Two stories about AI and work are running side by side, and they do not match. One is the story of viral warnings, booed commencement speakers and surveys showing widespread dread. The other is the story told by the labour statistics, which keep refusing to show the collapse that the first story predicts. A run of analyses this year has tried to reconcile the two, and the answer is more specific, and less dramatic, than either camp tends to admit.
The occupations that were supposed to vanish
Start with the roles most often named as AI's first casualties. Rather than shrinking, several have expanded. Recent tallies point to roughly 7 percent more software developers than in 2022, about 10 percent more radiologists and around 21 percent more paralegals. Federal data has also shown unemployment in AI exposed occupations running lower, not higher, than in less exposed ones, with no sign of a large migration of workers towards supposedly safer jobs.
Where the damage is real
That does not mean nothing is happening. Research from Stanford's Digital Economy Lab found roughly a 16 percent decline in entry level jobs within AI exposed occupations across 2024 and 2025, with the hit concentrated almost entirely on workers aged 22 to 25. Older staff in the very same roles saw employment grow. The cleanest way to read the data is that AI is eroding the bottom rung in tasks that can be fully automated, while augmenting and even rewarding more experienced workers above it.
Adoption is still early
Part of the gap between fear and outcome is simply that deployment remains thin. By recent counts only about one in five US companies use AI in any business function at all, which makes economy wide displacement implausible for now. A Federal Reserve study found that growth in coding employment had slowed by around 3 percent since ChatGPT's release, a real dent, but one set against employment for coders that is still rising overall.
The honest summary is that the disruption is narrow, generational and early rather than broad and finished. The pressing risk is not mass unemployment but a difficult transition: wage stagnation in exposed roles, jobs redefined faster than training can keep up, and a hollowed out entry level that makes it harder for young workers to get started. Treating that as the problem, instead of arguing about an apocalypse that the data has not delivered, is where the useful policy work begins.
