Agile Software Management with Cognitive Multi-Agent Systems

Konrad Cinkusz

and Jarosław A. Chudziak

Faculty of Electronics and Information Technology, Warsaw University of Technology, Warsaw, Poland

{konrad.cinkusz.stud, jaroslaw.chudziak}@pw.edu.pl

Keywords:

Agile Methodologies, Multi-Agent Systems, Large Language Models, Software Project Management,

MAS-LLM Integration.

Abstract:

This paper explores the integration of cognitive agents powered by Large Language Models (LLMs) into

software project management within the Scaled Agile Framework (SAFe). We introduce the CogniSim frame-

work, an ecosystem where virtual agents operate in a simulated software environment to fulﬁll key roles in IT

project development. Emphasis is placed on the adaptability of these agents to the Scrum methodology, partic-

ularly in decision-making and problem-solving. By combining LLMs with Multi-Agent Systems (MAS), we

focus on improvements in project management, development processes, and Agile methodologies. Through

simulations and case studies, we demonstrate advancements in task delegation, communication, and project

lifecycle management, highlighting the potential of LLM-augmented MAS to manage software projects with

increased precision and intelligence. Our ﬁndings provide insights into essential components for an effective

cognitive multi-agent ecosystem, including Dynamic Context techniques and Theory of Mind for enhanced

agent collaboration, laying the groundwork for future research in this ﬁeld.

1 INTRODUCTION

The complexity and scale of modern software sys-

tems necessitate advanced approaches to software en-

gineering. Agile methodologies like SAFe have be-

come standard practices, emphasizing iterative de-

velopment, customer collaboration, and ﬂexibility

(Dingsøyr et al., 2012). However, effectively manag-

ing large-scale, complex projects within these frame-

works remains challenging (Perkusich et al., 2020).

Multi-Agent Systems offer a promising solution

to these challenges (Talebirad and Nadiri, 2023;

Chudziak and Wawer, 2024). MAS consist of

networks of autonomous agents that collaborate to

achieve deﬁned objectives within their environment

(Cruz, 2024). In software engineering, each agent

can manage speciﬁc aspects of the development life-

cycle, such as requirements gathering, code gener-

ation, testing, or deployment. The characteristics

of MAS—reactivity, proactiveness, and social abili-

ties—make them well-suited for managing complex

and dynamic software engineering environments (Li

et al., 2023).

Simultaneously, Large Language Models, such as

GPT-4, have transformed natural language process-

https://orcid.org/0009-0001-5709-1172

https://orcid.org/0000-0003-4534-8652

ing by generating human-like content across various

formats (Guo et al., 2023). In software engineer-

ing, LLMs can automate routine tasks like code com-

pletion, documentation, and debugging, reducing er-

rors and boosting productivity (Barua, 2024). In-

tegrating LLMs with MAS creates cognitive multi-

agent ecosystems that leverage the strengths of both

technologies (Singhal et al., 2023). For instance,

agents responsible for code generation can utilize

LLMs to remain aligned with the latest programming

paradigms and libraries, ensuring that the software re-

mains current and resilient (Guo et al., 2023).

Effective collaboration among agents is crucial in

these ecosystems. Orchestration and choreography

paradigms offer strategies for coordinating agent in-

teractions (Cruz, 2024), while frameworks like FIPA

(IEEE Foundation for Intelligent Physical Agents

(FIPA), 2012) provide speciﬁcations for agent com-

munication and interoperability. Techniques from

symbolic knowledge management, such as nonmono-

tonic logic and belief logics, enable agents to handle

incomplete and evolving information, enhancing their

decision-making capabilities.

PASSI (Process for Agent Societies Speciﬁcation

and Implementation), shown in Figure 1, is a com-

prehensive methodology designed to guide the devel-

opment of FIPA-compliant multi-agent systems from

Cinkusz, K. and Chudziak, J. A.

Agile Software Management with Cognitive Multi-Agent Systems.

