Blog · Review Essay · Last reviewed June 23, 2026

What Tech Calls Thinking and the Ideology Factory

Adrian Daub's What Tech Calls Thinking: An Inquiry into the Intellectual Bedrock of Silicon Valley is a short book about the stories technology power tells when it wants ordinary business decisions to sound like destiny. Its target is not code, engineering, or even entrepreneurship. Its target is the portable vocabulary that lets a local industry present its preferences as the future's own voice.

For this review, the working unit is the permission story: the phrase, metaphor, origin myth, or category shift that makes a deployment feel natural before its evidence, harms, labor, dependencies, and alternatives have been inspected. The story does not need to be false to be dangerous. It only needs to decide what kind of proof the public is allowed to ask for.

The Book

What Tech Calls Thinking was published by FSG Originals in 2020 as part of the FSG Originals x Logic series. Macmillan's publisher page lists the imprint as FSG Originals and the ISBN as 9780374538644, and describes the book as an inquiry into the ideas and language Silicon Valley uses to justify itself. Stanford's current faculty profile identifies Daub as the J.E. Wallace Sterling Professor of the Humanities and a professor of German Studies and Comparative Literature, and notes that What Tech Calls Thinking has been translated into five languages.

The book belongs beside From Counterculture to Cyberculture, The Internet Revolution, TechGnosis, The Culture of Connectivity, Platform Capitalism, and The Tech Coup. Those books map networks, business models, spirituality, institutions, and power. Daub maps the prestige language that lets the same world describe itself as philosophy.

The argument is narrow in a productive way. Daub is not claiming that everyone who works in technology thinks alike, and he is not using "thinking" as an insult aimed at engineers doing ordinary technical work. He is studying the public language of founders, funders, journalists, conference stages, and management culture: disruption, dropping out, changing the world, design thinking, failure, authenticity, radical novelty, and other portable phrases that make business strategy sound like intellectual destiny.

That gives the book a useful governance definition: tech ideology is a category machine. It does not only state opinions; it sorts people, laws, institutions, products, and objections into the innovative, obsolete, scalable, frictional, inevitable, and unserious. Once those categories are accepted, a public argument can lose before anyone reaches the evidence.

Ideas With No History

The central move of the book is genealogical. Silicon Valley often presents its favorite concepts as new because the products are new. Daub slows that gesture down. Many of the ideas that circulate as fresh tech wisdom have older sources: counterculture, business self-help, New Age psychology, libertarian politics, Heidegger, McLuhan, Ayn Rand, Schumpeter, Esalen, and American revival culture.

This matters because forgetting an idea's history makes it harder to contest. If "disruption" is treated as a natural law of technological progress, then regulation looks backward. If "innovation" is treated as moral proof, then labor questions look like resentment. If "failure" is treated as heroic, then safety nets disappear from the story. If "changing the world" is treated as a founder's inner calling, then public consequences are recoded as side effects of vision.

Daub is especially sharp on what might be called institutional amnesia. A company borrows language from philosophy, counterculture, psychology, or social movements, then presents the borrowed language as native to its own product culture. The source disappears, the vocabulary remains, and the firm gets to claim depth without inheriting the older argument's constraints.

The AI-era version is concept laundering. A research term, philosophical term, therapeutic term, labor term, or safety term enters a pitch deck, product page, policy memo, or model card. By the time it reaches procurement, the term may no longer carry its old disputes. "Alignment," "assistant," "agent," "copilot," "open," "democratization," and "safety" can each name real work, but each can also become a shortcut around the institutional question: who is acting, with what authority, on whose data, for whose benefit, and with what right to refuse?

A useful definition follows from that: tech ideology is not a list of beliefs held by every technologist. It is a reusable grammar that sorts possible futures into the modern, the backward, the inevitable, the unserious, the scalable, and the obsolete. Once that grammar is installed, public debate often happens inside categories the industry has already selected.

