Authors and currents: what the thesis rests on

That the performance of a technology is decided in its human fabric rather than in its technical merit may be a comfortable opinion or a finding. The difference lies in where the claim comes from, and this chapter answers the question the method tends to leave hanging: what does all of this rest on? It is not required reading before the applications, and readers coming from The thesis of the bridge will find developed here names that appeared there in miniature, such as Tavistock, Latour or Kenya.

What follows is not a literature review. A review orders a field to show that one knows it; this chapter orders a field to show what an argument hangs from. There are eight currents, and two things matter about each: its core idea, reduced to what it contributes and without the deference of the academic summary, and above all the place it occupies in the structure of the thesis. The bibliographic record of each source, with its verification status, is in the Bibliography; what is offered here is the scaffolding.

Before taking them one by one, it helps to see the shape of the whole, because that shape is what gives the argument its force. The eight currents come from disciplines that do not keep company (postwar industrial sociology, the anthropology of science, information engineering, the economics of innovation, the social psychology of organizations, among others) and were written over seven decades, with no dialogue among them. And yet they converge. That traditions so distant reach the same point is what turns the claim into a finding. The scaffolding is not a collection of supporting citations but that convergence, ordered as a structure.

The foundation: the organization as a sociotechnical system

At the origin is the current that gives its name to everything else. In the early fifties, Eric Trist and Ken Bamforth, of the Tavistock Institute, studied the mechanization of English coal mining and ran into an anomaly that founded a discipline: an objectively superior technology lowered productivity.¹ The explanation was not in the machine. Mechanization had dismantled the social organization of the work, the small, self-regulating teams that shared out tasks, covered absences and sustained morale underground, without putting anything in place to fulfill that function. From that finding comes the principle that orders the discipline. Every productive organization has a social subsystem (people, roles, culture, relations) and a technical one (tools, processes, technology), and its performance does not depend on optimizing each separately but on their joint optimization. The two subsystems are optimized together or fail together: improving only one, however real the improvement, can degrade the whole.

The word “social” invites a misunderstanding, that of softness, and it has to be defused from the start. The social subsystem is not climate or motivation but the structure by which work is organized: who decides, who covers for whom, how autonomy is shared. The finest proof came from later comparisons within the same mining. One and the same longwall technology performed differently depending on how it was organized: when tasks were fragmented, performance fell; when the group’s self-regulation was preserved, it did not. The machine was identical in both cases, and what moved the result was the design of the organization of work. It is the cleanest demonstration that the social subsystem is an engineering variable and not a residue to be administered.

The later tradition of sociotechnical design turned that finding into method. With the ETHICS method, Enid Mumford added the operative corollary the principle had left implicit: if the social subsystem is part of the system, the design has to involve those who are going to use it.² Not out of courtesy or in pursuit of consensus, but because only they know the real process the system will have to support. User participation is, in that sense, requirements engineering. This current is the foundation of the argument: it justifies the central claim, every organization is a sociotechnical system, and the first rule of the method, read the social before touching the technical.

Mumford’s principle also has a face the present makes sharp: the institutional mechanisms through which that involvement is exercised in the face of AI. Recent literature documents them in the works councils that negotiate AI use in the German tech industry, enabled by the reform of its works constitution act.³ Participation stops being only a design technique (the workshop, the co-design of link

1. and takes on a layer of power: who has the formal right to weigh in on what a system decides about work. It is the natural extension of fair design toward the institutional, with a caveat the method does not hide. Those cases come from high-union-density contexts, and in the Latin American SME fabric the mechanism is rarely formal: the instrument changes, not the principle of giving the worker a voice.

