Josh Miller, CEO of The Browser Company, said this week that AI agents have almost no traction outside the tech industry, a claim backed by the raw usage numbers. OpenAI disclosed last month that its Codex and ChatGPT Work agents together reach about 10 million weekly users. ChatGPT and Gemini each average roughly 1 billion monthly active users. Agent adoption, on that scale, is a rounding error.
Miller made the argument in a viral X post fired off during a family trip to Europe, then defended it in an interview on August 6, 2026. His frame is that the industry has convinced itself agents are the future while the general public has never touched one and does not care.
The gap matters because the AI labs have spent billions training models capable of far more than chat. Agents are the commercial thesis for that spend — the mechanism by which reasoning models and tool use turn into revenue that justifies the training runs. If consumers do not adopt them, the unit economics of frontier training get harder to defend.
Key facts
- 01OpenAI's Codex and ChatGPT Work agents draw roughly 10 million weekly users combined, per OpenAI figures shared last month.
- 02ChatGPT and Gemini each average around 1 billion monthly active users — orders of magnitude larger than agent adoption.
- 03Anthropic's Claude Code and Claude Cowork are seeing adoption similar to OpenAI's agents, per people close to the company.
- 04Josh Miller sold Arc-maker The Browser Company to Atlassian last year for $610 million; his prior startup Branch was bought by Facebook in 2014.
- 05Miller says every top AI lab but one cited the 2013 Spike Jonze film Her as its product vision during acquisition talks last summer.
People close to Anthropic told Wired that Claude Code and Claude Cowork are seeing adoption in a similar range to OpenAI's agents. That means the two leading labs, combined, are looking at a user base roughly two orders of magnitude below their chatbots. No agent has had a ChatGPT moment.
Miller's diagnosis is that "AI agent" is a category label invented by the industry, not a product consumers ask for. He points to his own company's most-used feature in the Dia browser: a personalized morning briefing with a greeting, a to-do list pulled from calendar and email, and a piece of art. The feature is powered by an agent under the hood. Users, he argues, neither know nor need to know that.
His track record gives the critique weight. Miller's previous product, the Arc browser, was acquired by Atlassian last year for $610 million. His first company, Branch, was bought by Facebook in 2014. He also served as the White House's first director of product under President Obama.
Miller blames groupthink at the frontier labs for the flat consumer response. He said that during acquisition conversations last summer, nearly every lab he met with cited the 2013 Spike Jonze film Her as its guiding product vision. He said one lab out of the group he met with did not.
“It was wild to me. I think every single lab but one mentioned the movie Her as the way they articulated their vision.”— Josh Miller, CEO of The Browser Company
The specific complaint is that companies are shipping capabilities — an agent that can navigate a website, an agent that can write code — rather than products designed around what a non-technical user actually wants to do when they open a laptop. Those capabilities read as demos, not finished products. The framing echoes a recurring theme in recent AI coverage, from Google Maps turning its Ask Maps feature into an ordering and booking agent to Naïve raising $28.5 million to let agents run entire companies. Both are capability-forward, not consumer-frame.
There is a counterweight worth naming. Reporter Maxwell Zeff, who conducted the interview, said he personally finds ChatGPT Work and Claude Cowork more useful than their predecessors for tasks like building pitch decks and organizing datasets. The pool of people who want to build their own productivity tools is real, just small. And Miller's argument is convenient for someone selling an AI-powered browser that hides its agent plumbing.
Still, the numbers are the numbers. Ten million weekly agent users against a billion monthly chatbot users is not a distribution problem the labs can wave away with a better prompt library. It is a product problem — the kind that gets solved by naming a job the user already wants done, not by shipping a more capable harness. Whichever company builds the first agent-powered product that a non-technical user opens by name and reaches for daily will define the next phase of consumer AI. So far, no one has.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
