“Describe me a strategy based on EMA crossover with an RSI filter.”

ChatGPT delivers the logic in 10 seconds. I open TSLab, assemble the blocks. In 15 minutes — a ready robot.

Sounds like a dream. But does it work in practice?

For the past month I’ve been testing the combo: AI for idea generation, visual builders for assembly. Here’s the reality.

What the AI + Builder Combo Promises

The idea is simple:

  1. AI generates a strategy (ChatGPT, Claude) — You describe the idea in text, AI gives the logic: entry/exit conditions, filters.

  2. Builder assembles the robot (TSLab, Designer, fxDreema) — You transfer the logic to blocks, run a backtest, robot is ready.

Experiment: 10 Strategies from ChatGPT → TSLab

I asked ChatGPT to generate 10 simple strategies.

Prompt:

“Suggest a simple indicator strategy for daily stock trading. Use only classic indicators (SMA, EMA, RSI, MACD, Bollinger Bands). Describe entry and exit conditions.”

Results out of 10 strategies:

  • 3 showed profit on backtest (>20% annual)
  • 5 were around zero (+5% to -5%)
  • 2 were unprofitable (-10% and -15%)

Problem #1: AI Doesn’t Understand Market Context

ChatGPT generates logically correct strategies. But it doesn’t know: the instrument’s specifics, the current market regime, or your trading style.

Example: I asked for a strategy for BTC/USDT (crypto, high volatility). ChatGPT suggested a 2% stop-loss. On crypto, 2% is noise. The bot got stopped out 20 times a day.

Conclusion: AI needs very precise direction. “Strategy for a volatile asset with 5-10% daily swings” gives better results than just “crypto strategy.”

Problem #2: Builders Limit Complexity

Claude can generate complex strategies with adaptive parameters and ML filters. But the visual builder doesn’t support that.

Conclusion: AI can generate a strategy more complex than a builder can assemble.

Problem #3: AI Hallucinates Indicators

ChatGPT sometimes suggests indicators that don’t exist in the builder. TSLab doesn’t have Ichimoku out of the box. fxDreema doesn’t have OBV.

Conclusion: You need to know which indicators exist in your builder. Otherwise AI will propose what can’t be implemented.

What Works: The Right Prompts

After a month of testing, I found the formula for a working prompt:

Bad prompt: “Come up with a trading strategy”

Good prompt: “Suggest a strategy for hourly EUR/USD candles (forex). Use only these indicators: SMA, EMA, RSI, MACD. Average pair volatility 50 pips per day. Goal: 3-5 trades per week. Stop-loss up to 30 pips.”

Real Workflow: How I Use AI + Builder

Step 1: Generate ideas via AI — get 5 ideas, pick the best 2.

Step 2: Assemble in builder — 15-20 minutes per strategy.

Step 3: Backtest — run on 3 years of history. If it fails, go back to AI for adjustments.

Step 4: Optimization via AI — “Backtest showed Sharpe Ratio 0.8. How to improve?”

Bottom line: AI doesn’t replace the analyst. But it accelerates hypothesis generation.

Future or Hype?

It’s not the future. It’s a tool.

AI + builders won’t replace a quant programmer. But they’ll speed up the work.

When it makes sense: Rapid prototyping, generating variations, explaining others’ strategies.

When it’s pointless: Production-ready systems, complex strategies (ML, arbitrage), deep understanding of algo trading.

My opinion: A useful tool for beginners and experienced traders alike. Lowers the entry barrier. But not a silver bullet. If you want deep understanding — learn programming.

Useful links: