Executive Summary

The People-Process-Technology (PPT) framework, descended from Leavitt’s diamond model and refined through decades of sociotechnical systems research, has competently carried digital transformation theory and practice through three generations of IT-enabled change. Its core insight, that organisational outcomes emerge from the joint optimisation of interdependent social and technical subsystems, remains analytically sound. What PPT does not anticipate is a class of technology that can itself act: systems that pursue goals, exercise judgment, coordinate with humans and other systems, and adapt their behaviour within guardrails. Generative and, especially, agentic AI present this challenge directly. They are not simply more capable tools occupying the Technology corner of the triangle; they perform processes rather than merely enable them, and they occupy roles that overlap functionally, though not legally or morally, with those of human workers.

Drawing on actor-network theory (ANT) and sociomateriality, which treat agency as distributed across human and non-human actors and organisational life as constituted in sociomaterial practices, this conceptual paper argues that AI transformation is best understood as the first organisational transformation, at scale, in which agency itself becomes a designable and governable variable. We distinguish three modes of AI deployment, automation, augmentation, and delegation, and show that PPT remains adequate for the first, strains under the second, and fails for the third. In its place, we develop a sociotechnical assemblage view, framed as Actors, Practices, Infrastructure and Governance (APIG), in which agency and accountability are treated as distributed properties of sociomaterial systems rather than attributes of isolated “people” or “technology.” We conclude by outlining implications for AI governance, organisational design and future research on human-AI work systems.

1. Introduction

The People-Process-Technology framework is one of the most enduring devices in the digital transformation toolkit. It distils decades of socio-technical thinking into a simple claim: meaningful and sustainable change emerges when organisations align their people (skills, culture, structures), processes (ways of working, decision rights) and technology (tools, platforms, infrastructure) around a coherent strategy. Practitioners and scholars alike have used PPT to diagnose transformation failures, plan change programmes and explain why “technology projects” often falter when they neglect human and organisational factors.

Over time, PPT has been extended but not fundamentally questioned. Commentators have added “Data” as a fourth element, producing People-Process-Technology-Data (PPTD), to reflect the centrality of data as a substrate for digital transformation. Others have elaborated sophisticated socio-technical variants, but the underlying assumption remains that people are agents, processes encode their decisions, technologies execute them, and data flows through this system as an input and output.

The rise of artificial intelligence, and particularly generative and agentic AI, challenges this assumption in a new way. In many contemporary deployments, AI systems do not simply execute predetermined rules or provide optional advice; they plan, sequence and adapt actions within organisational workflows, and they do so at scale. In urban governance, for example, AI systems are already redistributing discretion across institutional levels and roles, changing how accountability and oversight must be conceived. In professional domains such as radiology, clinicians report re-crafting their work and identity around AI diagnostic systems whose recommendations they both rely on and contest. These are not straightforward “tools” in the classical sense.

This paper advances three claims.

1. Not all AI deployments equally strain PPT; many instances of AI-as-automation remain well handled by the original framework.
2. A particular mode of AI use, AI-as-delegation or agentic AI, breaks PPT’s implicit assumption that agency is monopolised by humans, requiring us to treat agency as distributed across human and non-human actors.
3. Actor-network theory, sociomateriality, distributed cognition, routine dynamics and affordance theory provide a more suitable theoretical basis for understanding AI transformation as a reconfiguration of distributed agency than PPT does.

On this basis, we propose an alternative conceptual frame, Actors, Practices, Infrastructure and Governance (APIG), and argue that AI transformation is best understood as the deliberate redesign of APIG, rather than as the alignment of People, Process and Technology around human actors.

The paper proceeds as follows. Section 2 revisits PPT and its extensions, highlighting the often implicit assumptions about agency. Section 3 introduces a three-mode typology of AI deployment and shows where PPT holds, strains and fails. Section 4 outlines the theoretical foundations in ANT, sociomateriality, distributed cognition, routine dynamics and affordance theory. Section 5 develops the APIG framework and contrasts it with PPT. Section 6 discusses implications for AI governance and organisational practice. Section 7 concludes with directions for future research.