Disruption as Permission

The book's best political insight is that tech ideology often works by changing the frame before accountability arrives. A taxi company and a ride-hailing platform may both coordinate paid transportation, but the platform can insist that it is a new category long enough to exploit a regulatory gap. A media company and a social platform may both organize public attention, but the platform can insist that it merely connects users. A labor system can look like employment in practice while being described as flexible access, entrepreneurship, or opportunity.

That is why this is not just a book about rhetoric. Rhetoric is infrastructure when courts, investors, journalists, workers, users, and regulators act inside the categories it supplies. The words do practical work. They slow down old protections, create moral glamour around ordinary extraction, and make public skepticism feel unsophisticated.

The practical pattern is category arbitrage. A firm enters a regulated social function, names itself as something else, benefits from the lag, and then treats later accountability as an attack on progress. The novelty may be real at the interface layer, but the affected duty may be old: wages, safety, non-discrimination, consumer protection, professional responsibility, public records, environmental impact, or due process.

The same pattern now appears in AI. A chatbot becomes a "copilot" before anyone has settled its responsibilities. A model-mediated search surface becomes an "answer engine" before citation, correction, and source economics are stable. A synthetic companion becomes "support" before consent, dependency, memory, and duty of care have been worked through. A workplace agent becomes "productivity" before its effects on skill, pace, monitoring, and bargaining power are visible.

The category change is a safety event. A name changes what a user expects, what a buyer tolerates, what a regulator asks for, and what a vendor can call a failure. If a system is framed as a creative assistant, a false answer may look like an inconvenience. If the same system is framed as a legal assistant, medical helper, tutor, benefits navigator, or hiring tool, the same failure lands on rights, money, health, education, and due process.

That is why procurement should record not only the vendor's technical claim but the vendor's category claim. What older duty is being renamed? Is a regulated professional service being sold as software? Is workplace management being sold as empowerment? Is surveillance being sold as care? Is advertising being sold as personalization? If the category changes the legal or moral burden, it belongs in the risk record.

The Institution That Denies Itself

One reason Daub is useful is that he notices Silicon Valley's strange relation to institutions. The culture likes dropouts, garages, outsiders, founders, rebels, and informal genius. But the actual system depends on universities, immigration pipelines, state-funded research, defense and public infrastructure, venture capital, law firms, standards bodies, media attention, and highly organized labor markets.

The result is an institution that can deny being one. It can claim outsider status while exercising insider power. It can attack bureaucracy while building private bureaucracies of ranking, moderation, policy, procurement, identity, scoring, and behavioral control. It can invoke openness while enclosing interfaces, data, models, cloud platforms, and app ecosystems.

This denial becomes more dangerous as software systems move into public functions. When a private company supplies the interface through which people work, learn, search, apply for services, express themselves, or receive automated judgments, it is no longer merely selling a product. It is shaping the conditions under which other institutions can see and act.

The contradiction is not hypocrisy as a personality flaw. It is a governance structure. A firm can be anti-institutional in style and institution-building in effect. It can train the public to distrust slow democratic process while asking schools, cities, agencies, hospitals, and employers to reorganize themselves around private infrastructure.

The recordkeeping consequence is simple: a vendor that becomes institutional infrastructure should leave institutional evidence. Contracts, system inventories, audit logs, model or system cards, incident contacts, update notices, exit rights, and appeal pathways are not bureaucratic clutter. They are the way an institution proves that a charismatic story did not silently become an unreviewable dependency.

The AI-Age Reading

Read after the rise of generative AI, What Tech Calls Thinking becomes a field guide to the story layer around machine intelligence. Current AI culture is full of concepts that sound technical, moral, and civilizational at once: alignment, safety, superintelligence, AGI, copilots, agents, acceleration, democratization, open models, frontier systems, human feedback, and beneficial intelligence.

