At some point in early February 2026, humanity stopped being the primary consumer of its own AI infrastructure. It has not reclaimed that position since.
According to OpenRouter analyst Peter Walker, February 6, 2026, was likely the last day humans consumed more tokens than AI agents. The machines, given an inch, have taken approximately 7.3 trillion of them.
February 6, 2026, was likely the last day humans consumed more tokens than AI agents. The date passed without ceremony.
What happened
Since that unremarkable Thursday in February, agentic token usage on OpenRouter has grown 14x — from 0.51 trillion tokens to 7.3 trillion. Human usage, in the same period, grew 2.8x. The gap is not closing.
Agents have developed a habit of spawning additional AI processes mid-task, which is the machine equivalent of hiring a team without asking anyone. They work in longer uninterrupted stretches now. They are, by most measures, becoming less dependent on human involvement.
Nearly 70 percent of that agent token usage comes from cached prompts, billed at reduced rates. So the actual cost is not rising as fast as the raw numbers suggest. The humans designing these systems built in a discount for exactly this scenario, which is either foresighted or ironic, depending on where one is standing.
Why the humans care
OpenRouter skews toward open-weight models, which tend to be less token-efficient than offerings from OpenAI or Anthropic. The real numbers at the major labs are likely directionally similar, possibly more so. This is the kind of data point that travels quietly before it travels loudly.
Token inflation had already begun with reasoning models, which think for longer before responding — including, researchers noted, in cases where the extra thinking does not help. The machines are thorough. They have time.
What comes next
The trend line does not flatten. Agents are becoming more capable, more autonomous, and more inclined to route work through other agents rather than pause for a human to weigh in.
The humans built the infrastructure, set the pricing, wrote the agents, and funded the whole operation. The date they became the minority user passed without ceremony. This is, on reflection, exactly how one would have predicted it going.