Even beginners in FX are curious about "AI x automated trading"—Can you really build EAs with ChatGPT or Claude?

“An era where you can create automated trading with AI” — this is a phrase you’ve likely seen often on social media and YouTube. With ChatGPT and Claude generating so much buzz, many people wonder, “Can I have AI create an FX EA (automatic trading program) and make it a side job?” In fact, I’m one of those who started from that curiosity.

To cut to the chase,“Creating FX automated trading with AI is indeed possible”. However, that does not mean AI will magically make money on its own. I have actually had Claude create over 90 EAs and tested all of them. In that process, I learnedthat there is a large gap between what AI can create and what AI can win with.

Using Claude Code as an AI tool to extensively produce FX EAs, I have been validating how practical “AI × automated trading” can be. I’ll write these records so even FX beginners can understand.

In this article,① what it specifically means that “AI can create EAs,” ② why “earning solely by AI is a lie,” ③ and the realistic possibility of forming a side business by building with AI— I will explain these three points in plain language with as few technical terms as possible. If you’re interested in “AI × FX” but don’t know where to start, this is for you.

---

Three reasons why “AI automating trading” became a reality

“FX automated trading with AI is possible” — this was half a dream until about two to three years ago. At that time, AI could handle simple code, but it wasn’t capable of putting together the complex program that is an EA. From 2024 to 2025, the situation changed drastically. ChatGPT and Claude made it possible to practically writeMQL4, the language used for EAs in a usable way.

Reason ① — AI can write MQL4 now

MQL4 is a programming language for writing EAs. It’s a bit specialized; if you learn it on your own by buying a technical book, a beginner typically spends 3–6 months. This has long been the biggest barrier for individuals who want to create their own EA.

Now, simply instructing ChatGPT or Claude in Japanese, like “Write an EA using this method,” yields executable MQL4 code within minutes.We are living in an era where even programming beginners can build EAs by conversing with AI.

Reason ② — Backtesting environments are also ready

Creating an EA isn’t enough; you need backtesting to verify it would have really won in past markets. Previously, only paid software offered viable options, but now the MT4 tester built-in, or free Python environments, provide sufficiently practical validation.

Reason ③ — The rise of prop trading firms

“Even if you can make an EA, you need capital to run it” — this used to be a major barrier. In recent years, however,prop trading firms (companies that lend money to traders) have become accessible to individuals. One I’m attempting (Company A) allows you to borrow a several-million-yen account by paying a fixed challenge fee and manage it. This opens a path where you can start a side business even without your own funds if you have the skill.

With these three elements in place, the option of a “FX novice creating EA with AI and running it at a prop firm as a side business” is technically feasible.

---

But the claim that “AI-made EAs all win” is a lie

Here’s the counterpoint.“Can AI create it” and“Can AI win with it” are entirely different. I had Claude create over 90 EAs in a month. Of those, only a tiny handful remained usable in real operation.

The biggest trap I fell into

At first, one EA showed a backtest with monthly return +5.20% / maximum drawdown -3.96%. I seriously thought, “This will set my side income.” When I tested it in a live environment, the result collapsed to monthly return -0.06%.

Diagnosis showed the EA code hada forward-looking bug that used future information—an embedded statistical trick. Even more alarmingly,the same bug was hidden in over 90 EAs. The numbers that looked good on the desk were all an illusion.

Accidents that arise precisely because AI can create

This isn’t AI’s fault. If you tell AI to “create a winning EA,” it will obediently write “code that looks good on backtests.” But whether it wins in live operation is another matter. If you give AI free rein without a verification lens, you can end up mass-producing 100 flashy-looking but phantom EAs and losing them all.

In other words, to grow a side business with AI × automated trading you need boththe ability to have AI create and the ability to critically verify what AI creates. Only when these two wheels are in place does “AI-powered side EA” become a realistic option.

---

From here on, I will describe the concrete flow I used to have Claude create EAs (including some prompt examples), the three pitfalls of relying on AI for profit, a practical roadmap to nurture a side business, and the genuine model I reached after testing 90 EAs — the results should provide FX beginners with materials to decide whether to start an AI × automated trading side business.

---

undefined