AI agents flop with everyday users despite Silicon Valley hype

Josh Miller, CEO of The Browser Company, triggered a wave of recognition this week when he posted on X that "nobody is really using AI Agents" outside the tech community, despite the industry's conviction that agents will "totally transform how we work and live our lives." His candid observation—that "the general public dgaf" (doesn't give a fuck)—resonated widely among Silicon Valley insiders who have been perplexed by the public's indifference to their latest obsession. Miller told an interviewer that he fired off the post while half asleep on a family trip to Europe but stands by the argument. The core frustration is that, after years of development and billions in investment, agents remain marginal.

The numbers back up his claim starkly. OpenAI disclosed that its Codex and ChatGPT Work agents collectively have about 10 million weekly users. People close to Anthropic report that its Claude Code and Cowork agents are seeing similar levels of adoption. By contrast, ChatGPT and Gemini each average around a billion monthly active users. Agents, in other words, represent a rounding error—a tiny fraction of the user base for traditional chatbots. This is a problem for the AI labs because most consumers are still using generative AI for simple tasks like looking up information and chatting, while Silicon Valley has trained models capable of far more. Agents were supposed to unlock that capability and justify the investment, but adoption has flatlined.

Miller's deeper argument is not simply that agents have failed to catch on—it's that the industry has misunderstood what it is selling. "No one wants AI agents, because AI agents aren't a thing," he says. "It is an invented frame made up by our industry to collectively refer to something." The problem is framing: the tech industry is marketing a technology category rather than a solution to a real human problem. He argues the path forward is to build products that solve concrete needs—making someone calm, focused, and productive—without even naming the agent machinery inside.

He offers a concrete example from his own work. The Browser Company's most popular feature is a personalized morning briefing in its AI-powered browser, Dia. When users open their laptop, they see a homepage with a greeting, a daily to-do list populated automatically from their calendar and email, and random tidbits designed to spark joy—such as a piece of art. Technically, Miller says, this feature was only made possible by an AI agent. But the user does not need to know that, and arguably should not. The product succeeds because it solves a specific problem (starting the day organized and inspired), not because it advertises itself as cutting-edge agent technology. Miller's career lends weight to his argument: his browser Arc developed a cult following and was acquired by Atlassian for $610 million; his earlier startup Branch was bought by Facebook; and he served as the White House's first director of product under President Obama. His point—that successful AI products will embed agents invisibly rather than brand them explicitly—offers a possible road map for an industry facing an adoption crisis.