Everyone is asking AI to write code, but today I did something even more thrilling—getting it to write a piece of code that can automatically spend real money and place orders with zero human supervision.

First, I laid out the decision logic card: I thought through the boundaries, the worst-case scenarios, and the kill conditions before I dared to touch it. I initially wanted Codex to set up the actuators by itself, but the dispatch channel was unstable all day, so I switched to building it incrementally in my own main thread.

Before going live, I ran 7 rounds of adversarial code reviews, and hard-squeezed out 26 fatal P0 bugs. Starting from processes that could place a cascade of wrong orders by concurrency chaos, and tightening round by round—from concurrency edge cases to clock drift—until the only remaining issue was a corner-case aging vulnerability that would trigger after 14 days.

After deploying to the VPS, another real canary caught a bug in one evening that had passed a full-green day test and only showed up at night.

Tonight at 21:30, the first order was finally released for real.

Honestly, this is just a small, paid learning experiment. The daily exposure is around 10u, and I set a 30u hard kill line. I’m using small money to get real data and experience stepping on pitfalls—not trying to scale profit-making.

There aren’t any results yet. All I can say is: the code is live—waiting to validate tonight.

For an AI-written system that spends money, how far can it hold up through this review gauntlet?