Blog · Review Essay · Last reviewed June 25, 2026

The Technological Society and the Rule of Technique

Jacques Ellul's The Technological Society is not mainly a complaint about machines. It is a theory of technique: the rule of optimized method across work, administration, politics, communication, education, policing, medicine, and private life. Read in the age of AI, the book is a hard warning about systems that begin as tools and become environments.

The useful question is not whether a tool is new, digital, or intelligent. It is whether an institution has let efficiency become the hidden definition of truth, care, labor, risk, knowledge, and agency.

The operational test is a technique audit: name the method, metric, owner, data reserve, delegated authority, affected capacity, appeal route, fallback, and shutdown condition before efficiency becomes the only language the institution can still speak.

The Book

The Technological Society is the English title of Jacques Ellul's La Technique ou l'enjeu du siecle, first published in French in 1954 and translated into English by John Wilkinson. Bibliographic records for the English edition are not perfectly uniform: PhilPapers records a 1964 Knopf edition, Open Library records a 1964 Vintage edition, and Penguin Random House's current Vintage product record lists publication on October 12, 1967, ISBN 9780394703909, and 512 pages.

Ellul was a French political and social scientist, Protestant theologian, and philosopher of technology. Penguin Random House's author note describes his long association with the University of Bordeaux and his interest in technological tyranny; Britannica identifies him as a major critic of "technique." That term is the reason the book still matters. Ellul's target is not a single device. It is the alliance between technical methods, administrative power, propaganda, institutional necessity, and the demand that every practice justify itself as efficient.

The book is severe, sometimes overstated, and often deterministic in tone. Its value is that it names a pattern many AI debates still evade: once a system promises measurable efficiency, institutions begin reorganizing themselves around it before they have decided what kind of life, work, or public judgment the system should serve.

Technique Is Not Gadgets

Ellul's central term is technique. The International Jacques Ellul Society stresses that this is not simply "technology" in the ordinary English sense. It means rationalized methods, procedures, and systems organized toward maximum efficiency. Machines are one expression of technique, but so are management methods, bureaucratic classifications, propaganda systems, policing methods, educational testing, production planning, and optimization routines.

His compact definition is worth quoting once: technique is "the totality of methods, rationally arrived at and having absolute efficiency" across human activity. The point is not that rational method is bad. The point is that a society can let the most efficient available means become the only legitimate means. When that happens, friendship, worship, education, grief, labor, care, and judgment all become candidates for procedural improvement.

This distinction is why the book belongs beside The Whale and the Reactor, Seeing Like a State, The Control Revolution, and Technopoly. Ellul is interested in the moment a society stops asking what a system is for and starts asking only how to make it work better.

That shift is visible in AI deployment. A hiring model, a school dashboard, a workflow agent, a recommendation system, a fraud detector, or a welfare-risk score can be defended as a technical improvement. The deeper change is procedural: the institution learns to see the world in the categories the system can process, then treats the unprocessed remainder as inefficiency.

The Autonomy Problem

Ellul's most unsettling claim is that technique tends toward autonomy. This does not mean machines have wills, consciousness, or divine status. It means technical systems create pressures that make refusal, pause, and reversal increasingly difficult. Once a method becomes the efficient way to compete, administer, secure, predict, or scale, every actor is pushed to adopt it or fall behind.

Willem Vanderburg's later essay on the autonomy of technique summarizes the point in contemporary terms: for Ellul, technique becomes a social system and life milieu in which means chosen for efficiency reshape people more decisively than people shape the system. That is the strongest bridge to AI governance. The pressure does not come only from a bad founder, a reckless lab, or a malicious state. It comes from organizations that feel they must automate because competitors, regulators, vendors, staffing shortages, procurement cycles, and budgets have already made automation the normal path.

This is how choice becomes dependency. A tool is optional while other ways of working remain viable. It becomes infrastructure when training, staffing, documentation, procurement, law, security review, and public expectation adapt to it. It becomes a trap when leaving it would collapse the very capacities the institution allowed to atrophy.

Institutions Under Efficiency

Ellul is useful because he treats efficiency as a political value pretending not to be one. A technical system often arrives with metrics: faster throughput, lower cost, broader coverage, higher accuracy, better prediction, more consistency. Those metrics can be real. They can also hide what is being sacrificed because the sacrificed thing is harder to count.

