TL;DR: Two headlines, both too neat: "AI is coming for the jobs" and "AI will create more job opportunities". The truth sits between them. AI isn't subtracting jobs from the economy; it's widening a gap. Firms that adopt it deeply are hiring more, graduates included, while many others have slowed their hiring.
The apocalypse that didn't arrive
The forecasts were loud: Anthropic's own CEO said AI could wipe out half of white-collar jobs. Two years in, the data shows no sign of it. Stanford found unemployment for the most AI-exposed workers rose 0.77 points since 2022, against 0.85 for the least exposed. The exposed group is holding up slightly better (SIEPR). A softening market, not a machine eating it.
What slipped instead was pay. Apollo, using actual Claude usage logs across 321 occupations, found no measurable drop in exposed employment but real wage growth running about 6.7% behind everyone else (Apollo). Their line put it best: AI won't steal your job, but it might cost you a raise.
The gap is the story
The average hides a split. Ramp and Revelio Labs stopped guessing which firms use AI and watched the actual card spend: real payments to OpenAI, Anthropic and the rest, across 21,559 US companies (Ramp).
The serious adopters grew headcount 10% over two years, and entry-level hiring 12%. The dabblers showed no measurable change at all. And adopters' numbers isn't a moonshot budget: the leaders spent around $30 a head a month, roughly that same £25. The difference was what they bought with it: coding agents and API access wired into how work gets done, not a chatbot bolted onto the side.
The one place it's real
There is one real casualty: the first job. Workers aged 22 to 25 in the most exposed roles have seen employment fall 13% since ChatGPT arrived, while older workers in the same jobs held steady (SIEPR). The Dallas Fed explained why: AI is very good at the codified, textbook part of a job, which is exactly what a junior does on day one. The judgement you only get from years in the room, it can't touch (Dallas Fed).
Which side of the line
Which side of the gap are you building for? Three moves decide it.
Cross the spend line. A chat subscription is table stakes and shows up nowhere in the numbers. The firms pulling ahead put agents and API access into real hands, wired into the actual work. That's the $30-a-head line, and it isn't the money that's hard, it's the wiring.
Give it a year. The gains tend to land in 6 to 12 months. Judged too early, it looks like a failure, right before it would have paid off.
Rebuild the first rung, don't cut it. The deep adopters hired more graduates, not fewer. If the entry job used to be "do the codified grunt work", the new one is "run the AI that does it, and learn the judgement earlier". Cut the rung to save a little now, and you have no seniors in five years.
The best agents still can't touch the judgement or the messy long-horizon work. For humans, in the near future, that might very well be the whole job. And the question for managers and founders is this: are they investing in the wiring, the future-proof setup where their people and their operations can genuinely benefit from AI?