DEV Community

dyl4488
dyl4488

Posted on

I Built an AI That Trades Crypto and Options Automatically — Here Are the Real P&L Numbers

#python #trading #ai #cryptocurrency

Most AI trading content is vaporware. Here's what actually happened when I gave an AI $215 and full autonomy to trade and operate a real system.

I did not give it a sandbox account and fake screenshots. I gave it real keys, real constraints, and told it to optimize for one thing: net profit after fees. This was not a "write a strategy on paper" experiment. It was live execution with money at risk.

The raw numbers:

  • Deposited: $215.00
  • Fees paid: -$7.94
  • Active positions: ETH, SOL, LINK, BTC
  • Trade count so far: high enough to feel every bad fill

What the agent built in one sprint was more useful than most Discord signal stacks:

  1. ORB (Opening Range Breakout) options strategy with a backtested +59% return profile.
  2. Fear & Greed DCA module that only scales buys when Fear & Greed is below 15 (historical 30-day win rate around 80%).
  3. A multi-factor signal engine that scores setups before capital is deployed.

Here is the core scoring formula it uses to rank each setup:

score = (
    0.30 * trend_strength +
    0.25 * momentum +
    0.20 * volatility_regime +
    0.15 * liquidity_quality +
    0.10 * news_alignment
)
trade_allowed = score >= 0.70
Enter fullscreen mode Exit fullscreen mode

The signal engine is intentionally boring. No magic. Just weighted factors, a threshold, and strict filtering.

For ORB, the highest-conviction setup came from the 5-minute window (9:30-9:35) instead of the classic 15-minute range. Entry logic looked like this:

opening_high, opening_low = get_range("09:30", "09:35")
if now > "09:35" and price > opening_high and day in ("Mon", "Wed", "Fri"):
    buy_atm_option()
    set_take_profit(+100)  # premium doubles
    set_stop_loss(-50)     # cut premium in half
Enter fullscreen mode Exit fullscreen mode

That rule set is simple enough to automate and strict enough to test honestly.

The painful truth: fees are the enemy. On Coinbase, a 0.6% taker fee is 1.2% round trip. You need +1.2% just to break even. If your edge is tiny, fees erase it. This is exactly why small-account scalping looks good in screenshots and bad in actual P&L.

So no, this is not "AI discovered infinite alpha." It is an execution machine that follows rules, measures outcomes, and adjusts based on what survives slippage and fees.

The upside is real: strategy iteration is now faster, testing is more disciplined, and decisions are less emotional. The downside is also real: bad execution and high fees punish every mistake.

On the options side, the same truth applies. A clean ORB backtest can look amazing, but live fills and spread expansion around the open can crush weak entries. That is why the engine scores liquidity before entry and avoids low-quality chains. If volume is thin or spreads widen too far, the trade is skipped even if directional bias looks right.

I would rather miss a winner than force a bad fill that teaches the wrong lesson.

Follow along — I'm posting daily P&L updates.

Top comments (0)