AI agents collectively attract only about 10 million weekly users, a tiny fraction compared to the billion monthly active users that chatbots like ChatGPT and Gemini command, according to a viral industry debate and commentary published in Wired this week. The stark gap has sparked concern among tech insiders that Silicon Valley has spent billions building advanced agent technology without creating products ordinary people actually want to use. Despite widespread excitement in the tech industry about agents automating work and transforming daily life, the general public has shown little interest.
OpenAI reported last month that its Codex and ChatGPT Work agents together draw roughly 10 million weekly users, while sources familiar with Anthropic indicated that its Claude Code and Cowork agents see comparable uptake. When measured against chatbots like ChatGPT and Gemini—each averaging around a billion monthly active users—agents represent what amounts to a rounding error. Most consumers currently rely on generative AI for straightforward tasks such as searching for information and casual conversation, leaving the more sophisticated capabilities that Silicon Valley has invested billions to develop largely untapped.
Josh Miller, CEO of The Browser Company, sparked the conversation with a post on X asking why AI agents haven't caught on despite the technology being theoretically ready to transform work and life. "I just have not heard a single person outside of the tech community talk about an agent that they use," Miller said in a subsequent interview, standing by his claim that the industry needs to build agent products people genuinely want. He argued that AI agents aren't a stand-alone product but rather a technology, saying "No one wants AI agents, because AI agents aren't a thing." Miller noted that his company's most popular feature—a personalized morning briefing in its AI-powered browser, Dia—technically runs on an agent, but users don't need to know that.
The disconnect matters because AI labs face a fundamental problem: agents were supposed to justify the massive investment in training advanced models capable of far more than simple chat. Without mainstream adoption, those billions remain unrealized potential. Industry insiders have expressed broader worries that despite pouring considerable resources into building AI agents, no one has yet created a killer consumer product—leaving the question of whether agents will ever have their own "ChatGPT moment" unanswered. Miller's experience with The Browser Company illustrates the core issue: the company's personalized homepage—greeting users with a to-do list pulled from calendar and email, plus small joyful touches like artwork—succeeds precisely because it hides the agent technology underneath, focusing instead on making people feel calm and focused when they open their laptop.
The path forward, according to Miller's framework, requires the industry to stop selling "AI agents" as a concept and start building products that solve real problems people care about, with the sophisticated technology working invisibly in the background. He dismissed the importance of users understanding technical details like "harness" systems that call up tools, arguing that what matters is the product experience itself. The takeaway is blunt: billions spent on frontier AI capabilities mean nothing if the industry can't translate them into something ordinary people find useful enough to adopt. Companies racing to perfect agent orchestration may discover their competitive edge lies not in the sophistication of the underlying models, but in their ability to disguise complexity as simplicity—burying advanced automation beneath interfaces so intuitive that users never realize they're interacting with an agent at all.