The Best AI Agents Are the Ones You Don't Notice
2 September 2026
I read Maxwell Zeff’s piece in Wired yesterday on why normal people aren’t using AI agents.
ChatGPT and Gemini sit around a billion monthly active users each as plain chatbots, while actual agent products such as Codex, ChatGPT’s Work agent, Claude Code and Cowork are pulling something like 10 million weekly users between them.
Against that base, agents are basically a rounding error. The figures are approximate, and monthly chatbot users aren’t directly comparable with weekly agent users, but the adoption gap is striking.
The article quotes Josh Miller, who runs The Browser Company, the makers of the Dia browser.
He says something provocative: “AI agents” isn’t really a product category at all. It’s an invented frame the industry made up to describe a bundle of underlying capabilities.
His point was blunt: the technology is ready for agents to transform how we work and live, but the general public doesn’t care.
He also mentioned that nearly every AI lab leader he’s talked to cites the movie Her as their north star for what an agent should be. That seems like the wrong framing to him.
Build for the customer, not the demonstration
The core argument is that tech companies keep shipping the most impressive thing their models can technically do instead of a product built around what a customer actually wants.
Dia browser’s own most popular feature makes the point. It’s a personalised morning briefing, a to-do list and some curated content meant to create a small daily moment of usefulness.
Most people using it have no idea it runs on agents.
They don’t need to know.
The best technology has always worked this way. It solves a real problem so well that you stop noticing the technology at all.
Nobody thinks about the engine when a car starts. Nobody thinks about the algorithm when a search result is right.
The agents that win won’t be the ones people talk about. They’ll be the ones people forget are even there.
Reference
