Blog · Review Essay · Last reviewed June 25, 2026

In the Age of the Smart Machine and the Work Made Visible

Shoshana Zuboff's In the Age of the Smart Machine is one of the cleanest pre-internet books for understanding why workplace technology is never only a productivity tool. Computerization makes work visible, abstract, measurable, searchable, and governable. The machine does not just do tasks. It creates a record of the work, and that record changes what managers can know, what workers are allowed to learn, and where authority settles.

The sharper definition is this: a smart machine is a work system that performs, mediates, and records labor at the same time. Its political effect depends on who can use that record: workers learning the process, managers disciplining the process, vendors training the next model, or institutions rewriting the job around the trace.

The Book

In the Age of the Smart Machine: The Future of Work and Power was published by Basic Books in 1988. Open Library lists the 1988 Basic Books edition at 468 pages, with the subject matter centered on automation, machinery in the workplace, organizational effectiveness, and the social and economic aspects of automation. Hachette's current Basic Books listing presents the book as a study of the pitfalls and promises of computerized technology in business life.

Zuboff was studying computer-mediated offices, factories, professional settings, executive work, and craft workplaces before the consumer internet became ordinary life. Her official page describes the research as multi-year studies of office, factory, professional, executive, and craft workplaces moving from traditional to computer-mediated task environments. That timing is the book's advantage. It catches the transition while digital systems are still visibly entering the workplace, before later platform language made monitoring, metrics, dashboards, and data trails feel natural.

The book is not a simple anti-computer argument. It asks what kind of social relation a computer system installs. A machine can be used to automate: remove discretion, replace labor, standardize action, and tighten control. It can also be used to informate: produce new knowledge about work that could make workers more capable, more collaborative, and more able to exercise judgment. The tragedy Zuboff tracks is that institutions often choose the first path while speaking in the language of the second.

Current Context

As of June 25, 2026, Zuboff's smart-machine fork is no longer only a theory of office automation. It is visible in generative workplace assistants, call-center coaching tools, warehouse and logistics dashboards, hiring systems, productivity suites, coding agents, meeting summarizers, employee-wearable programs, and manager-facing analytics. Some tools are branded as AI and some are ordinary workflow software; the important boundary is whether the system materially changes work, records work, or makes the record actionable.

The policy vocabulary has caught up with part of the problem. The ILO treats algorithmic management as a workplace reality that can organize, assign, monitor, supervise, and evaluate work. The European Commission's Joint Research Centre tracks digital monitoring, algorithmic management, AI tools, and platformisation across EU workplaces. GAO's 2025 digital-surveillance report treats workplace monitoring as a matter of physical health and safety, mental health, and employment opportunity, while the NLRB's 2022 memorandum described electronic surveillance and automated management as potentially relevant to workers' organizing rights.

Legal timing still matters. The EU AI Act's Annex III lists many employment, recruitment, worker-management, task-allocation, monitoring, and performance-evaluation systems as high-risk. The published Article 113 timeline says the Regulation generally applies from August 2, 2026 with exceptions, while the European Commission's implementation page says the May 7, 2026 AI Omnibus political agreement sets December 2, 2027 for certain stand-alone high-risk systems and August 2, 2028 for product-embedded high-risk systems. Either way, employers are already deploying smart-machine records before the full compliance regime is settled.

The current lesson is not that every workplace AI tool is abusive. It is that the record must be governed before it becomes ordinary evidence. A system that helps a worker draft, route, summarize, code, coach, schedule, or diagnose should not quietly become a personnel file, productivity score, replacement map, or vendor training stream without a new review, notice, and contestable record.

Informating

The key concept is informating. Zuboff's official book page describes it as the process by which digitalization translates activities, events, social exchange, and objects into information. Google Books summarizes the practical fork: computerization can either automate in ways that dehumanize work, or informate by giving workers knowledge for critical and collaborative judgment.

A useful definition for 2026 is this: informating is the double action by which a system performs or mediates work while also producing a formal record of that work. The record is not a passive afterimage. It becomes material for supervision, learning, audit, metrics, training data, prediction, scheduling, litigation, and institutional memory. The machine changes the task because it also changes what can later be known about the task.

The governance test is therefore directional: does information flow only upward to managers and outward to vendors, or also back to the people doing the work? If the worker cannot inspect, correct, learn from, or challenge the record, informating has become a polite word for extraction.