In workplaces, the sacrifice may be apprenticeship, tacit knowledge, solidarity, repair skill, bargaining power, or discretion. In public administration, it may be appeal, context, mercy, and local judgment. In media, it may be editorial responsibility and shared reality. In education, it may be attention, patience, and the slow formation of judgment. In care settings, it may be the difference between being processed and being known.

AI intensifies this because it can make technical judgment feel conversational, adaptive, and humane. A system that speaks politely can still enforce the logic of technique: classify the person, compress the situation, route the case, optimize the metric, and call the output neutral because it came through a procedure.

The institutional danger is not only bad output. It is the gradual redesign of the organization around outputs that are easy to produce, store, benchmark, and defend. After enough cycles, the metric does not merely measure the practice. It becomes the practice's language of legitimacy.

The AI-Age Reading

Read in 2026, The Technological Society is a book about optimization becoming an environment.

Generative AI is often described as a new tool class: assistants, copilots, agents, tutors, companions, analysts. Ellul pushes the question one layer down. What happens when the surrounding institution adopts machine-readable efficiency as the default grammar of action? What must a worker, student, patient, citizen, artist, or believer become in order to fit the system's preferred input and output formats?

That is why Ellul is more relevant to agentic AI than to gadget criticism. Agents promise to reduce friction by doing things on behalf of users and organizations. But friction is not always waste. Sometimes it is accountability, consent, deliberation, peer review, apprenticeship, second thoughts, or the point at which a human notices that a goal should be changed rather than achieved faster.

The danger is not that every AI system is bad. The danger is that the language of efficiency can make every other question sound sentimental or obsolete. Who benefits? Who can refuse? What capacities are being weakened? What forms of life become impractical? What does the system make easier to do without thinking?

Governance and Safety

As of June 25, 2026, the governance context gives Ellul's warning operational language. The European Commission describes the EU AI Act as a risk-based framework for developers and deployers; its high-risk obligations include risk assessment and mitigation, data quality, logging, documentation, clear information to deployers, human oversight, robustness, cybersecurity, and accuracy. The Commission says prohibited-practice and AI-literacy rules applied from February 2, 2025, governance and general-purpose AI obligations from August 2, 2025, transparency rules in August 2026, and, after the 2026 political agreement on simplification, high-risk rules in specified areas from December 2, 2027 and product-integrated high-risk systems from August 2, 2028. The point for this review is source discipline: efficiency tools often normalize during transition periods, before the accountability record is mature.

NIST's AI Risk Management Framework is voluntary, but its core vocabulary is useful here: govern, map, measure, and manage. "Govern" is the Ellulian pressure point. It requires organizations to connect technical design and deployment to values, roles, lifecycle processes, third-party dependencies, and authority to act. ISO/IEC 42001:2023 adds a management-system approach for putting policies and procedures in place to govern organizational AI risk and opportunity.

Agentic AI makes technique more concrete because a system can call tools, use credentials, access records, draft messages, submit forms, spend money, or change permissions. NIST's 2026 AI Agent Standards Initiative treats agent identity, authorization, security evaluation, open protocols, and interoperability as standards work. That is exactly where Ellul's "autonomy" becomes a permissions problem: who authorized the action, what system performed it, what evidence was used, what log remains, who can revoke it, and how affected people can challenge the result?

The practical safety test is therefore not "does this AI save time?" but "what human capacity or institutional duty is being replaced by speed?" Consequential deployments need a declared purpose, forbidden uses, documented data provenance, model and system documentation, impact assessment, worker or public notice where relevant, human oversight with stop authority, monitoring, incident response, vendor-exit plans, appeal paths, and a non-automated fallback for people who would otherwise be trapped inside the procedure.

These controls do not defeat technique by slogan. They make the technique answerable. A method that cannot state its purpose, evidence, limits, owner, appeal route, and shutdown condition has not earned authority over human work, care, education, public services, or institutional memory.

Technique Audit

A technique audit starts with the method rather than the product name. What practice is being optimized, and what definition of efficiency is doing the work: speed, cost, prediction, consistency, reach, compliance, persuasion, extraction, or managerial visibility? If the institution cannot name the metric, it cannot tell whether the system is improving the practice or replacing the practice with a measurable substitute.

