When AI Gets Too Expensive: Why Companies Are Rehiring the Humans They Replaced

A growing number of companies are discovering that the AI tools meant to cut costs are now rivaling -- and in some cases exceeding -- the salaries of the workers they replaced. According to Zylo's 2026 SaaS Management Index, organizations spent an average of US$1.2 million on AI-first applications in 2025, more than double the prior year.

Key Takeaways

• Token pricing is the culprit. Subscription models have given way to usage-based token pricing, and the math doesn't favor lean budgets. Nvidia's VP of Applied Deep Learning Bryan Catanzaro told Axios the cost of compute now exceeds the cost of his team. Glean CEO Arvind Jain called it unprecedented: "This is the first time ever that I can remember that technology costs the same as people."
• The AI boomerang is real. A Robert Half survey of 1,365 hiring managers found 34% of companies that cut staff after AI adoption have already reinstated those roles or similar ones. Gartner projects 50% of businesses that eliminated customer service or operations roles will rehire by 2027. Klarna's AI chatbot savings evaporated after customer dissatisfaction forced them to bring back their human workforce.
• Microsoft cancelled Claude Code licenses after roughly six months, redirecting engineers to GitHub Copilot CLI -- widely attributed to unexpectedly high agentic AI costs.
• The cost curve will drop -- eventually. Gartner projects inference costs on advanced AI models will fall nearly 90% by 2030. But analysts warn that agentic AI consumes far more tokens per task, so lower unit costs won't necessarily mean lower total bills.

In Canada, about 60% of the workforce -- roughly 12 million workers -- are in jobs highly exposed to AI-driven transformation, though only 6.3% of AI-adopting businesses reported a decrease in employment so far.

Read the full article on Money.ca