Socio-technical aspects of Agentic AI

Praveen Kumar Donta, Alaa Saleh, Ying Li, Shubham Vaishnav, Kai Fang, Hailin Feng, Yuchao Xia, Thippa Reddy Gadekallu, Qiyang Zhang, Xiaodan Shi, Ali Beikmohammadi, Sindri Magnússon, Ilir Murturi, Chinmaya Kumar Dehury, Marcin Paprzycki, Lauri Lovén, Sasu Tarkoma, Schahram Dustdar

Abstract

Agentic Artificial Intelligence (AI) represents a fundamental shift in the design of intelligent systems, characterized by interconnected components that collectively enable autonomous perception, reasoning, planning, action, and learning. Recent research on agentic AI has largely focused on technical foundations, including system architectures, reasoning and planning mechanisms, coordination strategies, and application-level performance across domains. However, the societal, ethical, economic, environmental, and governance implications of agentic AI remain weakly integrated into these technical treatments. This paper addresses this gap by presenting a socio-technical analysis of agentic AI that explicitly connects core technical components with societal context. We examine how architectural choices in perception, cognition, planning, execution, and memory introduce dependencies related to data governance, accountability, transparency, safety, and sustainability. To structure this analysis, we adopt the MAD-BAD-SAD construct as an analytical lens, capturing motivations, applications, and moral dilemmas (MAD); biases, accountability, and dangers (BAD); and societal impact, adoption, and design considerations (SAD). Using this lens, we analyze ethical considerations, implications, and challenges arising from contemporary agentic AI systems and assess their manifestation across emerging applications. The paper further identifies open challenges and suggests future research directions, framing agentic AI as an integrated socio-technical system whose behavior and impact are co-produced by algorithms, data, organizational practices, regulatory frameworks, and social norms.

Introduction: Beyond Generative AI

While Generative AI has dominated recent discourse, the field is rapidly evolving toward Agentic AI—systems designed not just for content generation, but for autonomous action. These agents can perceive their environment, reason over long horizons, coordinate with other agents, and execute multi-step plans with minimal human intervention. However, a critical "socio-technical gap" exists: technical research often ignores the societal, ethical, and governance implications of these systems. This paper argues that Agentic AI must be viewed as an integrated socio-technical system where technical choices are inextricably linked to human values and institutional structures.

The MAD–BAD–SAD Framework

To bridge the gap between architectural design and societal impact, the authors introduce the MAD–BAD–SAD construct. This framework provides a structured taxonomy for analyzing the non-technical dimensions of agentic systems, moving from background definitions to future directions.

Structural overview of the socio-technical aspects of Agentic AI
  • MAD (Morality, Autonomy, Dilemmas): This dimension focuses on the motivations for deployment and the moral tensions inherent in independent action. It highlights the "moral crumple zone," where responsibility for failure becomes dangerously distributed among developers and users.
  • BAD (Biases, Accountability, Dangers): This examines the "dark side" of autonomy, including how biases propagate through reasoning pipelines, the "black-box" nature of decisions, and security vulnerabilities like data poisoning.
  • SAD (Societal, Adoption, Design): This analyzes the integration of agents into human environments, focusing on public trust, workforce displacement, and the necessity of human-centered design.

Architecture is Policy

A central thesis of this work is that technical modules are not value-neutral. The design of an agent's internal architecture directly influences its societal impact.

The core architecture of an Agentic AI system

As shown in the architecture, an agent relies on modules for Perception, Cognition & Planning, Execution, and Memory.

  • Perception: Can encode societal biases from training data, which are then amplified during planning.
  • Cognition & Planning: Hallucinations in this module are not just factual errors but physical risks. In Agentic AI, a planning error can lead to unsafe physical actions (e.g., a robotic arm malfunction), making "grounding" a critical safety requirement.
  • Memory & Knowledge Base: Decisions here are not just about storage efficiency but involve complex data governance and privacy considerations.

Applications and Case Studies

The socio-technical implications of Agentic AI manifest uniquely across various high-stakes domains:

  • Healthcare: Agents are evolving from diagnostic tools to autonomous "care coordinators" managing chronic diseases, raising significant accountability questions.
  • Industry 5.0: The focus is shifting from pure efficiency to a human-centric model where agents and humans co-produce work, requiring new models of collaboration.
  • Earth Observation: Autonomous agents on satellites can decide which data to downlink, bypassing bandwidth bottlenecks but introducing risks related to data selection bias.
  • Networking: In resource-constrained environments like UAV/IoT management, agents replace static rules with dynamic reasoning, necessitating robust safety frameworks.

Future Directions and Challenges

Current governance and legal frameworks are often ill-equipped for systems that evolve beyond direct human control. The paper identifies several critical paths for future research:

  1. Anticipatory Governance: Developing frameworks to predict and mitigate risks before deployment.
  2. Human-AI Collaboration: Moving from "replacement" to "complementarity," where the division of labor is clearly defined yet flexible.
  3. Global Cooperation: Establishing interoperable international standards for systems that operate across digital and physical borders.
  4. Technical Robustness: Focusing on "alignment-aware goal formulation" to ensure agents do not find "shortcuts" that violate human safety norms.

By framing Agentic AI as a socio-technical system, we can ensure that the move toward greater autonomy does not come at the expense of accountability, safety, or public trust.

Summary updated

Abstract

Agentic Artificial Intelligence (AI) represents a fundamental shift in the design of intelligent systems, characterized by interconnected components that collectively enable autonomous perception, reasoning, planning, action, and learning. Recent research on agentic AI has largely focused on technical foundations, including system architectures, reasoning and planning mechanisms, coordination strategies, and application-level performance across domains. However, the societal, ethical, economic, environmental, and governance implications of agentic AI remain weakly integrated into these technical treatments. This paper addresses this gap by presenting a socio-technical analysis of agentic AI that explicitly connects core technical components with societal context. We examine how architectural choices in perception, cognition, planning, execution, and memory introduce dependencies related to data governance, accountability, transparency, safety, and sustainability. To structure this analysis, we adopt the MAD-BAD-SAD construct as an analytical lens, capturing motivations, applications, and moral dilemmas (MAD); biases, accountability, and dangers (BAD); and societal impact, adoption, and design considerations (SAD). Using this lens, we analyze ethical considerations, implications, and challenges arising from contemporary agentic AI systems and assess their manifestation across emerging applications, including healthcare, education, industry, smart and sustainable cities, social services, communications and networking, and earth observation and satellite communications. The paper further identifies open challenges and suggests future research directions, framing agentic AI as an integrated socio-technical system whose behavior and impact are co-produced by algorithms, data, organizational practices, regulatory frameworks, and social norms.
BibTeX
@misc{donta2025socio,
  author = {Praveen Kumar Donta and Alaa Saleh and Ying Li and Shubham Vaishnav and Kai Fang and Hailin Feng and Yuchao Xia and Thippa Reddy Gadekallu and Qiyang Zhang and Xiaodan Shi and Ali Beikmohammadi and Sindri Magnússon and Ilir Murturi and Chinmaya Kumar Dehury and Marcin Paprzycki and Lauri Lovén and Sasu Tarkoma and Schahram Dustdar},
  title = {{Socio-technical aspects of Agentic AI}},
  year = {2025},
  eprint = {2601.06064},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.06064},
}
Figure