In a single stretch of reporting this month, you could read that artificial intelligence is destroying jobs faster than it creates them, that it is generating more than two jobs for every one it eliminates, and that it is not coming for your job at all. Each claim arrived with data attached. The contradiction is real, but it is less a sign that someone is wrong than a clue that these numbers are measuring different things.
Start with the alarming set. The technology sector shed about 63,000 workers in June alone, pushing its layoff rate to the highest level in two decades. Oracle cut roughly 21,000 roles and told regulators, in a formal filing, that deploying AI had reduced its workforce. The outplacement firm Challenger, Gray & Christmas found AI cited in a record 40 percent of announced cuts in May. Skeptics note that "AI" also makes a flattering label for ordinary cost-cutting, one that sounds like transformation rather than retrenchment, but the direction of travel is not in doubt.
Now the reassuring set. A Nomura analysis of employment cases across Asia found India adding roughly 2.6 AI-related hires for every job lost or displaced, with a similar pattern region-wide. Google's ATLAS study, the broadest yet, reports that AI has reached 68 percent of occupations but touches only about a fifth of the tasks within them, and completes fewer than one interaction in ten on its own. The phrase that captures it is "wide but shallow." For now the technology augments far more than it automates, which is why overall unemployment has barely moved.
Both pictures can be true because the pain is not spread evenly. The Nomura figures come with a catch: the jobs created are rarely the ones destroyed. A customer-support worker replaced by a chatbot cannot simply step into an AI-engineering role, and the gap between the two is where a two-tier labour market forms. A Stanford study found employment for workers aged 22 to 25 in the most exposed jobs down about 16 percent relative to their older colleagues. The bottom rung of the ladder, the codifiable junior tasks, is exactly what today's models do best.
Inside companies, the shift looks less like demolition than renovation. Drawing on 3.6 million employee records, the analytics firm Visier found overall hiring down 24 percent, yet the churn beneath that number is fierce: the share of AI engineers hired jumped 251 percent while data scientists fell 32 percent, within the same growing department. Roles that resist automation, such as teachers, lawyers, architects and security staff, hold their value because they carry judgment and accountability that a fast answer cannot. Mid-career workers, once seen as expensive, are suddenly prized for the experience a model lacks.
Even the people already using AI report something messier than the sales pitch. A survey of 4,595 Quebec union members found only 44 percent gained productivity while 26 percent lost it to "AI slop" they had to redo, and the benefits tracked existing advantages: 55 percent of postgraduates felt more productive, against 22 percent of those with only a high-school diploma. All of which points to the question the aggregate numbers hide. Whether AI destroys or creates jobs may matter less than how fast displaced workers can move into new ones, and who ends up capturing the gains. That part is not a forecast about the technology. It is a choice.