Adapting teams to Human–AI collaboration: purposive programs and distinction-based task performance Open Access

Workplaces across industries are undergoing a fundamental transformation through the rapid diffusion of Generative Artificial Intelligence (GenAI). Organizations that have traditionally relied on human intellectual input for decision-making and innovation now face tools that are accessible to all and capable of producing sophisticated outputs across a wide variety of domains. This technological shift destabilizes traditional sources of competitive advantage: while access to data, expertise, and knowledge once distinguished organizations, such resources are increasingly embedded in publicly available AI systems (Azagury and Moore, 2024; Berg et al., 2023; Ganuthula and Balaraman, 2025). As a result, organizations are compelled to reconsider where their core strengths lie and how they can leverage human–AI collaboration to remain competitive in environments shaped by uncertainty and accelerated innovation cycles (Csaszar et al., 2024).

Research has increasingly examined the economic, ethical, and technological implications of AI adoption in organizations (Buxmann et al., 2021; Morley et al., 2020; Wilson and Daugherty, 2018). Yet, most existing approaches to assessing the added value of AI in business are grounded in what we will refer here to as “traditional AI” – systems designed to classify, predict, and optimize on the basis of existing data through predefined rules or statistical models. These approaches frequently remain either overly abstract, focusing on broad trends in digital transformation, or excessively narrow, concentrating primarily on technological affordances and risks. Research by Dell’Acqua, Mollick, and colleagues (Dell’Acqua et al., 2025, 2023) highlights the ongoing need for a systematic examination of how GenAI reshapes organizational work practices, particularly at the team level, where collaboration, coordination, and decision-making converge.

Recent theoretical advances have begun to address this gap. Carter and Wynne (2024) synthesize team decision-making literature to identify mechanisms and boundary conditions that shape effective human–AI teamwork, while Westover (2025) introduces metrics for quantifying human–AI synergy and strategies for adaptive collaboration. Although these contributions advance understanding of effective human–AI teams, they do not provide a systems-theoretical account of how organizations can deliberately program the conditions for such effectiveness. Existing frameworks remain either tool-centric or behaviorally oriented, emphasizing individual adaptation without addressing the organizational programming logic that sustains collaboration over time.

This gap highlights the need for a perspective that captures the structural conditions of organizational adaptation. Recent systems-theoretical work reconceptualizes AI not as a tool or agent, but as an organizationally embedded decision program participating in communicative processes (Roth and Lien, 2026), while Watson et al. (2025) frame human–AI collaboration as structural coupling between operationally closed yet interdependent systems.

This perspective shifts the focus from AI as an external tool to its role within communicative structures of organizational reproduction. It enables analysis not only of whether human–AI collaboration is effective, but how such effectiveness can be generated and stabilized. Building on this, the paper conceptualizes purposive programming as a mechanism for deliberately designing and guiding these structural couplings.

The research project “KI im Bildungsbereich der WKW” (AI in Education at the Vienna Chamber of Commerce and Industry) provides an empirical basis for addressing this gap (Rupprecht et al., 2025). The project examines both the future competencies demanded by labor markets in light of GenAI and the effects of GenAI use on competence development. A key experimental element involved student teams tasked with solving complex challenges either with or without the support of GenAI tools. The experiments yielded an important insight: team quality is a decisive factor in determining outcomes. High-performing teams that used GenAI produced more creative and professional results, whereas low-performing teams struggled to achieve comparable outcomes – regardless of whether they used GenAI or not. The findings point to the necessity of linking AI integration with strategies for strengthening team dynamics.

Grounded in social systems theory, this paper develops a conceptual framework for such strategies by articulating distinctions such as competence, teamwork, and organizational programs. Luhmann's systems theory enables the analysis of social phenomena without reliance on ideological or value-laden assumptions. This approach provides organizations with a strategic lens for interpreting and responding to technological disruption, supporting decision-making processes that are informed by systemic interrelations and operational contingencies rather than normative prescriptions.

We therefore pose the following research question: Which strategies of purposive programming support the development of robust team dynamics that maximize competitive advantages in GenAI-assisted task-solving scenarios?

Building on this research question, the paper contributes to both organizational systems theory and management scholarship on AI adoption, while offering orientation for practitioners seeking to foster effective and future-proof collaboration between humans and machines.

More specifically, it advances three interrelated contributions. First, it extends Luhmann's theory of purposive programming to contexts of human–AI collaboration, thereby conceptualizing GenAI not merely as a tool, but as a structurally relevant element within organizational communication processes. Second, it provides a comparative empirical illustration that identifies process-level mechanisms distinguishing high- and low-performing GenAI-supported teams. Third, it develops a structured, distinction-based design pathway that translates systems-theoretical reasoning into actionable organizational strategy.

