When the Designer Leaves and Takes Their Mind With Them
Shadow AI Is Quietly Redefining Intellectual Ownership Inside the Architectural Studio
Imagine that your most productive designer submits their resignation tomorrow. What you may not realise is that Shadow AI has become a critical part of modern workflows, often existing beyond official systems. They leave behind an empty desk, drawings saved to the server, files organised in tidy digital folders. But on their personal device they carry something you cannot claim through any employment clause, something you would not even know how to describe: hundreds of prompts composed over two years of daily practice, AI agents trained on their specific design sensibility, automated workflows that allow them to accomplish in an hour what their colleagues need a full day to complete. The designer is gone. The gap remains.
This is not a hypothetical. It is happening now inside architectural practices around the world — including across the Arab world — in complete silence and in the total absence of any governing framework.
When the Designer Becomes Faster Than the Team
In virtually every architectural studio with more than five designers working today, there is likely at least one who possesses what the emerging academic literature describes as a “personal digital infrastructure”: a collection of AI prompts, automated workflows, and custom agents operating informally and entirely outside any sanctioned organizational policy. These tools allow that designer to produce a developed design concept in twenty minutes while the colleague at the adjacent desk requires three hours to reach the same point.
The problem is not the speed itself. It is the invisible friction that speed leaves behind. The team registers the gap without understanding its source. The project director relies on that designer’s outputs without knowing which tool produced them. The practice builds its reputation on the results of processes it cannot audit or reproduce.
Researchers Ross, Hibbert, and Moss identified and classified this phenomenon under the term “Shadow AI,” defining it as “the unsanctioned, informal use of artificial intelligence technologies within organisations,” while emphasizing that it represents a qualitatively distinct evolution from conventional Shadow IT — meaning unauthorized hardware or software — precisely because generative AI is characterized by opacity, autonomy, and probabilistic outputs. Those qualities make reviewing what the system produced yesterday considerably harder than reviewing a spreadsheet built with an unlicensed application.
Who Owns the Design Concept?
This is where the matter acquires legal and ethical dimensions that architects’ professional bodies are not yet equipped to address — not in the Arab world, and not in most countries anywhere. When a designer at your practice writes a prompt directing a generative model to develop an architectural concept for a client-funded project, to whom does the intellectual property of that concept belong?
The standard answer is: to the practice, under the terms of the employment contract. But that answer collapses when the question is pressed further. Does the practice own the prompt through which the concept was formed? Does it own the design thinking methodology the employee developed through two years of daily experimentation? Does it own the AI agent trained specifically on that designer’s sensibility?
The legal void here is wide and genuinely unsettling. Research conducted by Allen, Burton, Smith, and Wood revealed that employees who use unsanctioned tools within their organizations face penalties in subjective performance evaluations and reduced bonuses, even when their outputs are demonstrably excellent. The designer using a personalized, informal AI system therefore lives inside a compounded contradiction — producing better work while being penalized more severely for it. That contradiction does not encourage disclosure. It encourages concealment.
The Architectural Studio and the Social Learning of Digital Secrets
What makes this phenomenon more complex still is that it does not spread through announcement but through quiet observation. The less experienced colleague watches the senior designer complete a presentation model at an improbable speed, attempts to replicate the result, asks a casual question over lunch, learns the tool, and develops it further in turn. In precisely this way, the shadow infrastructure propagates through the practice the way any design sensibility does: through imitation and social reinforcement, not through formal training.
This is precisely the dynamic documented by Marcel and Clark in a startup context, where they demonstrated that Social Learning Theory accounts for the spread of Shadow IT tools through three mechanisms — observational learning, behavioral modeling, and peer reinforcement. When the absence of any clear formal policy is added to these mechanisms, the architectural studio becomes a productive environment for the proliferation of undocumented tools.
The irony is that studio culture — inherently experimental and oriented toward creative risk — cultivates this dynamic more readily than almost any other institutional environment. The studio rewards initiative and tolerates ambiguity; it does not reward bureaucratic caution. That same climate is precisely what makes building a personal AI prompt library feel entirely natural and professionally coherent — until its author departs.
When AI Becomes an Instrument of Control Rather Than Capacity
A parallel trajectory exists that carries equal concern: what happens when management decides to formalize AI and introduce it as an official organizational system. Research conducted by Monod and colleagues examining an AI-assisted sales tool at a Chinese firm reveals a disquieting dynamic. A tool designed to empower workers by providing richer customer information was progressively appropriated by management for surveillance and control — managers used it to track the tone and speech patterns of employees, monitor compliance, and plan the partial replacement of human staff with full automation.
