35 lines
3.0 KiB
Markdown
35 lines
3.0 KiB
Markdown
type:: [[REVIEWS]]
|
|
tags::
|
|
year:: 2025
|
|
venue:: [[MODELS]]
|
|
full-title:: A Model-Driven Approach to LLM Agent Design
|
|
date-start:: [[05-05-2025]] - 16:45
|
|
date-submitted::
|
|
external-links::
|
|
status:: [[DONE]]
|
|
deadline-submission::
|
|
file:: [[@MODELS_2025_paper_19]]
|
|
parent::
|
|
todoist:: https://app.todoist.com/app/task/llma-uml-extending-the-uml-for-modeling-llm-agent-systems-6Xh6hgr37HGMWJjg
|
|
|
|
- ### [[Highlights]]
|
|
- ### [[Comments]]
|
|
- #.tabular
|
|
- ### Paper summary
|
|
- The paper presents an overview of technologies supporting multi-agent systems based on Large Language Models (LLMs). Authors propose a taxonomy consisting of different models distinguishing structural components from cognitive modules. Five different frameworks for developing multi-agent systems are compared with respect to the proposed taxonomy,
|
|
- ### Strengths
|
|
- + The paper is about an interesting and relevant topic
|
|
- + Overall, it is well written
|
|
- ### Weaknesses
|
|
- - Not always rigorous
|
|
- - Most of the content relies on non-peer-reviewed sources
|
|
- ### Detailed comments for authors
|
|
- The paper is about a relevant and interesting topic. Unexpert readers can learn a lot by reading the paper, even though I have some concerns that are related to the following issues:
|
|
- The paper is presented as proposing a "Model-Driven Approach," but its content is largely a survey of existing technologies and taxonomies rather than a distinct approach or methodology.
|
|
- The paper claims to provide practical guidelines and flexible scaffolds for composing architectures but lacks concrete details. The discussion on decomposing monolithic architectures into modular components remains abstract, limiting practical applicability.
|
|
- While the authors introduce nested chats as a mechanism for managing agentic interactions, the paper does not adequately address governance or mechanisms for resolving potential conflicts or ambiguities that may arise during collaboration or delegation.
|
|
- The authors mention specific agentic frameworks (e.g., Autogen, OpenAI Agents SDK, CrewAI, PydanticAI, LangGraph) without clarifying the selection criteria. Understanding these criteria is essential to appreciate the representativeness and comprehensiveness of the survey.
|
|
- A significant portion of references (34 out of 50) comes from arXiv. Although the topic evolves rapidly, it is essential to rely predominantly on peer-reviewed sources to strengthen the scientific robustness of the claims and taxonomy proposed.
|
|
- Minor
|
|
- Figure 5 lacks a legend explaining the color schema used. Including this would improve the readability and interpretability of the visual information presented.
|
|
- To summarize, the paper addresses a timely and relevant topic, providing a useful overview and taxonomy. However, the identified concerns, especially the mismatch between the title and the actual content, vague methodological guidance, and reliance on non-peer-reviewed sources, must be addressed to improve clarity, rigor, and utility of the paper. |