AI Can Cut Headcount, But the Real Cost May Be the Technology Itself
For months, the central question around artificial intelligence and jobs has been which workers it will replace. Now companies are discovering another bill, the token bill. The article cites reports of extreme examples, including an unidentified enterprise customer that reportedly ran up about $500 million in one month using Claude after failing to set proper limits, and OpenClaw creator Peter Steinberger, who said OpenAI tokens cost him $1.3 million in 30 days after about 100 Codex runs.
Those cases are outliers, but they illustrate a broader shift. AI can lift productivity, shorten workflows and help teams do more work with fewer resources, yet every prompt, answer, code review and document analysis carries a price. PwC data show that sectors more exposed to AI have recorded much stronger productivity growth than less exposed ones. Still, Uber’s chief technology officer recently said the company’s 2026 AI budget had already been exhausted early in the year, and NVIDIA has also reported very high computing costs.
The article says the market is moving from excitement about AI to more exact management questions, such as which model to use, which tasks justify an advanced model and how to measure return on investment. A KPMG survey found that 57% of managers expect humans to steer AI agents, underscoring the need for oversight. In practice, that is creating demand for workers and managers who understand both AI and its economic impact.
Elad Systems says it trains new employees to break tasks into smaller parts, match models to task complexity and use tools such as Cache and RAG to reduce unnecessary calls to expensive models. Yuval Markovich, head of development in the company’s digital division, said firms now want people who understand the business consequences of technology choices, not only how to build AI products. He argued that building systems that are smart, efficient and economical is now a major advantage. Priority is using AI in day-to-day operations as well, from analytics infrastructure to customer forecasting and service management. Michael Lorberbaum said AI has cut development time for ERP and BI systems by 20% to 40%, while internal models help employees prepare presentations, analyze spreadsheets, build apps and answer emails. The article’s bottom line is that AI may not simply replace workers, it may instead change the kind of workers companies need, and the real test is whether organizations can manage the technology well enough for it to pay off.