The most recent research, in turn, qualifies the very idea of joint optimization. Amanollahnejad et al. (2026) extend sociotechnical systems theory (STST) to AI adoption in 27 UK SMEs. They show that joint optimization is not a destination but a recursive negotiation made of episodic alignment, technical bricolage and episodic leadership, and in doing so they qualify any narrative of linear, irreversible maturity.⁴

The mechanism: why an artifact “works”

Sociotechnical theory diagnoses that there are two subsystems, but it does not explain why the articulation between them is achieved or lost. That mechanism comes from the actor-network theory of Bruno Latour, Michel Callon and John Law, whose thesis is more radical: the technical and the social are not two spheres to be coordinated but a single network of actors, human and non-human. An artifact does not “work” by virtue of its internal properties. It works when it achieves translation, that is, when it enrolls the actors around it and aligns their interests enough for each to recognize itself in it and find a reason of its own to sustain it.⁵ Success stops being an intrinsic property of the technology and becomes a relational achievement. Where translation fails, the artifact turns out neither good nor bad: it is simply ignored.

This current names the verb of the thesis, to translate, and founds one of its own concepts: translation in flesh and blood, the idea that translation is done by a person and not by a document. It also closes off an illusion of method, the idea that adoption is a step after design or a training problem to be solved later. If “to work” is to have translated, adoption is the very test of whether the design understood anyone. Taken together, the sociotechnical current and the actor-network make up the thesis of the bridge: the problem is one of two subsystems (Tavistock) and the solution is to translate between them (Latour).

Translation, moreover, is not a single act but a process with moments, which Callon described in his founding study. First one problematizes: the matter is defined so that the other actors need to pass through oneself. Then one interests, fixing each actor in that role; one enrolls, when they accept it; and one mobilizes, when they act accordingly. Naming those moments serves the method because it makes it possible to diagnose where an adoption breaks. An initiative may have problematized well and fail at enrollment, or enroll a few and mobilize no one. Translation is not won or lost as a block but in stretches, and each stretch is a concrete place to look.

The motor: leading the change, not decreeing it

Posed this way, the thesis of the bridge is a static diagnosis. It says there are two subsystems and that success lies in aligning them, but it does not say how to intervene in the social subsystem to get there. That verb is supplied by organizational change management, which is why this current is the motor of the thesis: if the sociotechnical current diagnoses, change management is the therapy.

The canonical skeleton comes from Kurt Lewin and has three moments: unfreeze (create the need and dislodge inertia), change (the real transition) and refreeze (stabilize the new so it does not revert). John Kotter made it operational in eight steps, running from urgency and the guiding coalition to early wins and anchoring in the culture.⁶ Juan Díaz Barrios supplied the substance beneath the procedure: an integral change is sustained by delegation, communication, collaboration, participation and learning, not by the announcement.⁷ The three share one idea: the obstacle to change is human (habit, fear, meaning), not technical. That links at once to Mumford’s participation and to the enrollment of translation, because getting the actors to recognize themselves in the new is, in practical terms, to unfreeze and freeze again.

Here it is necessary to stop and be honest, because this is the current most exposed to passing off codified practice as science. The “Lewin model” as it circulates is largely a later construction. As the rereading by Cummings, Bridgman and Brown documents, Lewin never wrote “refreeze” as a closed model; the idea stayed in a minor subsection of his work.⁸ Kotter’s eight steps, for their part, are prescriptive literature of managerial practice, not research with a control group. The argument uses them as useful maps of a process, not as strong empirical theory. The empirical weight of the thesis is carried above all by the sociotechnical current and by the field evidence at the end of the chapter. Change management supplies the verb, and it is better to say so before a sharp-eyed reader points it out.

The same reading extends to culture, which is where change usually gets stuck. Edgar Schein (2010) distinguishes three levels of culture: artifacts, declared values and tacit basic assumptions. A serious diagnosis has to reach the third and not stop at the mission poster. Henderson and Venkatraman (1993) and Gupta et al. (2020) stress the alignment of business, IT and culture, and Westerman et al. (2011) add leadership and cultural preparation before scaling digitally. Shokrollahi Yancheshmeh (2026) reminds us that technological change without cultural change usually fails when infrastructure and talent are already scarce. The bridge uses these frames as a descriptive map leading to sociotechnical systems theory, not as an alternative doctrine.