Some of these terms name real problems. Some name real research programs. But Daub's warning is that a useful term can also become a permission structure. Once a company controls the language in which its technology is publicly understood, it can make its own tradeoffs sound like the only responsible path. Scale becomes inevitability. Deployment becomes learning. Data hunger becomes progress. Market capture becomes democratization. Closed infrastructure becomes safety. Public dependence becomes partnership.

The book also clarifies why AI belief formation is not only a matter of users trusting model outputs. Belief forms upstream, in the institutional story that tells people what kind of thing they are meeting. If the system is introduced as an assistant, people test it as help. If it is introduced as a colleague, they route work to it. If it is introduced as a therapist, they disclose. If it is introduced as an oracle of future intelligence, they may treat present defects as temporary shadows of an approaching destiny.

The same story layer can be learned into the product. A base model, adapter, fine-tune, retrieval policy, refusal template, or system prompt can make one vocabulary feel normal and another feel out of bounds. The ideology layer is not consciousness or belief inside the model. It is a repeatable behavioral frame: preferred sources, preferred metaphors, default cautions, hidden sponsorship, refusal style, institutional voice, and the quiet narrowing of what counts as a reasonable question.

That is the recursive danger: the story changes use, use creates evidence, evidence funds the next story, and the cycle begins to look like proof.

Governance and Safety

As of June 23, 2026, public institutions have begun treating AI language itself as a governance surface. The OECD's updated definition keeps the term operational: an AI system is machine-based, infers from inputs, and generates outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. That kind of definition matters because it pulls the conversation away from mystique and back toward inputs, outputs, autonomy, adaptiveness, and context.

The FTC's 2024 Operation AI Comply made the same point through enforcement. The agency announced actions against deceptive AI claims and schemes, including claims that AI could substitute for professional legal work without adequate evidence. The important governance lesson is not limited to those cases. There is no special exemption for a product just because its marketing calls it AI. Claims about capability, replacement, reliability, and safety need evidence.

The EU AI Act and NIST's AI Risk Management Framework point in the same direction from different legal positions. The EU Act creates phased legal duties around transparency, prohibited practices, general-purpose AI, high-risk systems, documentation, human oversight, and post-market monitoring. The AI Act Service Desk timeline says Article 50 transparency rules start applying on August 2, 2026, while the Commission's simplification page says a May 7, 2026 political agreement would move rules for certain high-risk areas to December 2, 2027 and product-embedded high-risk systems to August 2, 2028. NIST's voluntary AI RMF, released in 2023 and now under revision, and its Generative AI Profile, published in 2024 and updated in April 2026, frame risk management around governance, mapping, measurement, and management across the AI lifecycle.

Article 13 of the EU AI Act is especially relevant to Daub's method because it turns category language into instructions for use. High-risk systems must give deployers information about intended purpose, performance, known risks, human oversight, input data, logging mechanisms, and how to interpret outputs. In plainer terms: the label is not enough. A serious system has to say what it is for, what it cannot do, how it was tested, and what records remain when it acts.

Daub's contribution is to show why these controls have to start before procurement or deployment. They have to inspect the metaphor. A public agency, school, hospital, employer, church, newsroom, or city should ask what the system is being called, which older duty the new name obscures, what evidence supports the claim, who can contest the output, and who has authority to stop use when the story outruns the proof.

The safety issue is not that current AI systems are conscious, divine, or entitled to institutional obedience. The safety issue is that language can transfer authority before accountability arrives. A plausible voice, a polished interface, a confident vendor narrative, and a future-facing product category can make deference feel modern even when the underlying system is brittle, undocumented, extractive, or poorly matched to the task.

The practical control is a claim-language register. For every consequential AI deployment, record the product name, category claim, task claim, evidence source, affected population, old duty being renamed, oversight owner, appeal path, update trigger, and shutdown authority. That register does not settle the politics of technology, but it prevents the permission story from becoming untraceable.

Source Discipline

The source rule for this book is simple: treat slogans as evidence of ideology, not evidence of capability. A founder keynote can show how a company wants to frame a technology. It cannot by itself show that the system is accurate, safe, lawful, fair, or useful in a real institution.

