AI Agents Are Silicon Valley's Biggest Bet. Almost Nobody Outside Tech Is Using Them.
Ten million weekly users sounds impressive until you compare it to the billion people using basic AI chat. A viral post from a veteran tech founder has sparked an uncomfortable conversation about why AI agents have failed to reach ordinary people.

Key points
- OpenAI reported roughly 10 million weekly users across its Codex and ChatGPT Work agents as of last month, compared to around one billion monthly active users for ChatGPT as a basic chatbot.
- Josh Miller, CEO of The Browser Company, whose AI browser Atlassian acquired in 2024 for $610 million, posted that the general public "doesn't give a fuck" about AI agents.
- Miller argues AI agents are a technical category, not a product, and that companies are shipping impressive demos instead of tools people actually want.
- The Browser Company's most-used feature is a personalised morning briefing that runs on agent technology, but users never see that label.
An AI agent is software that can carry out multi-step tasks on its own: book a meeting, search the web, write and run code, all without you clicking through each step. Silicon Valley has spent billions building them. Most people have never tried one.
That gap is now embarrassing enough that tech founders are saying so in public.
"Hot take, isn't it kinda crazy that nobody is really using AI Agents," Josh Miller, CEO of The Browser Company, wrote in a widely shared post this week. Miller has earned the right to be blunt. His browser Arc built a devoted following before Atlassian bought The Browser Company for $610 million last year. Before that, Facebook acquired his startup Branch in 2014. He also served as the White House's first director of product under President Obama.
He told Wired AI he fired the post off half-asleep during a family holiday in Europe. He stands by every word.
Why aren't ordinary people using AI agents?
Because no one has built a product that gives them a clear reason to. Miller's core argument is sharp: AI agents are a technical description of how software works inside, not something a real person would go looking for.
"No one wants AI agents, because AI agents aren't a thing," Miller said. "It is an invented frame made up by our industry."
His point lands harder when you look at the numbers. OpenAI says its Codex and ChatGPT Work agents together attract about 10 million weekly users. People familiar with Anthropic, the company behind the Claude family of AI models, say its Claude Code and Cowork agents are seeing similar figures. ChatGPT and Google Gemini, the simpler chat versions of those same products, each count roughly one billion monthly active users. Agents are a rounding error by comparison.
Most people use AI today for simple things: looking something up, drafting an email, summarising a document. Labs have spent billions training models that can do far more, and they need consumers to actually adopt agents to justify that investment.
What does a good agent product actually look like?
Like something people want to open, not a demo of what software can theoretically do.
The Browser Company's own data makes Miller's case. Its most popular feature ever is a personalised morning briefing inside its AI browser, Dia. Open your laptop and you get a greeting, a to-do list pulled from your calendar and inbox, and a small piece of art chosen to spark a moment of calm. Under the hood, an AI agent assembles all of it. The user never needs to know that.
"Who gives a shit if it looks like an AI agent?" Miller said. "No one uses AI agents."
He also identified a cultural problem inside the labs themselves. When The Browser Company was exploring acquisition offers last summer, Miller met leaders across the top AI companies. Almost every one of them cited the 2013 film Her, in which a man falls in love with an AI voice assistant, as their product vision.
"I think there is sort of a lack of diversity of opinions and convictions," he said.
The latest wave of agent products lets people build small personal software tools: pitch decks, organised spreadsheets, automated workflows. Useful for a certain kind of power user. Not a mass-market product.
What should ordinary people take from this?
If you have tried an AI agent and found it confusing or underwhelming, that is the industry's problem, not yours. The tools are genuinely early, genuinely demo-flavoured, and mostly aimed at developers and productivity enthusiasts.
The practical advice: if you use ChatGPT or Google Gemini for everyday tasks, keep doing that. The agent versions will improve, and the best ones, if Miller is right, will eventually stop calling themselves agents at all.