In order to situate our conceptual contribution, it is necessary to outline the current scholarly discussions that inform the relationship between organizations, technological transformation, and team-based performance. The literature on Generative AI has rapidly expanded, yet most work remains either technology-driven, offering limited theoretical grounding for understanding how organizations can structurally adapt to the new realities of human–AI collaboration or productivity related (Berente et al., 2021; Brynjolfsson et al., 2025; Dell’Acqua et al., 2025; Noy and Zhang, 2023). Social systems theory – in the tradition of Niklas Luhmann – provides a framework that helps investigate the operational logics by which organizations reproduce themselves, adapt to environmental changes, and regulate complexity. This perspective provides the basis for a distinction-based and conceptual analysis of how strategies may be formulated when the conditions of competence, teamwork, and organizational advantage are reshaped by GenAI.

Broadly speaking, Artificial Intelligence (AI) refers to machines that simulate cognitive tasks once performed exclusively by humans (Collins et al., 2021). In this paper, we will refer to “traditional AI” when we mean those that have been primarily designed to classify, predict, or optimize based on predefined rules and statistical models applied to existing data. By contrast, Generative AI (GenAI) represents a paradigmatic shift. It is not confined to reproducing patterns from its training data one-to-one but is capable of generating novel artifacts – texts, images, or code – that appear as original contributions.

This generative quality fundamentally transforms the ways in which knowledge is represented, communicated, and made actionable. To describe GenAI merely as a “search engine” would obscure its performative dimension, reducing it to the retrieval of pre-existing information. A more fitting analogy can be drawn from the early days of the internet: had it been characterized solely as a “digital library,” such a label would have overlooked the extensive reconfiguration of societal, organizational, and communicative practices that followed. Similarly, reducing GenAI to a tool for information retrieval would neglect its paradigmatic significance and its capacity to reshape epistemic, social, and organizational practices.

In the context of this paper, GenAI is therefore not regarded as a neutral instrument but as a communicative force that directly affects the dynamics of team-based work and organizational strategies (Luhmann, 2013). By altering the very conditions under which competence, creativity, and decision-making are enacted, GenAI compels organizations to reconsider how they program their operations and how they sustain competitive advantage in environments marked by technological ubiquity and uncertainty.

Recent research on Generative AI has increasingly examined its impact on strategic decision-making and team performance. Recent research suggests that Generative AI increasingly influences strategic evaluation and decision-making, including the assessment and comparison of alternative strategic paths (Doshi et al., 2025). Beyond the strategic level, studies further indicate that GenAI can enhance productivity, accelerate knowledge work, and improve creative output under certain conditions (Dell'Acqua et al., 2023; Brynjolfsson et al., 2025). At the same time, evidence indicates that the performance effects of AI are contingent upon coordination quality, prior expertise, and task structure (Noy and Zhang, 2023).

At the team level, collaboration technologies have long been shown to influence creativity and knowledge integration, depending on shared mental models and communication practices (e.g. Zhang et al., 2025). More recent work highlights that human–AI collaboration may enhance task performance while simultaneously affecting intrinsic motivation and perceived ownership (Wu et al., 2025).

While these studies provide valuable insights into performance and productivity outcomes, they largely remain tool-centric: AI is treated as an independent variable whose effects are measured against predefined performance metrics. Less attention has been paid to how organizations structurally adapt their decision premises and programming logic when generative technologies render knowledge production abundant.

This paper shifts the focus from AI impact to organizational programming. Rather than asking whether GenAI improves performance, we ask how purposive programs and distinction-based strategies shape the conditions under which GenAI becomes a multiplier rather than a destabilizer of teamwork.

From a systems-theoretical perspective, organizations are understood as social systems constituted by communication. They reproduce themselves by reducing environmental complexity through decision premises, such as structures, programs, and personnel (Luhmann, 2018). Within this framework, strategies are themselves organizational programs – codified expectations that guide decision-making under conditions of uncertainty. Luhmann does not explicitly employ the term strategy in his organizational theory, particularly in Organisation und Entscheidung (Luhmann, 2013), his analysis centers on decision-making, the reduction of complexity, and the establishment of decision premises rather than on strategic planning in the conventional sense. In a Luhmannian sense, strategy can be defined as a paradoxical form of meta-communication that guides and frames organizational decision-making by structuring how organizations talk about their future (Rasche and Seidl, 2020). Strategies gain their relevance precisely because organizations operate in environments marked by contingency: alternative courses of action are always possible, and strategies provide a rationale for selecting among them.

