Model AI-exposed work units
AI Supply / Demand Simulator
This simulator models AI as a supply shock for specific work units, not as a clean replacement forecast for whole jobs.
The default scenario is student study resources: notes, summaries, quizzes, and generic explanations. Those used to carry value because they were scarce, effortful, and locally useful. Once AI can generate a passable version on demand, the generic resource becomes much harder to sell. The work around it doesn't disappear, but the paid part moves toward feedback, diagnosis, accountability, and course-specific judgement.
What the model shows
The chart starts with a simple market for one exposed work unit. When AI capability rises and marginal cost falls, the supply curve shifts outward. The unit price drops. Quantity often rises. Human value only holds if demand expands enough, or if the remaining human part is scarce enough.
You can scrub the shock stage, replay the transition, switch between price, demand, and human-residual views, and open the advanced controls for baseline demand, human supply cost, elasticity, rigidity, and value capture.
That separation is useful because it keeps job survival and task devaluation apart. A tutor, writer, support agent, or developer may still be needed. But a specific unit inside the role can lose exchange value quickly.
Why demand still matters
Cheap production doesn't always mean fewer humans. If lower prices unlock much more demand, the market can expand. If demand is mostly fixed, the exposed unit gets compressed. James Bessen's work on AI and demand is useful here: technology can expand employment in some markets and reduce it in others depending on demand elasticity.
The simulator lets that variable move. Student notes tend to show a severe price collapse because demand for generic summaries is limited once everyone can make one. Support replies often look softer because cheaper replies can mean more customers served.
What stays scarce
The scarce part is usually less tidy than the automated part. It is context, trust, judgement, relationships, accountability, taste, and the ability to notice what is actually wrong.
That isn't always a nicer job. It can be harder to measure, harder to train, and more emotionally loaded. But it is where the value tends to move when AI turns a generic work unit into abundant supply.