Silicon Valley Loves AI Agents, but Most People Still Don’t Use Them

AI agents may be Silicon Valley’s next big bet, but mainstream users have not embraced them. Despite intense investment from OpenAI, Anthropic and other labs, the public still uses chatbots far more than agent-style tools, forcing the industry to confront a basic problem: the technology may be impressive, but it has not yet become a must-have product for everyday people.

The disconnect is now becoming hard to ignore. In tech circles, AI agents are being used to automate work, navigate websites, write code and even build personal software. Outside that bubble, however, most people have little reason to seek them out. That gap is fueling a growing debate over whether “agents” are the right product category at all, or merely an internal technical capability in search of a consumer-friendly form.

Josh Miller, chief executive of The Browser Company, pushed that argument into the open this week with a post on X that went viral inside the tech world. His central point was blunt: the industry keeps acting as if AI agents are already a mass-market phenomenon, while ordinary users largely do not care.

Miller argued that the AI industry may be overestimating public demand for agents, saying the real challenge is not building more capable systems, but creating products people genuinely want to use.

That view is resonating because it reflects a broader anxiety across the sector. The biggest AI labs have spent billions teaching models to do far more than answer questions. Yet consumers continue to gravitate toward the simplest version of the technology: chat. If agents are the business model many companies are counting on, their adoption remains far below the level needed to justify the hype.

Why AI agents have not gone mainstream yet

AI agents have not gone mainstream because most consumers have not found a compelling reason to adopt them. For the average person, the value of a chatbot is obvious: ask a question, get an answer. The value of an agent, by contrast, is often abstract, hidden behind technical jargon and unclear workflows.

That distinction matters. Chatbots are easy to understand because they mimic conversation. Agents, however, are usually marketed as systems that can plan, call tools and complete tasks on a user’s behalf. In practice, that often means a product is capable of impressive behind-the-scenes actions, but the user experience still feels experimental.

Industry leaders are hoping agents become the next major interface for AI, but that transition has not yet happened. The public has already adopted generative AI in large numbers for search, writing help and casual conversation. But when it comes to more advanced automation, adoption narrows quickly.

One reason is that many current tools are built around capability demonstrations rather than everyday habits. They can do things that look striking in a demo, but those functions do not always solve a common consumer pain point. In consumer technology, novelty can attract attention; usefulness builds habits.

How many people are actually using AI agents?

Only a relatively small number of people are using AI agents regularly compared with chatbot services. OpenAI recently said its Codex and ChatGPT Work agents together have roughly 10 million weekly users. Anthropic, according to people familiar with the company, is seeing similar adoption for its Claude Code and Cowork products.

Those figures are notable, but they still look modest next to the scale of the most popular chatbots. ChatGPT and Google’s Gemini each have around a billion monthly active users on average, according to widely cited industry estimates. Against that backdrop, agent usage is tiny.

That difference highlights the current shape of the AI market:

• Chatbots are becoming a mass-market habit.
• Agents remain a niche tool for technical or highly motivated users.
• Consumer demand is concentrated in simple, immediate use cases.
• Enterprise experiments are moving faster than everyday consumer adoption.

What Josh Miller says the industry is getting wrong

Miller’s argument is that the industry is focusing too much on the label “AI agents” and not enough on the actual experience people want. In his view, the market does not need more abstract talk about autonomous systems. It needs products that make people feel organized, calm and productive the moment they open their laptop.

Miller says the term “AI agent” is itself part of the problem. To him, it is a framework invented by the industry to describe a broad set of technical capabilities. Consumers, he suggests, do not care about the architecture underneath the experience. They care whether a tool is useful, intuitive and pleasant to use.

Miller’s position is that users should not have to think about whether a product is powered by an agent at all; what matters is whether it helps them get started, stay focused and finish work more easily.

That perspective is informed by his own product work. The Browser Company, which Miller leads, built Arc, a browser that developed a devoted following before being acquired by Atlassian in 2025 for $610 million. His first startup, Branch, was acquired by Facebook in 2014. Before entering startups, he served as the first director of product at the White House under President Barack Obama.

Those credentials give weight to his critique. Miller is not arguing from the sidelines; he is speaking as someone who has helped build products that found an audience by rethinking a familiar category. He believes the same principle now applies to AI.

How The Browser Company is using AI differently

The Browser Company’s experience suggests that users respond better to helpful features than to the idea of an agent. Miller says the most popular feature in the company’s AI-powered browser, Dia, is a personalized morning briefing that appears when users open their laptops.

That briefing includes a greeting, a to-do list assembled from calendar and email data, and small curated touches meant to make the experience more engaging. The important part, Miller argues, is not that an agent helped create the experience. It is that the result feels immediately useful.

In other words, the technology is invisible and the benefit is obvious. That may be the clearest clue yet to how AI products will reach ordinary users: not through loud branding around autonomy, but through quietly helpful functionality embedded into tools people already use.

The browser example also underscores a larger product lesson. People are often willing to use AI when it is folded into a familiar workflow. They are less likely to experiment with a standalone “agent” whose value proposition is uncertain.

