If you've read your own job description lately and quietly counted how many lines a chatbot could handle, that's a fair question. The worry deserves a better answer than either reassurance or a frightening headline.
Here is my view. A job is a bundle: tasks, judgment, relationships, and being accountable when something goes wrong. AI is getting good at pieces of that bundle. In the near term, I think most of the change looks like people working alongside these tools, with experienced people checking the output and owning the call. Over a longer stretch, AI may take on much more of the work. I won't pretend otherwise, and I can't promise everyone a supervisor role at the end of it.
A task-based study gives us a useful way to separate possible changes from actual job losses. The ILO's refined exposure index scores work task by task and treats transformation as more likely than full replacement. Exposure is not the same as a lost job, and it doesn't guarantee safety either.
One workplace study helps because it is concrete. Researchers following 5,172 customer support agents found that an AI assistant raised issues resolved per hour by about 15 percent on average. The gains were uneven. The most experienced, highly skilled agents got small speed gains and slightly lower quality. That is one setting, not a forecast for the economy.
Take a hypothetical. Picture a bookkeeper at a twelve-person landscaping company. Reconciling receipts takes them six hours a week. Suppose a tested AI tool can help prepare that work. The bookkeeper knows which unusual payment arrangements need checking and what a convincing but wrong reconciliation looks like. Some of that knowledge can be supplied to the tool too. The useful question is what still needs checking, by whom, and at what cost.
The honest objection is that employers don't pay for bundles forever. If tools save enough time across several tasks, an accounting firm may redistribute the remaining work and eliminate a position. "Your job is more than its tasks" is cold comfort to the person who gets cut. I think that risk is real, and nobody should wave it away.
That's exactly why I'd rather you run the experiment before someone else runs it on you. Pick one task you do every week and write down how long it takes now. Use an approved tool and invented or public examples first. If your workplace permits a real trial, check every result and log the time spent correcting it. Then discuss the result with your manager, including any time saved, time lost, and work you could do differently. You'll learn more about your job's future than a headline can tell you.
AI-assisted editing. Published with Rob Price's approval.