2. People-Process-Technology: Origins, strengths and limits

2.1 Origins in Leavitt’s diamond and socio-technical systems

PPT has deep roots in organisational theory. Leavitt’s “diamond” model conceptualised organisations as an interdependent system of people, tasks, structure and technology, arguing that change in one element necessarily reverberates through the others. Sociotechnical systems research in the Tavistock tradition similarly emphasised the joint optimisation of social and technical subsystems, cautioning against technology-led change that neglected work design and human needs.

Modern PPT formulations distil these ideas into a pragmatic framework: successful digital transformation requires coordinated changes in people (roles, skills, culture), processes (workflows, decision rights, control mechanisms) and technology (applications, infrastructure, data platforms). Consulting and practitioner literature repeatedly emphasises that technology alone cannot deliver value; benefits materialise when process and people are aligned with new capabilities.

2.2 Extensions: Data and the evolving triangle

As data has become more central to digital business models, a common extension has been to explicitly add Data as a fourth element, creating PPTD. Proponents argue that data is the “lifeblood” or “connective tissue” of digital transformation, shaping and being shaped by people, processes and technology. In AI contexts, several frameworks explicitly foreground data governance and quality as preconditions for effective AI.

These extensions, however, remain conservative: they add boxes to an existing schema without altering its basic metaphysics. People are still the only recognised agents; processes are patterns of human work; technology, including AI, is infrastructure; data is an object to be processed. The point is not that this is wrong, but that it embeds assumptions that are not innocent once AI systems begin to act in ways that look, functionally, like agency.

2.3 The implicit agency assumption

PPT’s categories presuppose that:

• People decide, judge and bear responsibility.
• Processes encode and stabilise those decisions as routines and rules.
• Technology executes, enables and accelerates processes without independent goals or judgment.

This was a reasonable approximation when “technology” meant information systems that stored data and executed deterministic logic under human control. It is more tenuous when “technology” includes learning systems that not only execute but also shape criteria for judgment, allocate tasks, and evolve their behaviour. As AI deployments shift along this spectrum, we need finer distinctions about what kind of AI we are talking about.

Three Modes of AI Deployment

We distinguish three ideal-typical modes of AI use in organisations: automation, augmentation, and delegation/agency. In practice, deployments may combine elements of all three, but the typology clarifies where PPT’s assumptions hold.

AI-as-automation: Executing Predefined Tasks

In AI-as-automation, systems carry out well-specified tasks with little or no discretion. Examples include robotic process automation scripts, rules-based decision engines, and machine-learning models embedded in fixed workflows (e.g., credit scoring in a tightly regulated lending process). The criteria for correct performance are defined ex ante by humans. The system maps inputs to outputs in a stable way, with exceptions escalated back to people.

• Technology executes processes that people designed.
• Processes encode human judgment.
• People remain the locus of agency and accountability.

Data quality and governance matter, but the underlying picture of human-centric agency is not seriously challenged.

AI-as-augmentation: Assisting Human Judgment

In AI-as-augmentation, systems produce recommendations, forecasts, or drafts that humans use as inputs into their own decisions. Diagnostic support tools in radiology provide probability scores or flagged regions on medical images, but radiologists remain accountable for the diagnosis. Generative "copilots" for software development, legal drafting, or knowledge work produce candidate outputs that human professionals review, edit, and integrate.

Here, cognition becomes more clearly distributed. Distributed cognition research shows that even before AI, cognitive work in complex settings such as ship navigation or air traffic control was performed by networks of people, artefacts, and procedures, not individuals alone.

PPT can be stretched to cover these cases, Technology supports People in Process, but it starts to obscure as much as it reveals.