In the context of technological disruption through Generative AI, this conceptualization is especially useful. Traditional competitive advantages – based on access to data, knowledge, or expertise – are increasingly commodified as they become embedded in widely accessible AI systems. Consequently, organizations must recalibrate their strategies to create differentiation through other means, such as encouraging team competence, shaping innovative forms of collaboration, and cultivating distinctive modes of human–AI interaction. A Luhmannian approach frames this not as a matter of normative orientation (“what organizations should value”) but as a matter of functional adaptation (“which distinctions and programs enable continued organizational reproduction under changed conditions”).

Thus, in this paper, strategies of purposive programming are conceptualized as organizational mechanisms for guiding team dynamics (especially in AI-assisted environments). Such strategies can enable organizations to stabilize expectations, coordinate complex tasks, and align competencies with evolving market requirements, while at the same time preserving the systemic autonomy needed to navigate unpredictable conditions.

In Luhmann's terminology, purposive programs (Zweckprogramme; output-oriented programs) refer to a goal-oriented form of organizational programming. Unlike conditional programs (input-oriented programs), which specify fixed reactions to predefined conditions (“if X, then Y”), purposive programs establish a desired end state while leaving open the paths by which it may be achieved (Luhmann, 2018).

For instance, an organizational purposive program might be framed as “increase market share by 5%.” The statement sets a measurable outcome but does not prescribe how exactly this should be accomplished, allowing for a range of possible strategies and decisions that remain aligned with the organizational aim. In team contexts, such programs function similarly: they define a shared objective, such as “develop a concept paper that creatively addresses the client's problem,” without dictating each communicative step. This openness enables teams to mobilize their autopoietic communication processes, adapt to contingencies, and innovate – while also making space for the integration of new tools like GenAI without disrupting systemic closure.

Rather than prescribing how GenAI should be deployed, purposive programs allow teams to explore multiple pathways and incorporate AI tools when they are seen as meaningful within the ongoing communication of the team. This ensures that adaptation to new technological environments does not rely on rigid rules but on flexible goal-setting that can accommodate unforeseen developments. Moreover, purposive programs do not attempt to control communication directly but instead provide a reference point that the system internalizes and operationalizes in its own way.

In this way, purposive programs operate less as instruments of control and more as orientations for communication by providing a shared reference point. This openness, however, also entails a displacement of responsibility: decisions are framed not only against present circumstances but also in relation to future purposes, thereby justifying courses of action through reference to the program rather than to individual choice.

While purposive programs offer organizations a flexible orientation framework for dealing with uncertainty, they do not by themselves specify how persons can navigate the complexity of their own decision-making. To complement this, comparative-systemic work and the work on multifunctional organizations (see section “2.5 Distinction-based Strategic Approaches”) provides distinction-based methods that enable teams and organizations to reflect on their own communicative processes. By formulating guiding questions and contrasts, this approach makes latent assumptions visible and supports decision-making even under conditions of uncertainty. In doing so, it links directly back to the responsibility dimension of purposive programming: decisions cannot be avoided but must be justified, and distinction-based observation offers a structured way of taking responsibility for them.

By creating a department or project group, the organization sets structural expectations: decisions are delegated to a defined subset of persons ^[1], within a certain thematic or functional boundary.

Thus, departments and projects act as structural decision premises that shape what kind of decisions are possible and by whom.

From a systems-theoretical perspective, persons emerge as communicative constructions that ensure consistency, predictability, and role-based expectations within social interaction. They do not exist as living or thinking entities in communication but serve as reference points that mediate between the temporal flow of communication and the opacity of individual consciousness. The concept of person enables the attribution of multiple roles, allowing communication to draw on both the current and potential positions of participants (Luhmann, 2018).

Within teams, this enables members to coordinate despite the complexity of individual consciousness, since communication can attribute stable identities and multiple roles to participants. In this way, the systems-theoretical concept of person supports robust team dynamics by facilitating reliable interaction and flexible role-taking.

The affinity between distinction-based approaches and Luhmann's social systems theory lies in their shared emphasis on observation through difference. For Luhmann, systems constitute themselves by drawing distinctions – between system and environment, relevant and irrelevant communications, or decision and non-decision – and by recursively operating on these distinctions. Distinction-based methods align with this logic by making such guiding differences explicit and usable as tools for self-observation.

