Software engineer Jake Nestler has developed a trading AI agent based on Anthropic's Claude. The bot mimics the author's trading style, analyzes risks, and autonomously determines position sizes. This was reported by Bloomberg.
During a month of testing, the agent achieved a 7% return, compared to 4.5% for the S&P 500 over the same period. In its first week, the algorithm refrained from buying Nvidia shares at their peak, saving the owner from a $10,000 loss. However, the strategy exhibited high volatility, with drawdowns reaching 22%.
Nestler trained the model for two weeks using the Alpaca platform. Initially, Claude displayed excessive conservatism, opting only for blue-chip stocks. The developer had to manually adjust the settings to encourage the AI to make riskier trades.
The popularity of AI agents among retail traders continues to grow. Tools like OpenClaw allow users to manage bots via Telegram or WhatsApp. Major platforms such as OKX, Bybit, and Kraken have already implemented interfaces to simplify interactions with automated systems.
It's Not All Clear-Cut
Experts caution about the risks involved. According to Jay Malaviya, co-founder of the Kairos terminal, profitable algorithms rarely become publicly available, and social media often spreads fake screenshots of returns—typically to promote malicious software.
Nestler himself is not yet ready to entrust real assets to the bot. In prediction market experiments, his agent quickly lost the entire deposit. He likened AI trading to slot machines, where outcomes are largely determined by luck.
Recall that in April, Anthropic launched an experimental product called Claude Design, an AI-based design generation tool.