Concepts of our own: the vocabulary of the bridge

At Distribuidora Norte, the “last purchase” field meant something different at each branch, and what the salespeople knew about customers on the verge of leaving had no name in any system. Neither problem was technical. Both were problems of words: one that meant too many things, and a kind of knowledge that had none.

On ground where business and technology misunderstand each other all the time, because they use the same words for different things, a precise vocabulary is already half the translation done. This chapter gathers the terms with which the paradigm thinks and writes. They are not textbook definitions but operational senses, built for the work: each one names something that, without the word, would go unnoticed. When a term has an academic owner, it is flagged and referred to Authors and currents.

Readers who have gone through The four links have already seen several of these terms at work. In the case narrative they appear in context; here they are defined for reference. The two readings complement each other: the story teaches the method and the vocabulary makes it precise.

Core terms

Sociotechnical bridge

The paradigm, and the practice that embodies it, that combines in a single competence the ability to read the human fabric of an organization and the ability to build its technical system, without delegating either one. Coordinating two specialists is project management. The bridge is something else: a competence that takes on the translation instead of outsourcing it. → The thesis of the bridge

Translation in flesh and blood

The operational version of translation in actor-network theory.¹ In that theory, to translate is to align the actors’ interests so that an artifact stabilizes. “In flesh and blood” means that translation is not left to a process or to a requirements document passed from hand to hand, but to an integrated practice that preserves the meaning of the human problem all the way to the code. It stops being a step, which degrades at every handover, and becomes a continuous practice.

Translation gap

The structural gap in the market between those who understand people but do not build and those who build but do not read people. That is where digital transformation projects die, and that is the space the bridge occupies. → The thesis of the bridge

The loop of the four links

The method: sociology of organizations, software engineering, data engineering, AI architecture and governance and, once again, the person. What is distinctive is not the four disciplines, which exist separately on the market, but the fact that they form a closed cycle without handovers, so that the sociological hypothesis is not lost at the edges. → The four links

Sociotechnical diagnosis

The deliverable of link 1 and the precondition for everything else. It is a reading of the organization done before touching the technology: it separates the declared problem from the real one, maps power, incentives and resistance, and rules on where AI pays off and where it only destroys value. → The diagnostic method

Integrated pre-sales

Pre-sales (discovering the problem, designing a solution, sizing it and proposing it before anything is signed) carried out by the same person who will later build it. Committing scope and price is the moment when the translation gap costs the most, and separating whoever sells from whoever builds forces the hypothesis to be translated twice. It is translation in flesh and blood applied to the commercial phase, with the negative finding (“not here”) as an honest part of the scope. → The diagnostic method

Sociotechnical maturity

The degree to which an organization is ready for AI to add to it rather than subtract from it. It is not measured by infrastructure alone: it is multidimensional (strategy, culture, talent, governance and data quality), an idea anchored in research on AI readiness.² The IMIA model adapts and extends those categories to make them measurable, and its full definition lives in IMIA — the maturity instrument. Measuring maturity first is rule number one of the method.

From silo to architecture

Most organizations do not reach AI from an excess of failed technology but from scarcity: management by spreadsheet, data in unconnected silos, manual processes, decisions without evidence. The real opportunity, and one within reach today, is a constructive arc: connect the silos, build an information architecture, govern the data and only then add AI and agents where the organization is ready. The counterintuitive point is that starting from little or no digitization does not reduce risk but concentrates it. It forces the technical and the human subsystems to be built at the same time, which is exactly what the bridge knows how to do. → The bridge applied to small businesses

The adoption ladder

If from silo to architecture says where an organization starts, the ladder says where it climbs to, and that it climbs one step at a time: from everyday office tools to the systems that organize operations, from there to AI that anticipates rather than merely records and, on today’s step, to agents that execute end to end. The cost of having skipped the human reading grows with each step, which is why the paradigm does not age with the technology; the thesis develops that slope. → IMIA — the maturity instrument

Analytical ladder

Data also climbs one step at a time, in parallel with the adoption ladder. The descriptive step says what happened: how much was sold, how many orders came in. The diagnostic step explains why: which product dropped, which branch drifted. The predictive step anticipates what will happen: which customer will leave, which stock will run out. And the prescriptive step, at the top, recommends what to do: restock this, call that customer.

