The Silicon Salary: Why Humans are Suddenly the Low-Cost Option
AI tools got so good that companies couldn’t stop using them — and now the bills are out of control. Uber burned through its entire 2026 AI budget by April because engineers were using agentic coding tools that charge per every step the AI “thinks,” not per seat. The average engineer cost $150–$250 a month. Heavy users hit $2,000. Microsoft quietly pulled back Claude Code access for the same reason. Meanwhile, an MIT study found that for 77% of vision-based tasks, a human is still the cheaper option. So the math just doesn’t work the way the “automate everything” pitch promised. The companies getting this right aren’t banning frontier AI tools. They’re setting budgets at the task level, matching the model to the job, and putting humans back in the workflows where cost and judgment actually matter.