DOI: 10.5220/0013153000003890

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 1, pages 385-392

ISBN: 978-989-758-737-5; ISSN: 2184-433X

385

Figure 1: The PASSI design process (IEEE Foundation for Intelligent Physical Agents (FIPA), 2012).

initial requirements to code implementation. It inte-

grates object-oriented software engineering principles

with agent-based system design, offering a structured

approach for building systems that involve peer-to-

peer agent interactions. By referencing PASSI, we

situate our framework within a lineage of method-

ologies that emphasize standardized communication

protocols and semantic interactions—ensuring inter-

operability, scalability, and adherence to established

best practices. Our approach extends these funda-

mentals by leveraging Large Language Models, en-

abling cognitive agents to augment traditional PASSI-

based design principles with enhanced adaptability,

richer context-awareness, and more ﬂexible decision-

making capabilities within complex, evolving soft-

ware environments.

Reinforcement learning further improves agent

collaboration by allowing agents to learn optimal be-

haviors through trial and error (Sutton and Barto,

1998), helping them adapt to new challenges and opti-

mize interactions in dynamic, unpredictable environ-

ments (Du et al., 2023). Dynamic Context techniques

enhance the capabilities of LLM-augmented MAS by

allowing agents to adapt their behavior based on real-

time data (Du et al., 2024), while incorporating the

Theory of Mind (Li et al., 2023; Kosinski, 2024) en-

ables agents to predict and understand the actions and

intentions of others, improving overall cooperation

(Kim et al., 2024).

In this paper, we introduce the CogniSim frame-

work, where cognitive agents powered by LLMs col-

laborate within a simulated software environment, as-

suming key Agile roles in product management, ar-

chitecture, development, and testing. The frame-

work aligns with Agile methodologies, focusing on

the Scaled Agile Framework (Scaled Agile, Inc.,

2024) and emphasizing the adaptability of agents to

Scrum. By simulating complex workﬂows and envi-

ronments, including potential integration with LeSS

(LeSS Company B.V., 2024), DaD (Project Man-

agement Institute, 2024), and Nexus (Scrum.org,

2024), CogniSim enables efﬁcient decision-making

and context-aware actions. Through simulations and

case studies, we highlight how the integration of

LLMs and MAS could improve task delegation, com-

munication, and lifecycle management, illustrating

the potential to manage software projects with in-

creased precision and intelligence.

2 FRAMEWORK OVERVIEW

Managing modern software projects requires inno-

vative solutions that address the growing demands

for adaptability, coordination, and efﬁciency. The

CogniSim framework provides such a solution, in-

tegrating cognitive agents powered by Large Lan-

guage Models within a Multi-Agent System to opti-

mize project management processes in Agile method-

ologies like Scrum and SAFe.

2.1 Concept of CogniSim

CogniSim could operate within a multi-layered sys-

tem architecture that integrates both internal and

external software ecosystems, forming a cohesive

bridge between them. As shown in Figure 2, Cog-

niSim interacts with external systems through various

APIs, ensuring the cognitive agents can access real-

time data and make informed decisions.

Within the internal environment, agents are struc-

tured into distinct roles, each reﬂecting critical re-

sponsibilities commonly found in Agile software

management. Product Owners manage backlog prior-

itization, DevOps Engineers oversee CI/CD pipelines,

and the Development Team implements features

based on user stories. By augmenting these roles with

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

386

LLM-driven capabilities, CogniSim automates rou-

tine tasks and optimizes efﬁciency.

Figure 2: Multilayer concept showing how system is work-

ing with external environment.

2.2 Design of CogniSim

CogniSim integrates cognitive agents, driven by

LLMs, to emulate human-like decision-making, au-

tomate tasks, and streamline workﬂows through in-

telligent collaboration. Each agent learns and adapts

based on environmental interactions, improving over

time via machine learning techniques. By simulating

human team dynamics, these agents enhance commu-

nication, coordination, and overall project efﬁciency

(Li et al., 2023).

2.3 Components of the CogniSim

Framework

Key components include:

• Cognitive Agents. AI-driven entities capable of

processing natural language, learning from data,

and interacting with other agents and stakehold-

ers.

