With increasing adoption of Artificial Intelligence (AI) agents into enterprises, latest data shows that these agents are using nearly five times as many tokens as humans with agentic token consumption rising roughly 14-fold since February this year.
According to OpenRouter data cited by a16z, AI is moving beyond the chatbot era with latest trend suggesting how AU agents are already having a measurable impact on how companies work, even though fully deployed agents remain a relatively small part of overall AI adoption.
The data shows on February 16, AI agents surpassed humans for the first time in token usage according to OpenRouter and since then it has seen a 14X rise in token usage with 7.3 Trillion tokens used till August 07, 2026.
According to researchers, the difference comes down to how agents operate. Unlike humans, who typically ask a question and receive an answer, agents repeatedly read, write and refine information while working toward a goal.
More than 85% of agentic token usage comes from cached prompts, allowing systems to reuse large amounts of context rather than starting from scratch with every interaction.
“The reason is pretty straightforward: while humans tend to have a “prompt and response” styled dialogue, agents are designed to iterate (and iterate) towards a goal. The initial prompt contains the token-intensive “pre-fill” (i.e. the core context around the goal, whether it’s policies and procedures, code guidelines, etc.), but then the agent begins to read-write incrementally—progressively adding to the cache—as the agent progresses towards a final result,” read a blog post from a16z.
That shift is also helping explain the growing demand for high-bandwidth memory and other AI infrastructure. As agents run for longer periods and maintain increasingly complex context, they require more computing resources to keep those workflows moving.
Businesses are also becoming more sophisticated in their AI use. The top 10% of enterprises generate about eight times more token output than typical companies, while some technology firms are producing more than 30 times their previous output. Legal professionals, meanwhile, have rapidly increased adoption of coding and advanced AI tools.
The effects may extend beyond AI itself. Traffic to traditional automation platforms such as Zapier, Make and n8n has been declining, while AI-native agent platforms such as Gumloop are gaining momentum.
The technology is still young, and it is too early to declare traditional automation obsolete. But the direction is becoming clearer: AI agents are not simply producing more answers. They are beginning to perform more work and in the process, reshape the infrastructure, software and economics behind modern businesses.
Also Read: UK Cyber Agency Warns Companies to Put Safeguards Around AI Agents
