Build · AI · 24 Aug 2026
Agentic trading: a rules-based strategy
I have built and successfully run the trading agent for almost 2 months now. That brings a 4.46% return home, compared to the S&P 500's 2.10% for the same period.
Two months ago, I handed one of my strategies to an AI agent and let it run the code. Here's what I learned.
The setup: a rules-based plan on 7 US mega-cap tech names, where every rule lives in Python — position sizing from available cash, a +3% take-profit on each buy, a cap so no single name exceeds 25% of capital, and a rule that only adds on a genuine dip. Zero hand-trading.
What surprised me most wasn't the return — it's up low-single-digits over two months, through a real drawdown and recovery — it's the iteration. I could add a rule, see exactly how it would've behaved on my own book, refine it, and redeploy, in minutes. My strategy became something I version like software.
And because the rules are code, execution carries no emotion. No hesitation on a red day, no chasing green. The system just follows the plan.
The numbers so far: 16 cycles, 25 positions open, 40+ closed — every closed one at its take-profit target. The honest caveat: there's no stop-loss by design, so winners get harvested and laggards are held until they recover. Discipline, not magic.
Agentic trading turned a strategy in my head into a strategy I can test and improve. Happy to share what I've built.