AI-as-delegation/agency: Autonomous Goal Pursuit

The third mode, AI-as-delegation or agentic AI, is where PPT breaks. In this mode, systems are assigned goals and constraints and empowered to plan, act, and adapt within those boundaries, often orchestrating multiple tools and interactions.

• Agentic AI “co-pilots” that can read emails, draft responses, schedule meetings, and update CRM records based on high-level instructions.
• Algorithmic management systems that allocate tasks, set performance benchmarks, and adjust work sequencing.
• Autonomous AI agents that can initiate marketing campaigns, adjust pricing, or manage incident response workflows.

Here, the categorisation of AI as “Technology” becomes misleading. These systems are better seen as actants, non-human actors enrolled in organisational networks who make a difference to outcomes.

Theoretical Foundations: Distributed Agency and Sociomaterial Systems

To make sense of AI-as-delegation, we draw on five strands of theory that decentre individual human agency and foreground distributed, sociomaterial systems.

Actor-network Theory: Non-humans as Actants

Actor-network theory (ANT) advances three moves that are directly relevant:

• It treats agency as an effect of networks.
• It refuses the human/non-human divide.
• It rejects pre-given distinctions between “social” and “technical.”

Under this lens, an agentic AI system is straightforwardly an actant, not merely a resource.

Sociomateriality: Constitutive Entanglement

Sociomateriality argues that the social and material are not separable subsystems but are “constitutively entangled” in practice. AI systems are not external add‑ons to pre‑existing processes, but constitutive elements of the practices that make up work.

Distributed Cognition: Systems of Humans and Artefacts

Distributed cognition research reframes cognition as a property of systems comprising people, artefacts, and environments, rather than of isolated minds.

Routine Dynamics: Ostensive and Performative Participation

Routine dynamics research reconceptualises organisational routines as generative systems with both ostensive (abstract, schematic) and performative (concrete, enacted) aspects.

Affordance Theory: What AI Systems Make Possible

Affordance theory provides a vocabulary for talking about what technologies afford, the action possibilities they enable or constrain, without collapsing into technological determinism.

From PPT to APIG: A Sociotechnical Assemblage View

Synthesising these strands, we propose reframing AI transformation around a sociotechnical assemblage comprising Actors, Practices, Infrastructure, and Governance (APIG).

Actors: Human and Non-human Agency

Actors include all entities that can make a difference to organisational outcomes: employees, customers, AI agents, algorithms, documents, devices, platforms, and regulatory frameworks.

Practices: Sociomaterial Routines of Work

Practices are the recurring sociomaterial activities in which work is accomplished and are constituted by the entangled actions of human and non-human actors.

Infrastructure: Models, Data, Platforms and Integration

Infrastructure comprises the material substrate of AI-enabled work: models, datasets, data pipelines, compute resources, integration layers, etc.

Governance: Allocating Rights, Obligations and Oversight

Governance encompasses rules, norms, structures, metrics, and oversight mechanisms that allocate decision rights and accountability across actors.

How APIG Handles the Three AI Modes

For AI-as-automation, actors are mainly human; practices are lightly reshaped routines. For AI-as-augmentation, actors include AI systems; practices become sites of sociomaterial co-production. For AI-as-delegation, actors include AI agents; infrastructure affords autonomous orchestration and adaptation.

APIG does not discard PPT’s concern with alignment but reframes what is being aligned: not static categories, but living assemblages of actors, practices, infrastructures, and governance regimes.

6. Implications for AI Transformation and Governance

6.1 Designing for Distributed Agency

First, AI transformation should be framed as designing distributed agency, not deploying tools. Organisations need to make explicit choices about:

• Which decisions and actions remain strictly human;
• Which are augmented by AI;
• Which are delegated to agentic systems, with clear escalation paths and override mechanisms.

This requires mapping Actors and Practices, not just processes and systems, and understanding where non‑human actants already “make a difference” to outcomes. It also requires humility about control: as empirical work on algorithmic management and diagnostic AI shows, professionals often experience AI as both support and constraint, altering their discretion in ways that require careful negotiation.

