Narrator: What if the first billion-dollar AI company in a country didn't come from the expected tech hubs, filled with brilliant researchers and venture capital? What if it came from a remote, windswept island town of just half a million people, a place largely overlooked by the tech industry? This isn't a hypothetical. In 2020, Nasdaq acquired Verafin, a fraud detection company, for $2.75 billion, making it Canada's first AI unicorn. Its home wasn't Toronto or Montreal, but St. John's, Newfoundland. This surprising outcome revealed a critical blind spot for many experts: the true economic disruption of AI isn't just about building better prediction algorithms. It's about understanding and redesigning the systems in which those predictions operate.

In their book, Power and Prediction: The Disruptive Economics of Artificial Intelligence, authors Ajay Agrawal, Joshua Gans, and Avi Goldfarb argue that we are in "The Between Times"—a period where AI's technical power is proven, but its widespread economic impact has yet to be realized. The key to unlocking this potential, they contend, lies in moving beyond simple applications and embracing a complete system-level rethink of how industries and businesses function.

Narrator: The authors argue that the adoption of any transformative technology, from electricity to AI, happens in three stages. To illustrate this, they use the historical parable of electricity's adoption in manufacturing. The first stage involved "point solutions," where a factory owner would simply replace a single, large steam engine with a single, large electric motor. The layout of the factory remained the same, and the productivity gains were minimal.

The second stage brought "application solutions," where individual machines were redesigned with their own smaller electric motors. This offered more flexibility but was still constrained by the old factory design. The real revolution, the "system solution," came when entrepreneurs like Henry Ford realized that electricity's flexibility allowed for a complete redesign of the factory itself. They could build single-story plants outside of expensive cities and create the assembly line, a system that fundamentally changed manufacturing and unlocked massive productivity gains.

Today, AI is in a similar state. Many companies are implementing point solutions, using AI to automate or improve a single existing task, like using an AI to read medical scans. While useful, this approach barely scratches the surface. The true power lies in asking how cheap, accurate prediction allows for the redesign of the entire system—the hospital, the insurance company, or the supply chain.

Narrator: While AI-driven decisions can optimize one part of a system, they can create chaos in another. The authors call this the "AI bullwhip" effect. Imagine a restaurant that uses an AI to perfectly predict daily customer demand for its menu items. Instead of placing a consistent weekly order for ingredients, it now places highly variable daily orders with its supplier.

The supplier, facing this new unpredictability, implements its own AI to manage its inventory. This chain reaction continues up the supply chain, from the distributor to the grower, with each step amplifying the volatility. A small, smart change at the restaurant's end creates a massive, chaotic swing at the other end of the whip. The system becomes less reliable for everyone.

To manage this, organizations have two choices. The first is to increase coordination. Think of a coxswain in an eight-person rowing boat, who gathers information from all rowers and coordinates their rhythm to act as one. The second is to increase modularity. The authors tell a parable of two watchmakers. One builds a watch piece by piece; an interruption forces him to start over. The other builds the watch in stable sub-assemblies, or modules. When interrupted, he only loses a small amount of work. Similarly, an organization can design its systems to be modular, shielding other departments from the unpredictability of an AI-driven decision.

Narrator: For centuries, prediction and judgment have been bundled together in human decision-making. A seasoned doctor, for example, both predicts the likelihood of a disease and judges the best course of treatment. AI fundamentally breaks this bond. It decouples the act of prediction from the act of applying judgment.

The authors share the story of basketball legend Michael Jordan, who broke his foot early in his career. The team's doctors made a prediction: there was a 10% chance that playing on the foot would lead to a career-ending injury. That was the prediction. The judgment, however, was about what to do with that information. The team owner, Jerry Reinsdorf, judged that a 10% risk of losing his star player was far too high. Jordan, on the other hand, had a different judgment. When asked if he'd take a headache pill that had a 10% chance of being fatal, he famously replied, "It depends how f**king bad the headache is."

For Jordan, the value of playing and competing for a championship outweighed the risk. This is the essence of judgment: weighing the payoffs of different outcomes. AI provides the prediction, but humans provide the judgment. This decoupling is where power shifts, as the person or entity that controls the judgment—deciding what to do with the prediction—ultimately holds the power.

Narrator: Because AI enables the redesign of entire systems and the decoupling of judgment, it inevitably threatens existing power structures. Those who stand to lose influence or profit will resist change, even if it's for the greater good of the organization.

The classic example is Blockbuster's failure to adapt to the rise of Netflix. It wasn't that Blockbuster's management didn't see the threat. In fact, they launched their own subscription and on-demand services. The problem was internal. A huge portion of Blockbuster's revenue, especially for its franchisees, came from late fees. The new subscription model, which had no late fees, directly threatened that income stream. The franchisees, who held significant power within the organization, resisted the change. The board ultimately sided with the old model, and Blockbuster collapsed.

This illustrates a critical lesson for AI adoption. The greatest barrier may not be technological, but human. When a new AI-driven system threatens to make an existing role or department less powerful, resistance is almost guaranteed.

Narrator: A common fear is that AI will perpetuate and amplify human biases. This is a valid concern, as AI learns from data generated by a biased world. However, the authors argue that this is precisely why AI offers a unique opportunity to reduce discrimination.

Consider a study on racial disparities in knee pain. For years, doctors observed that Black patients reported more pain than white patients with similar-looking X-rays. The prevailing medical judgment was that the disparity was due to factors outside the knee. But when researchers trained an AI on the same X-rays, the AI learned to see things the human radiologists had missed. It found subtle indicators in the images that explained nearly five times more of the racial pain gap. The problem wasn't the patients; it was the biased human system for reading X-rays. The AI didn't just spot the bias; it provided a path to fix it.

Unlike human bias, which is hard to pin down and even harder to change, algorithmic bias is scrutable. You can audit the code and retrain the model with better data. As Sendhil Mullainathan notes, "Changing algorithms is easier than changing people." The challenge, therefore, is not just to build unbiased AI, but to redesign the systems around it to be fairer, a task that AI itself can help achieve.

Narrator: The central message of Power and Prediction is that the AI revolution will not be about simply replacing tasks. It will be about reimagining and rebuilding entire systems. The companies and leaders who succeed will be those who adopt a "system mindset," looking beyond the immediate application of an algorithm to see how cheap, accurate prediction can fundamentally change the way value is created and delivered. This involves decoupling prediction from judgment, anticipating shifts in power, and designing new, more efficient, and fairer systems from the ground up.

The book leaves us with a profound challenge. It asks us to look at our own organizations and industries not as a collection of fixed processes, but as a dynamic system of decisions. The most important question is not "What tasks can AI automate?" but rather, "If prediction were nearly free and perfectly accurate, how would we redesign our entire system to solve our most fundamental problems?" Answering that question is the true work of the AI era.