• Communication Protocols. Standardized proto-

cols for message exchange and coordination.

• Decision-Making. Agents select optimal ac-

tions using rule-based, data-driven, or hybrid ap-

proaches.

• Collaboration Tools. Interfaces that facilitate in-

teraction among agents and between agents and

human team members.

2.4 Integration with Agile

Methodologies

CogniSim enhances Agile practices by automating

routine tasks and offering data-driven insights. In a

SAFe context, CogniSim coordinates multiple teams,

manages dependencies, and aligns work with the

project vision. Figures 3 and 4 show how agents col-

laborate with human stakeholders during Program In-

crement (PI) preparation and Scrum iterations. Cog-

nitive agents, representing roles traditionally held by

humans, assist in backlog reﬁnement, sprint planning,

code implementation, and quality assurance. Over

time, agents improve their performance by actively

learning from feedback loops provided during the de-

velopment process.

2.5 Roles and Responsibilities of Agents

Figure 3: Role of cognitive agents in the preparation phase

for Program Increment (PI).

In CogniSim, cognitive agents simulate roles such as:

• Project Manager: Manages communication with

clients, ensuring requirements are captured and

prioritized.

• DevOps Engineer: Maintains CI/CD pipelines,

deployment processes, and collaborates with QA.

• QA/Test Engineer: Deﬁnes test cases, automates

tests, and ensures product quality.

• Development Team: Implements features, coor-

dinates with UX and system teams, and integrates

feedback.

• UX Designer: Designs user interfaces, working

with development to ensure usability.

Agile Software Management with Cognitive Multi-Agent Systems

387

• System Team: Maintains build/testing environ-

ments, ensuring integration efforts run smoothly.

• Customer Representatives: Provide continuous

feedback throughout development.

This automation frees human teams to focus on

strategic activities, while agents handle routine tasks

and offer real-time insights.

2.6 Beneﬁts of CogniSim

Integrating cognitive agents yields several beneﬁts:

• Enhanced Decision-Making: Data-driven in-

sights improve choices throughout the project life-

cycle.

• Increased Efﬁciency: Automating routine tasks

frees humans for more strategic work.

• Improved Collaboration: Natural language pro-

cessing and role simulation improve communica-

tion.

• Scalability: Coordinating multiple teams and

managing dependencies supports large-scale Ag-

ile implementations.

2.7 Use Case Illustration

In a virtual environment, cognitive agents manage

project management, DevOps, QA, and development

roles. These agents automate code generation, testing,

and deployment while adhering to Agile practices like

Scrum and SAFe, optimizing workﬂows and reduc-

ing time to market. Figure 4 shows cognitive agents’

roles during a Scrum iteration, ensuring continuous

integration, feedback, and quality deliverables.

Figure 4: Cognitive agents’ roles during Scrum iteration ex-

ecution.

3 METHODOLOGY

This section outlines the methodology employed to

integrate cognitive agents powered by Large Lan-

guage Models into Agile software project manage-

ment using CogniSim. We describe the agent-based

system design, the development environment, and

how virtual agents simulate key roles within SAFe.

3.1 Structure of a Single Virtual Agent

Each virtual agent in CogniSim represents a self-

contained unit powered by an AI-driven LLM core.

Figure 5 illustrates the structure of a typical agent.

Figure 5: Structure and functionalities of a single virtual

agent in CogniSim.

Agents perform:

• Task Management: Organizing and prioritizing

tasks based on requirements.

• Decision Making: Using LLM capabilities to

evaluate solutions and provide data-driven deci-

sions.

• Quality Assurance: Automated checks ensuring

outputs meet quality standards.

• Review and Feedback: Offering insights on per-

formance and outcomes for continuous improve-

ment.

This modular approach supports scalability, ﬂexi-

bility, and improved efﬁciency throughout the project

lifecycle.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

388

Figure 6: CogniSim MAS powered by LLMs (Cinkusz and

Chudziak, 2024).