6.2 Governance as a Design Domain, Not an Afterthought

Second, governance must be treated as a design domain. Recent research on AI governance highlights tensions between innovation, capability and institutional responsibility, arguing that AI systems cannot be adequately governed through traditional, actor‑centric frameworks alone. Regulatory developments such as the EU AI Act and national AI laws increasingly require organisations to evidence risk management, transparency and human oversight for high‑risk and high‑impact AI.

Within APIG, Governance is where boards and executives specify which AI modes are permissible, what forms of monitoring and evidence are required, how ownership–control–responsibility gaps will be closed, and how human actors can contest or override AI decisions. This aligns with emerging guidance that positions AI governance as a standing component of corporate governance, not as a one‑off technology policy.

6.3 Rethinking Leadership, Capability and Change

Third, AI transformation demands new leadership capabilities. Studies of AI in public administration, healthcare and digital government show that effective adoption depends less on technical prowess than on leaders’ ability to navigate shifting boundaries of discretion, accountability and public value. In agentic AI settings, leaders must become designers and stewards of distributed agency: clarifying roles for human and non‑human actors, ensuring practices remain aligned with institutional purposes, and adapting governance as systems and contexts evolve.

This has implications for change management. Traditional narratives focus on “bringing people along” with technology changes; APIG suggests that change programmes should explicitly surface and renegotiate patterns of agency across actors, practices, infrastructure and governance. Resistance may not be to technology per se but to perceived losses of discretion or shifts in responsibility caused by AI systems, issues that require governance responses as much as training.

6.4 Research Agenda

For scholars, APIG and the distributed‑agency lens suggest several avenues:

• Empirical studies of AI‑as‑delegation. Detailed ethnographies and case studies of agentic AI in work systems, examining how agency is distributed, how professionals experience and reshape their roles, and how governance mechanisms work in practice.
• Comparative analyses of AI modes. Systematic comparisons of AI‑as‑automation, augmentation and delegation across sectors to test the claim that PPT fails most clearly in the third mode and to refine the boundaries between modes.
• Design and evaluation of governance mechanisms. Investigations into how organisations implement governance for distributed agency, including risk controls, oversight structures, audit practices and regulatory compliance under AI‑specific regimes.
• Conceptual integration with legal and ethical theory. Further development of how distributed agency frameworks intersect with doctrines of responsibility, liability and rights, in light of emerging legal scholarship on AI and accountability.

7. Conclusion

This paper has argued that while the People–Process–Technology framework remains useful for many forms of digital and AI‑enabled change, it is increasingly misaligned with a world in which non‑human actors can be designed and deployed as agents in organisational work systems. Rather than adding another box to PPT, we suggest that AI‑as‑delegation/agency breaks the framework’s implicit assumption that agency is the sole preserve of humans and that technology merely executes.

Drawing on actor–network theory, sociomateriality, distributed cognition, routine dynamics and affordance theory, we proposed an alternative assemblage view framed as Actors, Practices, Infrastructure and Governance (APIG). Under APIG, AI transformation is understood as the reconfiguration of distributed agency and accountability across human and non‑human actors, sociomaterial practices, technical infrastructures and governance regimes.

Conceptually, this moves AI transformation from “tool deployment” to “agency design.” Practically, it suggests that boards, executives and policymakers should ask less “what tools are we buying?” and more “how are we allocating and governing agency in our sociotechnical systems?” For researchers, it opens a programme of work on how agentic AI reshapes routines, professions, governance and institutions.

PPT, we have suggested, is not so much wrong as exhausted. Its categories no longer carve nature at its joints once organisations can deploy systems that act, learn and coordinate. APIG, grounded in established sociotechnical traditions, offers a more promising starting point for understanding, and governing, the next wave of AI transformation.