The agentic extension of the thesis

None of the eight currents in the previous chapter ever saw an AI agent: Tavistock is from 1951, Latour from 2005, Rogers’s diffusion from the sixties. The book has been arguing that the thesis does not age with the technology, but so far it asserted this without building out what shifts when the technical system stops being an artifact that is used and becomes an actor that decides and executes on its own. This chapter builds that extension. It does not correct the eight currents of Authors and currents: it rereads them under agency. What it adds —which is nearly everything— are constructs of the paradigm, not attributable to any author. Where recent academic literature arrives, on its own, at the same place, I flag it as validating convergence —not as a source this is derived from.

Consider the limit: if Distribuidora Norte, on its second pass, delegated route assignment to an agent without auditable rules, the failure would no longer be a salesperson ignoring an alert —recoverable— but dispatches executed thousands of times with no witness. That shift of role is what this chapter is about.

From the artifact that is used to the actor that decides

The bridge thesis was built on an implicit image of technology: an artifact that a person uses, operates or consults. A billing system, an ERP, even a predictive model, do not decide: they record, order, suggest. The decision —and with it the responsibility— always stays on the human side. The predictive model flags a customer as at risk; it is the salesperson who decides to call. That frontier —the human as the last filter of every case— is the silent assumption on which the whole method rests.

The agentic layer breaks that assumption. It is the rung of the adoption ladder at which software stops suggesting and starts to decide and execute end to end: it approves a refund, routes a complaint, prioritizes a case file, triggers an action, without anyone looking at each case. It is not a leap in the model’s power; it is a leap in role. The technical system stops being something that is used and becomes something that acts.

Here appears the gap that none of the eight currents fills entirely. Actor-network theory already treated non-humans as actors —a breakwater, a speed bump, a key act in the network, enroll others, resist—.¹ But Latour’s non-humans act without deciding: they do what their material inscription fixed for them, they do not choose among courses of action according to a criterion they evaluate in the moment. The theory provides a vocabulary for the non-human that does; it has none for the non-human that decides.

What the paradigm calls the deciding actor fills that gap: the AI agent is an actor in the actor-network sense —it takes part in the network, enrolls, produces effects—, but of a kind that theory did not typify, because it decides among alternatives and executes its decision without a human filtering it case by case. It has no interests or intention —it is not a subject—, but it does exercise a delegated discretion: it operates within a margin of choice that a human entrusted to it. The sociotechnical problem stops being “do people adopt this artifact?” and becomes “who answers for what this actor decides, and by what rules does it decide?”. It is a conceptual decision, not a finding: extending the vocabulary of the actor preserves the symmetry that makes the theory powerful —humans and non-humans in the same network— and adds only what is missing, the margin of decision.

The thesis: it does not break, it shifts

Faced with the agentic layer, the temptation is to declare the method obsolete: if the system decides on its own, what is the point of the human reading, the control, the joint optimization? The answer of this extension is the opposite, and it is its thesis. What the paradigm calls the displacement of control: neither Tavistock’s joint optimization nor Latour’s translation breaks when the actor decides; they shift one level upward. Human control stops being exercised over each decision and comes to be exercised over the rules by which the agent decides —what it resolves on its own, when it is obliged to escalate to a person, what it leaves on record, how it is revoked—. It is the move from governing the decision to governing the governance of the decision. The primacy of the human does not disappear: it changes object.

The thesis was latent, scattered, in three places in the method —the fourth link (“human control moves one rung up, from the decision to the governance of the decision”), the concept judgment in the design, not in each transaction and the potential dimension of IMIA—. The three say the same thing from three angles; naming it as displacement of control makes it the axis of the extension.

It remains to specify what is preserved. Joint optimization said that the social and the technical subsystem are optimized together or fail together. Under agency, the technical subsystem now decides within the social: the interdependence does not loosen, it intensifies, because an agent badly coupled to the human fabric does not produce a tool no one uses —the classic failure, recoverable—, but decisions effectively taken that have to be undone. And translation now includes translating human judgment into rules for the agent to execute, an act that is itself sociotechnical, because it decides what judgment is delegated and what is not.

