Why AI Erodes Your Work Identity Before You Feel It

A woman who runs a small analytics team told me last week that her work feels lighter and less hers at the same time. The reports go out faster now. The judgment calls that used to feel like the center of the job arrive half-finished from a tool, and she signs them. She is not afraid of losing her job. She is afraid she is becoming someone any competent person could swap in for.

What the research shows

There was a study presented this spring at a conference on how people work with computers, from a team led by the researcher Upol Ehsan, that followed expert workers for a full year as AI moved into their daily tasks. These were people whose judgment is the job: specialists in cancer care, and engineers who build software. The researchers watched something they named intuition rust settle in. The early gains were real. The tool made the work faster and the output cleaner. Underneath that, the fine-grained judgment that takes years to build began to dull from lack of use.

Then came a slower shift they called identity commoditization: the specialist gradually reduced to a person who oversees an AI, distinct expertise flattened into a generic role that, on paper, anyone could fill. The harms were quiet at first. The researchers describe a cascade, from a silent drift in behavior, to a real decline in skill, to a kind of existential unease about what was left.

This lines up with an older idea from work psychology. A 2022 paper by Eva Selenko and colleagues argued that a job does more than pay us. It hands us a self: the tasks and categories that let a person say, with some continuity over time, this is who I am and what I do well. When AI takes over the tasks that anchored that self, the line between who you were good at being and who you are now gets harder to hold. It is the same thread running through what the research on how reliance erodes self-belief keeps finding. Lean hard enough on the tool, and the felt sense of being capable starts to thin out.

Work carries a lot of weight for most of us, which is why this is worth taking seriously.

This is close to the identity threat that leaders have been learning to spot in their teams. Here it is pointed back at the person doing the spotting.

What it doesn’t tell us yet

This is one study, and a close-up one. It followed a small number of people in two specific fields, cancer care and software, watched over a year rather than measured with a clean before-and-after. There are no effect sizes, no broad sample, nothing that proves this happens to everyone who works alongside AI. The Pew figure is older, and it is about how central work is to identity in general, not about AI at all. So treat the cascade as a careful warning grounded in close observation, not a settled law. The early dashboard wins are exactly where the cost hides, which is its own version of the gap between AI adoption and real productivity.

One thing to notice in your work today

Notice the gap between this went faster and this was mine. When an output comes back fine but not in a person’s own voice, and it gets signed anyway, that is the small moment the study is pointing at. It is not a problem to fix today. It is a thing worth tracking over a quarter. Are the calls that used to feel like the heart of the work still the ones being made by the person, or have they quietly become things to approve. The noticing is the whole practice here. The work stays figure-out-able, one signed-off decision at a time, as long as the through-line still feels like a person’s own.