Distinction-based approaches provide a way of focusing attention on what matters in order to establish a “state of better” in the most direct way, surfacing alternatives, and enabling reflection on the guiding premises of action. In this sense, they complement purposive programming by adding an additional layer of strategic orientation that addresses how teams interpret their objectives, distribute responsibility, and integrate novel technological tools such as GenAI.

Distinction-based questions allow organizations to promote adaptive decision-making under conditions of uncertainty. From this perspective, distinctions are not static dichotomies but dynamic tools that enable systems to observe themselves and their environments in new ways (Stingl de Vasconcelos and Belcredi, 2019). Distinction-based methods not only serve analytical functions but also foster collective sense-making in situations where no clear precedent exists (Belcredi and Stingl de Vasconcelos, 2021). By posing distinction-based questions – for example, “Let's assume with your next move the team is closer to a state of better – what is different?” – teams are encouraged to consider the agreed useful distinctions established in the “state of better” as contrast to an earlier stage. In the context of GenAI-assisted collaboration, where the tendency may be either to over-rely on or to dismiss the technology, reflexive questioning can promote a balanced and extensive approach that may integrate AI into human communication without subordinating one side to the other.

Several systemic tools also align with distinction-based approaches. Among them are the Systemic Structural Constellation methods developed by Insa Sparrer and Matthias Varga von Kibéd – constructivist-systemic instruments designed to open new perspectives through structured observational distinctions (Sparrer, 2006; von Kibed, 2012). These methods have been widely applied in organizational consulting and team development, particularly through the work of Elisabeth Ferrari, who has further advanced their dissemination via targeted media and training formats (Ferrari and Rühl, 2013).

A further extension of distinction-based strategies can be drawn from Roth & Sales' recent work on multifunctional organizations. Roth and Sales demonstrate how system-theoretical distinctions serve not only as analytical instruments but also as structuring devices for organizational observation and orientation (Roth and Sales, 2025). Distinctions function as operational markers through which organizations identify relevant differences, position themselves across multiple functional environments, and allocate attention and resources accordingly. In the context of GenAI-assisted teamwork, this implies that guiding distinctions – such as explorative/optimizing, formal/interpretive, or human/technical – do more than facilitate reflexive sense-making. They simultaneously shape the communicative structures that determine which clear states teams establish, how they assign roles, and integrate technological inputs. Roth and Sales' contribution thus reinforces the strategic role of distinction-based approaches as organizational mechanisms that couple reflexivity with structured pathways for decision-making under conditions of uncertainty.

Teams navigating GenAI environments imply a move away from simple oppositions such as GenAI versus human intelligence, toward distinctions that highlight complementarities and hybrid forms of value creation. In other words, distinction-based strategies can reframe collaboration as a communicative practice in which human and machine contributions are co-constructed, rather than competitively juxtaposed.

This paper adopts a mixed conceptual and empirical design, combining systems-theoretical reasoning with controlled practice-based experiments. Conceptually, it draws on Luhmann's organizational theory to articulate how purposive programming can orient team communication and decision-making in GenAI-supported environments. Empirically, it analyzes a series of structured, practice-based workshop experiments, in which student teams solved authentic cases under different conditions: some teams were allowed to use GenAI tools, while others were explicitly excluded from such support. These experiments are part of the internal research stream of the project and are complemented by an external research stream comprising discourse and network analyses. The experimental component was designed and implemented by Patrick Rupprecht, in collaboration with Isabel Rodenas, both researchers at the Vienna University of Applied Sciences (FHWien der WKW). The overall project leader was Tilia Stingl de Vasconcelos Guedes (Rupprecht et al., 2025).

The study is embedded in the institutional project “KI im Bildungsbereich der WKW” (Artificial Intelligence in the Educational Sector of the Vienna Chamber of Commerce and Industry). The project's overarching purpose was to develop research-based recommendations to support WKW educational institutions in their use of GenAI (project period: January 2024 to June 2025). In brief, the project sought to (1) understand how human–machine interaction reshapes competence acquisition; and (2) monitor the societal discourse, with particular attention to communication skills and competence requirements in AI-supported educational contexts. Central to the study was the guiding research question: How can AI be meaningfully deployed in WKW educational institutions, and which competencies should be cultivated so that people can work more effectively with AI? (Rupprecht et al., 2025)

Two research streams were designed to complement each other by combining an outward-looking perspective on societal and labor market demands with an inward-looking perspective on educational practice.

The external research comprised a discourse and network analysis to map the evolving public discourse around GenAI competences. The objective was to clarify which GenAI-related competences are in demand on the labor market and what requirements this imposes on employees. This stream was conducted by a Vienna-based research agency. Their findings provided boundary conditions and triangulation for the internal experiments by specifying competence profiles and discourse dynamics relevant to educational curriculum design.

