JUDGMENT判断AI has not taken jobs yet. It is already taking the entrance — and those two facts do not cancel each other out. Set the two sets of September numbers on the same table and the conclusion is uncomfortable: answering "the work is going away" with "nobody got laid off" is like weighing yourself with a measuring cup. Wrong unit.

FACT事实On September 1 the Federal Reserve Bank of New York published its regional business survey: AI adoption reached 61 percent among service firms, up from 40 percent a year earlier, and 51 percent among manufacturers, up from 26 percent. In the same week, China's Ministry of Human Resources and Social Security ran four online hiring fairs under its Hundred Days, Ten Million campaign; the artificial-intelligence fair alone drew nearly 8,000 employers and close to 120,000 openings, advertised under titles such as large-model algorithm engineer, AI product manager, data-mining engineer and AI architect. Neither set of numbers looks like a slump. The valuable half sits on the back pages. Among firms that have adopted AI, the median share of service-sector employees actually using it is 17 percent. Three-quarters call their investment minimal to modest; only 5 percent of service firms treat it as a major strategic commitment. Over the past six months, 4 percent of service firms cut staff because of AI, against 1 percent a year earlier, while manufacturers reported none for a second year; 15 percent hired fewer people than they otherwise would have, and 13 percent hired more. More than a third are retraining. One line near the end of the report deserves to be copied out on its own: among firms that have not adopted AI, cost was the least-cited reason — about half said their kind of work simply does not lend itself to it.

SPECULATION推演Three misalignments. First, the subtraction and the addition land at different heights. Fifteen percent hiring less against 13 percent hiring more nets out near zero, and aggregate statistics see nothing unusual. But the roles left unfilled are mostly junior ones, while the roles being added are deployment, governance and integration — the subtraction lands at the bottom of the ladder, the addition above the middle. The total is flat; the entrance is tilted. Second, adoption rate and depth of use are out of proportion. This is a trial, and the inference cuts both ways: every judgment now circulating about employment, including the optimistic ones, is an early signal. Third, the retraining points sideways, not upward. The content is tool literacy, routine-task automation and prompting; the stated goal is doing the current job better. Firms treat AI as an add-on rather than a reorganization, and an add-on does not assign anyone a new position. The people squeezed out stay where they are.

OPINION观点Do not take "nobody was laid off" as comfort. Watch two other numbers instead: the count of junior postings, and how long a new hire takes to deliver something on their own. The first is already moving. The second appears in no official statistic, and it decides whether there is still a mid-level job to fill three years from now. One finding deserves its own line: half of the firms that have not adopted AI say their work does not suit it. The boundary of AI is drawn by the work itself, not by the budget. Whether an industry gets rewritten depends less on what it spent than on how much of its work can be written down as a path. An organization is a set of information paths, and where the path is clear, the node stops being essential.

---

Further Reading