People have finally started saying to someone’s face that AI is taking jobs. The CEO of Anthropic said bluntly in a TV interview that over the next one to five years, as many as half of entry-level white-collar jobs could be wiped out. This isn’t some distant future from a sci-fi movie—it’s already happening in job postings, performance reviews, and office seating plans. “Entry-level” doesn’t usually mean decision-making roles. It often means process-driven work like organizing information, screening resumes, writing basic copy, taking meeting notes, responding to standard support tickets, and checking spreadsheets.

Interestingly, this works a lot like sector rotation: where the hot money moves is never evenly distributed. The job market is the same. The roles that get hit first are always the ones that involve standardized, repeatable work done at a screen and keyboard. Take e-commerce customer service, for example. During major sales events, companies used to hire extra staff, with teams of ten taking turns handling inquiries. Now, they can feed common questions, return and exchange policies, and response templates to AI, then get by with two or three people and a review process. The rest move on to handling unusual complaints and complex after-sales issues. Or take junior legal assistants: they used to review dozens of contracts a day and flag risky clauses. Now, a tool can do an initial pass, and people only need to review the high-risk items. It’s not that jobs disappear overnight; it’s that the output of the same job gets repriced.

Weekend trading volume is only 30% of the weekday level, so even a little volume can move the market. The same logic applies to hiring: a role that used to be split among several people can now produce as much as several people with just a little help from efficiency tools. Companies do the math simply: with the same budget, if a task can be done by “a person + tools,” they won’t hire according to the old staffing model. That’s why entry-level roles shrink first: their tasks are easiest to break down into standard steps and easiest to measure.

Friends in traditional industries have been asking lately whether they should switch careers. I tell them not to rush into signing up for a course. First, ask whether your work can be broken down by a single prompt. If it can be split into “input—process—output,” such as formatting reports, rewriting content in bulk, basic translation, or routine Q&A, start thinking about a backup plan. What can’t be broken down is more valuable: work that requires on-the-ground judgment, accountability, negotiating with people, cross-department coordination, or making decisions with ambiguous information—those are things tools can’t replace for now. We’ll come back next week and see if this holds up.