San Francisco's Most Honest Chatbot Is a Man With a Phone

On the corner of 6th and Folsom, in the part of San Francisco where every third billboard promises to replace your engineering org by Q3, there is a billboard for a chatbot called ChatTJB. It describes itself as a leading chat interface powered by AI. It looks, from the sidewalk, exactly like the seventeen other things trying to sell you inference that week.

The billboard cost about $6,000. It is, by the standards of its block, the most honest object in the neighborhood.

Because ChatTJB is powered by AI. It’s just that here, AI stands for Average Individual. The Average Individual is Tucker Bryant, 32, a former Google employee, conceptual artist, and keynote speaker whose initials are, conveniently, TJB. There is no model. There is no cluster. There is a man, and there is his phone, and when you submit a query to the leading chat interface, Tucker Bryant reads it and types back — slowly, sometimes cryptically, depending on when he is awake and how motivated he is feeling about your question.

The website does not hide any of this. It is, if anything, the most forthcoming product page in the history of the category. ChatTJB describes itself as “artisanal intelligence, handcrafted by a single human being.” It is not a large language model. It is a large language experience — an LLE, not an LLM, a distinction the About page treats with the gravity of a term sheet. The About page also clarifies, for anyone still reaching for their wallet, that ChatTJB is public art and a critique, that it is not artificial intelligence, and that it is not an investment.

That anyone needed to be told this last part is, of course, the entire point.

The critique is gentle and it lands like a piano. San Francisco is a city where “powered by AI” has become a kind of ambient background hum, a phrase that means everything and therefore nothing, attached to search, to sandwiches, to companionship, to the thing that reads your Slack and the thing that reads your bloodwork. Bryant’s wager was that you could put those three words on a billboard, staple them to a man with a phone, and watch a meaningful number of people simply trust it. He was correct. He is now buried in the proof.

He posted about it on Instagram on Sunday, August 1. It went, in the industry term, viral. What happened next is where the piece stops being a critique of AI hype and starts being an involuntary reenactment of it.

Demand exploded. ChatTJB has received more than 30,000 requests. At peak, Bryant was looking at roughly 5,000 queries in a single hour — a genuinely respectable inference load, if the inference in question weren’t one man reading his phone on a Tuesday. There is no autoscaling group for this. There is no way to spin up a second Tucker Bryant. The single-operator large language experience had, in the language of every deck ever presented to a16z, achieved product-market fit, and product-market fit for a joke is indistinguishable from a crisis.

So Bryant did the only thing left to do. The thing every AI company does when the model can’t keep up.

He asked for volunteers.

Sit with that for a second, because it is the best part and nobody involved appears to have planned it. Faced with overwhelming demand, the chatbot whose entire artistic thesis is “handcrafted by a single human being” quietly onboarded a human backend. The single operator became several operators. The Average Individual became, discreetly, a Distributed Average of Several Individuals, load-balanced across whoever was free to type. The branding did not change. The billboard still says one guy. The website still says handcrafted. Somewhere behind the curtain, additional hands.

This is not a bug in the art. This is the art completing itself. Every “powered by AI” product you have ever used is, at some layer, a person you were not told about — a contractor labeling data, a reviewer in the loop, a moderator absorbing the queries the model refused. ChatTJB started as a parody of the label and, at scale, spontaneously grew the exact thing the label is usually hiding. Bryant built a monument to the lie and then, under load, was forced to become the lie, on purpose, in public, still telling everyone the truth on the About page. We would call it too on-the-nose if the nose weren’t the entire sculpture.

The reaction has been the strangest part of all, which is to say it has been nice. This is a city with a documented, load-bearing exhaustion around AI advertising — the billboards, the launch threads, the founder who slides into your replies to tell you his agent can now feel. And here comes a piece of AI advertising, on a real billboard, for a real “chatbot,” and the tech-weary locals took one look at the man behind the curtain and, in the words of one widely-circulated reaction, decided this rules. Warmly. Sincerely. The internet, briefly, agreed on something, and the thing it agreed on was a $6,000 sign for a service that answers you personally, badly, and honestly.

It turns out the feature people were missing this whole time was disclosure. ChatTJB does exactly what it says, which in this market registers as a wild, almost illegal act of transparency. You ask it a question. A human being — average, individual, handcrafting your response at whatever speed a human handcrafts things — gets back to you when he can. There is no upsell. There is no context window. There is a guy, and he read your message, and that was always going to be the most a chatbot could honestly promise.

He can’t promise it much longer. The volunteers are typing. The average is distributing. Somewhere in SoMa, a man is watching his art scale, which is the one outcome he was warning everybody about.

Sources: