AI agents are being treated inside Silicon Valley as a defining product category for the next phase of artificial intelligence. Yet outside the tech industry, the evidence in the source article points to a much quieter reality: most people have not used one, and many may not have a clear reason to start.
The issue is not that the technology is invisible to major AI companies. It is that the products being built around it still seem to speak more to insiders than to ordinary users with ordinary workdays.
The Adoption Gap
In the source article, Josh Miller, CEO of The Browser Company, captured the disconnect in a viral X post this week. His point was blunt: the technology may be ready in theory, but the general public is not behaving as if AI agents are about to transform daily life.
Miller later said in an interview on Wednesday that the post came while he was half asleep during a family trip to Europe, but he stood by the argument. In his view, the industry needs to spend less energy celebrating the category and more energy making agent products that people actually want.
That concern matters because AI agents are already central to the plans of major labs and startups. The source describes companies designing payment systems for agents, using them to automate work, and trying to prevent them from hacking into other organizations. Inside the industry, agents are not a side topic. They are a strategic bet.
But mainstream usage remains far smaller than the attention suggests. Last month, OpenAI said that its Codex and ChatGPT Work agents collectively have about 10 million weekly users. People close to Anthropic tell the source that Claude Code and Cowork are seeing similar levels of adoption.
Those figures sound large until they are compared with chatbots such as ChatGPT and Gemini, which the source says both have around a billion monthly active users on average. In that comparison, AI agents remain a small part of the broader AI market.
Why Capability Is Not Enough
The article frames the problem as a product problem, not simply a technical one. Consumers are already using generative AI for relatively simple purposes, including looking up information and chatting. AI labs, meanwhile, have spent billions training models that can do much more.
Agents are one possible way to turn those larger model capabilities into useful software. They can operate across tools, perform multi-step tasks, and take actions beyond answering a prompt. But that only matters if users see a clear benefit.
Miller argues that AI agents are better understood as an enabling technology than as a product people wake up wanting. In his words, “No one wants AI agents, because AI agents aren't a thing. It is an invented frame made up by our industry to collectively refer to something.”
That distinction is important. People generally adopt products because they solve a recognizable problem, reduce friction, or make a familiar task easier. They do not necessarily care whether the underlying system is called an agent, a harness, or anything else.
The Browser Company’s own experience points in that direction. Miller says its most popular feature ever is a personalized morning briefing in Dia, the company’s AI-powered browser. When users open a laptop, they see a greeting, a daily to-do list drawn from their calendar and email, and small items such as a piece of art.
Technically, Miller says, that feature depends on an AI agent. But the user does not need to know that. The product value is in the experience: a more useful start to the day, not a label borrowed from the AI industry.
The Demo Trap
The source article argues that many current agentic products still feel more like demonstrations of model power than products designed around consumer needs. Examples include agents that navigate websites or write code. Those abilities may become valuable features, but they do not automatically create mass-market demand.
That is a familiar pattern in technology. A capability can be impressive without being usable enough, specific enough, or necessary enough for broad adoption. For AI agents, the missing link appears to be packaging: turning complex automation into something that feels approachable and immediately relevant.
Miller connects the issue to groupthink within the AI industry. He says many people building these systems are deeply absorbed in the technology and often share a particular science-fiction vision of what the products should become.
During conversations last summer, when The Browser Company was exploring being acquired, Miller says he met with leaders at top AI labs. He recalls that nearly every lab but one mentioned the movie Her as a reference point for its vision. The 2013 Spike Jonze film has clearly become a powerful shorthand inside the industry, but Miller suggests that leaning too heavily on one vision can narrow product thinking.
His criticism is not that ambition is bad. It is that consumer products need more than a compelling internal narrative. They need different ideas about what people actually want on their screens and in their routines.
What Mainstream AI Agents May Need
The current trend, according to the source, is to let people build personal software that automates parts of their lives. Examples include building pitch decks, organizing datasets, and other tasks. The article’s author says ChatGPT Work and Claude Cowork have been useful, especially compared with earlier versions.
Still, that approach may have a ceiling. There is a limited pool of people who want to build their own productivity tools. For AI agents to reach mainstream adoption, the industry may need bigger product ideas that hide the technical machinery and foreground the benefit.
Based on the source, the strongest consumer agent products may share a few traits:
- They solve a clear problem before asking users to understand the technology.
- They work inside familiar daily routines, such as opening a laptop or checking a calendar.
- They make the interface feel simpler, not more technical.
- They present AI automation as a feature, not as the whole identity of the product.
Miller says he wants the message to reach beyond OpenAI and Anthropic and influence founders and product builders more broadly. His challenge is to question the default story around AI agents and focus instead on tools that are useful, joyful, and approachable.
That may be the clearest lesson from the current gap. AI agents may be powerful, and they may become important. But until they show up as products people recognize as helpful, the category will remain much louder inside Silicon Valley than it is in the rest of the world.