Research

Democratizando a fabricação digital em Fab Labs: efeitos cognitivos das interfaces agênticas no projeto
João Ariedi (João Carlos Ariedi Filho), Frons Educação, Brazil · Dorival Campos Rossi, São Paulo State University (UNESP), Faculty of Architecture, Arts, Communication and Design, Bauru, Brazil

9th International Congress Media Ecology and Image Studies (MEISTUDIES), 16–19 November 2026, Mesa 06. Expanded abstract accepted 20 August 2026; full text (5,600 words, 17 references, in Portuguese) submitted 7 October 2026; asynchronous presentation delivered 31 October 2026.

Why the full text is not published here yet

MEISTUDIES routes accepted full texts to one of its publication outlets: the congress ebook (Ria Editorial) or a partner journal (Razón y Palabra; Hypertext.net, Scopus-indexed). That editorial decision is pending, and journals require that the text not have been published elsewhere in full. Until the venue is decided, this page carries the abstract and an extended summary written for it; the full text and its DOI will be linked here the moment they appear.

Abstract

The diffusion of agents built on large language models has accelerated two central activities of cognitive work, information synthesis and content production, and shifted the bottleneck toward judgment. This article discusses what that shift means for design work in Fab Labs, spaces created to democratize digital fabrication that depend on a fragile cycle of person-to-person training around equipment such as the laser cutter. A theoretical essay grounded in the literature on work and automation, situated learning, knowledge management and distributed cognition, and in the authors' Fab Lab practice, it examines a configuration in which the user describes in natural language what they want to fabricate, an agentic layer produces and verifies the vector design, and the human evaluates, adjusts and approves. Five axes organize the discussion: the new mode of operating machines, gains, possible losses, risks, and learning through the agent. We argue that the change is not a paradigm shift but a syntagm transition: whole links of the activity chain are absorbed by the agentic layer, and decision becomes the core human contribution. Every gain on the absorbed axis casts a shadow on the human axis, and the sharpest consequence is a reproduction problem: decision-oriented profiles thrive in the new regime, yet such profiles are formed by the very practice the agent absorbs. Learning through the agent is therefore the condition for democratization to survive its own success. It closes with the agenda of a forthcoming master's study that will test these claims in a university Fab Lab.

Keywords: agentic interfaces; digital fabrication; Fab Labs; situated learning; tacit knowledge

Extended summary

The configuration. In a digital fabrication lab, the vector drawing that feeds the laser cutter is the central design image, and until now it was drawn. In the configuration the article discusses, it is described: the user says in natural language what they want to fabricate, an agentic layer produces the drawing and checks it against machine and material constraints, and the human evaluates, adjusts and approves. For a media ecology audience the point is immediate: the agentic interface is not a new tool but a new medium interposed between intention and matter. The design image is written before it is seen. Fab Labs were created to democratize personal fabrication (Gershenfeld, 2005), yet they depend on a fragile training cycle: the knowledge needed to design for and operate a laser cutter is tacit, situated and passed person to person (Lave & Wenger, 1991; Nonaka & Takeuchi, 1995). When the trained user leaves, the lab starts over. That cycle, not the machine, is what agentic mediation promises to attack.

Thesis. Democratization through agentic interfaces is real, but it is not access to machines: it is a redistribution of competences, and every gain on the axis the agent absorbs casts a shadow on the axis that stays human. Borrowing the structural distinction between the axis of substitution and the axis of combination (Saussure; Jakobson), the article argues that the change is not a 'paradigm shift' but a syntagm transition: whole links of the design activity chain are absorbed by the agentic layer, and decision becomes the core of the human contribution. The text is a discussion, not a report of results; it is the essay that opens the empirical research.

