Bots and Agents: 61.8% of Web Traffic and 5 Times More Tokens on OpenRouter

Cloudflare measures that the majority of internet traffic already comes from bots and AI agents, with humans accounting for only 38.2% of the volume. In enterprises, a16z, citing OpenAI Signals, observes a surge in the use of Codex for legal tasks, increasing by 108 times between February 2026 and June 2026. On OpenRouter, agents now consume 5 times more tokens than humans after a 14-fold increase, a dynamic driven by iterative functioning and a cache that concentrates more than 85% of the tokens according to the platform.

Cloudflare: 61.8% of web traffic generated by bots

Cloudflare indicates that in June, traffic produced by bots and AI agents surpassed that of humans on the internet. At this point, humans represent only 38.2% of the traffic, compared to 61.8% for bots. This shift positions automated activity at the forefront of online flows.

In the enterprise, Codex is particularly advancing in law

According to a16z, which relies on OpenAI Signals, the use of Codex for legal tasks in enterprises exploded between February 2026 and June 2026, with the number of users increasing by 108 times. Since February, this trend has also affected other professions: sales and recruitment (41x each), marketing (26x), and healthcare (24x). Still according to a16z, the technology sector is not the fastest to adopt the most effective tools. Knowledge workers are becoming more sophisticated in many areas, with the movement being particularly pronounced in the legal field.

On OpenRouter, agents consume 5 times more tokens

OpenRouter, an interface that aggregates access to multiple AI models, is witnessing a rise in the power of agents. On this platform, token consumption by agents has surpassed that of humans, before being multiplied by 14. Currently, agents consume 5 times more tokens than human users.

The agents' cache concentrates more than 85% of the tokens

a16z emphasizes that humans prefer question-and-answer exchanges, while agents iterate towards a goal: an initial request that is highly contextual, followed by incremental reads and writes. This process is accompanied by a cache enriched over the course of execution. According to OpenRouter, more than 85% of the tokens consumed by agents come from this cache, a mechanism presented as the source of most of the relative increase in their consumption. However, this reading is specific to OpenRouter's data, which may not necessarily reflect public usage.