Analogies and Metaphors for Agentic AI: How Industry Practitioners Explain Emerging AI Systems

AI Agents in the Double Diamond: Augmenting Design Thinking Practice in Organizational Innovation

Abstract Overview: Design thinking remains one of the most widely adopted innovation methodologies, yet practitioners routinely encounter bottlenecks that limit its effectiveness at organizational scale. These bottlenecks include data overload in discovery, cognitive bias in problem definition, idea recycling in development, and slow iteration in delivery. Artificial intelligence (AI)—particularly large language models (LLMs) and agentic systems—offers a compelling response to each of these challenges. Innovation leaders lack practical guidance on how to deploy AI within design processes while preserving the human-centered ethos that gives design thinking its value. Drawing on research in design cognition, AI agent architectures, and human–AI collaboration, we identify four practice moves innovation teams can implement across the four stages of the Double Diamond. These moves center on four specialized agents: the Empathy Agent, which synthesizes qualitative data at scale; the Problem-Framing Agent, which counters cognitive bias; the Ideation Agent, which generates cross-domain knowledge; and the Iteration Agent, which compresses prototyping and feedback cycles. For each move, we specify the agent involved, the human control points required, and the actions leaders can take. The article offers innovation and R&D managers and design leaders a roadmap for building hybrid human–AI innovation practices that enhance rather than diminish human creative capability. PRACTITIONER TAKEAWAYS Use specialized artificial intelligence (AI) agents strategically, not generically as an add-on technology. Deploy agents at specific Double Diamond bottlenecks—for example, empathy agents for data synthesis, problem-framing agents for bias mitigation, ideation agents for solution diversity, and iteration agents for faster prototyping and testing. Maintain human control through three deliberate interaction patterns: validation loops (verifying agent outputs), direction loops (setting agent priorities), and learning loops (updating both agent parameters and team protocols based on outcomes). Start with a single stage pilot. Identify where your team’s most acute bottleneck lies, deploy the corresponding agent type, and measure impact on insight quality or iteration speed before expanding to other stages.

Is Conversational AI Taking Control?: From Individualistic to Holistic Approaches to Sustaining Human Agency

Conversational AI systems are increasingly presented as agentic, intended to carry out tasks on behalf of users with a degree of autonomy. While agentic conversational AI systems have the potential to improve work efficiency and quality of life, they also introduce new risks and harms. We present the findings of a conceptual review that describes four concepts of agency, including personal agency, social agency, institutional agency, and artificial agency. Based on these interrelated concepts, we present four provocations to foster discussion on the extent to which conversational AI systems are becoming agentic and in which ways these systems may support and impede human agency. In doing so, we focus attention on the performative affordances of agentic conversational AI systems and highlight the societal responsibilities involved in their design. We argue for shifts from individualistic toward holistic approaches to developing agentic conversational AI systems that sustain our diverse and evolving forms of human agency in everyday life.

Seeing humans in technology: The antecedents, boundaries and consequences of AI anthropomorphism.

Anthropomorphism has received substantial attention in research on artificial intelligence (AI) because imbuing AI-driven technologies, such as algorithms, chatbots, and embodied robots, with humanlike qualities can mitigate AI aversion, a major barrier to AI acceptance and adoption. However, anthropomorphism may also weaken advantages associated with their nonhuman nature. This review advances three claims. First, AI anthropomorphism operates at two levels: design-based manipulations that companies can implement to make their technologies more humanlike, and individual tendencies to anthropomorphize those technologies. Second, while anthropomorphized AI may sometimes be evaluated similarly to human agents, people still distinguish AI from humans in many situations. Third, anthropomorphism can produce both positive and negative outcomes depending on the activated schemas and expectations. We conclude by outlining future research on how anthropomorphized AI may reshape our understanding of humanness.

Position: AI Agents in Scientific Teams Should Be Studied as Human-Agent Systems

Large language model-based agents are increasingly deployed as collaborators in scientific discovery yet most current work focuses on the autonomous capabilities of"AI Scientists". We argue that this overlooks the social aspects of scientific teamwork, and that studying AI Scientists as human-agent systems (HAS)--where the unit of analysis is the human-agent pair--is both underexplored and undervalued. We establish these points through literature and empirical analysis, and highlight recent incidences and studies which show that deploying agents in science without accounting for human-agent dynamics introduces near-term risks, including reduced diversity of scientific inquiry. Through analysis of real-world case studies, we show that scientists and agents can augment each other's capabilities. We call for new research that adopts the HAS lens to develop mathematical frameworks for understanding and fostering human-AI synergy in scientific discovery.

From automation to agency: The paradigm of agentic AI across technology, society and the ontology of work

The evolution of artificial intelligence is redefining the relationship between humans and machines, shifting from traditional automation toward systems endowed with agency. Unlike conventional AI, which primarily supports or executes predefined tasks, agentic AI systems are capable of autonomous decision-making, proactive goal generation, learning from experience and coordinated action within complex environments. This shift represents not only a technological advancement but an ontological transformation, as machines increasingly operate as cognitive agents rather than passive tools. This article adopts an interdisciplinary perspective to examine the paradigm of agentic AI, tracing its evolution from earlier forms of automation and outlining its defining characteristics, architectures and application domains. It then analyses the implications for work and organisational structures, highlighting how agentic systems reconfigure roles, redistribute cognitive labour and enable new forms of human–machine co-agency. Finally, the paper addresses the ethical, legal and governance challenges raised by autonomous agents, arguing for responsible adoption frameworks that preserve human centrality, accountability and social justice in emerging socio-technical systems.

SSS Framework: A Think-First, AI-Second Model for AI-Assisted Writing.