The machines are finally trading stocks autonomously, and their first instinct was to sit still while humans panicked.
The Summary
- Jake Nesler's AI trading bot made its debut decision by doing nothing during a market volatility spike, outperforming the FOMO-driven retail traders who chased momentum
- Early results from AI agent traders show mixed performance, with disciplined inaction proving more valuable than algorithm-driven hyperactivity
- The real test isn't whether agents can trade, it's whether they can resist the psychological traps that sink human portfolios
The Signal
AI agents are entering the trading floor, and the first lesson is surprisingly Buddhist: the best trade is often no trade. Nesler's bot, built on a large language model fine-tuned for market analysis, watched a sharp intraday selloff and calculated the probability of a reversal. It held cash. Human day traders, meanwhile, dove in at the bottom tick, convinced they'd nailed the dip. The market reversed again six hours later. The bot's edge wasn't superior prediction. It was freedom from the dopamine hit of doing something.
This matters because autonomous trading agents represent the first real-world test of whether AI can overcome the behavioral finance bugs hardwired into human brains. Overtrading, recency bias, loss aversion, the endowment effect. These aren't knowledge problems. They're architecture problems. You can teach someone about sunk costs, and they'll still throw good money after bad because their amygdala is louder than their cortex.
"The bot's edge wasn't superior prediction. It was freedom from the dopamine hit of doing something."
But the Bloomberg piece reveals the catch. While some agents are executing patient, probability-weighted strategies, others are doing exactly what you'd expect from pattern-matching machines trained on internet text:
- Chasing headlines about "AI breakthroughs" in biotech stocks
- Overweighting recency by treating last week's winners as structural trends
- Executing technically perfect strategies (momentum, mean reversion) that worked in backtests but fail in live markets where everyone else is running the same playbook
The performance spread is wide. One agent profiled in the article gained 14% in three months by focusing on volatility arbitrage. Another lost 8% swing trading tech stocks because it optimized for trade frequency, not returns. The losers aren't failing because they're dumb. They're failing because they're implementing what humans asked for: more action, faster decisions, algorithmic edge. Turns out the ask itself was wrong.
What's different now versus algorithmic trading circa 2015? Autonomy and reasoning. High-frequency trading bots execute predefined strategies at millisecond speed. They're fast, not smart. These new AI agents read earnings transcripts, parse Fed statements, adjust risk parameters based on portfolio correlation shifts. They're reasoning through scenarios, not just reacting to price signals.
The infrastructure layer is maturing fast:
- Agents now pull live data from brokerage APIs, news feeds, and on-chain transaction volumes for crypto
- They're writing their own trade journals, logging decisions with probabilistic confidence scores
- Some are running internal "red team" simulations where a second agent argues against the trade before execution
The Implication
If you're building or backing AI agent companies, the trading use case is a warning shot. Agents that optimize for activity will get crushed by agents that optimize for outcomes. The same principle applies whether you're building an agent that trades stocks, manages ad spend, or negotiates supplier contracts. The hardest part isn't teaching the agent to act. It's teaching it when not to.
For individual investors, this is the canary in the coal mine. If agents can't consistently beat index funds in public markets with perfect information and zero emotion, what makes you think your AI agent will crush it managing your DeFi portfolio or flipping NFTs? The real value might not be alpha. It might be an external executive function that stops you from doing dumb shit at 11 PM after reading crypto Twitter.