On the execution side, the Agile Manifesto (2001) and the DORA research gathered in Accelerate (Forsgren et al.) bring that change down to short cycles with verifiable delivery, which link 2 develops. They complement the sociotechnical reading of link 1 without replacing it, and they are a reminder that the product evolves while users test it in their work: treating it as a project that “ends” on the day it goes into production destroys adoption.

From this current, the argument uses one corollary more than the models themselves: the rereading of resistance. In ordinary language, “resistance to change” names an irrational obstruction to be overcome. Read from the social subsystem, it is information: it says what the change threatens, whether a control being lost, a competence ceasing to be valued or a source of power written in no org chart. Whoever treats it as an obstacle loses the data; whoever reads it as a symptom finds the point in the social subsystem the design did not contemplate. Well-led change does not crush resistance: it deciphers it.

The descent into the AI era: human-centered AI

The three previous currents were born before contemporary AI, and one was needed to carry the sociotechnical principle into the era of models. That is human-centered AI. Ben Shneiderman proposes combining high automation with high human control and shows that they are not a trade-off: the best-designed systems are at once more autonomous and more governable.⁹ Virginia Dignum systematizes responsible AI (transparency, accountability, values built into design), and Cathy O’Neil documents the reverse: the concrete harm of opaque models that hide biases at scale and become unauditable just when they decide most about people’s lives.¹⁰

The core idea of the current is that AI should amplify people, remain under their control and give them confidence, not isolate or replace them. It is the direct translation of the sociotechnical thesis into the AI era, and it underpins link 4 of the method and two of its own concepts: AI as an extension of judgment and value in the person, not in the model. It is also where the theoretical scaffolding meets the normative frame. The auditability and human control that HCAI (human-centered AI) treats as good design are, on the regulatory plane, what instruments such as the EU AI Act or the NIST AI Risk Management Framework require. Theory and governance ask for the same thing in two different languages.

The most precise form of Shneiderman’s proposal is a change of axis. Common sense imagines a single lever, where more automation means less human control and vice versa. His framework splits that lever into two independent axes: the machine’s degree of automation and the person’s degree of control. The desirable quadrant is neither high automation with low control nor the reverse, but both high: very capable systems that the person still governs. That separation dissolves the false dilemma between power and control, and it is the technical version of the principle of joint optimization: one does not choose between the machine and the person, one optimizes them together. What makes it possible is explainability. A system that cannot be explained cannot be controlled and, therefore, is not adopted: it is tolerated, and abandoned at the first doubt.

The age of agents puts that quadrant to the test, because a system that executes on its own seems to leave no room for human control. The answer, developed in The agentic extension, is that control moves from each decision to the design of the rules by which the agent decides, and that the more autonomous the system, the higher the control axis has to be.

Between this current and actor-network theory there is a tension worth marking, because they hold up the argument from opposite assumptions. The actor-network is symmetrical by method: it treats humans and non-humans with the same vocabulary and refuses to privilege some over others in advance. Human-centered AI is asymmetrical by principle, because it puts the person at the center as the ultimate criterion of value. Far from cancelling each other out, they divide the work. The actor-network describes how a system actually stabilizes, who enrolls and which interests get translated; human-centered AI indicates the configuration toward which it should stabilize: the one that gives judgment back to the person. One says how the world is, the other where it is worth moving it. It is the same division that later governs the use of Rogers: description and norm are different tools, and they should not be confused.

The material requirement: data as fitness for use

A thesis on AI cannot stop at organization and judgment. AI feeds on data, and data have a current of their own that the argument needs. Richard Wang and Diane Strong established that data quality is not technical accuracy but fitness for use in a context, and they broke it down into dimensions that go well beyond correctness: intrinsic, contextual, representational and accessibility.¹¹ Thomas Redman supplies the economic and management side. He documents that poor data quality is above all a process problem, not a technology problem, and that it costs most organizations a far from trivial share of their revenue, on the order of 15% to 25%.¹²

This current holds up link 3 of the method and the concept of the data that lies: a piece of data can be technically impeccable and still lie, because its meaning depends on a human context that was lost at some point. Defining quality by fitness for use, rather than by accuracy, moves the problem from the database to the organization. Data are of poor quality when they do not serve the decision someone has to make with them, and that is not fixed with a script. Data quality is, at bottom, a social problem that shows up in a table.

