Poet runs one-man 'AI' chatbot on SF billboard as art criticism

Tucker Bryant, a poet and former Google product marketer, has placed a billboard near Folsom and Sixth streets in San Francisco advertising ChatTJB as "the leading chat interface powered by AI*"—with a small footnote that reads "*human intelligence." The joke conceals a genuine operation: Bryant himself is the entire inference stack, manually answering every message sent through the ChatTJB interface.

On August 4th, AI investor Olivia Moore, a partner on Andreessen Horowitz's AI investing team, announced the project on X. Moore reported that Bryant had been speaking with thousands of users per day and had paused new usage after demand overwhelmed the operation. A screenshot Moore posted showed the ChatTJB administrator interface displaying 2,146 answered messages and 30 pending—a snapshot of the workload at that moment. Moore also confirmed that Bryant was handling image-generation requests by drawing them by hand. The daily user figure has not been independently corroborated; public traces on a San Francisco discussion three days before Moore's post show people describing conversations with the service and sharing image requests, establishing that the service attracted attention without verifying Moore's thousands-per-day claim.

Bryant describes ChatTJB as something fundamentally different from the software products its interface imitates. On his art portfolio, he calls it "the worst LLM in the world" and explains that the project uses poetic, cryptic, and intentionally bad answers to criticize what he terms "cognitive surrender." ChatTJB is performance art packaged with the visual language of a consumer AI product. That framing aligns with Bryant's career: he studied at Stanford University and later worked in product marketing at Google, developing go-to-market strategies and helping launch and position products, according to a December 2025 interview. He now works as a poet, artist, and innovation speaker, and his professional biography describes his practice as designing experiences around timing, interactivity, friction, and constraints.

ChatTJB applies those design principles directly. Response time depends on whether Bryant is awake, available, and willing to answer. Image generation requires him to pick up a pen. The growing queue—visible in the administrator interface—transforms the hidden economics of inference into a tangible human backlog. In the 2025 interview, Bryant had warned that instant answers from LLMs can remove users from the creative struggle that develops judgment and voice. ChatTJB forces that friction back into the interface, ensuring each response carries the limits, taste, and inconsistency of one person. The timing of the project proved receptive: AI companies had already covered San Francisco with advertisements selling agents, models, and infrastructure, with slogans including "Stop hiring humans" and "Own your inference." Bryant copied the format, then reversed its underlying pitch. The distribution itself followed a startup playbook: Moore, with her standing as an AI investor at a prominent firm, carried the project into the online audience that evaluates new AI products. ChatTJB gained initial attention by looking investable long enough for users to discover the person at the center.

The project's inability to scale is its central mechanism and its meaning. Conventional AI services spread the cost of a model across large volumes of automated requests. ChatTJB adds one unit of human labor for every answer, making rising demand operationally punishing. The pause described by Moore functions as evidence of the constraint Bryant designed into the work. That same constraint makes ChatTJB nearly impossible to copy without destroying its message: hiring a support staff, automating replies, or routing prompts to an actual model would increase throughput while erasing the single-author relationship users came to test. Bryant has built an AI interface whose appeal depends on preserving the bottleneck that prevents it from operating like an AI startup.