This shift from empowerment to control is not an anomaly. It is a recurrent pattern that emerges when management discovers the monitoring affordances that AI tools make available. In the architectural studio context, this means that the partner who decides to “regularize” AI use and integrate it formally may find — whether intentionally or not — that the resulting system monitors designers’ productivity, their conceptual approaches, and the speed of their client responses.
Holmström and Hällgren add an analytically precise dimension to this picture, distinguishing between different organizational AI contexts based on two axes: the degree of algorithmic management and the degree of transparency. Architectural studios in which personal AI tools operate typically fall into what they describe as a low-transparency zone — where neither the user fully understands how outputs are generated, nor will the colleague who inherits the work afterward be able to reproduce it.
When the Expert Leaves the Room Empty
Perhaps the most consequential dimension of this phenomenon surfaces at the moment of departure. When an employee who has assembled a complex personal digital infrastructure leaves an architectural practice, they take with them not only their expertise and professional memory — which every resignation involves — but also the operating logic of a workflow that was never documented, a methodology that was never written down, and outputs that cannot be reviewed or reproduced because no one understands how they were built.
Massingham contributed a rare longitudinal study on the organizational impact of knowledge loss, finding that the gap left by the departure of skilled employees does not close with time, and that incoming staff do not offset the loss of departing experts. When the departing figure is one who carried an elaborate personal digital infrastructure, the gap compounds: the practice loses human expertise and technical infrastructure simultaneously.
The consequences extend beyond the individual. If that designer occupied a central position in the practice’s internal communication network — a dynamic identified by Younis, Ahsan, and Chatteur, who linked central network position to patterns of collective departure — their resignation may initiate a successive wave of exits among connected colleagues, accompanied by a collective loss of undocumented organizational knowledge.
Toward a Digital Studio Policy: What Needs to Happen
At the global scale, several prominent architectural firms have begun addressing this matter seriously, even where transparency about the specifics of their policies remains limited. Zaha Hadid Architects, whose early and sustained commitment to algorithmic and computational design is well established, operates a model in which digital design tools constitute part of a shared institutional infrastructure rather than individual property — an approach that emerges in part from the integrated, database-dependent character of the firm’s design method. Bjarke Ingels Group similarly applies increasing institutional attention to documenting computational design processes and workflow automation in ways that ensure transferability and reviewability, driven by the operational demands of managing projects across multiple continents with teams that do not always share personnel.
Most architectural practices, however — and certainly the large majority of practices across the Arab world — lack any institutional framework that addresses this territory. Research by Nault and Ruhi in the field of knowledge management suggests that the answer does not lie in prohibiting personal AI tools but in connecting them to the organization’s shared knowledge infrastructure — meaning that the designer integrates their AI routine into a documented system the practice can access, audit, and update, rather than leaving it as a hidden file on a personal device.
In practical terms, the digital studio policy that is now overdue needs to address four foundational concerns. First, disclosure: the working environment must encourage the surfacing of personal tools rather than their concealment, which will only occur where psychological safety is genuine and punitive evaluation practices are absent. Second, documentation and versioning: designers should be expected to document their prompt libraries and update them within a shared system, in the same way that project drawings are documented and version-controlled. Third, audit trails: significant AI-generated outputs should be traceable and reviewable. Fourth, workflow portability: a designer’s digital routine should be transferable to a successor without loss of continuity.
Arab architects’ professional associations have not yet incorporated any provision addressing AI into licensing requirements or codes of professional ethics — a notable absence at a moment when the pace of change is accelerating. The professional ethics framework that governs how an architect stamps and assumes responsibility for drawings has not yet adapted to accommodate the question of whether it is professionally acceptable to stamp a drawing whose generative concept was produced by a tool the designer cannot explain to anyone.
The architectural studio has always been a place where ideas entangle and inform one another, where relationships form through shared critique and mutual learning. What Shadow AI introduces today is not a threat to that culture but a mirror held up to the absence of shared agreement on a foundational question: who owns what we produce together when our production is increasingly generated by invisible tools? The practices that answer that question honestly and with clear policy — before they are forced to by circumstance — will hold a competitive advantage measured not in the speed of their outputs but in the durability of their institutional memory.