Many organizations believe they are on the predictive step because they bought an AI tool, but in practice they operate on a poor descriptive one: each area calculates “sales” its own way and nobody audits the definitions. Buying prediction on top of broken description only automates the error faster, with a nicer chart. That is why the diagnosis has to name the real analytical step before promising models. → Value metrics

Institutional memory

Before talking about data architecture, something more basic is needed: the organization must be able to give consistent answers to the questions that come up every week. “How many active customers do we have?” should not return three different numbers depending on who asks. Spreadsheets circulating by email, answers that “live in Juan’s head” and Excel versions where nobody knows which one is current do not reveal a lack of technology but a lack of reliable memory. Modeling on top of that, even in the most expensive cloud, only moves the disorder somewhere more visible.

Warning terms (what the thesis combats)

Automating the error faster

What happens when technology is mounted on a broken process or dirty data: the problem is not fixed, it is executed at greater speed and scale. It sums up why order matters: first understand, then automate.

The data that lies

Technically “correct” data (well typed, no nulls) that nonetheless does not say what people believe it says, because its meaning depends on a human context that was lost. One example is a “status” field that means collections to one team and shipping to another. The concept puts into practice the idea of Wang and Strong³: data quality is contextual and depends on use; it is not an isolated technical attribute. Almost always, data that lies is the trace of a badly incentivized human process. → Authors and currents

Sociotechnical debt

By analogy with technical debt, it is the liability that builds up every time technology is installed without resolving the social subsystem: resistance nobody worked through, processes nobody understood, users nobody involved. It does not show up in the code, but it is collected in zero adoption, sabotage and rejection, and it stays invisible until nobody uses the “perfect” system. Like all debt, it accrues interest: the later it is recognized, the more it costs to settle the human subsystem that was skipped.

Erosion of the scaffold

The silent decay of the path by which an organization develops its own experts, when AI takes over the entry-level tasks (the “easy” ones) that used to be the novice’s practice ground. Unlike unemployment from automation, which is visible and debated, it is a loss of future capacity that no usage metric records: productivity improves while the scaffold falls away. The academic owner is Beane, with the three C’s of craft learning (challenge, complexity, connection). It is a design risk and not a destiny, because the novice’s challenge can be deliberately preserved. → Psychology of adoption

Shadow AI (symptom, not crime)

The ungoverned use of generative AI tools outside any policy, often with internal data. Security frames it as a risk to be banned; the paradigm reads it as a diagnostic symptom. It is the same old workaround, now with a language model instead of a parallel spreadsheet, and like any workaround it exposes a latent demand. The answer is to read that need and govern it: inventory the use cases, classify the risk and channel them toward governed alternatives. That way shadow AI stops being only a risk and becomes an adoption roadmap. → Sociotechnical hygiene

The pre-sales handoff

The common consulting pattern in which whoever sells and sizes is not whoever builds. Someone closes scope, price and promise, and only afterward hands the project to a delivery team that never heard the client. It is the translation gap in its most expensive form, because the mismatch gets locked into the commitment. The symptom repeats itself: in the first week, the team finds out that what was sold is not what the organization needed, and the contract is already signed. → integrated pre-sales

Technological solutionism

The belief that every social or organizational problem has a solution that can be bought as a tool. It is the direct antithesis of the bridge, because it confuses having the technology with solving the problem. (It is a widely used term, adopted here as a recurring target and not as an invention of our own.)

Vanity metric vs. value metric

The vanity metric measures what is easy to measure and looks good on a dashboard: number of models, registered users, “AI usage”. The value metric measures what changes the day of the person in link 1: an hour freed up, a better decision, a gap that closes. The bridge measures value.

Metric anarchy

When each area calculates the same quantity its own way and two dashboards make two different truths official, the failure is not in the tool: what is missing is governance of meaning. Sales as the commercial team sees them are not the same as billing as administration sees it, and if nobody agrees on that, every new dashboard just adds a third version of the truth. The cure is not a neater chart. It involves naming an owner for each indicator, defining in a single sentence what each number means and retiring the indicators nobody uses to decide anymore. → Value metrics

Agile theatre

Agile rituals without increment or learning: the daily meeting that changes nothing, the sprint that ends in a demo nobody uses, the retrospective where everyone nods. The organization adopted the vocabulary of agility but not its method. The antidote is not more meetings but three concrete commitments: a shared Definition of Done (finished means usable in operations), reviews in which people from operations take part and not only the technical team, and a Product Owner able to prioritize within forty-eight hours when needed. → The four links

Technological lock-in

Dependence on a vendor that makes leaving costly, or impossible. Before signing, the verdict asks three simple questions. Can we export our data if we leave? Who owns the code that sets us apart? What happens if the vendor raises prices by forty percent next year? If the answers are uncomfortable, the problem is not technical but one of negotiation and contract design.

Value terms (the promise)

Value in the person, not in the model

The measurement principle of link 4: the deliverable is never a model but a better human decision. An impeccable model that changes no decision is worth zero; a modest system that gives a team back an hour a day is worth a great deal.