That definition has two consequences that make it a management problem. The first is that quality is relative to the decision. The same dataset can be excellent for one decision and dreadful for another without a single value changing, because what changed was the use; there is no “quality” data in the abstract. The second is that quality is determined upstream, in the process that generates the data, not downstream, in the database that stores it. A piece of data is the trace of a human act: someone filled in a field according to a criterion. If that act is badly designed or badly incentivized, no later cleaning recovers what was never recorded. That is why data quality is not tackled with data quality tools, but by intervening in the process that dirties the data.

The prior condition: maturity and absorptive capacity

Before choosing what technology to adopt, an organization has to be in a position to absorb it. That is the current of maturity. (A note on order: the scaffolding is laid out in sequence of construction, from foundation to mechanism and motor, but in the method this current comes first, because measuring maturity before buying is rule number one of the thesis.) Jan Jöhnk, Malte Weißert and Katrin Wyrtki systematized AI readiness as a multidimensional phenomenon. From a qualitative study they distilled 18 readiness factors grouped into five categories: strategic alignment, resources, knowledge, culture and data.¹³ The lesson is that having technology is not enough; the organizational conditions to sustain it are needed. Those factors ground the concept of sociotechnical maturity, the rule of measure maturity first and the theoretical basis of the IMIA model.

But a list of factors describes a state, and maturity is a process. The dynamic mechanism Jöhnk lacks had been supplied thirty years earlier by Wesley Cohen and Daniel Levinthal with absorptive capacity: an organization’s ability to recognize the value of external knowledge, assimilate it and apply it depends on the knowledge it already has, and is therefore cumulative and path-dependent.¹⁴ That reveals something the snapshot of factors misses. An organization with no history of data or analytics cannot “buy AI” and absorb it all at once, because it lacks the prior knowledge that would make it legible. Maturity is not the inventory of factors present (the snapshot) but the learning capacity accumulated over time (the film). That is why the diagnosis that opens the method reads trajectory, not only state.

The mechanism also hides a property that inverts the usual intuition about adoption: absorptive capacity reinforces itself. Since it depends on prior knowledge, each investment in learning widens the base that makes the next learning cheaper. By the same token, its absence perpetuates itself: whoever has not accumulated cannot recognize the value of what they lack and so does not invest in acquiring it. Maturity cannot be skipped with a purchase, because what is missing is not the tool but the trajectory that makes it usable, and that trajectory is not sold separately.

The pace: adoption as a social process

So far the scaffolding explains why adoption is achieved or lost, but not at what pace or in what order it happens. Everett Rogers showed that the adoption of an innovation is not an event but a social process that unfolds over time and traces an S-curve. He also offered two lenses the argument uses again and again. The first divides the population into five categories according to when they adopt (innovators, early adopters, early majority, late majority and laggards), each with different motives. The second explains speed, which depends on five perceived attributes of the innovation: relative advantage, the best predictor, compatibility, complexity, trialability (being able to try it) and observability (seeing its results).¹⁵ The decisive word is the adjective. These are attributes perceived by the adopter, not objective technical properties of the artifact. It is the actor-network thesis in another key: what Latour framed as relational, Rogers specifies as perceived, because the weight lies not in the artifact but in the mind of whoever decides to adopt it.

The value of this lens is that it makes the phrase “adoption fails” diagnosable. An AI initiative does not stall because the technology is bad but because, for its real users, it scores low on compatibility (it clashes with how they work), on complexity (they do not understand it) or on observability (no one sees the result). The lens also explains a persistent pattern that Geoffrey Moore called the chasm: the pilot thrills innovators and early adopters and dies on crossing to the early majority, because it was designed for those who tolerate friction and not for those who demand that the thing simply work.¹⁶ From that pattern comes a rule of the method: design for the early majority from the start, not for the enthusiast.