This is a useful distinction because many AI deployments combine both tendencies. A support chatbot may help a worker answer questions, but it may also record every interaction, score every response, and train a replacement pipeline. A clinical assistant may reduce documentation burden, but it may also make a nurse's reasoning more legible to billing, compliance, and management systems. A coding assistant may accelerate work, but it may also shift expertise from apprentice learning into opaque vendor tooling.

Informating names the double action. The system acts on the world and writes the world down. Once the record exists, it becomes available for comparison, optimization, discipline, sale, audit, training, and institutional memory. That is why the politics of workplace AI starts before anyone asks whether the model is accurate. It starts when work becomes data, and when the data becomes a managerial claim about what the work really was.

The Workplace Becomes Text

Zuboff's most durable insight is that computer-mediated work creates an electronic text of organizational life. Tasks, exceptions, timings, messages, decisions, machine states, customer interactions, and worker actions become recordable traces. This is not a neutral mirror. The record changes what is worth noticing.

Before computerization, much work remained embodied, tacit, local, and partly opaque to management. A skilled worker knew the machine by sound, a clerk knew which exception mattered, a technician understood a process through touch and sequence. Digital systems convert parts of that knowledge into symbols, fields, dashboards, logs, tickets, and alerts. Some of that conversion is genuinely useful. It can reveal failures, share knowledge, and reduce dangerous dependence on one person's memory. It can also flatten practice into metrics and make local judgment look like noise.

This is where the book sits beside arguments about legibility and classification. A workplace becomes easier to manage when it becomes easier to read. But reading is not the same as understanding. A dashboard can show throughput while hiding fatigue. A score can show compliance while hiding fear. A model can summarize performance while missing the craft knowledge that made the work resilient.

The electronic text is therefore a power object. It is the place where a worker's action can become evidence, where a manager's summary can become institutional memory, and where a vendor's schema can decide which parts of work are preserved. The workplace becomes not only searchable but editable by the categories of the system. That is the early form of what later appears as algorithmic management.

The AI-Age Reading

Read in 2026, In the Age of the Smart Machine is a prehistory of AI labor politics. The machine in the title is no longer only a mainframe, factory system, office terminal, or enterprise database. It is also a model, an agent, a workflow assistant, a surveillance stack, and a vendor service that observes work while helping perform it.

That matters because AI makes informating more intimate. Older systems captured transactions and process states. New systems capture drafts, prompts, hesitation, style, emotional tone, correction patterns, tacit preferences, and traces of reasoning. The worker is not only operating software. The worker is teaching, verifying, and being profiled by software that may later be used to reorganize the job.

The book also clarifies the apprenticeship problem. If institutions use AI mainly to automate visible tasks and capture invisible knowledge, they may consume the practices that train future experts. The system can look efficient in the short run because it has absorbed years of human skill. Then the organization discovers that it has weakened the slow, social, error-correcting pathways that produced that skill in the first place.

The stronger AI-age reading is that every workplace assistant has two products. One is the immediate answer, draft, summary, schedule, ticket, code patch, or recommendation. The other is the work trace: prompts, edits, approvals, overrides, exceptions, rejected completions, and time saved or lost. If that second product flows only to management and vendors, informating becomes extraction. If workers can inspect, contest, learn from, and govern the record, informating can still support shared intelligence. The same distinction belongs in AI system inventories, audit trails, and procurement records, not only in ethics language.

This is the bridge to the site's recurring concern with dashboards and recursive reality. The system records work; the record becomes the basis for management; workers adapt to the record; future systems train on the adapted behavior; and then the organization treats the resulting data as a neutral picture of work. Zuboff helps name the first turn of that loop. The AI era makes the loop faster and harder to see.

Governance and Safety

As of June 25, 2026, the governance question around smart machines at work is not only whether AI tools comply with model-safety norms. It is whether the workplace record they create is lawful, contestable, useful to workers, and bounded by institutional rights. The EEOC, DOJ, CFPB, and FTC joint statement on automated systems made the baseline explicit in 2023: existing civil-rights, consumer-protection, fair-competition, and equal-opportunity laws still apply when organizations use AI or automated systems.

The U.S. Department of Labor's 2024 AI Best Practices roadmap frames workplace AI around worker empowerment, job quality, rights, privacy, and economic security. Its listed practices include governance structures, meaningful human oversight for significant employment decisions, transparency to workers, worker input, protection of labor and employment rights, training, and worker-data security. Those are Zuboff's fork translated into current governance language: information can democratize work only if workers have access, voice, training, and power over the record.

