Jake Trefethen

Jake Trefethen leads AI at Public, the brokerage for people who take investing seriously, and runs its generative AI offerings, including the research bot Alpha. He joined us with Public's head of product, Emily Kurtz, in Episode 8 (August 15, 2025). Jordan had built a ChatGPT trading bot on Public's new API, and Jake and Emily reviewed it on air.

What Public does

Public offers stocks, bonds, treasuries, options and crypto, plus yield products, across brokerage accounts and IRAs. Emily explained that the new API is a "fourth front end" for people who want programmatic execution instead of the app. It launched as a REST interface for equities and options, with crypto, bonds, money movement, WebSockets, MCP and Python SDKs on the roadmap.

Jake's side is the AI layer. Public launched Alpha, originally powered by GPT-4, a couple of months after ChatGPT came out, and it was one of the first ChatGPT plugins. Since then Public has shipped six or seven more AI features, from real-time event and earnings call summaries to explaining why your portfolio is moving.

Key ideas from the conversation

• Send the model a picture of the chart (11:41). Jake asked Jordan if he had tried multimodal prompts. In his experiments, sending a price chart image, even several time frames with indicators overlaid, gets good visual price action analysis.
• Time frames still trip models up (13:57). When he tested it, the model struggled to tell a two-minute chart from a two-hour one. Historic context helps, and million-token context windows now fit a lot of it.
• Don't trust the full context window (14:36). His rule of thumb is that quality degrades around 75% of the stated window. For scale, he said The Great Gatsby is about 60,000 tokens.
• Generated assets (20:44). Public was experimenting with custom indexes built from natural language, such as a GLP-1 basket or an anti-tariff hedge. Jake sees this filling the gap between cheap ETFs, pricier thematic ETFs and a personal wealth adviser.
• GPT-5 was a letdown because of the hype (25:52). After months of people equating GPT-5 with AGI, a router change and a slightly different model was always going to disappoint.

The the models have improved a lot over the last couple years, but it does a super good job actually of doing, like, visual price action analysis.

— Jake Trefethen, Episode 8

Should PMs vibe code?

In the news segment we discussed Shopify asking product managers to vibe code in interviews. Jake, who has a technical background himself, thinks PM prototypes built in tools like Replit or Cursor mostly end up as throwaway code. That is still valuable: one person can get from idea to something executable without a round of meetings. But he expects some pushback before PM code reaches production at a brokerage.

And so a lot of it ends up being throwaway code, but that can be very helpful in iterating through design cycles really quickly, not having to involve a bunch of different parties communication overhead and getting from idea to something that's executable super quickly with one person.

— Jake Trefethen, Episode 8

To your point of a signal, hiring somebody who's supposed to be inspiring others to ship software products, if they haven't played around with a a vibe coding tool and tried to build something themselves, I'd definitely question their curiosity a little bit.

— Jake Trefethen, Episode 8

What we built for the episode

Jordan built the Public API trading bot in Replit in three to six hours. Every 30 seconds it sends the price of IBIT, the Bitcoin ETF, plus portfolio, trade history and risk settings to ChatGPT, which decides whether to buy or sell through Public's API. Prices came from Polygon.io. After about a dozen trades it had turned $300 into $299. Jake pointed out one reason at 16:54: ChatGPT didn't know the account only got two round-trip trades a day, so it wasn't optimizing for that. This is not investment advice. For more, see our AI in finance page.

FAQ

Who is Jake Trefethen?

Jake Trefethen leads AI at Public, including all of its generative AI offerings such as Alpha. He has a technical background and works as a product manager, and he joined Built This Week in Episode 8 with Public's head of product, Emily Kurtz.

What is Public's Alpha?

Alpha is Public's investment research bot. Jake said it was originally powered by GPT-4, launched a couple of months after ChatGPT came out, and was one of the first ChatGPT plugins.

Can LLMs analyze stock charts from images?

Jake said multimodal models do a really good job of visual price action analysis when you send them a picture of the chart, even across several time frames and with indicators overlaid. In his own tests they still had trouble telling time frames apart.

What are Public's generated assets?

Generated assets is a feature Public was experimenting with that lets investors build custom indexes from natural language, for example a GLP-1 basket or an anti-tariff hedge.