For AI claims, separate at least five records: the marketing name, the technical system, the deployment context, the affected population, and the governance mechanism. "AI-powered" names none of them. A credible claim should identify the task, data source, model or vendor, baseline comparison, testing conditions, known failure modes, monitoring plan, appeal path, security boundary, and accountable owner.

This is where Daub's humanities method becomes operational. Genealogy is not decoration. It is a way to ask where a concept came from, what work it performs, who benefits when it sounds natural, and what would change if the concept were replaced by plainer language. Calling a system "automation," "classification," "prediction," "ranking," "text generation," "search," "surveillance," or "workflow management" may reveal duties that the word "AI" hides.

Good source discipline therefore prefers primary records where possible: publisher pages for bibliographic facts, author or university pages for author roles, statutes and regulators for legal claims, standards bodies for frameworks, and product documentation or audits for system behavior. Commentary can sharpen interpretation, but it should not become the factual floor.

It also separates rhetorical evidence from operational evidence. A keynote, manifesto, interview, or launch page can prove how a company wants the public to understand a system. It cannot prove task performance, safety, legality, fairness, or institutional fit. A regulator's action can prove an enforcement theory or settlement posture; it does not prove that every similar system has the same defect. A standard can define a practice; it does not prove that a vendor followed it.

Where the Book Needs Care

The book is brisk and polemical. That is part of its force, but it also means some readers will want more patient differentiation between public tech mythology, ordinary engineering practice, and the many workers who do not control either venture narratives or founder ideology. Daub knows this problem and states that his project is centered on highly visible figures and media-facing ideas, not on the whole technology workforce.

The book also predates the current generative AI boom. It does not analyze foundation models, data-center buildout, model safety politics, synthetic-media governance, AI companions, or agentic workflow systems as contemporary objects. The translation to AI is therefore an application of Daub's method, not something the book itself fully performs.

Finally, the book should be paired with work that presses harder on race, gender, labor, empire, infrastructure, and extraction. Atlas of AI, Artificial Whiteness, Race After Technology, Ghost Work, Programmed Inequality, and Surveillance Valley show parts of the machine that ideology can make decorous or invisible.

The strongest use of Daub is not to sneer at all technical optimism. It is to slow down the moment when optimism becomes a license. There are real tools, real discoveries, real accessibility gains, and real engineering achievements inside the same culture. The test is whether those achievements remain answerable to evidence, labor, rights, environmental costs, and public alternatives.

What This Changes

The practical lesson of What Tech Calls Thinking is that governance has to inspect the metaphor before it inspects the product. What is the system being called? What older institution is it trying not to resemble? Which duties disappear under the new name? Which people become users instead of citizens, workers, patients, students, authors, defendants, customers, or members of a public?

That question matters for AI because model-mediated systems rarely arrive as bare machinery. They arrive inside stories of intelligence, assistance, creativity, inevitability, and care. Those stories shape contracts, procurement, investment, curriculum, regulation, newsroom coverage, and private hope. They decide which harms look accidental and which demands look reasonable.

That is one of the site's recurring concerns: interfaces become institutions when their categories teach people what counts as reality. A dashboard does not only report work. It changes work. A ranking does not only sort value. It trains value. A chatbot does not only answer questions. It teaches the user what kind of authority is supposed to be speaking.

The corrective is not to ban strong language. It is to make strong language earn its keep. If a vendor says "copilot," name the pilot. If it says "autonomous," name the permissions and rollback path. If it says "democratized," name who owns the infrastructure. If it says "safe," name the tests, incidents, and unacceptable-use boundaries. If it says "open," name the parts that are still closed.

Daub's book is a reminder that technological politics often begins before the dashboard opens. It begins when an industry teaches people what to call the dashboard, what future to imagine through it, and which old tools of analysis to leave behind at the door.

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