✦ ArchUp Editorial Insight
The article under analysis does not begin with architecture. It begins with a departure — a designer who resigns and carries away something the employment contract was never written to describe. That framing is structurally precise, because the actual subject is not artificial intelligence; it is the liability transfer that occurs when an organization benefits from a personal digital infrastructure it never formally acknowledged, never documented, and never governed. The designer who built a private prompt library, trained custom agents on two years of client-facing work, and automated workflows that made them measurably more productive than their colleagues was, in effect, constructing a parallel operational system inside the studio — one that the practice extracted value from daily while bearing none of the institutional cost of maintaining, auditing, or transferring it. When that designer leaves, the organization discovers that what it treated as individual performance was in fact shared infrastructure held under a single person’s credentials, on a personal device, outside any version-control system the practice controls. The loss is not the person. The loss is the undocumented system the person had become. Research by Massingham confirms that this gap does not close with time and is not offset by incoming staff — a finding that becomes structurally acute when the departing figure occupied a central position in the practice’s internal collaboration network, since Younis, Ahsan, and Chatteur’s work on voluntary turnover demonstrates that central connectors do not leave alone; their exit reorganizes the network around them. What the article identifies as a “disclosure problem” is more precisely a psychological safety deficit produced by exactly the punitive evaluation dynamic Allen, Burton, Smith, and Wood documented: the employee who performs better through unsanctioned tools is penalized in subjective assessments, which drives the tool underground rather than into institutional memory, which guarantees the knowledge loss the practice most needs to prevent. The Shadow AI condition in the architectural studio is therefore not a technology governance failure. It is the predictable outcome of an incentive structure that simultaneously extracted productivity from personal digital infrastructure and punished the transparency that would have made that infrastructure transferable — a pattern this archive identified at the organizational scale in The Studio That Was a Person, where tacit knowledge held by a single practitioner was revealed as a solvency risk only at the moment of succession, and in Architectural Attention Economy, where individual cognitive protocols generated institutional value that remained unaccountable to any shared system. The four-part policy framework the article proposes — disclosure, versioning, audit trails, workflow portability — is technically sound, but each of its components requires a prior condition that the article names but does not fully pursue: the organization must first remove the punishment for surfacing the tool before it can expect the tool to be surfaced. Without that structural change, disclosure norms remain aspirational, versioning remains voluntary, and audit trails document only the workflows employees were willing to admit to. The deeper question the article raises for architectural practice specifically is one of professional liability: the architect who stamps a drawing assumes legal responsibility for its content, but no equivalent accountability mechanism exists for the generative prompt that produced the concept the drawing encodes. Arab professional associations have not yet written that clause. When the consequence of that omission materializes — in a disputed concept, a failed handover, a client claim against a practice that cannot reconstruct how its own design was produced — the party who absorbs it will, with structural consistency, be the one who had no seat at the governance table where the policy silence was maintained.
References
Ross, J.A.J., Hibbert, L., and Moss, E.J. “Shadow AI: Governance, Risk, and Organisational Resilience.” International Conference on Artificial Intelligence, Computer, Data Sciences and Applications (ACDSA), IEEE, 2025.
Allen, D., Burton, F.G., Smith, S.D., and Wood, D.A. “Shadow IT Use, Outcome Effects, and Subjective Performance Evaluation.” SSRN Electronic Journal, 2017.
Silic, M., and Back, A. “Shadow IT — A View from Behind the Curtain.” Computers and Security, 2014.
Marcel and Clark, M. “A Social Learning Theory Analysis of Peer-to-Peer Problem Solving and Shadow IT Emergence: A Case Study Approach.” 13th International Conference on Awareness Science and Technology (iCAST), IEEE, 2025.
Nault, K., and Ruhi, U. “From Knowledge Management to a Learning Organization: Strategic Use of AI Tools.” IEEE International Conference on Technology Management, Operations and Decisions (ICTMOD), IEEE, 2024.
Monod, E., et al. “From Worker Empowerment to Managerial Control: The Devolution of AI Tools’ Intended Positive Implementation to Their Negative Consequences.” Information and Organization, 2024.
Holmström, J., and Hällgren, M. “AI Management Beyond the Hype: Exploring the Co-Constitution of AI and Organizational Context.” AI and Society, 2021.
Khoza, L.T. “Managing Knowledge Leakage During Knowledge Sharing in Software Development Organisations.” SA Journal of Information Management, 2019.
Massingham, P.R. “Measuring the Impact of Knowledge Loss: A Longitudinal Study.” Journal of Knowledge Management, 2018.
Jennex, M.E. “A Proposed Method for Assessing Knowledge Loss Risk with Departing Personnel.” VINE, 2014.
Younis, S., Ahsan, A., and Chatteur, F.M. “An Employee Retention Model Using Organizational Network Analysis for Voluntary Turnover.” Social Network Analysis and Mining, 2023.
Hickok, M., and Maslej, N. “A Policy Primer and Roadmap on AI Worker Surveillance and Productivity Scoring Tools.” AI and Ethics, 2023.