Each current, reread under agency

Actor-network. Translation in flesh and blood held that it is a person, not a document, who aligns the actors of an organization. The agentic layer adds an actor —the deciding one— and, with it, a twist: the agent is not only enrolled, it also enrolls. When it prioritizes case files, officials reorganize their work around its ordering; when it approves refunds, the fraud area adjusts its controls to what the agent lets through. That agentic enrollment is translation read in reverse: we do not only translate the world into the agent by giving it rules; the agent, once operating, translates us —it sets the pace, the format and the distribution of the human work that remains around it—. Hence the sociotechnical reading does not end when the agent goes into production: the agent changes the organization that had been read.

Human-centered AI. Shneiderman holds that high automation and high human control are not a trade-off.² The agentic layer seems to contradict him —if the agent decides on its own, does control not drop by definition?—, and the answer is no, if one understands where it is exercised. “High human control” stops meaning “a person reviews each decision” —which at scale would be impossible or would turn it into a rubber stamp— and comes to mean that a person designs, audits and can revoke the rules by which the agent decides. Shneiderman’s quadrant holds; what moves is the point of application of control: from the case to the rule. And Dignum and O’Neil come in with more force, not less. Accountability mattered when the model suggested, and it is structural when the agent decides, because it no longer falls on the person who pressed the button: there was none.³ The harm at scale that O’Neil documents changes in nature: an opaque model that suggests badly produces suggestions a human can still discard, whereas an agent that decides badly produces decisions already executed.⁴

Diffusion. The adoption ladder already places the agentic layer on its highest rung; the rereading adds a nuance about two of Rogers’s perceived attributes.⁵ Observability is inverted: the better an agent works, the less it is seen —it decides well, in silence, asking for nothing—, so that the attribute Rogers counted as a driver of adoption becomes a problem of governance, because what is not seen is not audited. And trialability —trying things out little by little— gets complicated when what is being tried out executes: an agentic pilot is not innocuous like one that suggests, it already acts. The descriptive and non-normative use of Rogers that the method maintains becomes sharper under agency: to diffuse an ungoverned agent quickly is to diffuse an actor that decides badly at the speed of the S-curve.

Tacit knowledge. Here the extension touches its most delicate nerve. The displacement of control supposes translating human judgment into rules, but Polanyi’s paradox —“we know more than we can tell”— says that the core of expert judgment is tacit and not fully codifiable.⁶ If that is true, not all judgment can be turned into rules: what is not codified is what makes the expert an expert. Hence what the paradigm calls the delegation frontier: the limit, mobile and always partial, between the judgment that can be carried over into auditable rules —thresholds, explicit criteria, typified cases— and the judgment that cannot, because it is tacit, contextual or loaded with conflicting values that no rule resolves without deciding for us. The frontier is not technical but sociotechnical: defining what falls on each side is the human reading of the agentic layer, and to design an agent is, above all, to decide well where to place it —and to oblige it to escalate to a person everything that falls on the tacit side—. Weick adds the final edge: if the organization enacts the categories by which it measures the world, an agent that decides on those categories not only reproduces the blind spots of constructed meaning, it executes them and, by deciding, reinforces them, because its decisions become tomorrow’s data.⁷

Governance rises with autonomy

From all of the above follows a rule: what the paradigm calls the proportionality rule. The governance a system requires is proportional to the margin of decision delegated to it. Automating a fixed-rule task —a report, an upload, a calculation— asks for little governance: it does not decide, it executes what is prescribed. An actor that decides among alternatives demands high governance: usage policies, risk management, auditable rules, escalation and revocation paths. Autonomy without proportional governance is not opportunity: it is delegation in the dark.

This does not redefine IMIA; it founds it. Its potential dimension already distinguishes fixed-rule automation from agentic automation and asserts that the agentic component raises the bar of the governance dimension; the proportionality rule is the principle of which that distinction is a case. That is why “high agentic potential over low governance is not opportunity, it is risk” is not a loose warning, and the governance gate —the integration level is not reached with governance below the floor— is proportionality turned into a hard constraint: the higher up the ladder, the higher the governance floor for the leap to count as maturity and not as exposure.

