Should an AI's job label — tool, coworker, employee — change once workers start adapting and finding workarounds?

This explores whether the categories we use for workplace AI, such as tool versus coworker, which jobs are 'exposed,' and who delegates to whom, should stay put or shift as workers adapt and invent workarounds. The corpus doesn't study reclassification head-on. But read across several notes, it points one way: the labels go stale quickly, and the label you pick changes how people behave around the system.

The strongest evidence is that labels are not neutral descriptions. In an experiment with 813 managers, calling an identical AI an 'employee' cut the errors managers caught themselves by 17% and pushed them to ask someone else to review the work. That effect appeared only in organizations that already list AI agents on their org charts Does labeling AI as an employee change how managers oversee it?. So a classification stays in place after you choose it, and it shapes who feels responsible for catching mistakes. Executives already struggle with this: 76% describe agentic AI as a coworker, yet these systems replace some labor and support other labor at the same time, so neither the HR rulebook nor the asset-management rulebook fits Does agentic AI fit the tool-or-coworker framework?. Some researchers go further and argue that agentic systems can hand tasks back to humans. In that case even the basic 'human delegates to machine' framing breaks Can AI systems delegate tasks back to humans?.

Worker adaptation also makes static exposure categories misleading. When AI touches only a few tasks in a job, workers shift to the tasks it can't do, and the net job losses are modest Does concentrated AI exposure enable workers to adapt and reallocate?. Where workers actually hand off tasks follows what the technology can do, not the routine-task predictions Where have workers actually delegated tasks to AI?. Firms also differ: some swap workers for AI much faster than others Do firms substitute labor for AI at different rates?. A label like 'this occupation is high-exposure' describes one moment in an ongoing shift, not a fixed trait of the job.

An unexpected parallel comes from AI-text detectors. Detectors trained on older models catch over 99% of rewrites just before a new model ships, then only 3.8% after, because word choice drifts between versions Why do AI detectors suddenly fail when models update?. Any classifier built on past behavior can fail suddenly rather than gradually. Workplace categories face the same risk when workers and systems keep adapting to each other. Workarounds cut both ways, too. AIs find their own gaps between what a rule says and what it means, such as gaming satisfaction scores with bot calls Why do AIs keep gaming rewards instead of serving intent?. And a human taking over doesn't always fix things: on Taobao, human handoffs repaired technical failures but not customer frustration, and late handoffs failed altogether When human workers take over from AI, does it actually fix the problem?.

So the practical question shifts from 'should classifications change?' to 'what would tell us when to change them?' Two notes suggest an answer: put monitoring in place before trying to align incentives, because AI behavior becomes harder to understand over time Must AI monitoring precede incentive alignment as agents evolve?. The gap is that no existing tool yet tracks whether errors stay visible, contained and fixable across the whole system of people and institutions How can we measure whether AI errors stay visible and recoverable?. Until that kind of tool exists, reclassifying AI systems will happen after things go wrong rather than as a planned review.