Jordan built a trading bot in Replit that lets ChatGPT decide when to buy and sell, and places the trades through Public's newly launched brokerage API. It traded one stock, IBIT (the Bitcoin ETF), with a decision every 30 seconds. Total build time was three to six hours. He showed it in Episode 8 (August 15, 2025) to two people from Public: Emily Kurtz, head of product, and Jake Trefethen, who leads AI.
Public had just released its API, which Emily described as a "fourth front end" for people who want to trade programmatically. Jordan is no day trader; he manages his own portfolio and that's it. That was the point: test how easy it is for a non-expert to get a working trading app on the API.
And then ChatGPT actually does some analysis and decides whether or not I should buy or sell.
— Jordan Metzner, Episode 8
The bot "went a little bit crazy," buying and selling over and over. On the day of recording it traded four shares and lost about 11 cents before Jordan switched it off for the show. Overall it made roughly 10 to 12 trades and turned $300 into $299. Jordan's view: breakeven on that many trades is "actually pretty good," especially "because I don't know what I'm doing." Jake noted ChatGPT didn't know the account only got two round-trip trades a day, so it wasn't optimizing for that. Some pages, like AI predictions, didn't fully work.
This is not investment advice. I would not trade real I would not trade real money at scale using this algorithm, but the cool part is you could trade it against the history and you kind of see how well it would perform.
— Jordan Metzner, Episode 8
Disco mode was a hit. Jake said it should be a default on all browsers.
Jordan built a second bot months later, for prediction markets: see the Kalshi prediction market trading bot. More on the industry: AI in Finance.
Jordan vibe coded a web app in Replit that pulls the price of one stock every 15 seconds, sends the price, portfolio, holdings, trade history, buying power and risk settings to ChatGPT every 30 seconds, and lets ChatGPT decide to buy or sell through Public's API.
Yes. Public had just launched a REST API for programmatic trading. Replit already knew much of it, and Jordan pasted in pieces of Public's docs when needed. He created a Public API key and an OpenAI API key.
Ours didn't, but it didn't lose much either. It made about 10 to 12 trades and turned $300 into $299. Jordan was clear it is not investment advice and said he would not trade real money at scale with it.
Jordan put the total build time at somewhere between three and six hours.
Public's AI lead Jake Trefethen suggested sending price-chart images across multiple time frames to a multimodal model. Jordan also wanted multi-model consensus, WebSockets for streaming prices and a market-hours API.
