Arctic Wolf, a cybersecurity company, cut 250 jobs this week. The announcement cited a push to "leverage AI and automation" across the business. No other explanation was offered. The same day, Torsten Slok, chief economist at Apollo Global Management, published the fifth blog post in a week arguing that AI will be a net creator of employment, not a destroyer of it. Meanwhile, ADP's April payroll data showed US private-sector hiring at its strongest pace in over a year, with construction employment growing, driven in part by the building of data centres to support the AI boom. Three data points, three very different stories about what AI is doing to work.
The case Slok is making is built on the Jevons paradox, an idea from 19th-century economics. When the cost of something falls, consumption of it tends to rise, and that increased consumption can generate more total activity than was lost to efficiency. Slok's version: when AI reduces the cost of a task, demand for the broader job that contains that task can actually grow. He cites radiology as his test case. A decade ago, AI was supposed to eliminate radiologists. Instead, the number of radiologists has grown and their salaries, now exceeding $500,000 a year in some US markets, have risen sharply. The explanation, in Slok's telling, is that cheaper imaging analysis made the overall field more affordable and accessible, expanding the total market for diagnostic services.
It is a serious argument and deserves to be taken seriously. But Slok himself conceded in an interview this week that AI will "disproportionately impact" software and programming industries, and that the geographic and sectoral effects will be uneven in ways that matter for real workers and real communities. The Jevons paradox describes a direction of travel, not a guarantee. It tells you that falling costs can expand markets; it does not tell you when, for whom, or whether the new jobs pay what the old ones did, or appear in the same places, or require similar skills.
The Arctic Wolf case illustrates the asymmetry. When a company like Arctic Wolf automates functions and eliminates 250 positions, the people who held those positions face an immediate and concrete harm: no income, an uncertain job market, the need to find comparable work in a labour market that is, in some sectors, contracting. The new jobs that Slok's theory predicts will emerge, more radiologists, more construction workers building data centres, more roles in sectors where AI-enabled expansion creates demand, may be real, but they are not available to the former Arctic Wolf employees on Monday morning. The gap between macro-level optimism and individual-level consequence is where the real human cost of this transition lives.
The ADP data is worth taking seriously too, even if it complicates the narrative. April payrolls grew more than expected, with construction employment rising noticeably. The LA Times and analysts at Mizuho Securities point to data centre construction as a meaningful contributor: AI investment is real, the infrastructure required to run it is being built, and that building requires workers. This is not a hypothetical future job-creation effect; it is happening now. The question is whether it scales to compensate for displacement happening in other sectors, and whether those construction jobs are accessible to, say, the office workers, customer service staff, and software testers who are losing positions elsewhere.
What is becoming clear is that aggregate employment numbers are not the right instrument for measuring what is happening. The labour market can be simultaneously creating jobs in construction and logistics while destroying jobs in cybersecurity, software development, legal research, and content creation. The totals can net out at something that looks stable or even positive while the disruption to specific workers, sectors, and communities is severe. The people in the growing sectors and the people in the shrinking sectors are not the same people. They are not in the same cities. They do not hold the same qualifications.
The debate between "AI destroys jobs" and "AI creates jobs" is, in this sense, a false binary. Both processes are underway. The more useful question is about distribution, timing, and whether the institutions of democratic economies, unemployment systems, retraining programmes, labour law, and tax policy, are equipped to manage a transition that is faster and more uneven than most prior technological disruptions. On that question, this week's events offer little reassurance.