Agentic information systems

Source: Florian Holldack, Leonardo Banh and Gero Strobel; Electronic Markets (open access) · 2026-01-02

Recent advancements in artificial intelligence (AI) have catalyzed the emergence of agentic information systems (IS), which exhibit autonomous behavior and advanced cognitive capabilities. Unlike traditional IS, which functioned primarily as reactive tools supporting humans, agentic IS can make decisions independently, act in unstructured environments, and even delegate tasks to humans. This paradigm shift fundamentally transforms the human-IS relationship, questioning the long-standing assumption of human agentic primacy in IS research and practice. In this article, we provide a conceptual overview of agentic IS, delineating their defining characteristics and situating them within the broader evolution of IS. We introduce key archetypes of agentic IS, explore novel patterns of delegation and interaction between humans and machines, and discuss the socio-technical implications of these developments. Furthermore, we highlight the challenges and risks associated with integrating agentic IS from an individual, organizational, and societal perspective, emphasizing the need for nuanced understanding to harness the potential while addressing emerging complexities.

Depending on the employed conceptualization of agency, there are various agentic IS archetypes that encompass a spectrum of functions, from assisting humans in ambiguous tasks with enhanced capabilities to automating comprehensive work procedures (see Table 1). From an entity-focused agency perspective, these IS artifacts differ in their individual levels of agency, which enables them to assume novel, more complex tasks without human intervention or initiation. From a relational perspective, the degree of human-agentic IS integration differentiates them with higher relational agency representing more close collaboration.

Employing the entity-focused or weak relational perspective, humans as well as IS artifacts can possess some degree of agency, resulting in varying novel human-IS constellations and corresponding archetypes. Considering the conventional distribution of agency (i.e., based on the traditional IS paradigm), the assisting agentic IS archetype emerges, wherein humans possess the agency primacy, while IS artifacts are positioned on the lower ends of the agency continuum (Burton-Jones & Straub, 2006). Thus, rights and responsibilities regarding decision-making and actions are primarily attributed to humans (Baird & Maruping, 2021; Demetis & Lee, 2018). Within this archetype, IS artifacts primarily function as reactive tools requiring human initiation and intervention, making them suitable for assisting humans during task completion and decision-making (Berente et al., 2021). For instance, generative AI-based applications (e.g., ChatGPT or Claude) can assist users in optimizing text fluency, generating images for social media, or supporting the identification and correction of coding errors (Banh & Strobel, 2023; Feuerriegel et al., 2024). While these artifacts exhibit some degree of agency, such as interpreting user prompts and generating corresponding outputs, their agency remains substantially limited to human-provided tasks and scopes.

However, as agentic IS acquire higher degrees of agency, the autonomous agentic IS archetype manifests, which challenges or even reverses the traditional roles by attributing lower levels of agency to humans (Adam et al., 2024; Guggenberger et al., 2023). Within the reversed role, human agents become increasingly dependent on agentic IS artifacts' decision-making capabilities (e.g., autonomous trading agents), as human influence and control over the procedure and overall outcome is significantly reduced (Kar et al., 2024; Rinta-Kahila et al., 2023; Zhang et al., 2021). These artifacts are increasingly capable of autonomously handling ambiguous, long-term task within intricate environments, thereby reducing or even removing the necessity for human initiation and intervention (Demetis & Lee, 2018; Schuetz & Venkatesh, 2020). Moreover, decisions made by agentic IS artifacts can lead to social actions that actively influence the behavioral space of human agents, with some artifacts even prescribing specific actions, potentially resulting in a complete reversal of traditional supervision and subordinate roles (Lanz et al., 2024; Stelmaszak et al., 2024). Thus, agentic IS artifacts can be employed to direct other agents (including humans) by prescribing tasks, tracking and evaluating overall progress and even disciplining them through rewards or replacement (Kellogg et al., 2019; Möhlmann et al., 2021).

The assisting and autonomous archetypes both primarily describe situations in which one entity holds superiority in agency. Nevertheless, within the agentic IS paradigm, the agency of human agents and IS artifacts can converge, with both entities possessing high degrees of agency, leading to the collaborative agentic IS archetype (Braun et al., 2023; Jakob et al., 2024). In these instances, agentic IS can function as teammates or colleagues working in collaboration with human agents in order to achieve shared objectives (Lyytinen et al., 2020; Waardenburg & Huysman, 2022). Thus, both agents can take on rights and responsibilities for tasks and problems in a similar manner while actively communicating with each other (e.g., generative agents in software engineering) (He et al., 2025). This transforms agency primacy into a dynamically shifting property, preventing clear distinction and attribution to specific entities (Fabri et al., 2023; Murray et al., 2020).

