AI and the Organization of Work

Week 12 · November 16, 2026

When should an organization give decisions to people, algorithms, or some combination of the two?

Opening Hook

A manager introduces an AI system that can forecast demand, assign tasks, flag unusual behavior, and recommend promotions. The system makes some decisions faster and more consistently than a person could. Yet workers still hold local knowledge, notice exceptions, and decide how much care and initiative to bring to the job.

If the manager transfers more authority to the system, its recommendations may improve. But workers may also stop gathering information, exercising judgment, or speaking up when the system is wrong.

Here is the question this session is about: when should an organization give decisions to people, AI, or some combination of the two?

Tasks Are Not Jobs

AI typically changes particular tasks: drafting, prediction, classification, scheduling, monitoring, or diagnosis. A job is a bundle of these tasks plus relationships, responsibilities, and decision rights. Automating one task can remove work, create new work, or make complementary human tasks more valuable.

To understand AI and employment, begin with tasks and complementarities rather than treating every job as simply automated or untouched.

Choose a job you know. Which tasks are easiest for AI to perform, and which human tasks become more valuable if that happens?

Decision Authority and Effort

Athey, Bryan, and Gans use the economics of delegation to study who should hold decision authority when both a person and an AI system may have useful information. Giving an AI more authority can improve the immediate decision, but it can also weaken the person’s incentive to acquire information or exert effort.

Better AI does not automatically imply giving AI every decision: authority also changes human incentives to learn, investigate, and contribute.

When should an AI recommendation be binding, and when should a worker be able to override it? What incentives does each rule create?

Monitoring, Trust, and Control

AI lowers the cost of observing, predicting, and evaluating worker behavior. That can improve coordination and detect genuine problems. It can also make intensive control easier, narrow attention to measurable behavior, and signal distrust. The organizational question is therefore not only what AI can observe, but what should be measured, who can contest a judgment, and how the system changes cooperation.

AI changes the cost of control, but the workplace effects depend on how authority, accountability, and the right to challenge a decision are designed.

How would you design an appeals process for a workplace decision made partly by AI? Who should bear the burden of explaining or correcting an error?

Why might giving a more capable AI system greater decision authority reduce a worker's contribution?

  1. Because a capable AI is always wrong.
  2. Because the worker has less reason to gather information or exercise judgment.
  3. Because decision rights disappear when AI is introduced.
  4. Because workers always oppose new technology.

Review Cards

Connections

Builds on: Session 11, where monitoring could improve incentives but also signal distrust and crowd out cooperation.

Sets up: Project presentations and final synthesis, where students connect a fact about work to a mechanism involving incentives, authority, or power.

Reading Guide

Required

Charles I. Jones. “A.I. and Our Economic Future.”

What to look for: The channels through which AI could transform economic growth and work, as well as the limits and risks that complicate simple forecasts.

Bring to class: One claim about AI’s economic effects that depends on an assumption about what tasks people or machines can do.

Optional

Aaron Chatterji, Daniel Rock, and Eduard Talamàs. “Transformative AI and Firms.”

Use this broader piece to think about firm boundaries, knowledge, investment, and competition when AI capabilities change quickly.