From direct manipulation to conversational operation. Operating a laser cutter traditionally is a chain of direct manipulations, most of them tacit: which kerf allowance to use, which speed stops MDF from charring, how to hear that something is wrong. Norman (2013) named the distance between intention and available actions the gulf of execution, and the distance between what the artifact shows and what the user understands the gulf of evaluation. Agentic mediation reshapes that geography asymmetrically: the gulf of execution collapses, since intention reaches the cut file without the user mastering the toolchain, while the gulf of evaluation widens, since the user must now judge a proposal they did not build. The regime eases exactly what used to form the operator and demands exactly what the operator only acquired by being formed. The interface becomes an interlocutor with codified expertise; operating becomes negotiating; responsibility migrates from executing correctly to accepting responsibly. Seen through distributed cognition (Hutchins, 1995), what changes is the topology of the system: what circulates between user, agent, machine and material is no longer motor skill and parameter knowledge but intentions, proposals and verifications. One caveat keeps the case honest: the agent never fires the machine. A wrong proposal costs material, machine time and possibly safety, so the Fab Lab is a microcosm where the gap between looking right and being right cannot be ignored.

Gains. Four, each real and each priced. Acceleration: the agent compresses the synthesis and production links, and the bottleneck migrates to formulating intention and evaluating proposals (Eloundou et al., 2023; Acemoglu & Restrepo, 2019). Access: the vector-design barrier falls for anyone who can describe what they want, bringing in the artisan who knows the material but not the tool, the teacher with a lesson plan but no file, the student from another field. If the first democratization gave access to machines, the second gives access to design. Organizational memory: the interface is an instrument of externalization in Nonaka and Takeuchi's sense; constraints, allowances and recurring errors become rules the agent applies, and the knowledge survives the departure of the key user. Error prevention as embedded pedagogy: the checks encode the community's accumulated 'why nots', and if the agent explains a refusal, every interception is a learning moment. That last gain depends on a design condition the article returns to: the same check that teaches when it explains merely substitutes when it stays silent.

Losses. Four shadows of those gains. The ladder loses its lower rungs: legitimate peripheral participation (Lave & Wenger) presupposes a periphery of simple tasks, and those are precisely the most codifiable tasks the agent absorbs first; the lab may discover too late that it produced users without producing members. Atrophy of embodied competence: parameter intuition and feel for the material, the body's knowledge Merleau-Ponty describes, are cultivated by direct, repeated operation and silently stop being cultivated. Loss of productive failure: the piece that did not fit taught about allowances, the burnt edge about power; the Fab Lab literature itself had already warned against the impressive, empty 'keychain' the beginner produces on a first visit (Blikstein, 2013), and an interface that delivers valid artifacts without the struggle can multiply keychains. Loss of reconstructibility: the artifact arrives without its reasons unless the agent supplies them; knowledge becomes consumable rather than reconstructible, an externalization that never completes into internalization.

Risks. Where losses meet design and governance decisions. The scaffold that is never removed, re-enacting Bainbridge's (1983) ironies of automation: the better the agent handles the common case, the less prepared the human is for the uncommon one, which is exactly where their decision matters. Miscalibrated trust, the hinge of the article: the human-as-decider is the regime's safety mechanism, and that mechanism presupposes the very evaluative competence the regime erodes. A decider who approves what they cannot judge is not deciding but ratifying. Homogenization: shared priors flatten the paradigmatic variety of what a lab produces; access democratizes and output converges, against a maker culture born of variety and, in Brazil, of the gambiarra. Concentration: whoever codifies the rules governs the practice, so the externalized knowledge demands curation and an open process of change. An empirical warning gives weight to all of this: in a long-horizon benchmark of pedagogical agents (Lee et al., 2026), agents performed reasonably within each turn but failed almost uniformly to compose sessions into a coherent formative sequence, and the agents that always answered scored worst on pedagogical criteria. In the article's terms, agents handle the paradigmatic register, selection, well and the syntagmatic register, composition toward an end, poorly.

