Futurum Group CEO Daniel Newman posted on X that AI agents are now using five times more tokens than people. The numbers come from an Andreessen Horowitz (a16z) chart that tracks OpenRouter usage. As of August, agents have processed 7.3 trillion tokens while humans have used 1.4 trillion. Newman says the gap could grow to ten times or more.
Why most of the traffic is recycled
a16z notes that over 85 % of the tokens used by agents are from cached prompts. That means the agents are largely re‑reading text they have already seen rather than generating new content. The cache helps speed up responses but also inflates the token count.
Why it matters
For everyday users, the surge in agent token usage shows that AI tools are being called on repeatedly for the same information, which can make them faster but also hides how much new content they actually create. It also suggests that costs tied to token counts may grow faster than user‑generated queries, affecting pricing for services that charge per token.
