H., Bethard, S

    Prasad, A., Koller, A., Hartmann, M., Clark, P., Sabharwal, A., Bansal, M., Khot, T.: ADaPT: As-Needed Decomposition and Planning with Language Models. In: Applying Cognitive Design Patterns to General LLM Agents 13 Duh, K., Gomez, H., Bethard, S. (eds.) Findings of the Associ...

    2024

  29. [36]

    In: 2024 2nd International Conference on Foundation and Large Lan- guage Models (FLLM)

    Renze, M., Guven, E.: Self-Reflection in LLM Agents: Effects on Problem-Solving Performance. In: 2024 2nd International Conference on Foundation and Large Lan- guage Models (FLLM). pp. 476–483. Dubai (Nov 2024)

    2024

  30. [37]

    Journal of Experimental & Theoretical Artificial Intelligence16(4), 261–293 (Oct 2004)

    Schut, M., Wooldridge, M., Parsons, S.: The theory and practice of intention re- consideration. Journal of Experimental & Theoretical Artificial Intelligence16(4), 261–293 (Oct 2004)

    2004

  31. [38]

    In: Proc

    Shao, Y., Jiang, Y., Kanell, T.A., Xu, P., Khattab, O., Lam, M.S.: Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models. In: Proc. of 2024 Conf. of the North American Chapter of the Association for Compu- tational Linguistics: Human Language Tec...

    2024

  32. [39]

    In: 37th Conference on Neural Information Processing Systems (Mar 2023)

    Shinn, N., Cassano, F., Berman, E., Gopinath, A., Narasimhan, K., Yao, S.: Re- flexion: Language Agents with Verbal Reinforcement Learning. In: 37th Conference on Neural Information Processing Systems (Mar 2023)

    2023

  33. [40]

    In: Twenty-Ninth International Joint Conference on Artificial Intelligence

    Silva, L.d., Meneguzzi, F., Logan, B.: BDI Agent Architectures: A Survey. In: Twenty-Ninth International Joint Conference on Artificial Intelligence. vol. 5, pp. 4914–4921. Yokohama/Virtual (Jul 2020)

    2020

  34. [41]

    Autonomous Robots (Aug 2023)

    Singh, I., Blukis, V., Mousavian, A., Goyal, A., Xu, D., Tremblay, J., Fox, D., Thomason, J., Garg, A.: ProgPrompt: program generation for situated robot task planning using large language models. Autonomous Robots (Aug 2023)

    2023

  35. [43]

    Sypherd, C., Belle, V.: Practical Considerations for Agentic LLM Systems (Dec 2024), http://arxiv.org/abs/2412.04093, arXiv:2412.04093 [cs]

    2024 arXiv

  36. [44]

    Human Factors: The Journal of the Human Factors and Ergonomics Society 45(1), 61–76 (2003)

    Taatgen, N.A., Lee, F.J.: Production compilation: A simple mechanism to model complex skill acquisition. Human Factors: The Journal of the Human Factors and Ergonomics Society 45(1), 61–76 (2003)

    2003

  37. [45]

    Machine Learning5, 299–348 (1990)

    Tambe, M., Newell, A., Rosenbloom, P.S.: The problem of expensive chunks and its solution by restricting expressiveness. Machine Learning5, 299–348 (1990)

    1990

  38. [46]

    In: Gluck, K

    Taylor, G., Wray, R.E.: Behavior Design Patterns: Engineering Human Behavior Models. In: Gluck, K. (ed.) 2004 Behavioral Representation in Modeling and Sim- ulation Conference. Arlington, VA (2004)

    2004

  39. [47]

    Oxford University Press (1983)

    Tulving, E.: Elements of Episodic Memory. Oxford University Press (1983)

    1983

  40. [48]

    Psychological Review80(5), 352–373 (1973)

    Tulving, E., Thomson, D.M.: Encoding specificity and retrieval processes in episodic memory. Psychological Review80(5), 352–373 (1973)