Learning through the agent. Beyond socialization (master to apprentice) and internalization of explicit material (the manual nobody reads), agentic mediation opens a third mode: dialogic, at the moment of need, in the learner's language, about the piece on the screen. The promise is real and the danger is documented: obtaining answers is not building competence, and an agent that delivers the perfect cut file on every request can be an excellent executor and a terrible teacher, with the two indistinguishable from the artifact. Five design principles follow: explain while doing; expose the reason behind every check; withhold strategically, since the right answer is sometimes a question back or a suggested test on scrap; remove the scaffold as competence forms; and let externalized rules be read, not only executed. These converge on the article's sharpest formulation, the reproduction problem: the regime favors decision-oriented profiles, but deciders are formed by the peripheral practice the agent absorbs. If agentic mediation is not also a place where deciders are formed, it will consume the stock of deciders it inherited and not replace it. Learning through the agent is therefore the condition for democratization to survive its own success.

Analysis: the syntagm transition as key. A transition is paradigmatic when it replaces the elements occupying the positions of a stable chain; digitalization was the exemplary case, since synthesis, decision and production remained the activities of design work and only their medium changed. A transition is syntagmatic when it alters the chain itself: links are suppressed, merged or created, and the meaning of the remaining links is redefined. That is what agentic interfaces do to fabrication design. The key explains the distribution of gains and shadows: gains concentrate on the paradigmatic axis of selection, shadows on the syntagmatic axis of composition, and the regime compresses the axis on which it was good to be formed while overloading the axis on which no one is being formed any more. Each gain has its named tension: memory with ossification, access with dependence, error prevention with loss of the formative error, acceleration with homogenization. None is an argument against agentic mediation; together they are the map of what interface design and lab governance must manage, and each is an observable variable.

Research agenda. The master's project this essay opens will develop and validate an agentic interface for a university Fab Lab's laser cutter, organized on four fronts: access (does mediation raise beginners' success rate and shorten time to first valid cut against traditional training?), artifact quality (does increased use come without degrading blind expert evaluation or flattening variety?), knowledge retention (does codification reduce dependence on key users, and is what is codified still read and revised?), and competence profiles, the front that contains the others: with synthesis and production absorbed, do the competences that distinguish successful users become decisional in nature, and does the mediation form those competences or merely presuppose them?

Selected references

Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks. Journal of Economic Perspectives, 33(2), 3–30.
Bainbridge, L. (1983). Ironies of automation. Automatica, 19(6), 775–779.
Blikstein, P. (2013). Digital fabrication and 'making' in education. In FabLab: Of machines, makers and inventors. Transcript.
Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs. arXiv:2303.10130.
Gershenfeld, N. (2005). Fab: The coming revolution on your desktop. Basic Books.
Hutchins, E. (1995). Cognition in the wild. MIT Press.
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge University Press.
Lee, U., et al. (2026). EduClaw-Bench: A long-horizon benchmark for pedagogical LLM agents with simulated learners. arXiv:2608.03206.
Nonaka, I., & Takeuchi, H. (1995). The knowledge-creating company. Oxford University Press.
Norman, D. A. (2013). The design of everyday things (rev. ed.). Basic Books.
Rossi, D. C., Contini, G. C., & Abiko, K. P. (Eds.). (2023). FAB LAB: O futuro da fabricação digital no Brasil. FAAC-UNESP.

Use of generative AI

Generative AI tools were used in preparing the article (structuring and editing, citation-reference integrity checks, file generation; reference prospecting, each reference verified at source). The object of discussion, the syntagm-transition thesis, the argumentative structure, the design principles and the theoretical interpretation are the authors' own, who take full responsibility for the content.

A cultura organizacional na transformação de organizações tradicionais em organizações AI first: um estudo de caso em um laboratório corporativo de inteligência artificial
Juliana Santos Egea (lead author) · João Ariedi (João Carlos Ariedi Filho, printed as "Joao Carlos Ariedi") · Ricardo Amaral Silva · Ailton Donizetti Fabri — all São Paulo State University (UNESP), Bauru School of Engineering (FEB), Brazil

XXXIII SIMPEP, Simpósio de Engenharia de Produção, Bauru, 3–6 November 2026. Area 6, Organizational management; sub-area 6.1, Organizational behavior. Expanded abstract (six pages, in Portuguese) accepted 10 August 2026 and presented at the symposium. Second author of four.

