Blog · Review Essay · Last reviewed June 15, 2026

Power and Prediction and the System Redesign Problem

Ajay Agrawal, Joshua Gans, and Avi Goldfarb's Power and Prediction is a business-economics book with a political consequence: if AI makes prediction cheap, the decisive question becomes who redesigns the system of decisions around it.

The Book

Power and Prediction: The Disruptive Economics of Artificial Intelligence was published by Harvard Business Review Press in 2022. HBR Store lists Ajay Agrawal, Joshua Gans, and Avi Goldfarb as authors, gives the publication date as November 15, 2022, and lists the book at 288 pages. Google Books and VitalSource list ISBN-10 1647824192 and ISBN-13 9781647824198 for the Harvard Business Review Press edition; Amazon uses the same ISBN-10 as its product identifier.

The book is a sequel to Prediction Machines, but its emphasis shifts. The earlier book made AI legible as cheap prediction. This one asks what happens when prediction changes enough that existing workflows stop making sense. The authors' answer is economic, but the implications are institutional: decision systems have owners, beneficiaries, excluded parties, and failure modes.

The Decision Is the Unit

The book's useful move is to make the decision the basic unit of AI analysis. A prediction does not govern by itself. It enters a decision: approve the loan, route the patient, reorder the warehouse, assign the worker, flag the student, recommend the video, dispatch the driver, escalate the ticket, deny the claim. In each case, prediction and judgment are distributed across people, machines, rules, incentives, and interfaces.

This matters for Spiralism because AI becomes powerful less through isolated cleverness than through institutional attachment. A score becomes authority when a bank, hospital, platform, school, or agency reorganizes action around it. A chatbot becomes labor policy when a call center changes staffing. A recommender becomes culture when a platform builds attention markets around it. The model is only one component in a larger circuit of belief and action.

System Redesign

The Schwartz Reisman Institute at the University of Toronto summarizes Goldfarb's argument around the "Between Times": after AI capability has become visible, but before organizations have fully redesigned around it. That phrase is valuable because it resists both hype and dismissal. The most important consequences may arrive after the first wave of task automation, when institutions redesign roles, accountability, data collection, standards, and customer expectations around machine prediction.

That is where Power and Prediction becomes more than a management book. System redesign is a power struggle. If AI changes the cost of prediction, someone still chooses the objective, the tolerance for error, the review process, the budget, the rights of affected people, and the gains from efficiency. The book's title is exact: prediction redistributes power when institutions build new routines around it.

The Agent Reading

Read in 2026, the book is also a guide to AI agents. An agent is not just a better prediction machine. It is prediction tied to action: retrieving files, drafting responses, choosing next steps, filling forms, triggering tools, and updating records. That makes the decision unit more visible, because every agentic workflow has a boundary where suggestion becomes execution.

NIST's AI Risk Management Framework treats trustworthy AI as something to incorporate into design, development, use, and evaluation. The European Commission describes the AI Act as a risk-based framework with rules for high-risk systems, transparency, and general-purpose AI models. Those frameworks make the governance point that Power and Prediction mostly leaves implicit: once AI reaches the system level, responsibility cannot remain at the level of individual prompts or model outputs.

Where the Book Needs Care

The book's clarity is also its limitation. "Decision" can sound clean, as if organizations are chains of discrete choices waiting to be optimized. Many institutions are messier than that. They contain habits, conflicts, hidden labor, legal constraints, defensive paperwork, procurement compromises, budget pressure, and workers who know when the official decision model is fiction. AI can optimize the visible decision while worsening the surrounding world.

The book also needs a stronger labor and public-interest reading. System redesign may generate productivity, but it can also convert discretion into surveillance, compress work into dashboards, and shift error onto people with the least bargaining power. In public services, the question is not only whether AI improves prediction. It is whether affected people can understand, contest, and survive the decision system built around that prediction.

What This Changes

Power and Prediction gives this archive a practical way to read AI adoption without treating models as magic. Ask where prediction enters the organization. Ask what decision it changes. Ask what judgment remains human, what judgment disappears, and what judgment is silently moved to a designer, manager, vendor, or policy office.

The book's best lesson is that AI disruption is not only a capability story. It is a redesign story. Institutions will decide whether prediction becomes a tool for better judgment, a cover for automation bias, a weapon of labor control, or a way to make old authority appear newly technical. The political work begins where the economic analysis ends: in the redesign of the system that turns prediction into action.

Sources

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


Return to Blog · Return to Books