No current calls for as much caution as this one, because it is the one that most easily slides from description into prescription. Rogers is a diffusionist: he tends to assume that adopting is good and that the only problem is speeding it up. The thesis of the bridge does not accept that assumption, because the Kenya experiment and O’Neil’s work show that adopting without judgment can amplify harm. That is why Rogers is used descriptively, to understand how and how fast something spreads, and never normatively, because adoption is not an end in itself. The normative filter is set by the human judgment that human-centered AI defends; without it, the S-curve may be spreading harm at scale, and faster. Nor should Rogers’s most-cited figure be over-read: his synthesis of studies attributes to the five perceived attributes between 49% and 87% of the variance in adoption rates, but that is an aggregate range from the literature, not a constant measured in an experiment.

Finally, Rogers contributes two ideas the method treats as design levers rather than passive descriptions. The first is reinvention: adopters rarely take an innovation as it comes but modify it as they adopt it, and that adaptation, far from being a deviation to correct, is often the condition for the adoption to last. The second is that two of the five perceived attributes, observability and trialability, are not fixed properties of the artifact. Design can manufacture them, by making visible a result that was hidden or by letting people try without commitment. Read this way, Rogers’s attributes stop being a thermometer of why something is not adopted and become a list of interventions. Each low attribute is a place where design can work before the S-curve decides on its own.

The substrate: tacit knowledge and the construction of meaning

The seven previous currents form a flow. The eighth is not a step in that flow but the ground on which all of them rest: tacit knowledge and the construction of meaning. Michael Polanyi put it in a phrase that became famous, “we know more than we can tell”: much expert knowledge is tacit, embodied in practice and impossible to codify fully in a manual or a document.¹⁷ Karl Weick supplied the complement. Organizations do not find the meaning of events, they construct it, retrospectively, socially and guided by plausibility rather than accuracy. They also enact the environment they later measure (in Weick’s vocabulary, they act on it and so partly create it), so that the categories with which they record the world end up shaping the world they see.¹⁸

These two ideas are the epistemic godparents of the book’s own concepts, which is why this current runs beneath the others. Translation in flesh and blood is Polanyi carried to the bridge. Reading an organization rests on tacit knowledge that does not travel by document, so translation has to be done by a person and not by a deliverable; for the same reason, actor-network translation has to be embodied. The data that lies is Weick. Data do not tell the truth on their own: they signify within a situated act of sensemaking. When that context is lost, or when the organization enacted the category it measures badly, the data remain impeccable and still deceive. Wang and Strong give the normative side (what quality data are); Weick’s sensemaking gives the descriptive side (why their meaning is fragile). Anchoring the book’s vocabulary in these two currents takes it out of metaphor: it stops being a happy image and becomes a defensible concept.

Here too the position has to be declared. Resting on Polanyi’s tacit knowledge sets up a known tension with Nonaka’s knowledge management, for which much of the tacit can be made explicit and codified. The thesis takes sides knowingly. What can be codified gets codified, and that is what systems, documentation and models are for, but the core of organizational judgment remains tacit, and that is why translation is done by a person. Codification is not denied; what is denied is that it exhausts knowledge.

Each of these ideas has an internal structure, and it is there, more than in the famous phrase, that it starts to operate. In Polanyi, tacit knowledge has a “from-to” form: one relies on particulars one cannot state, such as the touch of the hand or the reading of a gesture, in order to attend to a whole one does recognize. That is why one knows more than one can say, and why that knowledge is passed on by living alongside the practice, not by reading about it. In Weick, sensemaking is a cycle: the organization enacts the environment, selects a plausible interpretation of what happened and retains it as a frame for next time. The consequence for data is severe. An organization does not measure a world that was already there but the categories it enacted, and those categories then hand it back a world that confirms what it already believed. Data can lie not only by losing context but also by having been built not to see.