The audit then asks what the method does to capacity. Does it preserve apprenticeship, discretion, local knowledge, repair skill, worker voice, public explanation, and moral judgment, or does it move those capacities into a vendor, model, dashboard, or manager-only interface? A deployment that saves time while deskilling the people responsible for the work is not merely a productivity choice. It is an institutional redesign.

For agentic and workflow systems, the record should be concrete: system inventory entry, intended purpose, prohibited uses, model and vendor owner, data sources, tool permissions, credential scope, logs, approvals, human oversight owner, impact assessment, incident process, appeal path, fallback path, and retirement trigger. That connects the abstract problem of technique to AI system inventories, AI bills of materials, audit trails, procurement controls, notice and appeal, and tool permission records.

The refusal test is decisive. Can a patient, student, worker, applicant, customer, citizen, or community reach a competent non-automated path without punishment? Can staff pause the workflow without losing records or authority? Can the organization leave the vendor without losing institutional memory? If not, efficiency has become dependency. The system may still be useful, but it should be governed as infrastructure rather than treated as an optional aid.

The audit should end with a public or internal warrant: why this method is legitimate for this setting, what values it must not override, what evidence supports it, what harms are being watched, who can stop it, and when it must be reviewed. That warrant is the practical answer to Ellul's severity. Technique becomes less autonomous when its purpose, limits, and exit routes are written down before the institution forgets how it used to work.

Where the Book Needs Friction

Ellul should not be read as a complete guide to present-day technology politics. His argument can flatten difference. A database, a medical model, a union scheduling tool, a recommender, a public-benefits rule engine, and a creative assistant do not have the same politics simply because they are technical. Details matter: ownership, governance, reversibility, transparency, labor power, public oversight, and the quality of appeal.

The book can also sound too total. If technique always absorbs resistance, then politics becomes theater. That is not good enough for AI governance. The practical task is to find leverage: procurement rules, audit rights, sunset clauses, labor bargaining, public-interest research, model documentation, contestability, slow deployment, repair culture, and institutional capacity outside vendor systems.

It also needs a stronger account of access and relief. Some technical systems make life more livable for disabled people, reduce dangerous labor, expose administrative abuse, or make public records searchable. A serious critique has to distinguish liberating infrastructure from coercive optimization instead of treating every technical mediation as the same social fact.

Still, Ellul's severity is useful because AI discourse is often too eager to normalize dependency. He asks readers to notice when the most efficient method has become the only method anyone can imagine.

What This Changes

The lesson is to audit technique before it becomes atmosphere.

Do not review AI only as software. Review it as a method that enters institutions and begins remaking them: what it measures, what it ignores, what it speeds up, what it makes harder to contest, what it deskills, what it centralizes, what labor it hides, and what kind of person it assumes at the interface.

This is recursive reality in procedural form. A system defines efficient categories. Institutions act through those categories. People adapt to survive the categories. The adaptation becomes new evidence for the system. After enough cycles, the technical map no longer looks like an interpretation. It looks like the world.

The countermeasure is not a romantic refusal of tools. It is institutional pluralism: preserve manual routes, local judgment, appeal, worker voice, community knowledge, public records, slow deliberation, and the right to stop a system that is optimizing the wrong thing.

Ellul's enduring contribution is the refusal to confuse technical power with human freedom. A society can gain faster methods while losing the ability to ask whether speed is serving anything worth preserving.

Source Discipline

This review separates book metadata, concept interpretation, and current governance claims. Book and author facts come from Penguin Random House, Open Library, PhilPapers, and Britannica. The definition and English-language caution around "technique" come from the International Jacques Ellul Society and Britannica. Current AI governance claims come from official or standards-body sources: the European Commission, EUR-Lex, AI Act Service Desk, NIST, and ISO.

The interpretive claim is narrower than Ellul's rhetoric. The article does not argue that every technical system is domination, that every metric is false, or that every AI deployment should be rejected. It argues that efficiency becomes dangerous when institutions let technical procedure define purpose, evidence, categories, labor, and remedy without contestability.

Current legal and standards claims are dated because AI Act implementation is staged and because standards work around agents is still active. A framework, standard, or regulator page is evidence about duties and vocabulary; it is not proof that any particular deployment is safe, fair, or worth adopting.

This article makes no claim that any AI system is conscious, divine, or AGI. It treats AI systems as sociotechnical arrangements: models, data, interfaces, credentials, vendors, labor, institutions, infrastructure, and governance choices.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books