Why the full text is not published here yet

The research belongs to the lead author, and the paper's citable form is the SIMPEP proceedings (anais), expected after the event. Until the proceedings appear with ISBN and page range, this page carries an English abstract and extended summary written for it; the proceedings entry will be linked here when it is published.

Abstract

Written in English for this page; the paper itself carries no abstract.

Artificial intelligence has become one of the main drivers of digital transformation, and organizations across sectors have set out to become AI-first: organizations that treat AI as a priority element in executing processes, supporting decisions and generating value. Reaching that state demands more than technology investment; it demands organizational change capable of sustaining AI in everyday work, and organizational culture sits at the center of that change. This study analyzes the trajectory of a corporate AI laboratory charged with spreading AI practice across companies of the same business ecosystem, asking how organizational culture shaped the consolidation of the laboratory's initiatives in the transformation of traditional organizations into AI-first ones. An applied, exploratory, qualitative single case study, it draws on document analysis of the laboratory's materials and on semi-structured interviews with employees involved in its initiatives, analyzed through Bardin's content analysis. Results show that the transformation did not depend on technology adoption alone but on values, behaviors and practices that favored experimentation, continuous learning and the systematic use of AI; leadership support, cross-area collaboration, shared experience and continuous learning were the decisive enablers, while resistance to change, uneven maturity across areas, weak standardization, insufficient team integration and difficulties in prioritization and governance were the main obstacles. The main contribution is an Organizational Maturity Framework for AI-First Organizations, grounded in the case evidence and articulating the cultural, organizational and managerial dimensions that sustained the transformation.

Keywords: artificial intelligence; AI first; culture; organizational maturity; digital transformation

Extended summary

Problem. Organizations of every sector have started a journey toward becoming AI-first: incorporating artificial intelligence as a priority element in how processes run, how decisions are supported and how value is generated. The paper's starting observation, following Raisch and Krauss (2021) and Tarafdar, Beath and Ross (2023), is that this transformation requires more than infrastructure; it requires organizational changes able to sustain the incorporation of AI into work. Kane et al. (2019) found that digitally mature organizations distinguish themselves above all by a culture of innovation, experimentation, collaboration and continuous learning, and Schein and Schein (2022) define culture as the set of shared assumptions that guides how people interpret problems, learn and respond to change. In AI adoption contexts, a culture favorable to innovation strengthens collaboration, knowledge sharing, experimentation and adaptability, the elements the literature treats as fundamental to the sustainable incorporation of new technology (Faraj, Pachidi & Sayegh, 2018; Jaiswal, Arun & Varma, 2022).

The case. A corporate AI laboratory responsible for disseminating AI practice across companies in the same organizational ecosystem. Its initiatives spanned training, experimentation, knowledge sharing, the development of intelligent agents, the deployment of automations, the definition of good practice and governance for AI use, and the standardization of processes that bring AI into organizational activity. The laboratory also mapped AI opportunities with leadership and technical teams and ran pilot projects that strengthened a culture of innovation and continuous learning, which made it a fitting setting to observe how culture shapes such a transformation. The research question: how did organizational culture influence the consolidation of the laboratory's actions in transforming traditional organizations into AI-first organizations?

Method. Applied, exploratory, qualitative research through a single case study (Yin, 2015) of the laboratory within a Brazilian business ecosystem, chosen to understand a contemporary phenomenon in its real context. Two data sources: document analysis of materials produced during the laboratory's implementation (institutional presentations, workshop records, training materials, technical documents, standardized processes, organizational indicators, action plans, records of the AI-opportunity mapping, pilot projects) and semi-structured interviews with employees directly involved in the laboratory's initiatives, purposively selected across areas and organizational levels to capture different perceptions of the cultural changes observed. Data were analyzed with Bardin's (2016) content analysis to identify thematic categories (leadership, continuous learning, cross-team collaboration, knowledge sharing, experimentation, trust in AI use, governance) and then confronted with the literature on culture, digital transformation and organizational AI adoption to identify convergences and divergences.