The EU AI Act puts employment and worker management in a high-risk frame. Annex III lists AI systems used for recruitment, targeted job ads, application filtering, candidate evaluation, decisions affecting work relationships, promotion, termination, task allocation based on behavior or traits, and monitoring or evaluating performance and behavior. That classification matters even outside the EU as a benchmark for seriousness: workplace AI can affect livelihood, dignity, rights, and bargaining power, so it needs governance before deployment, not apology after harm.

NIST's AI Risk Management Framework is useful because its core functions are operational: govern, map, measure, and manage. For a smart-machine workplace, mapping means identifying the task, affected roles, data flows, hidden labor, and decision points. Measuring means checking error, bias, accessibility, skill loss, surveillance effects, and worker experience. Managing means retention limits, audit trails, appeal paths, incident review, vendor accountability, and authority to pause use. Governance means those controls are owned by named people rather than implied by the interface.

The safety implications are concrete. Before a workplace AI system is adopted, an organization should state whether it is automating, informating, or doing both; whether work traces can be used for discipline, productivity scoring, training, promotion, firing, or replacement planning; whether workers and representatives can inspect the record; how accommodations and appeals work; how long data is retained; what the vendor can reuse; and who can stop the system when it distorts the work it claims to improve.

A minimum smart-machine record should name the system owner, vendor, affected roles, data sources, model or workflow version, purpose, decision points, worker notice, human-review authority, retention rule, vendor reuse rights, accessibility review, consultation record, incident owner, and stop condition. The record should separate assistance from assessment: a tool introduced to help write, summarize, route, or code should not become evidence for discipline or replacement without a fresh review.

Source Discipline

This page separates book facts, conceptual interpretation, and current governance claims. Basic Books/Hachette, Open Library, Harvard Business School, Google Books, Zuboff's official page, and the review record support the bibliographic and concept claims. DOL, EEOC/FTC, ILO, JRC, GAO, NLRB, the European Commission, the EU AI Act Service Desk, and NIST support the current workplace-governance claims checked for the June 25, 2026 review date.

The AI reading is an application of Zuboff's framework, not a claim that the 1988 book predicted today's model architectures or that AI systems are conscious, divine, or AGI. The claim is narrower: AI systems intensify the old smart-machine pattern because they help perform work while producing records that can govern, train, evaluate, and reorganize work.

For workplace claims, source discipline means naming the decision point and legal setting. A hiring screen, call-center coach, warehouse quota, meeting summarizer, productivity dashboard, coding assistant, and medical documentation tool create different records and risks. Vendor marketing is weak evidence; stronger evidence includes worker notices, job analyses, validation records, data-retention rules, appeal logs, representative-consultation records, and contract terms that let an employer audit, pause, or exit the system.

Where the Book Needs Friction

The book is strongest inside organizations. It is less directly about platform markets, generative media, global data extraction, or consumer surveillance than Zuboff's later The Age of Surveillance Capitalism. Readers looking for cloud monopolies, ad-tech prediction markets, biometric surveillance, content moderation labor, or foundation-model supply chains need other books beside it.

It also carries the optimism of its own fork. The informating path remains real, but it is politically harder than the concept can make it sound. Information does not automatically democratize work. New visibility can empower workers only if they have rights, bargaining power, training, time, institutional voice, and access to the record. Otherwise, transparency flows upward and discipline flows downward.

That limitation is useful rather than fatal. It keeps the reader focused on governance. The question is not whether a workplace AI system produces information. It will. The question is who can inspect it, correct it, learn from it, refuse it, and benefit from it.

The book also needs to be paired with accounts of race, gender, disability, class, and migration in labor markets. The electronic text does not fall on everyone equally. Workers with less bargaining power may be more exposed to monitoring, less able to challenge bad records, and more likely to see "learning" systems used as discipline. A democratic informating system requires more than access to data; it requires protection against retaliation and real power to change the workflow.

What This Changes

This book belongs in the catalog because it shows how a machine becomes an institution by making reality administratively readable. That is the bridge from early office automation to AI agents. The same pattern recurs whenever a system turns messy human practice into structured traces and then treats those traces as the place where truth lives.

The practical lesson is procedural. If AI is introduced into work, the record it creates needs governance: worker access, appeal rights, retention limits, audit trails, data minimization, clear ownership, and protections against using assistance logs as a quiet replacement map. The tool should increase the user's agency, not merely increase management's resolution.

In the Age of the Smart Machine is ultimately a book about a choice that keeps returning. Machines can make work more knowable in ways that expand human judgment, or they can make workers more knowable in ways that shrink it. AI raises the stakes because the same interface can help, observe, evaluate, imitate, and replace. The politics is in the arrangement.

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