Why the AI industry keeps leaning on demos

The AI industry keeps leaning on demos because the most advanced model capabilities are easier to showcase than to productize. It is simpler to impress an audience by having a model navigate a website or write code than it is to design a polished consumer service around those abilities.

That gap between demonstration and product is now one of the central tensions in AI. Labs have built systems that can do extraordinary things in controlled settings. But turning those abilities into something ordinary people return to every day is a different challenge altogether.

Miller sees groupthink as part of the issue. He says many people building these systems share a similar science-fiction vision of the future, which narrows the range of product ideas being explored.

He points to a recurring cultural reference inside the industry: the movie Her. According to Miller, multiple AI lab leaders have invoked the 2013 Spike Jonze film as shorthand for their product ambition. That, he argues, reveals a lack of diversity in how the sector imagines the future of AI interfaces.

Miller says the repeated use of Her as a reference point shows how concentrated the industry’s thinking has become, with many teams chasing a similar aesthetic and interaction model.

The result, he suggests, is a narrow product pipeline. Instead of exploring many distinct ways AI could improve daily life, companies keep returning to the same cinematic idea of a conversational, emotionally resonant digital companion.

What makes an AI product stick with consumers?

An AI product sticks when it solves a recurring problem in a simple, obvious way. That has always been true in consumer software, and AI is unlikely to be an exception.

For now, the strongest consumer AI products tend to do one of three things well:

1. Save time on a common task.
2. Reduce friction in a familiar workflow.
3. Produce an outcome that feels immediately better than the alternative.

Agents can theoretically do all three. But the category often asks users to imagine future savings rather than experience present value. That is a harder sell, especially when the user must first learn a new interface or trust a system to take action on their behalf.

There is also a trust problem. The more autonomy a tool has, the more questions users ask about reliability, accuracy, permissions and control. For enterprise teams or technically savvy users, that may be an acceptable tradeoff. For consumers, it can be enough to stop them from trying the product at all.

That is why many AI companies are now experimenting with what are effectively agentic features rather than full agent products. The strategy is to hide complexity, surface the result and let the user experience the benefit without having to understand the machinery.

How big is the gap between agents and chatbots?

The gap is enormous, both in adoption and in public awareness. Chatbots are now familiar to hundreds of millions of people. AI agents, by contrast, are still a concept many users have never encountered directly.

The numbers tell part of the story. A combined 10 million weekly users for OpenAI’s agent offerings is meaningful, but it is still a fraction of the audience for the company’s chatbot products. The same is true for Anthropic. Even with strong engagement among power users, agents have not yet achieved cultural visibility on the scale of chat.

The difference is also qualitative. Chatbots are general-purpose tools with a single clear entry point: type a prompt. Agents are usually framed around goals, task completion or workflow automation, which demands more context from the user and often more setup.

That extra complexity may not matter to builders and early adopters. But it matters a great deal to mainstream users, who generally prefer products that feel self-explanatory.

Why this matters for OpenAI, Anthropic and the rest of the field

This matters because agents are not just a product experiment; they are part of the industry’s future business model. AI labs have trained models that can do much more than answer questions, and they need consumers and companies to pay for those abilities.

If users continue treating AI mostly as a chatbot, the commercial upside of more advanced systems may be harder to unlock. That does not mean agents will fail. It means their path to mass adoption may look very different from the one many founders imagined.

For OpenAI and Anthropic, the challenge is not limited to shipping capable systems. It is about translating technical progress into user behavior. That usually requires:

• clearer use cases,
• more intuitive interfaces,
• less jargon,
• and stronger reasons to return daily.

The good news for the industry is that early adoption often underestimates future demand. Many technologies spend years as niche tools before breaking out. The bad news is that the current “agent” pitch may be failing for reasons that are not just about maturity, but about framing.

What happens next?

The next phase is likely to be a shift from standalone agents to embedded AI features that quietly automate useful work. That could mean browsers, writing tools, productivity suites and operating systems absorbing agent-like capabilities without advertising them as agents.

That approach aligns with Miller’s thesis. If users do not care about the label, then product teams should stop building for the label and start building for the feeling the product creates. The winning AI experience may not be one that announces itself as an agent at all.

The bigger lesson for the industry is that frontier capability does not automatically translate into consumer demand. Silicon Valley can become obsessed with a technical category before ordinary people even notice it exists. But for a technology to cross from buzzword to habit, it has to solve a real problem in a way that feels natural.

For now, AI agents are still waiting for their breakout moment. The question is whether that moment will come from better technology, better product design, or a complete rethinking of how the industry talks about the future.

As Miller sees it, the answer probably starts with dropping the hype and building something people actually enjoy using.

Bottom line

AI agents remain a major strategic priority for the tech industry, but most consumers still do not use them in any meaningful way. That is the tension now defining the category. The technology may be powerful, but until companies make it approachable, useful and emotionally resonant, the public is likely to keep choosing chat over agents.