If AI can do a few of your tasks, you can shift to the rest. What if it nibbles at every task instead?

This explores whether the shape of AI exposure inside a firm matters, meaning AI hitting a few tasks versus being spread across all of them, and what that does to workers' ability to move into other work. The corpus's most direct answer is that concentration cushions the blow. Task-level data on firms from 2010 to 2023 show that higher average AI exposure reduces labor demand. But when exposure is concentrated on a few tasks, workers can reallocate to the tasks AI isn't touching, and the net employment effect ends up modest Does concentrated AI exposure enable workers to adapt and reallocate?. Two firms with the same average exposure can therefore have very different outcomes.

The intuition is that a job is a bundle of tasks. If AI takes over one task in the bundle, the worker keeps the rest and shifts their time toward them. If AI grazes every task a little, nothing is left inside the firm to shift toward. So the average exposure number can mislead, because the distribution across tasks decides whether the exposure displaces people or just reshuffles their work.

What happens next depends on the firm too. Firms with higher AI exposure swap out online-marketplace freelancers for AI tools faster and more cheaply than less-exposed firms. That points to returns to scale in a firm's internal AI capability, not one uniform wave of technology adoption Do firms substitute labor for AI at different rates?. The pressure is also not spread evenly across kinds of work. Tasks people have actually handed to AI cluster in information-intensive occupations and follow what the technology can do, not the older 'routine task' predictions Where have workers actually delegated tasks to AI?. Reallocation is therefore a question about particular firms and occupations, not the labor market as a whole.

The cushion has limits. Moving to non-displaced tasks mostly means staying within skills you already have, and that is where AI's productivity gains show up. When workers used AI to learn something new, the gains disappeared and the learning suffered When does AI actually boost worker productivity?. If the tasks left over require different skills, AI demand is splitting around a technical core of Python, SQL and data analysis, and non-technical occupations diverge from it rather than converge Is AI creating common skills across jobs or deepening divisions?. Room to move is also uneven across people. Exposure concentrates among high-skilled, high-paid workers in male-dominated occupations but spreads across all skill levels in female-dominated ones, which leaves lower-paid women exposed with fewer resources to adapt Does AI exposure hit low-wage workers harder in some fields?. That finding is about who is exposed, not which tasks remain, so it doesn't show those workers lack tasks to move into. It does show the buffer isn't evenly distributed.

The corpus stops at employment. The reallocation finding says workers keep their jobs by moving to non-displaced tasks. It doesn't say whether those tasks are better or worse work than the ones AI absorbed.