"Are You Really Using AI for Trading?"
That's a question we've heard many times over the past few years.
The short answer is yes, but specifics vary.
Let's dive in.
First, let's agree on terms
What most people mean by AI is large language models (LLMs).
Yes, we've been using LLMs in our EAs for a while now.
They are part of an AI-driven filter in some of our EAs, such as EasyAI.
However, they do not trade on their own.
Since fully AI-driven trading is still unreliable, we use them only as a filter on top of time-tested logic that's proven to work over many years of live, verified trading.
In addition, we've been using some of the best available models with Deep Research/Deep Think functionality, which allows them to fully utilize their potential.
Do we use LLMs during research and development?
The answer is two-fold.
On the one hand, modern AI tools are great at assisting with market research.
What used to take weeks and a team of experts now takes days and one person coordinating AI agents.
On the other hand, vibe coding often produces code that's hard to maintain reliably.
So, while we are doing research and prototyping with ChatGPT/Claude/Gemini, when it comes to actual production-ready solutions that need to be 99.999% reliable on real-money accounts, we still do it almost entirely manually.
Do we use neural networks in our EAs?
More advanced users might argue that artificial neural networks qualify as AI.
The answer is yes.
We have used neural-net-based filtering in some of our EAs, such as Golden Pickaxe and Perceptrader AI, for several years.
We even pioneered a rare feature that allows users to optimize neural nets within MT 4/5 EAs.
But why aren't we using AI directly?
Can't we just plug in ChatGPT 5.2 Deep Research/Gemini 3 Pro Deep Think/Claude Opus 4.6 and let them make trading decisions, since they are so smart?
As reality has shown us time and again, modern AI models aren't good at "just do it".
If they aren't given clear instructions like "do this to achieve that", their "own" thinking tends to be heavily correlated and not always reliable.
In common words: "garbage in, garbage out".
Why exactly that is the case is a topic for another post.
But for now, we'll conclude that these models are only as good as the person guiding them allows them to be.
To conclude
Yes, we are heavily utilizing AI, but we are doing so selectively.
We use AI's strengths, such as speed and pattern recognition, while avoiding it in areas where its weaknesses outweigh its strengths.
And if you're ready to start leveraging all of that:
Best,
Val