These boundaries blur further with advanced human-IS interfaces that facilitate dense integration into human decision-making and acting (e.g., AI-based augmented reality glasses or even futuristic brain-computer interfaces), referring to the hybrid agentic IS archetype (Mayer et al., 2024; Rai et al., 2019). Instead of maintaining distinct and separate levels of agency, these heterogeneous teams exhibit conjoined agency, i.e., a shared capacity to act intentionally (Leonardi, 2023; Murray et al., 2020). However, the entity-focused and weak relational perspectives assume both the possibility of attributing agency to specific entities and the suitability of this approach for capturing the effects and implications of the emerging agentic IS paradigm (Dattathrani & De’, 2023; Schatzki, 2025). Consequently, these perspectives become increasingly inadequate for understanding the agency dynamics and implications within highly integrated human-agentic IS teams, thereby calling for analysis through the theoretical lens of sociomateriality (Cabitza et al., 2021; Leonardi, 2023).

The strong relational perspective builds on these limitations by focusing on the collaboration between human and IS artifacts (Fabri et al., 2023; Kuss & Meske, 2025) and thus facilitates a more comprehensive perspective on the collaborative and hybrid agentic IS archetype. It defines agency by the actions that are afforded or constrained through their relationship (Cecez-Kecmanovic et al., 2014; Leonardi, 2023). This shifts attention to the bilateral interplay between humans and agentic IS artifacts and their shared capacity to act within organizations leading to co-creation, framing agency as a systemic phenomenon rather than an entity-based attribute (Kuss & Meske, 2025; Waardenburg & Huysman, 2022). Therefore, this agency influences which entities are able to perform what actions and how those actions are conducted (Leonardi, 2023; Stelmaszak et al., 2024). Hence, more sophisticated human-IS collaborations (e.g., collaborative risk assessment in courts) influence the sociomaterial entanglement by affording new action potentials. Moreover, the relational perspective becomes increasingly relevant when considering networked businesses in which human agents collaborate with multiple artificial agents (Hofmann et al., 2024; Tsvetkova et al., 2024). In these environments, multi-agent systems demonstrate high collective agency even when individual entities exhibit low agency, requiring system-level analysis to capture their combined effects (Allmendinger et al., 2026; Stelmaszak et al., 2024; Tsvetkova et al., 2024). However, the strong relational perspective does not provide a framework for differentiating how agency is distributed across individual entities, thereby blurring responsibility boundaries and hindering accountable assignments of control (Kuss & Meske, 2025).

In light of agentic IS assuming novel roles, the traditional IS paradigm becomes insufficient for conceptualizing the multifaceted interaction and collaboration patterns between human and IS artifacts (Baird & Maruping, 2021; Stelmaszak et al., 2024). The theoretical framework of delegation has emerged as the prevailing alternative for describing and analyzing the relationships within the agentic IS paradigm (Grisold & Schneider, 2023; Guggenberger et al., 2023). Delegation is defined as the transfer of rights and responsibilities for making decisions and performing tasks to another agent (Baird & Maruping, 2021; Braun et al., 2023). The objective is aimed at enhancing productivity and flexibility by recognizing that a single agent may lack the resources, capabilities, or knowledge necessary to achieve complex, multi-step goals independently (Gur & Bjørnskov, 2017). This process requires metaknowledge, i.e., a comprehensive understanding of an individual’s capabilities, those of other agents, and the overarching objectives (Evans & Foster, 2011; Fügener et al., 2021a). While delegation has always been central within the IS paradigm, agentic IS artifacts enable extended forms of delegation and even reversed patterns toward humans (see Fig. 3). These patterns of delegation build on and extend concepts from related fields like HCI’s mixed-initiative interaction by specifically foregrounding the transfer of organizational rights and responsibilities.

User-invoked delegation is the most common form of delegation where human agents transfer rights and responsibilities to IS artifacts or other agents while retaining delegation ownership (Adam et al., 2024; Grisold & Schneider, 2023). The delegation procedure involves temporarily transferring (sub)tasks to agentic IS artifacts with or without direct human supervision (Alter, 2024; Mayr et al., 2024). Thus, humans commonly oversee task progression and maintain accountability for outcomes. Within this paradigm, there exists a spectrum of human supervision levels, thereby providing human agents with varying levels of control. The most restrictive is the “human-in-the-loop” approach, where a human agent manually approves all decisions or at least those of a critical nature made by the agentic IS artifact prior to execution (Grønsund & Aanestad, 2020; Jussupow et al., 2021).