The internal research consisted of two elements:

1. Baseline status assessment. Over 1,500 learners completed questionnaires on their use of GenAI and on GenAI-related task settings to capture the status quo of practices and perceptions within the educational settings.

2. Experimental study. Across six runs (December 2024–February 2025), 47 participants from a University of Applied Sciences (tertiary education), a Tourism School (upper secondary vocational education), and a Business School (upper secondary vocational education) took part in workshop experiments. We introduced practice-based learning settings, tasking student teams with solving authentic, domain-relevant problems with some groups required to integrate GenAI tools in their work, while others were not permitted to use GenAI at all. The design followed a pragmatic approach: tasks, team composition, and tool choices reflected realistic educational settings. Data sources included artifacts, live presentations, questionnaires, observations, and full-process video recordings. Participants spanned educational levels, from upper secondary school students (16–18 years) to part-time Bachelor students and full-time Master students (20–35 years).

Assessment focus: Observations drew on technical skills, accomplishments, and quality of results, but also on established teamwork criteria (e.g. shared goal clarity, role clarity and leadership, mutual accountability, social skills, open communication culture, strategic composition). These criteria were analytically linked to purposive programming: while the experiment design specified a shared end state, it left pathways open, enabling systematic comparison of how teams operationalized goals with and without GenAI support.

The task of this experiment was to analyze the potential installation of photovoltaic systems on suitable rooftops in Vienna, addressing technical, economic, social, and legal challenges while developing strategies, stakeholder communication approaches, and a final presentation.

Each session followed a standard structure:

1. Briefing: Participants received an identical problem scenario requiring the design of a concept and a 10-min presentation.

2. Work Phase: 90 min of collaborative problem-solving. Recorded via MS Teams.

3. GenAI group: Four laptops with ChatGPT Plus (GPT-4o and o1 models), no access to search engines.

4. Non-GenAI group: Four laptops with standard browsers (Google search), no GenAI tools.

5. Both groups: Access to MS Teams, PowerPoint, and Word for collaboration and content creation.

6. Participants completed a self-assessment questionnaire on their performance.

7. Participants had 10 min to present their solutions.

All sessions were observed by trained facilitators who completed structured observation forms. Additional data sources included:

1. participant self-assessment questionnaires on collaboration quality and perceived competence gains.

2. Observer field notes.

3. Video recordings of the entire work process.

4. Presentations.

The experiments provide direct insights into how GenAI alters team processes, decision-making, and competence development under controlled yet authentic conditions. By concentrating on this part of the study, the paper is able to link systems-theoretical reasoning with empirically observable team behavior, thereby illustrating how purposive programs can structure human–AI collaboration.

In addressing the research question – which strategies of purposive programming support robust team dynamics in GenAI-assisted task-solving – the experimental comparison enables the identification of process-level mechanisms that differentiate effective and ineffective programming under conditions of technological abundance. As these findings constitute the empirical foundation of the conceptual argument advanced in the following sections, we now turn to a more detailed account of the experimental design and its underlying conditions.

The experimental design was conceived as an analytical proxy for organizational teamwork. Although conducted in educational settings, the workshops reproduced core features of organizational collaboration, including time pressure, task complexity, distributed expertise, and collective decision-making under uncertainty.

Participants were recruited from institutions with a strong practice-oriented focus and ties to the Vienna Chamber of Commerce and Industry, ensuring that the tasks reflected applied problem-solving contexts rather than purely academic exercises. The heterogeneous composition of teams mirrored organizational contexts in which varying levels of expertise and digital literacy coexist.

From a systems-theoretical perspective, the teams can be understood as temporary organizational subsystems operating under purposive programs: a shared end state was defined while pathways remained open. This configuration supports analytical generalization regarding process mechanisms of human–AI collaboration beyond the educational context.

The study follows a comparative quasi-experimental design aimed at examining how access to GenAI influences the operationalization of purposive programs. Teams were either granted structured access to GenAI tools or excluded from them, while task complexity, time constraints, and available resources were held constant.

Rather than seeking statistical representativeness, the design supports analytical generalization (Yin, 2018) by identifying process patterns across cases. The experiment was structured to observe how decision premises and team configurations shaped coordination, responsibility distribution, and technological integration under comparable conditions.

Ethical Considerations: Participation was voluntary, and all data were anonymized prior to analysis. Formal ethics approval was not required, as the study was conducted within regular educational settings, involved no interventions beyond standard pedagogical practice, and posed no foreseeable risk to participants.

Across six experimental runs (December 2024–February 2025), 47 participants took part in the study. Teams consisted of four members, with one session comprising three due to absence. The sample included approximately 50% undergraduate participants (final-year vocational students, ∼18 years) and 50% postgraduate students (Bachelor's and Master's, ∼25 years), reflecting heterogeneous expertise levels.

Questionnaire and observational data were collected using Microsoft Forms, with descriptive statistics generated via its integrated functions. Observations were documented in structured templates during live sessions and complemented by video reviews conducted by an independent observer, with retrospective annotations compared to live notes.

Data was paraphrased and analyzed deductively using predefined competence dimensions derived from the external research (see Section 3.1.1), including technical expertise, communication, leadership, media literacy, ethics, and application skills. Artifacts and presentations were comparatively assessed in terms of coherence, completeness, novelty, and contextual adequacy, while questionnaire data were analyzed descriptively to identify patterns in competence development, ownership, and collaboration quality.

The analysis followed a cross-case comparison logic focused on theory-informed pattern recognition (Eisenhardt, 1989) rather than hypothesis testing.

Internal validity was strengthened through observer training and the use of standardized observation templates aligned with predefined teamwork criteria. The systematic comparison of live and retrospective observations served as a consistency check to reduce observer bias.

Data triangulation combined four sources: structured observation forms, video recordings, participant self-assessments, and produced artifacts.

While the study does not claim statistical generalization, reliability was enhanced through consistent task framing, identical technological conditions within groups, and transparent documentation of procedures. Although the educational setting limits external validity, the controlled comparative design supports robust analytical conclusions regarding GenAI-supported teamwork processes.

The following findings provide the empirical basis for answering our research question. Rather than measuring AI impact in isolation, the analysis focuses on how teams operationalize purposive programs under GenAI and non-GenAI conditions.

Team processes – rather than tool access – most strongly differentiated outcomes. High-performing teams (with or without GenAI) displayed early goal alignment, explicit task partitioning, role definition, and balanced participation throughout the work phase and final presentation. Conversely, weak outcomes were associated with unclear roles, uneven participation (e.g. two presenters carrying a GenAI team while others disengaged), or prolonged discussion without convergence on deliverables. These patterns closely map onto established teamwork factors such as shared goals, role clarity/leadership, mutual accountability, social competence, open communication, and purposeful composition. These criteria followed the teamwork factors outlined by Tarricone and Luca (2002), which also informed the observers' instructions.

Speed and Output: When supported by adequate collaboration infrastructure and prior familiarity with GenAI tools, GenAI-enabled teams matched or even outperformed non-GenAI teams in producing complete presentations. They frequently capitalized on GenAI's strengths in generating graphics and accelerating initial drafts. However, in contexts where AI literacy or coordination was limited, this speed advantage diminished.

Creativity and Novelty: When overall team quality was held constant, access to GenAI was associated with a modest but notable increase in novelty. A prominent illustration occurred during the MODUL session, where the GenAI-supported team proposed utilizing photovoltaic panel shadowing as a mechanism for building cooling – a creative extension beyond conventional solutions.

Critical Validation: A recurring weakness among some GenAI-enabled teams was insufficient validation of GenAI-generated outputs. Teams that systematically cross-checked results – for instance, by comparing alternative models or applying pragmatic “reality checks” – produced more robust outcomes. In contrast, uncritical reliance on GenAI suggestions often led to superficial or internally inconsistent content.

Motivation and ownership: Several higher-education participants viewed AI as a default tool; removing it reduced motivation and, in one non-GenAI team, contributed to lower focus and incomplete output. At the same time, a VBS participant in a GenAI team reported diminished ownership (“presenting something that did not really come from me”) suggesting that effortless content generation can reduce perceived authorship and engagement.

Results were robust across settings: (1) high-performing non-GenAI teams matched the output of GenAI-enabled teams; (2) GenAI teams with coordinated workflows and baseline literacy gained advantages in speed and novelty; (3) non-GenAI teams often excelled in scope control and time discipline when roles and goals were explicit; and (4) homogeneous cohorts were not inherently efficient – familiarity sometimes encouraged digressions.

In summary, team processes outweighed tool effects. Early goal alignment, clear role allocation, and mutual accountability were associated with higher performance, regardless of GenAI access. GenAI acted as an amplifier rather than a substitute: when collaboration was strong, it accelerated drafting and enriched visuals and ideas; when teamwork weakened, it neither corrected coordination deficits nor ensured quality.

When teams translated purpose into operational checkpoints (shared target images, “done” criteria, time boxes, validation routines), GenAI accelerated progress and broadened options. Without such premises, its advantage disappeared, and structured non-GenAI teams performed equally well. In short, GenAI shifts the bottleneck from production to selection: managers must define “what counts,” coordinate inputs, and verify adequacy.

The findings thus support a (purposive) program-centric rather than tool-centric design of work. Minimal conditional subprograms (e.g. plausibility checks, fact-checking, source control) institutionalize validation without over-determining the process. GenAI is most effective when embedded in purposive programs that set direction and criteria; it extends means but cannot replace the specification of ends.

Viewed through the systems-theoretical lens developed in this paper, the findings gain theoretical relevance. Organizations, understood as autopoietic systems of decision-communication (Luhmann, 2018), reproduce themselves not through the tools they employ but through decision premises that stabilize expectations and coordinate action under uncertainty. The empirical observations are consistent with this perspective: teams with aligned goals, clear roles, and explicit validation routines outperformed others, irrespective of GenAI access.

This pattern substantiates Luhmann's distinction between purposive and conditional programs. Conditional logic – relying on AI outputs – proved insufficient, whereas purposive programming enabled the effective integration of human judgment and AI-generated content. The shift of bottlenecks from production to selection and validation further highlights that, under conditions of generative abundance, effectiveness depends on the specification of ends and evaluation criteria.

This study contributes by showing that in GenAI contexts, competitive advantage no longer resides in access to tools but in the programming of observation and decision premises. It thereby extends existing research beyond tool-centric and behavioral accounts, offering a systems-theoretical explanation of how effective human–AI collaboration can be structurally enabled. While the empirical basis is illustrative rather than statistically generalizable, it reveals robust process mechanisms with analytical relevance for organizational design.

From a managerial perspective, the implication is clear: organizations should prioritize the deliberate design of decision premises (aligning goals, articulating selection criteria, and embedding validation routines) over mere tool adoption. Competitive advantage thus shifts from generative capacity to the disciplined structuring of observation, selection, and responsibility.

Accordingly, the study asked which strategies of purposive programming support robust team dynamics in GenAI-assisted task-solving. The findings indicate that robustness depends on the stabilization of shared decision premises. Effective purposive programming involves (1) early goal alignment, (2) the translation of abstract purposes into operational selection criteria, and (3) the establishment of validation routines that counterbalance generative abundance.

GenAI does not replace purposive programming but amplifies its importance by increasing the need for reflexive organizational capacities. Building on these insights, three complementary distinction-based approaches are proposed to translate purposive programming into practice: (1) a comparative-systemic practice that structures observation by differences (Belcredi and Stingl de Vasconcelos, 2021); (2) a systemic structural constellation format (9/12-Fields) from the SySt© institute to externalize decision premises in strategy work (Sparrer et al., 2012); and (3) the concept of the multifunctional organization (Roth and Sales, 2025) to align team programs with the multiple functional environments in which organizations actually operate.

Purposive programs specify ends (the “state of better”) while leaving pathways open. In comparative-systemic work, the openness is disciplined by distinction-based questioning: teams inquire into the minimal, observable differences that would indicate progress toward the purpose. These questions turn abstract aims into contrastive, testable criteria (more/less, earlier/later, worse/better, central/peripheral), which teams can then embed as sub-programs. In our experiment settings, GenAI helped produce alternatives fast; distinction-based instructions help select and refine them (Stingl de Vasconcelos and Belcredi, 2019). This complementarity is at the heart of a program-centric approach: ends steer; distinctions calibrate; tools execute.

Systemic Structural Constellations offer compact formats – such as the 9/12-fields arrangements –provide a structured spatial framework for externalizing decision premises. In strategy workshops, this format externalizes key premises in relation to time (on the one side four dimensions: past, present, future and long-term future) and space (on the other side three dimensions: external context, boundary, and internal context) and allows teams to examine their tensions in space, enabling teams to see how alternative distinctions reconfigure options. Teams then “walk” the canvas to test how a change in one field (e.g. moving goals to further fields or seeing different constrains in relation to resources) alters feasible next steps. Such tools can provide a concrete way to build the checkpoints that distinguish high-performing teams – without over-specifying pathways (Sparrer et al., 2012).

