AI doesn't just replace tasks — it also creates new ones. Are those new jobs actually worth as much as the ones lost?

This explores whether the new tasks AI creates are worth as much as the tasks it displaces, economically and for workers. The collection's most direct answer is that the balance isn't set by the technology. It depends on what firms choose to build. Acemoglu, Autor and Johnson sort AI into types by what it does to work, and they argue that automating expertise pays firms more than building AI that creates new, valuable tasks for people Why do firms build automating AI instead of pro-worker AI?. So the gap between new and displaced tasks isn't a random outcome. Profit incentives tilt it, and they tilt it toward automation. What's good for each firm adds up to too little investment in technology that makes workers more valuable.

Whether displaced workers can move into other work depends on how AI's reach is spread across a job. Firm-level data from 2010 to 2023 shows that high average AI exposure lowers demand for labor. When exposure is concentrated in just a few tasks, though, workers shift their time to the tasks AI doesn't touch, and net job losses stay small Does concentrated AI exposure enable workers to adapt and reallocate?. That's a useful distinction. A job that loses one task keeps its value by leaning on what's left, while a job where AI touches everything has nowhere to move.

Some of the 'new tasks' AI creates are less glamorous than the phrase suggests. Studies of AI-assisted work find that total time often doesn't shrink. It moves away from doing the work and toward writing prompts, reading outputs and checking them Does AI really save time, or just change how we spend it?. That counts as new work, but it isn't obviously more valuable, and it changes what people learn on the job. Related work suggests what may stay with people: checkpoints for actions that can't be undone or that other people will see, like sending an email. Users want to approve those even when the AI's output is good enough What makes people distrust AI agents they delegate to?.

There's also a clue about which tasks get displaced first. Wei's 'verifier's rule' argues that AI gets good at tasks whose answers are easy to check, and that organizations can make more tasks checkable by building test suites and answer keys up front Does task verifiability determine what AI systems will learn to solve?. If that's right, displacement follows checkability, not value. The tasks left to people tend to be the hard-to-judge ones, which may be valuable but are also hard to measure, price or reward.

The collection is thin here. It doesn't have studies that directly measure the wages or productivity of AI-created tasks against the ones they replaced. What it offers is the economic framework (who decides what gets built), evidence on reallocation (when workers can adapt), and a fine-grained look at what the new work actually looks like day to day.