News

This AI Chatbot Is Really Just a Bunch of People Typing

• By John K. Waters
• 08/19/2026

ChatTJB was supposed to be a joke about artificial intelligence. Then it ran into a very familiar technology problem: too many users.

The site looks like a conventional AI chatbot, but its original architecture was considerably simpler. A user typed a question. Tucker Bryant, a 32-year-old conceptual artist and former Google employee, read it, thought about it, and typed back an answer.

No large language model generated the response. On ChatTJB, AI stood for "Average Individual."

Bryant describes the site as a piece of satire and public art. Its About page dresses up the premise in the language of the AI industry, describing the service as a "single-operator large language experience" built on a "human-grounded inference layer." It then translates that description into ordinary English: a person named Tucker reads each question and writes back when he is awake and motivated.

For a while, that was literally how ChatTJB worked.

Then people started using it.

Bryant told the San Francisco Standard in early August that ChatTJB had received about 15,000 prompts from roughly 11,000 people over the course of a week. At peak traffic, he said, the site received about 5,000 prompts in an hour. He was sometimes spending as many as 10 hours a day answering them.

The result is an unexpectedly literal experiment in what happens when a human-powered chatbot has to scale.

The Billboard That Broke the Chatbot
ChatTJB began attracting wider attention after Bryant bought a billboard in San Francisco advertising it as "the leading chat interface powered by AI."

Bryant told the San Francisco Standard that the billboard cost about $6,000 for a month. The location at Sixth and Folsom streets was partially obscured by a tree, and Bryant himself described the sightlines as poor. That was part of the joke.

The joke fit comfortably into San Francisco's increasingly elaborate AI marketing culture, where cryptic billboards have become a recurring feature of the city's technology boom.
But ChatTJB had a different target.

Bryant told the Standard that he viewed the project as commentary on what he calls "cognitive surrender," the willingness to hand over more thinking to AI systems without necessarily questioning their answers. He also described the project as a spoof of the city's AI hype cycle.

ChatTJB's own design reinforces the point. Its About page claims that incorrect answers are a feature because they force users to decide whether they should trust what they are reading. It jokes that latency is evidence that somebody is actually considering the user's question.

Bryant did not pretend his answers were particularly authoritative, either.

He told the Standard that the responses were "very, very mid. Objectively mid."
That limitation was part of the premise. ChatTJB offered the familiar shell of an AI product while making the human fallibility underneath it impossible to ignore.

Then the Human Could Not Scale
The project's growing popularity eventually overwhelmed the premise.

ChatTJB's homepage now says Bryant can no longer conceivably keep up with the prompts and is turning the service into a community project. When the main chat reopens, users may be connected to approved human volunteers, whom the project also calls AIs (Average Individuals).

The site says one volunteer will remain with a conversation rather than passing users among multiple people. A small number of volunteers have already been testing the system by working through the backlog.

That shift creates an amusing inversion of the standard AI scaling problem.

Technology companies generally try to replace expensive human labor with software that can serve millions of people at once. ChatTJB started with a single person and became popular enough to require a distributed human workforce.

The transition also means the project has had to take the implications of strangers using a chat interface for more than jokes more seriously.

Bryant wrote on the site that some users have been sending earnest messages and staying for longer conversations. His earlier interview with the Standard included examples of people asking about personal situations, including one user who said they were on the first night of their honeymoon and did not feel relaxed.

Once volunteers are involved, those conversations become more complicated.

A Joke Develops a Trust and Safety Department
ChatTJB now says it has created an onboarding quiz for volunteers, a message triage system, crisis-response resources, and other safeguards. Bryant says volunteers will be added slowly because he wants the service to be safe and is prepared to revert the project or shut it down if the community model does not work.

The site's volunteer page similarly says that Bryant wants others to help answer questions, while reminding users that "sometimes a chat with an average human being really is all you need."

Its privacy rules have become more explicit as well.

ChatTJB's privacy notice, last updated Aug. 3, warns users that their messages are read and answered by human volunteers, not an AI model. It specifically advises people not to submit information they would not want another human being to read.

The policy states that ChatTJB collects conversation content, session identifiers, certain visit metadata, and email addresses when users provide them. For volunteers, it also collects information including responses written to users, reports, and activity data. The company says the information is used in part for abuse prevention, moderation, and crisis flagging.

Those are ordinary operational details for a digital platform. They are also a surprisingly serious endpoint for a project built around the idea of one tired person manually pretending to be a chatbot.

There Is Even a Pro Tier
ChatTJB has another familiar piece of startup infrastructure: a subscription.

According to the site's terms of service, ChatTJB Pro costs $5 a month.

Bryant does not exactly push the upsell.

On the site's homepage, he says the Pro tier was originally created as part of the joke, actively discourages people from subscribing, and says it mainly exists to give his fingers a break and help cover the billboard expense. He also warns that paying does not guarantee he will respond.

The Standard reported that ChatTJB had two paying Pro subscribers as of Aug. 4. Bryant characterized the payments as closer to donations to an art project than purchases of a serious premium service.

The site's terms are less whimsical where they need to be. They describe ChatTJB as a novelty conversational service and warn that its answers are not professional legal, medical, financial, or emotional advice.

The Joke Got More Interesting When It Stopped Working
The easiest way to describe ChatTJB is as an AI parody.

That explanation undersells what happened after people started using it.

The project recreated a chatbot interface while deliberately removing the technology that normally makes such interfaces scalable. Its bottleneck was not GPUs, data center capacity, or inference costs. It was the number of messages one human could read and answer before his hands hurt.

Bryant told the Standard that the project's popularity surprised him. What seemed to resonate, he said, was not necessarily the quality of his answers but the experience of getting help from an ordinary person.

Now ChatTJB is trying to preserve that experience while adding more people behind the screen.

That means vetting workers, routing sensitive messages, setting privacy expectations, managing abuse, and figuring out what happens when somebody treats a novelty chatbot as something more consequential.

Those are not entirely different from the questions large AI companies face when they put conversational systems in front of millions of users.

ChatTJB just arrived at them from the opposite direction.