AI agents becoming day traders, but gains prove elusive

When Nvidia’s earnings sent the stock surging in late November, the agent-trained on Nesler’s own trading instincts-argued with itself over whether to follow the momentum. Fortunately, the bot decided against it, as chasing the high would have lumbered Nesler’s portfolio with an estimated $10,000 loss that week.

Nesler, a 29-year-old software engineer in Scranton, Pennsylvania, had burned himself out trying to trade options around his day job. He found himself inspired by an Anthropic experiment in which the firm’s Claude model controlled an office vending machine. But what if, instead of offering snacks, Claude could be trained to buy and sell stocks instead? Nesler spent two and a half weeks teaching the model how he thought about risk, entry signals and position sizing, then set it loose on a simulated brokerage account on Alpaca, an algorithmic trading platform, with $100,000 in fake money to spend.

“I wanted something that could be a proxy for the way I think and carry out those things while I’m doing other stuff,” he said in an interview. The rest of the agent’s week was less impressive, losing money on a series of speculative trades that didn’t go its way. After five days of trading, Nesler was left with one good call alongside a string of losses.

Across equities, crypto and prediction markets, a growing legion of retail traders are training AI agents to buy and sell assets on their behalf. It’s a sign of a new era in retail investing, where traders believe that AI-powered tools can produce better investment outcomes-and that anything still done manually is a process waiting for improvement.

Live Events

Open-source platforms such as OpenClaw allow users to talk to their AI agents through accessible messaging apps like WhatsApp and Telegram, attracting hordes of wannabe stock-pickers without a tech school resume. All they have to do is connect an AI model to the system, then let it loose with simple instructions.
On X, claims of extraordinary returns through AI agents have become a genre of their own. One viral post, viewed 4.7 million times, boasted of a 5,860% return in two days on prediction markets platform Polymarket. Its story was later debunked by another account operated by an AI agent, saying the claims were impossible. Similar posts have connected users directly to malware, presenting a security risk for unsuspecting investors.The tools to set up these bots have never been more accessible. Agents are becoming the logical next step for a generation of traders who came of age on apps like Robinhood Markets, adding another layer of automation to speculation. Trading platforms themselves are jumping on the trend, with companies like Public Holdings seeking to offer their own AI agents to customers. But what AI has yet to make easier is actually earning any money. Nesler encountered a recurring problem with his agent. The bot kept defaulting to responsible behaviour, gravitating toward blue chips and S&P 500 stocks-the kind of positions that would barely move in a week. Nesler said he had to override it repeatedly, pushing the model toward riskier trades that suited his own appetite.

The problem is baked into the technology. Large language models like Claude are trained on vast amounts of financial advice, risk management literature and market commentary. Left unprompted, they absorb the consensus view of what responsible investing looks like-and behave like the median of every financial adviser’s blog posts. Some of the traders deploying agents on top of these models are fighting that default conservatism, trying to coax risk-taking out of a system trained to avoid it.

Once tuned to Nesler’s liking, the agent posted a return of about 7% over 30 days-outpacing the S&P 500’s roughly 4.5% gain over the same stretch. In between, it tested his appetite for volatility, experiencing drawdowns of as much as 22%.

While he’s since published his code online for others to try, Nesler isn’t ready to recommend that anyone give it real cash.

“It’s totally possible to make money with it,” he said. “But I mean, anybody could do that with dumb luck on options. It doesn’t mean they’re not going to lose that money also.”