3.2 Agent-Based System Design

The CogniSim framework, as shown in Figure 6,

employs agents driven by advanced LLMs to simu-

late human-like interactions and decisions, automat-

ing tasks traditionally handled by humans (Huang

et al., 2024). These agents operate autonomously, col-

laborating within a MAS to address all phases of the

software lifecycle, from backlog reﬁnement to risk as-

sessment. Figure 7 illustrates how the MAS powered

by LLMs processes tasks through agent interactions.

Figure 7: Multi-Agent System powered by LLMs contain-

ing simple tasks from a client and a supervisor agent outside

the MAS for checking compatibility.

3.3 Development Platform and Tools

CogniSim is implemented in Python, using Ope-

nAI’s GPT-4 and GPT-3.5 models, and LangChain

(LangChain, 2023) for LLM integration. Agents

are deﬁned using JSON, ensuring ﬂexibility. Tools

like Visual Studio Code and libraries such as tqdm,

pandas, and openai support development and test-

ing.

Each agent’s behavior within CogniSim is mod-

eled through a layered architecture combining LLM-

driven reasoning and structured interaction con-

tracts. Agents are instantiated from JSON con-

ﬁgurations that deﬁne roles, instructions, and con-

straints; once initialized, they utilize GPT-3.5 or GPT-

4 via LangChain to parse input contexts, generate re-

sponses, and reason about tasks. For instance, a Prod-

uct Management agent might process backlog reﬁne-

ment instructions and produce prioritized user stories,

while a System Architect agent could leverage LLM

outputs for architectural decisions, referencing code

snippets or design patterns.

3.4 Utilizing Virtual Agents in Software

Management

Virtual agents simulate key roles—analysts, design-

ers, programmers, testers, project managers—each

equipped with prompts that initiate interactions. For

instance, Figure 8 shows a prompt for a Business An-

alyst during PI Planning, guiding their tasks in align-

ment with Agile processes.

4 EXPERIMENTS AND RESULTS

A series of simulations were conducted to assess the

effectiveness of the CogniSim framework in replicat-

ing Agile processes within SAFe. The setup included

key phases such as Program Increment (PI) Planning

and Iteration Execution, with cognitive agents rep-

resenting roles like Release Train Engineer, Prod-

uct Management, and System Architect. Their in-

teractions were facilitated through a dialogue system,

aligning objectives, prioritizing features, and reﬁning

backlogs.

The research questions centered on evaluating

agent capabilities in communication, coordination,

and decision-making. The multi-agent simulations

emphasized communication between agents, reﬂect-

ing the collaboration required in Agile development.

During PI Planning, agents aligned objectives, as-

sessed technical feasibility, and identiﬁed dependen-

cies, closely mirroring real-world Agile practices.

Results showed that the cognitive agents success-

fully replicated communication and decision-making

processes inherent in Agile methodologies. They co-

ordinated effectively, shared information efﬁciently,

and aligned their objectives for better planning and

execution. The agents provided data-driven insights

for feature prioritization and risk mitigation, improv-

ing decision-making quality.

Comparisons between agent outputs and human

analysts indicated that agents performed at a high

level, often matching or exceeding human-like anal-

ysis in feature evaluation and risk assessment. These

Agile Software Management with Cognitive Multi-Agent Systems

389

As a Business Analyst, your role during the PI Planning event is to collaborate

with Product Owners to ensure that features are broken down into clear user

stories with well-defined acceptance criteria. You work closely with Development

Teams to clarify details and answer questions as they estimate and commit to

user stories for the Program Increment. Your focus is on ensuring that requirements

are understood, dependencies are identified, and that the stories are ready for

implementation. Please follow these steps:

1. Analyze the Program Increment goals and break down features into user stories.

2. Define acceptance criteria for each user story.

3. Identify dependencies and potential risks.

4. Communicate with Development Teams to clarify any uncertainties.

5. Update documentation and ensure everything is ready for implementation.

Figure 8: Natural language prompt demonstrating the role of a Business Analyst during PI Planning.