At this point the current crosses O’Neil’s, and the connection deserves to be made explicit. If an organization measures the categories it enacted, blind spots included, a model trained on those data does not learn reality. It learns the enacted categories and then reproduces and amplifies them at scale. A model’s bias is rarely born in the algorithm: it is born in the earlier sensemaking that fixed what gets measured and what gets ignored. The data that lies, once automated, becomes a bias that scales. For the method, the consequence is not obvious: the sociotechnical reading of link 1 is also bias control, because it audits the categories before the model inherits them without asking where they came from.

The evidence that confirms it

Without field evidence, the scaffolding would be only a theoretical edifice. Evidence is not theory, but it is part of the argument, and three pieces are enough.

The Kenya experiment showed that generative AI helped high-performing entrepreneurs, by around 15%, and harmed low-performing ones, by around 8%, because the latter followed generic advice without the judgment needed to filter it.¹⁹ It is the field proof that AI without human judgment to guide it can widen inequalities instead of closing them. UNCTAD’s Technology and Innovation Report 2025 documents, on another scale, that AI’s opportunity for development coexists with a “compute divide” that shuts out the smallest players: inclusion is not automatic, it has to be designed.²⁰ And CEPAL’s data show AI penetration below 4% in Latin America against more than 20% in Europe, along with a gap within each country as wide as the one between regions: in Brazil, 41% of large firms use AI against 11% of SMEs.²¹ That distance between the large firm and the SME, which the chapter on SMEs develops, is the terrain where the bridge works.

Where the thesis stands between hype and anti-hype

This academic scaffolding coexists with a noisier popular debate. It is worth placing, even though the thesis of the bridge need not take sides in it. At one extreme, technological optimism invites always seating AI at the table and reads it as an amplifier of agency. At the other, the critique dismantles Big Tech’s hype and teaches how to tell what AI can do from what it cannot.²² The bridge is indifferent to that dispute. If AI is as powerful as the optimists say, governing it is all the more urgent, because it is a multiplier with a sign; if it is as much smoke as the critics say, the diagnosis that separates where it pays off from where it does not is all the more necessary. The position that best fits the argument is the middle one: reading AI as normal technology, gradual and mediated by institutions, not as a singularity. It is the same stance that underlies adoption without teleology and the descriptive use of Rogers. Normality is not lukewarmness but the condition for institutional work, the bridge’s work, to be what decides the outcome.

How they are articulated (the scaffolding in one image)

   sociotechnical (Tavistock/Mumford) ── the problem is of TWO subsystems
              +
   actor-network (Latour)             ── success is to TRANSLATE and align actors
              =  THESIS OF THE BRIDGE
              ↓ enacted by leading the change
   change (Lewin/Kotter/Díaz B.)      ── the bridge is crossed, not only drawn
              ↓ applied to the AI era
   HCAI (Shneiderman/Dignum/O'Neil)   ── AI amplifies the person or harms them
              ↓ which imposes three requirements
   data quality (Wang&Strong)         ── without faithful data, AI lies fast
   maturity (Jöhnk; Cohen&Levinthal)  ── without capacity to absorb, AI subtracts
   diffusion (Rogers/Moore)           ── adoption is a social process, not an event
              ↓ confirmed by evidence
   Kenya / UNCTAD / CEPAL             ── without judgment or design, AI widens gaps

Beneath this whole structure run Polanyi’s tacit knowledge and Weick’s sensemaking. They are not a step in the flow but the epistemic substrate of the book’s own concepts that run through every link: translation in flesh and blood and the data that lies. Read this way, the chapter does not list eight authors. It shows a single conclusion held up by seven paths that did not know one another and by an eighth that grounds them from below. That is the strength of the scaffolding, and the reason the thesis does not depend on any one current.

See also: The agentic extension · The thesis of the bridge · The four links · Concepts of our own · The bridge applied to SMEs · Bibliography