Roth and Sales argue that organizations are multifunctional systems that selectively process multiple codes (economic, scientific, political, legal, educational, etc.). Strategy, therefore, must be functionally unbiased: it orients action while acknowledging that different functions imply different adequacy criteria. Bringing this lens to team-level purposive programming means that selection criteria cannot be purely “technical” (e.g. accuracy) or purely “economic” (e.g. cost); they must be multi-coded and explicitly ranked for the given case. For GenAI workflows, this entails constructing a code map for the task (e.g. legal compliance > reputational risk > economic efficiency > novelty) and then tying validation sub-programs to each code (e.g. legal review prompts; fact-check and citation provenance; stakeholder sensitivity checks; novelty screens). This functional layering supports the evaluation and decision-making within the work process with the 9/12-Fields structure and the comparative-systemic logic.

In light of the aforementioned, we suggest a practical pathway that operationalizes purposive programming for team settings working with GenAI. The pathway translates abstract ends into observable selection criteria, surfaces the relevant field of factors, leverages functional interdependencies, and anchors the resulting premises in day-to-day decisions and actions. The stages correspond to the model shown in Figure 1.

At this stage, teams work together to create a clear and concise description of the better state they want their work to achieve. This provides a shared sense of direction (a purposive orientation) without fixing a single path to reach it. The outcome becomes something that can be openly discussed and tested, without prescribing the exact means of achieving it.

This step also defines what “done” looks like by describing clear states and criteria, as well as the boundaries of the work (scope and timeframe). By setting this alignment, the team ensures that contributions from GenAI remain useful: the AI's output must always be selected, adapted, and put into context by the team.

In these stages, teams make their decision premises by using structured mapping methods (such as the 9/12-Fields approach). This process brings time horizons and contextual aspects to the surface and transforms resources, relevant factors and interdependencies into shared knowledge. Instead of imposing rigid rules, it keeps purposive programs open but disciplined through context-sensitive distinctions.

By applying a multifunctional lens – e.g. considering economic, legal, reputational, political, or scientific aspects – teams ensure that decisions are robust across different environments. A code-map that sets priorities (e.g. legal compliance > reputational risk > economic efficiency > novelty), combined with simple validation routines (fact-checks, plausibility checks, provenance checks), provides orientation for dealing with GenAI outputs.

At this stage, teams have established a concrete work path including clearly defined responsibilities, checkpoints, timeboxes, “done” criteria, etc. Validation loops (such as fact-checking and stakeholder sensitivity checks) can be deducted from this roadmap at any time of work in progress.

GenAI is used here as an amplifier – helping generate alternatives, drafts, visuals, and counter-positions. The real sharpness, however, comes from the team's comparative-systemic questions (e.g. “Now that we are at the state we want to be at: what are the criteria and checkpoints, that makes us know we have arrived?”)

The outcome is a commonly committed, controllable roadmap, giving orientation for day-to-day decision-making, where quality arises from disciplined selection and contextualization – not from simply having access to tools.

This pathway aligns with our empirical insight that team processes dominate tool effects: GenAI accelerates and diversifies means but does not substitute purpose, responsibility, or decisions. By linking purposive ends to explicit distinctions and code-aware validation sub-programs, teams convert GenAI's abundance into robust, context-adequate outputs and reclaim competitive advantage through collaboration quality and program design.

Our study does not examine the precise temporal sequencing of programming elements; instead, it offers a structured framework that orients teams toward explicit goal alignment, role clarity, and validation routines throughout the work process. In organizational settings, leadership roles and basic decision rights are typically established prior to task execution. Although communication norms evolve during collaboration, they rest on predefined structures of responsibility.

The experiment was practice-based and educationally anchored; while tasks and teams were realistic, broader organizational contexts may vary. Future research should (1) quantify the effect sizes of the four-stage design on quality and time-to-decision; (2) compare different code maps across sectors (e.g. healthcare vs. finance) using the multifunctional lens; and (3) test whether structured briefings reduce downstream rework by improving early premise alignment. These steps would refine purposive programming into a portfolio of teachable routines for human–AI collaboration.

In sum, robust GenAI-assisted teamwork is not primarily a tooling problem but a matter of programming of observation and selection. Comparative-systemic distinctions lead to decisions; purposive ends steer; systemic structures collect and arrange relevant aspects; and multifunctional perspectives add further valuable criteria in complex environments. Where teams establish this framework, GenAI may become a multiplier of quality and speed rather than a source of noise.

The authors used ChatGPT (OpenAI) solely for language and grammar refinement of the manuscript. No content was generated by the AI tool. The authors remain fully responsible for the accuracy and integrity of the work.

The authors acknowledge the project “KI im Bildungsbereich der WKW” (AI in Education at the Vienna Chamber of Commerce and Industry), which provided the context and empirical setting for this research.