ﬁndings suggest that LLM-augmented agents can en-

hance software project management, enabling more

efﬁcient data-driven decisions.

The simulations conﬁrmed that agents adhered to

their respective Agile roles, ensuring consistent goal

alignment. This supports the system’s potential for

improving productivity and quality in complex soft-

ware projects, meeting the study’s success criteria.

4.1 Output Quality Analysis

A comprehensive evaluation of the quality and effec-

tiveness of agent-generated content can be achieved

through the application of diverse measurement algo-

rithms. Techniques such as cosine similarity, code il-

lustrated in Figure 9, are instrumental in detecting re-

dundant or overlapping information, thereby inform-

ing strategies for generating outputs that are more var-

ied and contextually appropriate. Metrics examin-

ing lexical and syntactic diversity further ensure that

the system avoids producing uniform or overly pre-

dictable outputs. Goal achievement metrics, which in-

corporate keyword matching and semantic similarity,

provide a mechanism for determining the alignment

of content with predeﬁned objectives.

5 DISCUSSION AND FUTURE

WORK

This study explored integrating cognitive agents pow-

ered by LLMs into Multi-Agent Systems to enhance

Agile software project management frameworks like

Scrum and SAFe. The CogniSim framework demon-

strated how such technologies can automate complex

tasks, improve decision-making, and boost overall

productivity by simulating key software development

roles.

Future research should focus on scaling CogniSim

for larger teams and more intricate projects, evaluat-

ing performance using metrics like sprint completion

times, defect rates, and adherence to deadlines. En-

suring interoperability with existing CI/CD pipelines

and issue trackers will enhance practical applicability

(Horling and Lesser, 2004).

CogniSim’s agent architecture employs a modu-

lar design, where cognitive agents integrate LLM-

driven reasoning layers with domain-speciﬁc knowl-

edge modules and standardized interfaces. These

agents communicate using established protocols (like

FIPA ACL) and optional extensions tailored to Agile

processes, ensuring smooth collaboration and inter-

operability. Although certain workﬂows (e.g., contin-

uous integration and testing) operate through largely

stable pipelines, CogniSim supports dynamic adjust-

ments, allowing agents to reconﬁgure their pipelines

based on feedback, evolving requirements, or per-

formance metrics, thereby enhancing resilience and

adaptability throughout the software lifecycle.

Improving human-agent collaboration through in-

tuitive interfaces and adaptive MAS architectures will

further align the system with various Agile practices

(Guo et al., 2024). Incorporating adaptive learning

capabilities may help agents respond more effectively

to individual team member styles, enhancing overall

team efﬁciency. Recent advances explore the synergy

of logical reasoning, long-term memory, and collab-

orative intelligence within multi-agent LLM ecosys-

tems (Kostka and Chudziak, 2024).

Unlike existing multi-agent Agile management

frameworks such as PASSI (IEEE Foundation for

Intelligent Physical Agents (FIPA), 2012), which

rely on predeﬁned coordination rules and heuristic

decision-making, CogniSim integrates Large Lan-

guage Model-driven cognitive capabilities. This inte-

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

390

from sklearn.feature_extraction.text import TfidfVectorizer

from sklearn.metrics.pairwise import cosine_similarity

def compute_cosine_similarity(messages):

corpus = [msg['message'] for msg in messages]

vectorizer = TfidfVectorizer(stop_words='english')

tfidf_matrix = vectorizer.fit_transform(corpus)

similarities = []

for i in range(1, tfidf_matrix.shape[0]):

sim = cosine_similarity(tfidf_matrix[i-1], tfidf_matrix[i])[0][0]

similarities.append(sim)

return similarities

# Example call

similarities = compute_cosine_similarity([{'message': 'Hello world'}, {'message':

'Hello there!'}]),→

Figure 9: Code snippet for computing content similarity between messages using cosine similarity.

gration enhances adaptability, linguistic understand-

ing, and decision quality while aligning more closely

with widely adopted Agile practices. Maintaining

transparency in decision-making and trust remains a

priority, along with ensuring ongoing updates to ad-

here to ethical standards (Tariverdi, 2024).

Implementing CogniSim in real-world case stud-

ies will provide insights into its impact on productiv-

ity, quality, and team dynamics, while testing it across

diverse projects will assess scalability and effective-

ness (Chudziak and Cinkusz, 2024). Embedding pre-

dictive analytics within MAS could anticipate delays

or issues, enabling proactive corrective actions in dy-

namic environments (Lin et al., 2024).

6 CONCLUSION

Integrating cognitive agents powered by Large Lan-

guage Models within Multi-Agent Systems presents

a signiﬁcant step forward in Agile software project

management. The CogniSim framework demon-

strated that these technologies automate routine tasks,

enhance decision-making, and improve collaboration

among virtual team members, increasing efﬁciency

and adaptability (Dignum, 2009).

By simulating roles such as project managers, de-

velopers, and QA engineers, cognitive agents man-

age complex tasks traditionally handled by humans,

allowing human teams to focus on strategic activities

requiring creativity. Experiments showed that cogni-

tive agents effectively replicated communication and

decision-making processes integral to Agile method-

ologies, often matching or exceeding human perfor-

mance in certain tasks (Konar, 2000).

Challenges remain in scaling the framework for

larger projects, improving adaptability to unexpected

changes, and ensuring seamless human-agent collab-

oration. Addressing ethical considerations—data pri-

vacy, security, and transparency—is also critical as

these technologies become more integrated into de-

velopment processes (M

uller, 2020).

Future work will reﬁne the framework, validate

it in real-world scenarios, and ensure responsible in-

tegration, aiming to achieve beneﬁts like improved

decision-making and reduced operational costs (Rose,

2020). The advancements in MAS and LLMs within

Agile frameworks hold promise for more intelligent,

responsive project management practices, enabling

teams to adapt swiftly to evolving requirements and

deliver higher-quality software products.

By incorporating LLM-powered cognitive agents

into multi-agent systems, this research introduces

an approach for optimizing Agile software devel-

opment processes. It combines the adaptability of

cognitive computing with coordination strategies in-

herent to agent-based systems. The study’s pri-

mary contributions include demonstrating enhanced

decision-making capabilities, the automation of repet-

itive tasks, and improved communication within dy-

namic and complex project environments.

REFERENCES

Barua, S. (2024). Exploring autonomous agents through the

lens of large language models: A review.

Chudziak, J. A. and Cinkusz, K. (2024). Towards llm-

augmented multiagent systems for agile software en-

gineering. In Proceedings of the 39th IEEE/ACM In-

Agile Software Management with Cognitive Multi-Agent Systems

391

ternational Conference on Automated Software En-

gineering (ASE ’24), page 2, New York, NY, USA.

ACM.

Chudziak, J. A. and Wawer, M. (2024). Elliottagents: a

natural language-driven multi-agent system for stock

market analysis and prediction. In Proceedings of the

38th Paciﬁc Asia Conference on Language, Informa-

tion and Computation, Tokyo, Japan.

Cinkusz, K. and Chudziak, J. A. (2024). Communicative

agents for software project management and system

development. In Proceedings of the 21th International

Conference on Modeling Decisions for Artiﬁcial Intel-

ligence, Tokyo, Japan.

Cruz, C. (2024). Transforming competition into collabo-

ration: The revolutionary role of multi-agent systems

and language models in modern organizations.

Dignum, V. (2009). Handbook of Research on Multi-Agent

Systems: Semantics and Dynamics of Organizational

Models. Information Science Reference.

Dingsøyr, T., Nerur, S., Balijepally, V., and Moe, N. B.

(2012). A decade of agile methodologies: Towards ex-

plaining agile software development. Journal of Sys-

tems and Software, 85(6):1213–1221.

Du, H., Li, Z., Niyato, D., Kang, J., Xiong, Z., Huang,

H., and Mao, S. (2023). Diffusion-based reinforce-

ment learning for edge-enabled ai-generated content

services.

