Agents now burn more tokens than human users of AI

According to data from OpenRouter (and chartmaker Peter Walker), agents are using nearly 5x as many tokens as humans, and agent usage has grown ~14X since just February.

Not only do agents use far more tokens than people, they use tokens very differently. Agents do far more work with cached tokens than people do – a16z report

Just a while ago, I was having to debate a supposed new reality where people now have multiple open tabs of tasks to complete due to AI enabling the mass exploration of various things at a fraction of the cost.

The prevalent argument was that there's growing illusion of productivity whereas people are just doing more rather than achieving more.

I mean, I can see how AI would cause people to chase multiple ideas and how that might lead to being lost in hills of multiple tasks that completing any becomes difficult.

That said, I've personally held to the conclusion that this is a systems problem. Before AI, people chased what they could afford to chase, the system design here was constrained by "cost" but the second that friction was out of the picture, the flaw of those systems revealed itself.

Execution on the right things slowed because options expanded leading to derailed focus.

The system of work has been flawed, the effect of AI removing the cost barrier reveals this.

So it isn't that AI cannot bring about productivity, but that humans lack the work environment design that can accommodate AI effectively for growth.

Agents and the new age

We've been talking about agents for a good while now and the consensus has been, for this time period, that the future belongs to AI agents.

Anything you can think of will eventually have some form of agents running the front line.

Online purchases, this has been a leading focus because it's made the most sense that commerce will be the easiest to automate with AI and the most profitable option to execute early on.

That said, it comes as no surprise that agents now use up more tokens than humans when it comes to using AI.

It doesn't really matter what these agents are doing, be that coding, scraping the web, running hacks perhaps, handling customers for businesses or managing business accounts, whatever it is, doesn't really matter, what matters is that it is happening and the reason it is happening is that the human-in-the-loop of AI operations was never going to remain in the near-tech level.

Agents were always going to run most of the operations that required using the best parts of the technology.

The human role was inevitably going to be something way less demanding of using tokens.

The rise of compute demand is an indication of where things are headed. It isn't human usage that's causing this demand, but agents.

The future is agentic.

Posted Using INLEO