Does AI Change What You Do at Work?

Does AI change what you do at work? For most people the honest answer is yes, but not by taking the job away. A study built from patent based AI exposure measures matched to individual workers' own survey answers about their daily tasks, and to German administrative wage records, finds that AI reaches inside a job and swaps some tasks for others long before it touches headcount.

The task that actually shrinks

The researchers asked workers directly which of seventeen detailed tasks they perform on the job, then tracked how those answers moved as AI exposure rose in a worker's industry. The clearest decline is in gathering, investigating, and documenting information, the classic desk research task, which falls by about 3 percentage points as AI exposure increases by one standard deviation. Tools that surface data automatically, from screening software to predictive maintenance systems, take over the manual legwork of finding things out.

Other tasks people might expect to shrink, like selling or marketing, showed no measurable change at all. The decline concentrates in one specific task, not across the board.

The task that grows in its place

Whatever gets automated somewhere still needs watching. The same study found that AI raises the routine task share of a job overall, driven almost entirely by an increase in monitoring or operating machines and technical processes. That is the opposite of what industrial robots do to the same jobs: robots have historically reduced routine tasks, while AI increases them, because someone still has to oversee the output an automated system produces.

Most of this shuffling happens inside the same detailed occupation, not between one job title and another. An accountant or a claims handler keeps the job title. The daily list of what fills the hours changes underneath it.

Why skill decides who gets the better half of the trade

The reassignment is not neutral. High skilled workers in AI exposed industries saw teaching, training, and consulting tasks rise by about 6 percentage points, enough to fully offset the abstract tasks AI took away. A worker with the standing to explain, coach, or advise gets handed more of exactly that work as automation frees up the hours.

Low skilled workers in the same industries did not get a comparable replacement task. Six of eight abstract tasks studied showed a negative trend for this group, with no offsetting gain elsewhere. The same technology arrives in both jobs. Only one side has a task ready to absorb the shift.

What happens to pay once the tasks move

Task change shows up in the paycheck. Earnings for low skilled workers in AI exposed industries fell by about 2.5 percentage points as exposure rose over a five year window, whether they stayed in the job or switched to a different one. High skilled workers who stayed in their job saw earnings rise by about 1.5 percentage points instead, with no measurable loss even among those who changed jobs.

Roughly half of the workers who did leave an AI exposed industry moved to a different, less exposed one, evidence of real adjustment happening at the edges even while most people keep their job title.

Building the half of the trade that pays off

Job displacement is not the risk this data points to. Task drift is. A job that gets more monitoring, more coaching, or more judgment calls and less routine information gathering rewards the person who can show they are good at the parts that remain, not the person still proving the parts a tool now handles.

That is a case for showing evidence of the skills AI has not automated away, not for treating this as an organization wide transformation story like the one Gallup's own survey data tells about AI adoption. This study looks at the same shift from inside a single job, task by task, rather than from the top of an org chart, and the two readings agree on the same point: the gain goes to whoever can do the part of the job a model still cannot.