Du, H., Thudumu, S., Vasa, R., and Mouzakis, K. (2024). A

survey on context-aware multi-agent systems: Tech-

niques, challenges and future directions.

Guo, T., Chen, X., Wang, Y., Chang, R., Pei, S., Chawla,

N. V., Wiest, O., and Zhang, X. (2024). Large lan-

guage model based multi-agents: A survey of progress

and challenges.

Guo, Z., Jin, R., Liu, C., Huang, Y., Shi, D., Supryadi, Yu,

L., Liu, Y., Li, J., Xiong, B., and Xiong, D. (2023).

Evaluating large language models: A comprehensive

survey.

Horling, B. and Lesser, V. (2004). A survey of multi-agent

organizational paradigms. The Knowledge Engineer-

ing Review, 19(4):281–316.

Huang, X., Liu, W., Chen, X., Wang, X., Wang, H., Lian,

D., Wang, Y., Tang, R., and Chen, E. (2024). Under-

standing the planning of llm agents: A survey.

IEEE Foundation for Intelligent Physical Agents (FIPA)

(2012). Design Process Documentation Template.

IEEE FIPA DPDF Working Group.

Kim, A. G., Muhn, M., and Nikolaev, V. V. (2024). Finan-

cial statement analysis with large language models.

Chicago Booth Research Paper, Fama-Miller Working

Paper.

Konar, A. (2000). Artiﬁcial Intelligence and Soft Comput-

ing: Behavioral and Cognitive Modeling of the Hu-

man Brain. CRC Press.

Kosinski, M. (2024). Evaluating large language models in

theory of mind tasks.

Kostka, A. and Chudziak, J. A. (2024). Synergizing logical

reasoning, long-term memory, and collaborative intel-

ligence in multi-agent llm systems. In Paciﬁc Asia

Conference on Language, Information and Computa-

tion (PACLIC 38), Tokyo, Japan.

LangChain (2023). Langchain core api reference.

LeSS Company B.V. (2024). LeSS Framework.

Li, H., Chong, Y., Stepputtis, S., Campbell, J., Hughes, D.,

Lewis, C., and Sycara, K. (2023). Theory of mind

for multi-agent collaboration via large language mod-

els. In Proceedings of the 2023 Conference on Empir-

ical Methods in Natural Language Processing. Asso-

ciation for Computational Linguistics.

Lin, F., Kim, D. J., and Chen, T.-H. P. (2024). When

llm-based code generation meets the software devel-

opment process.

uller, V. C. (2020). Ethics of artiﬁcial intelligence and

robotics. In Zalta, E. N., editor, Stanford Encyclope-

dia of Philosophy, pages 1–70. Metaphysics Research

Lab, Stanford University.

Perkusich, M., Chaves e Silva, L., Costa, A., Ramos, F.,

Saraiva, R., Freire, A., Dilorenzo, E., Dantas, E., San-

tos, D., Gorg

onio, K., Almeida, H., and Perkusich, A.

(2020). Intelligent software engineering in the con-

text of agile software development: A systematic lit-

erature review. Information and Software Technology,

119:106241.

Project Management Institute (2024). Introduction to Dis-

ciplined Agile (DaD).

Rose, D. (2020). Artiﬁcial Intelligence for Business. Pear-

son FT Press.

Scaled Agile, Inc. (2024). SAFe 6.0 Framework.

Scrum.org (2024). Nexus Guide.

Singhal, K., Azizi, S., Tu, T., and et al. (2023). Publisher

correction: Large language models encode clinical

knowledge. Nature, 620:E19.

Sutton, R. S. and Barto, A. G. (1998). Reinforcement Learn-

ing: An Introduction. MIT Press.

Talebirad, Y. and Nadiri, A. (2023). Multi-agent collabora-

tion: Harnessing the power of intelligent llm agents.

Tariverdi, A. (2024). Trust from ethical point of view: Ex-

ploring dynamics through multiagent-driven cognitive

modeling.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

392