Three risks of its own

The agentic layer does not gradually aggravate the known risks: it introduces three that do not exist while the human filters each case. The first is execution at scale without a witness: the human in the loop did not only filter, they saw —each decision passed through a gaze—; when the agent decides on its own, that gaze disappears from the transaction and no one is a witness to the individual decision. The risk is not speed, which automating the error faster already named, but the absence of a witness, and governance has to restore, through logging, sampling and auditing, the witness the flow lost. The second is the error that repeats itself: a human who errs tends to notice and correct; an agent that decides badly because of a badly drawn rule repeats the same error in every identical case, without fatigue or doubt, until someone intervenes on the rule —which is why the auditing of the agentic layer watches the rule, not the case—. And the third is the dilution of accountability: when no one decided each case, the risk is that to the question of who answers, the answer is “no one” —neither the agent, which is not a subject, nor whoever designed it, nor whoever operated it—. The antidote is the other face of the displacement of control: if control moves from the decision to the rule, accountability moves with it —whoever designed, audited and maintains the rule answers—, and that is why those rules must be explicit, auditable and attributable to a person. Autonomy does not dilute human responsibility: it relocates it, and forces it to be made explicit precisely where it used to be implicit in the gesture of deciding.

The field converges

The constructs above were built from within the method, by rereading the eight currents; they depend on no external source. But it is worth noting that, as this extension was being written, the academic literature arrived on its own at the same reading: agentic systems are sociotechnical systems. That convergence is not the source of the argument —it is its strongest validation, because when works that do not know of each other describe the same object with the same apparatus, the object is real and not a rhetorical preference.

The first framework to treat agentic AI explicitly as a sociotechnical system closes its thesis almost in the words of this chapter: an agent’s behavior is co-produced by “algorithms, data, organizational practices, regulatory frameworks and social norms”.⁸ It is the deciding actor read in a network, and agentic enrollment, said by others. A second work formally updates the joint optimization of the Tavistock tradition to the age of AI and formulates it across four levels —individual, organizational, ecosystem and societal—:⁹ it gives the thesis its version at scale, because an agent that decides badly does not stay in the organization that deployed it, it scales to the ecosystem. And the finding that most shields the wager is empirical: evaluated over twenty workflows, state-of-the-art manager agents still fail to jointly optimize goals, constraints and runtime.¹⁰ Translated into the bridge’s vocabulary: joint optimization —coordinating the social with the technical— is precisely what agents still do not do well. The sociotechnical reading is not a task agentic AI is about to absorb; it is the residue that resists automation.

Context as an asset

There is an unexpected bridge between frontier agent engineering and the method, and naming it turns the artifacts of the first link into infrastructure for the agentic layer. The emerging craft of building useful agents does not turn on the model but on the context assembled for it: what it knows about the organization, in what vocabulary, with what memory of what has already happened —what has come to be called context engineering.¹¹ Read through the method, that is not an alien technical novelty: it is exactly what the bridge already produces. Institutional memory —repeated questions with consistent answers— and the project’s bilingual glossary —business and technical, defined once— are, literally, the context an agent needs to decide well.

Hence a proposition with diagnostic bite: the organization that cannot explain its process to a person cannot package it as context for an agent either. Organizational illegibility, which used to be paid for in zero adoption, is now also paid for in the impossibility of delegating —there is no context to give. The usual order is inverted: one does not start with the agent and then “give it context”; one first builds the legibility of the organization —the bridge’s work— and that legibility is the context. The bridge does not only set the agent’s guardrails: it manufactures the material with which the agent thinks.

Why the thesis does not age

The book had been asserting that the bridge thesis does not age with the technology; this extension turns that into an argument. The agentic layer does not leave the sociotechnical reading without work: it moves its object and raises its stakes. Joint optimization and translation shift upward; Shneiderman’s control changes its point of application; Polanyi’s tacit judgment does not become codifiable, it becomes the frontier one must know how to draw; Dignum’s accountability does not dissolve, it relocates and has to be made explicit. Each rung of the ladder makes the omission of the human reading more costly, and the agentic one is the most costly of all: at the top, a process no one understood is not executed more slowly —it is executed on its own, at scale, in series and without a witness—. The bridge is not an asset that AI with agency renders obsolete: it is, exactly, what that autonomy renders indispensable.

See also: Authors and currents · The bridge thesis · Concepts of our own · IMIA, the maturity instrument · Bibliography