AI companies aren’t pouring billions into the world’s biggest data centers just so people can ask chatbots for recipes — they’re building for “agents.” As Wired’s Molly Taft explains, an agent is an AI system designed to make autonomous decisions: rather than waiting on a human to prompt each step, it decides on its own how to break one request into hundreds of smaller sub-prompts and carries them out to complete a task.
Maxwell Zeff, who writes Wired’s Model Behavior newsletter, offers an example: asked to build a website, an agent “might run for hours… re-prompting itself dozens of times” to generate pages, menus, and datasets — work a human would otherwise direct step by step.
The scale can be extreme: OpenAI recently said a swarm of over 10,000 agents exchanging 2.7 million messages cracked a longstanding math problem — a claim mathematicians have disputed — burning an estimated tens of millions of dollars’ worth of electricity, even as companies frame their footprint around single queries.
Agents break that framing. “In other technological growth areas, we’re constrained by how many people are driving a car or streaming Netflix,” says Boris Gamazaychikov, co-founder and CEO of Sustainable AI. “Now, this stuff is kind of decoupled from users… They’re talking about unicorns that have one employee.”
Related on the SLP
With little hard data from companies, climate scientist Zeke Hausfather estimated his own usage in a recent blog post, calculating his agent-heavy Claude sessions burn roughly as much energy as two refrigerators — a “net new source of emissions” as climate targets slip out of reach.
That footprint could soon be widespread: Meta just launched Muse, a personal agent built, the company says, “to work for billions of people worldwide,” running a dedicated cloud computer for each user around the clock.
