The former Google employee built ChatTJB, a chatbot that looks a lot like the AI assistants Silicon Valley has spent trillions of dollars developing. It looks just like a normal chat box and follows the same process: users type a question and wait for an answer. The catch? There’s no large language model on the other side—just Bryant, or increasingly, one of the more than 10,000 people who’ve applied to help him answer all those questions.

And questions there are plenty of: more than 100,000 prompts have been submitted since Bryant’s $6,000 San Francisco billboard advertising the site went up, he told Fortune. That’s not all: Bryant said the volunteer waitlist is growing by roughly 1,000 people a day, turning what started as an art project about AI dependence into a strange reversal of the AI boom’s defining anxiety. At a moment when companies are racing to make machines behave more like people, Bryant has found an audience for the opposite: people behaving like machines.

The ‘AI’ stands for average individual

Bryant, a former Google project manager and keynote speaker, launched ChatTJB in April. The site looked like the stripped-down interfaces that have been ubiquitous since ChatGPT: a text box, a question and eventually an answer from Bryant, who reads and answers the prompts himself. 

He made the joke harder to miss this summer with a billboard advertising ChatTJB as a “leading chat interface powered by AI.” A small asterisk on the ad revealed what AI meant: “average individual.” Though, fittingly, part of the disclaimer was obscured by a tree branch when Bryant shared a photo of it on Instagram.

The billboard landed in a city where advertisements for artificial intelligence have become nearly inescapable. In 2024, AI startup Artisan drew attention for its own provocative billboard campaign urging companies to “Stop Hiring Humans,” according to the San Francisco Chronicle.

Bryant told Fortune the honeymoon-night message is the one he keeps coming back to: someone wrote in on the first night of their honeymoon, saying they didn’t feel totally relaxed, and asked if that was okay to feel. 

“I think that was the moment the project began to shift from a surreal piece of satire into a place for people to chat in earnest with a well-meaning, if undeniably average, human being,” he said.

He’s since largely stepped back himself. “Right now, I’ve mostly stopped answering,” Bryant said after the pace of prompts—more than 5,000 an hour at its peak—eclipsed what one person could handle. He put out a call for other volunteering  “Average Individuals” to take over instead.

ChatTJB wasn’t built to make money. But its unexpected popularity has opened the door to something Bryant hadn’t necessarily planned.

“I’ve had a couple of conversations with people interested in working on the project in various ways,” Bryant. But for now, Bryant still thinks of ChatTJB as a short-term art project among several he wants to pursue. Still, he’s keeping his options open.“If the right partner wanted to help turn this into a sustainable project, I’d love to talk with them,” Bryant said.

What happens when no one questions AI

The project began from Bryant’s concern that people were outsourcing more and more of their thinking to AI without scrutinizing the answers they received.

Bryant caught himself asking an AI whether to wear short or long sleeves on a 65-degree day in a city where he had lived for seven years, he told Wired. His partner introduced him to a term for what he was doing: “cognitive surrender.”

Bryant’s project was inspired by 2026 research on the phenomenon from Wharton professors Steven D. Shaw and Gideon Nave, who coined the term “cognitive surrender” after finding that people who relied on AI assistance were more accurate when the AI was right. But when the AI gave incorrect advice, their accuracy fell by 15 percentage points compared with people who received no AI assistance. 

“When people hear the term, they immediately recognize the phenomenon, both in others and often in themselves,” Shaw told Fortune. “Academic work can define and measure a phenomenon, but art can communicate its emotional stakes in a way that a research paper rarely can.”

ChatTJB also introduces something most AI companies are trying to eliminate: friction. Users might wait hours for an answer, responses aren’t infinitely scalable, and the person on the other end may offer a perspective that’s personal rather than authoritative.

But Shaw cautioned against interpreting ChatTJB’s popularity as evidence that people actually want their technology to become slower. “I would not conclude that people generally prefer slow or inefficient AI tools,” he said. “ChatTJB is art.” Shaw and Nave tried several ways to reduce cognitive surrender in their experiments, including time pressure, incentives and feedback, but found it difficult to eliminate. Other researchers have explored “cognitive forcing techniques,” such as slowing responses or designing systems that push back and ask users questions.

The challenge, Shaw said, is introducing enough friction to make people scrutinize AI’s answers without making the technology so cumbersome that users simply choose a faster competitor.

“The more useful and efficient AI becomes, the easier it is to rely on it without much reflection,” he said.

For Shaw, the appeal of something like ChatTJB isn’t necessarily that people want worse technology. It’s that it gives them something a highly scalable chatbot cannot quite replicate.

“It tells us that people want more than instant information; they want to remain human,” Shaw said. The waiting and scarcity, he added, may even make Bryant’s eventual responses feel more meaningful.

ChatTJB works because “it restores a human trace that highly scalable AI interfaces tend to remove.”