    1973

  41. [49]

    Transactions on Machine Learning Research (2024)

    Wang, G., Xie, Y., Jiang, Y., Mandlekar, A., Xiao, C., Zhu, Y., Fan, L., Anandku- mar, A.: Voyager: An Open-Ended Embodied Agent with Large Language Models. Transactions on Machine Learning Research (2024)

    2024

  42. [50]

    In: Advances in Neural Information Processing Systems

    Wang, P., Li, Z., Zhang, N., Xu, Z., Yao, Y., Jiang, Y., Xie, P., Huang, F., Chen, H.: WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Models. In: Advances in Neural Information Processing Systems. vol. 37, pp. 53764–53797. Vancouver (Dec 2024)

    2024

  43. [51]

    arXiv (Nov 2022)

    Wang, R., Jansen, P., Côté, M.A., Ammanabrolu, P.: ScienceWorld: Is your Agent Smarter than a 5th Grader? In: EMNLP 2022. arXiv (Nov 2022)

    2022

  44. [52]

    In: Proceedings of the 37th International Conference on Neural Information Processing Systems

    Wang, W., Dong, L., Cheng, H., Liu, X., Yan, X., Gao, J., Wei, F.: Augmenting language models with long-term memory. In: Proceedings of the 37th International Conference on Neural Information Processing Systems. pp. 74530–74543. NIPS ’23, Curran Associates Inc., Red Hook, NY, ...

    2023

  45. [53]

    In: ICLR 2023

    Wang, X., Wei, J., Schuurmans, D., Le, Q., Chi, E., Narang, S., Chowdhery, A., Zhou, D.: Self-Consistency Improves Chain of Thought Reasoning in Language Models. In: ICLR 2023. arXiv (Mar 2023)

    2023

  46. [54]

    Proceedings of the AAAI Conference on Artificial Intelligence38(17), 19189–19196 (Mar 2024)

    Wang,Y.,Chu,Z.,Ouyang,X.,Wang,S.,Hao,H.,Shen,Y.,Gu,J.,Xue,S.,Zhang, J., Cui, Q., Li, L., Zhou, J., Li, S.: LLMRG: Improving Recommendations through Large Language Model Reasoning Graphs. Proceedings of the AAAI Conference on Artificial Intelligence38(17), 19189–19196 (Mar 2024)

    2024

  47. [55]

    In: NeurIPS 2022

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q.V., Zhou, D.: Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. In: NeurIPS 2022. vol. 35, pp. 24824–24837 (2022)

    2022

  48. [56]

    MIT Press, Cambridge, MA (2000)

    Wooldridge, M.J.: Reasoning about Rational Agents. MIT Press, Cambridge, MA (2000)

    2000

  49. [57]

    Computer Networks and ISDN Systems 28, 167–178

    Wray, R.E., Chong, R., Phillips, J., Rogers, S., Walsh, W., Laird, J.E.: Organizing information in Mosaic: A classroom experiment. Computer Networks and ISDN Systems 28, 167–178. See: http://ai.eecs.umich.edu/cogarch0 (1995)

    1995

  50. [58]

    Journal of Artificial Intelligence Research19, 355–398 (2003)

    Wray, R.E., Laird, J.E.: An architectural approach to consistency in hierarchical execution. Journal of Artificial Intelligence Research19, 355–398 (2003)

    2003

  51. [59]

    In: Proceedings of the Conference on Language Modeling 2024

    Wu, Q., Bansal, G., Zhang, J., Wu, Y., Li, B., Zhu, E.E., Jiang, L., Zhang, X., Zhang, S., Awadallah, A., White, R.W., Burger, D., Wang, C.: AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation. In: Proceedings of the Conference on Language Modeling 2024. P...

    2024

  52. [60]

    Wu, Y., Liang, S., Zhang, C., Wang, Y., Zhang, Y., Guo, H., Tang, R., Liu, Y.: From Human Memory to AI Memory: A Survey on Memory Mechanisms in the Era of LLMs (Apr 2025), http://arxiv.org/abs/2504.15965, arXiv:2504.15965 [cs]

    2025 arXiv

  53. [61]

    In: ICLR 2025 Workshops: Reasoning and Planning for LLMs

    Wu, Y., Hu, Z.: LLMs Are Not Good Strategists, Yet Memory-Enhanced Agency Boosts Reasoning. In: ICLR 2025 Workshops: Reasoning and Planning for LLMs. Singapore (Mar 2025)

    2025

  54. [62]

    Xu, F., Hao, Q., Zong, Z., Wang, J., Zhang, Y., Wang, J., Lan, X., Gong, J., Ouyang, T., Meng, F., Shao, C., Yan, Y., Yang, Q., Song, Y., Ren, S., Hu, X., Li, Y., Feng, J., Gao, C., Li, Y.: Towards Large Reasoning Models: A Survey of Reinforced Reasoning with Large Language Mo...

    2025 arXiv

  55. [63]

    Xu, W., Liang, Z., Mei, K., Gao, H., Tan, J., Zhang, Y.: A-MEM: Agentic Memory for LLM Agents (Feb 2025), http://arxiv.org/abs/2502.12110, arXiv:2502.12110 [cs]

    2025 arXiv

  56. [64]

    In: Advances in Neural Information Processing Systems

    Xue, S., Huang, Z., Liu, J., Lin, X., Ning, Y., Jin, B., Li, X., Liu, Q.: Decompose, Analyze and Rethink: Solving Intricate Problems with Human-like Reasoning Cy- cle. In: Advances in Neural Information Processing Systems. vol. 37, pp. 357–385 (Dec 2024)

    2024

  57. [65]

    Yao, S., Yu, D., Zhao, J., Shafran, I., Griffiths, T.L., Cao, Y., Narasimhan, K.: Tree of Thoughts: Deliberate Problem Solving with Large Language Models (May 2023), http://arxiv.org/abs/2305.10601, arXiv:2305.10601 [cs]

    2023 arXiv

  58. [66]

    In: The Eleventh Interna- tional Conference on Learning Representations, ICLR 2023

    Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K.R., Cao, Y.: ReAct: Synergizing Reasoning and Acting in Language Models. In: The Eleventh Interna- tional Conference on Learning Representations, ICLR 2023. Kigali, Rwanda (2023)

    2023

  59. [67]

    Zaharia, M., Khattab, O., Chen, L., Davis, J.Q., Miller, H., Potts, C., Zou, J., Carbin, M., Frankle, J., Rao, N., Ghodsi, A.: The Shift from Models to Compound AI Systems (Feb 2024), http://bair.berkeley.edu/blog/2024/02/18/compound-ai- systems/ Applying Cognitive Design Patt...

    2024

  60. [68]

    Proceedings of the AAAI Conference on Artificial In- telligence 38(17), 19632–19642 (Mar 2024)

    Zhao, A., Huang, D., Xu, Q., Lin, M., Liu, Y.J., Huang, G.: ExpeL: LLM Agents Are Experiential Learners. Proceedings of the AAAI Conference on Artificial In- telligence 38(17), 19632–19642 (Mar 2024)

    2024

  61. [69]

    Proceedings of the AAAI Conference on Artificial Intelligence38(17), 19724–19731 (Mar 2024)

    Zhong, W., Guo, L., Gao, Q., Ye, H., Wang, Y.: MemoryBank: Enhancing Large Language Models with Long-Term Memory. Proceedings of the AAAI Conference on Artificial Intelligence38(17), 19724–19731 (Mar 2024)

    2024

  62. [70]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Zhu, F., Simmons, R.: Bootstrapping Cognitive Agents with a Large Language Model. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 38, pp. 655–663. AAAI Press, Vancouver (Mar 2024)

    2024

  63. [299]

    Lawrence Erlbaum, Hillsdale NJ (1991)

    1991

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.