My original idea was pretty simple: give AI market data, let a group of agents hunt for trading strategies, test the hell out of those strategies, and eventually use the good ones to make money. Coming up with ideas was never the hard part. AI will give you ideas all day. The hard part is stopping one bad idea from getting anywhere near the bankroll.

Sounding smart is not permission to spend money.

So I split the job up. AI can read, argue, suggest strategies, and explain what it sees. Regular software handles the stuff that cannot be allowed to improvise: budgets, position sizes, risk limits, account balances, and the paper broker.

AI gets ideas, not the keys

Right now no model gets brokerage credentials. It cannot move money, change the risk rules, turn on leverage, or quietly switch from fake trades to real ones. When I eventually use real money, the first version will recommend a trade and I will decide whether to place it.

That is not as sexy as yelling “autonomous hedge fund” on YouTube, but I can actually test it. Duplicate order? Block it. Old market data? Do nothing. One bot says buy while another says short the same coin? Stop and sort it out.

What the fake $500 is for

A paper account cannot prove I’ll make money forever. It can show whether the machine does what it was told. I can see whether it records every trade, includes fees and slippage, respects limits, exits when it should, and comes back correctly after a restart.

  • Save the whole bet. What strategy fired, when it fired, buy or short, entry, stop, target, size, fees, and slippage all get written down before the result is known.
  • Keep the ugly stuff too. Wins, losses, rejected trades, blocked trades, and missed trades all stay in the record. Otherwise the system only remembers what makes it look smart.
  • “I don’t know” is allowed. If the data is missing or two signals disagree, the answer can be uncertain or no trade. It does not get to invent confidence.

What has to happen before real money

Before this touches real money, it has to survive backtests and paper trading, handle normal failures, and show its math. We also have to check the actual exchange, fees, order types, and API. Then I start with recommendations I approve myself.

I am serious about making money. That is exactly why I am being careful about who—or what—gets permission to touch it.