I use AI all the time. But there are only so many questions a person can ask a chatbot in a day.
Eventually, we have to eat and sleep. And we all have plenty of other things to do besides sitting at a computer typing prompts.
That puts a natural limit on AI demand.
But AI agents are starting to remove that limit.
As I showed you last week, an agent can spend hours researching a topic, writing software or completing a complicated assignment. That means it can make dozens or even hundreds of requests to an AI model along the way.
And unlike us, agents can work around the clock. What’s more, companies can also deploy thousands of them at the same time.
This week’s chart shows just how dramatically that’s changing AI demand.
Machines Are Becoming AI’s Biggest Users
This week’s chart comes from Andreessen Horowitz and OpenRouter, a platform that allows developers to access hundreds of different AI models through a single service.
It tracks token usage on OpenRouter based on where the requests originate.
The dark red line shows human activity, which includes people interacting directly with an AI model. The blue agentic activity line represents AI systems working through assignments with limited human involvement. And the mixed category includes traffic that combines elements of both.
Take a look.
Source: Andreessen Horowitz/OpenRouter
At the beginning of September 2025, nearly all the activity on OpenRouter came directly from humans.
Agentic usage barely registered.
But as you can see, that began to change early this year. On February 6, AI agents consumed more tokens than humans for the very first time.
And since then, the gap has exploded.
During the latest week shown on the chart, agents consumed roughly 7.3 trillion tokens, while humans only used about 1.4 trillion.
In other words, AI agents are now consuming more than 5X as many tokens as people on OpenRouter. And agentic token usage has increased roughly 14X since February.
To be clear, this chart doesn’t represent every AI request in the world. It only tracks activity routed through OpenRouter.
The company also can’t look inside every application to see whether a person or an agent generated each request. Instead, it classifies activity using several clues, including how often tools are called, the number of turns involved and how much time passes between requests.
So we shouldn’t assume that agents are consuming five times as many tokens across the entire AI industry.
But it’s clear that — even as human token usage continues to climb — agentic usage has broken away from it entirely.
And the reason comes down to how differently agents work.
When you ask a chatbot a question, it generally gives you an answer and waits for another prompt.
But an agent can keep going.
It can search the internet, read documents, call outside tools, check its work and try again when something goes wrong. It might make dozens of requests before a person sees the final result.
And it’s not limited to coding and software development.
Companies are building agents to help find new customers, answer support questions, review financial documents and take care of countless routine tasks.
And every time an agent goes to work, it consumes more tokens.
That means a company with 10,000 employees might someday have tens of thousands of agents working alongside them. All of them consuming AI around the clock.
And that could completely change how much computing power the world actually needs.
Here’s My Take
This week’s chart gives us a very different way to think about AI demand.
Until recently, it’s been measured by how many people use AI and how often they use it.
But a single person could soon have dozens of agents working on their behalf. And every one of those agents will need computing power.
That’s why the rise of AI agents could make the infrastructure boom far bigger than human demand alone ever could.
Regards,
Ian King
Chief Strategist, Banyan Hill Publishing
Editor’s Note: We’d love to hear from you!
If you want to share your thoughts or suggestions about the Daily Disruptor, or if there are any specific topics you’d like us to cover, just send an email to dailydisruptor@banyanhill.com.
Don’t worry, we won’t reveal your full name in the event we publish a response. So feel free to comment away!