Results. The transformation did not depend exclusively on adopting new technologies but on incorporating values, behaviors and practices that favored experimentation, continuous learning and the systematic use of AI in organizational activity. Documents and interviews showed that the laboratory's initiatives (mapping AI opportunities with leaders and technical teams, workshops, training, communities of practice, experimentation with different AI tools, development of intelligent agents, deployment of automations, definition of usage standards and good practice) strengthened knowledge dissemination, widened trust in using AI and brought technology initiatives closer to the organization's strategic goals. Leadership support, collaboration across areas, shared experience and continuous learning were the decisive factors driving adoption and consolidating an innovation-oriented culture. On the other side, the study identified what slowed the process: resistance to change, differences in maturity between areas, limits to the standardization of practice, the need for greater integration between teams, and difficulties in prioritizing initiatives and consolidating governance for AI use.

Contribution: an Organizational Maturity Framework for AI-First Organizations. Built on these findings, the framework (the paper's Figure 1, a conceptual proposal by the lead author) represents the main cultural, organizational and managerial dimensions that sustained the organization's evolution: leadership, organizational learning, knowledge sharing, experimentation, governance, standardization and the incorporation of AI. It is offered as a guide for organizations seeking to accelerate their AI adoption journey, giving an integrated view of the cultural factors that favored transformation, and as a reference for managers building environments able to incorporate AI continuously and strategically.

Conclusion. The growing incorporation of AI shows that the challenges of digital transformation go beyond implementing technology and involve deep change in organizational culture. Understanding how values, practices and behaviors are developed and shared was essential for AI to be incorporated into organizational processes rather than remaining a tool. By analyzing the laboratory's trajectory, the study widens the understanding of the cultural factors that sustain the move from traditional to AI-first organizations and helps close a gap the authors identify in the literature: empirical evidence on how culture shapes the consolidation of AI initiatives. The consolidation of that process depended above all on building a culture aligned with innovation, continuous learning, knowledge sharing and the systematic incorporation of AI into work routines.

Where it sits in this research line. The case is the AI-first transformation João Ariedi advises on as an engineering consultant (two co-authors are from the companies involved). It is the publication where a decade of organizational-intelligence work meets the AI transition, and it pairs with the MEISTUDIES article above: that one asks what agentic mediation does to the competences of the person at the machine; this one asks what an organization must become for AI to take hold at all.

Selected references

Bardin, L. (2016). Análise de conteúdo. Edições 70.
Davenport, T. H., & Mittal, N. (2022). All-in on AI: How smart companies win big with artificial intelligence. Harvard Business Review Press.
Jaiswal, A., Arun, C. J., & Varma, A. (2022). Rebooting employees: Upskilling for artificial intelligence in organizations. International Journal of Information Management, 64.
Kane, G. C., Phillips, A. N., Copulsky, J. R., & Andrus, G. R. (2019). The technology fallacy: How people are the real key to digital transformation. MIT Press.
Schein, E. H., & Schein, P. (2022). Organizational culture and leadership (6th ed.). Wiley.
Tarafdar, M., Beath, C., & Ross, J. (2023). Building organizational capabilities for artificial intelligence. MIT Sloan Management Review.
Vial, G. (2019). Understanding digital transformation: A review and a research agenda. The Journal of Strategic Information Systems, 28(2), 118–144.
Westerman, G., Bonnet, D., & McAfee, A. (2014). Leading digital: Turning technology into business transformation. Harvard Business Review Press.
Yin, R. K. (2015). Estudo de caso: Planejamento e métodos (5th ed.). Bookman.