Introduction
Power and Prediction: The Disruptive Economics of Artificial Intelligence by Ajay Agrawal, Joshua Gans, and Avi Goldfarb offers a powerful framework for understanding the economic and strategic impact of artificial intelligence. Building on their earlier work, the authors focus on prediction as the central function of AI and explore how this capability alters decision-making, business design, and market dynamics. This summary presents the book’s core arguments in a transformative format for curious readers, entrepreneurs, and business leaders seeking to navigate the unfolding AI revolution.
The Core Thesis: AI as Prediction
The authors argue that AI’s most essential contribution is improved prediction. While the term “artificial intelligence” conjures images of sentient machines, the authors ground AI’s real value in its capacity to take known inputs and produce unknown outputs with increasing accuracy.
Prediction, as defined in the book, is the process of filling in missing information to enable better decisions. In the past, prediction was expensive and limited. Now, as the cost of prediction falls dramatically due to machine learning, it becomes embedded in new domains and workflows, disrupting traditional economics and organizational logic.
The Prediction-Decision Chain
One of the book’s most important conceptual tools is the “prediction-decision chain.” AI enhances prediction, but it doesn’t make decisions; humans or other systems still must interpret, act on, or override the results. A full decision pipeline includes:
- Input Data – Raw material for prediction (e.g., images, text, numbers)
- Prediction – What AI is good at (e.g., estimating prices, identifying patterns)
- Judgment – Contextual interpretation of predictions
- Action – Execution of decisions
- Outcome – Result that informs future decisions
AI replaces only the prediction node but influences the entire chain, shifting the design of tasks, processes, and even corporate hierarchies.
Three Phases of Adoption
Agrawal, Gans, and Goldfarb explain that AI adoption progresses in three phases, each with increasing degrees of transformation:
1. Point Solutions
Initially, AI is deployed in narrow use cases. For example, logistics companies use AI to predict optimal delivery routes, or banks use it for credit scoring. These are relatively low-risk applications that slot into existing structures and improve efficiency.
2. System-Level Integration
In this phase, businesses restructure workflows to take full advantage of AI’s prediction capabilities. For example, AI-powered radiology changes how medical images are collected, analyzed, and used across healthcare networks, not just in diagnostic centers.
3. Structural Transformation
At the highest level of disruption, industries reorganize themselves. Uber and Lyft didn’t just automate dispatch; they reimagined urban mobility by combining prediction with smartphones and geolocation. Similarly, Airbnb disrupted hospitality by predicting accommodation availability and demand across a global peer-to-peer network.
The Judgment Bottleneck
While AI improves prediction, it doesn’t eliminate the need for human judgment. In fact, judgment becomes more important in environments of uncertainty or ethical complexity. For example, in criminal justice or hiring, AI might predict recidivism or employee performance, but judgment is required to evaluate fairness, legality, and societal norms.
The authors introduce the idea of the “judgment bottleneck” — the human interpretive gap between prediction and action. Overcoming this bottleneck requires new skills, governance models, and sometimes regulatory interventions.
The Cost Dynamics of Prediction
One of the book’s most transformative insights lies in its economic analysis of falling prediction costs. Historically, reductions in cost (e.g., computing, storage) have led to massive business reinvention. The same applies to prediction:
- When prediction is expensive, it’s reserved for critical tasks (e.g., weather forecasting, financial modeling).
- As prediction becomes cheap, it is applied everywhere — from supply chains to customer service, education, and healthcare.
This cost reduction unlocks new use cases and displaces existing decision systems. Businesses that fail to adapt may become obsolete, much like how Kodak missed the digital photography wave.
The New Competitive Frontier
With prediction commoditized, competitive advantage shifts to complementary assets. These include:
- Data – The fuel for accurate prediction
- Judgment – Interpretation and context to act on predictions
- Execution Capability – Ability to act swiftly and reliably
- Trust and Reputation – Essential for human-AI partnerships in sensitive domains
This dynamic favors companies that combine AI with domain expertise, human-centered design, and agile execution. Simply having the best algorithm is no longer sufficient.
Implications for Managers
The book offers pragmatic insights for managers who are planning AI adoption:
- Start small but think big – Pilot narrow prediction tasks to build familiarity, but design with future integration in mind.
- Invest in human-AI interaction – Upskill staff to work with AI and ensure workflows are optimized for judgment and collaboration.
- Reimagine value – Consider not just efficiency, but how AI might enable new value propositions, business models, or markets.
The authors warn against both complacency and hype. AI is not magic, but it is transformative when combined with thoughtful strategy.
Government and Regulation
Governments also face new challenges. As AI affects public services, justice systems, and labor markets, policymakers must:
- Regulate misuse and bias
- Promote data portability and fairness
- Invest in public AI infrastructure
- Ensure education systems prepare workers for human-AI collaboration
The authors suggest that public and private institutions will need new frameworks for risk management, especially in sensitive areas like health, finance, and civil rights.
Case Studies and Applications
Throughout the book, real-world examples illustrate the shifting economics of AI. Key cases include:
- Retail: AI optimizes inventory and personalizes recommendations.
- Finance: Credit underwriting becomes faster and more inclusive, but also riskier if not properly regulated.
- Healthcare: Diagnostic predictions enhance triage but require ethical oversight.
- Transportation: Self-driving tech redefines the concept of a “driver” and challenges regulatory norms.
- Legal and Hiring Tools: Controversial use of prediction demands explainability and transparency.
Each case underscores the importance of integrating prediction with human judgment, system redesign, and stakeholder engagement.
The Role of Entrepreneurs
Entrepreneurs play a crucial role in unlocking the economic value of AI. While large companies may focus on optimizing internal processes, startups are more likely to create entirely new business models.
The authors highlight how disruption comes not from simply automating existing systems, but from rethinking what is possible when prediction is abundant and cheap. This opens doors for innovation in education, climate solutions, remote work, and even creative arts.
Conclusion: Beyond Prediction
Power and Prediction concludes with a challenge: If AI gives us more predictive power than ever before, how will we use it? The authors emphasize that the greatest opportunities — and the greatest risks — lie not in prediction itself but in what we choose to do with it.
Will AI be used to deepen inequalities, or to promote inclusion? Will organizations use prediction to control, or to empower? These are not technological questions — they are social, ethical, and strategic. And the answers will shape the next era of business and society.
As prediction becomes increasingly commoditized, the real power lies in rethinking systems — education, healthcare, transportation, government — with a design philosophy that embraces uncertainty, human values, and strategic boldness.
Further Reading
For readers interested in exploring the themes further, the authors recommend revisiting their earlier book Prediction Machines and studying developments in AI ethics, human-centered design, and computational economics.