AI as an extension of judgment (not as a substitute)

The paradigm’s position on the role of AI, especially in government. AI does not replace the people who do the work (the public servant, the small business): it amplifies their judgment. Good AI gives decision-making power back to the person; bad AI takes it away and leaves them out. Its basis is human-centered AI.⁴ → The bridge applied to the public sector

Judgment in the design, not in every transaction

How AI as an extension of judgment survives when software stops suggesting and starts executing on its own. If an agent approves a refund, routes a complaint or prioritizes a case file without anyone looking at each case, the old “human in the loop” (a person reviewing one decision at a time) stops being possible and, at scale, stops being desirable. Judgment does not disappear: it moves. Instead of filtering decisions, it defines, audits and can revoke the rules by which the agent decides: what it does on its own, when it has to escalate to a person and what it records for review. The extension of judgment thus becomes the governance of judgment. That is the answer to anyone who fears that agentic AI will make the human reading obsolete. It makes it more necessary, because without auditable rules autonomy is delegation in the dark.

Context as an organizational asset

The knowledge an AI agent needs to operate with judgment, understood as an asset of the organization and not as a technical input prepared at the last minute. That context is made of the institutional memory and the bilingual glossary the method builds from link 1 onward. A diagnostic consequence follows: if an organization cannot explain its process to a person, it cannot hand it to an agent either. Illegibility, which used to be paid for in zero adoption, is now also paid for in the inability to delegate. That is why the work starts by making the organization legible; that legibility is the context the agent later receives. → The agentic extension

Closing the gap (not just delivering profit)

The definition of economic value used for small businesses and government. The result is not limited to one client’s gain: it includes shortening the distance between the few who already use AI and the rest of the productive fabric. In Latin America, where AI penetration is below 4%,⁵ that gap is both the opportunity and the mission. → The bridge applied to small businesses

AI as a multiplier with a sign

The synthesis that reconciles two claims that seem to clash: the mission is to close the gap, yet the evidence shows that badly applied AI widens it. There is no contradiction, because AI multiplies whatever it finds. On order and judgment, it multiplies value; on disorder and without judgment to filter it, it multiplies the harm. In the Kenya experiment it helped those who were already doing well and sank those who were doing badly.⁶ What decides the sign is not the power of the model but the human reading that guides it. That is why the bridge is much more than an ethical add-on: it is what sets the sign.

Process terms (SME context)

The maturity IMIA measures and the adoption ladder provide direction and a snapshot of a moment. In SMEs, the real configuration tends to be less stable: alignment between the technical and the social is not a destination reached once, but a series of windows that open and close. Amanollahnejad et al. (2026),⁷ in a qualitative study of 27 UK SMEs, call this episodic alignment: bursts of progress when there is sponsorship, reliable data and a pilot that worked; setbacks when management changes, the key person leaves or the patches stop holding. Geographic caveat: the study covers the UK in 2023–2024; the mechanism helps us think, not predict Latin America to the letter.

The study yields operational terms the diagnosis uses:

• Productive misalignment. Friction does not always need to be removed: sometimes it shows that a role is badly defined or that some data is unreliable. It differs from toxic misalignment, in which an error gets automated or a resistance with a valid reason gets ignored.
• Culture-strategy misalignment. The strategy on the slide says “innovate with AI”; on the floor, whoever makes a mistake or tries something different gets a reprimand. The technology advances in the demo and dies in operations (Gupta et al., 2020; Shokrollahi Yancheshmeh, 2026).
• Sociotechnical bricolage. Improvised integration (exporting a spreadsheet, re-keying data by hand, patching two systems together) as the usual way of keeping operations going when there is no large technical team. It is nothing to be ashamed of: it signals how far one can automate without first connecting things properly.
• Organized epistemic uncertainty. It goes beyond the data that lies: the organization has no agreement on which evidence to use when a decision is irreversible. If three people would give three different numbers for the same question, it is not yet time to model.
• Champion dependency. All the digital momentum tied to a single person, with no backup and no documentation that would survive their departure.
• Scale misfit. Copying a large firm’s solution or a generic guide and making one’s own operations worse, because the context is not the same.
• Coffee economy. Legitimate decisions made in relationship: at the counter, on the phone, over coffee with the supplier. Standardizing everything for an algorithm can feel like losing what makes the SME valuable.

None of these terms is a theoretical ornament. They help keep in view, amid the jargon, what is being done and for whom. Sooner or later, they all return to the same question: does this serve the person on the other side? A vocabulary that did not answer that would itself be one more way of automating the error faster.

See also: The thesis of the bridge · The four links · The diagnostic method · Bibliography