"AI-powered" is the most expensive sticker in trading. It costs the vendor nothing and it costs you everything, because it is designed to stop the one question that matters: what does the AI actually do?
Ninety percent of the time the honest answer is "nothing." The "AI" is a marketing word bolted onto the same fixed indicator logic that has been sold for fifteen years — moving-average crossovers with a new label. The model never sees a chart, never makes a decision, never does anything except appear in the sales copy. And you cannot tell the real ones from the stickers by the word, because they both use the same word.
You are not behind for not knowing the difference. The whole point of the label is that it explains nothing. So let us tear it down — here is what genuine AI in an EA actually does, what it does not do, and the exact questions that separate a decision loop from a sticker.
The Label vs the Loop
An EA is a loop: it watches the market, decides, and acts, over and over. The only question that matters about "AI" is whether a model sits inside that loop, making or shaping the decide step — or whether it sits in the brochure while the same old if-then logic runs the actual trades.
A fixed EA's decide step is a frozen rule: "if fast MA crosses slow MA, buy." It was written once and never changes. A genuine AI EA's decide step routes the live market context through a model that produces a judgment — and that judgment can differ as conditions differ, the way a human analyst's would. The label tells you nothing about which of these you have. The loop tells you everything. The seven components a real AI trading system needs are all about that loop being genuine rather than decorative.
What the AI Actually Decides
When the model really is in the loop, here is the work it does — concretely, not magically:
Reads the Current Market Regime
Before a single order, the model assesses what kind of market this is right now — trending or ranging, expanding or compressing volatility, news-driven or quiet. A fixed EA cannot do this; it applies the same rule to every regime and hopes. The AI's first job is context: deciding whether the conditions even suit a trade, and frequently concluding they do not. The trades it declines are often where most of the edge lives.
Produces a Verdict Before the EA Acts
On a candidate setup, the model returns a judgment — take it, skip it, size it down — with reasoning attached. This is the difference between a signal and a decision: a signal is a number crossing a line; a decision weighs context. In a real AI EA you can read why it passed or rejected a trade, because the reasoning is part of the output, not a black box.
Manages the Position, Not Just the Entry
Entry is the easy part. The model also shapes what happens after — whether to hold through noise, tighten on a volatility spike, or step aside when the regime it entered on has changed. That ongoing judgment, applied to live conditions the original code never anticipated, is where an adaptive system separates from a static one.
Can Re-Prompt to Newer, Better Models
This is the structural advantage a fixed EA can never have. The decision layer is a prompt to a frontier model, so when a stronger model ships, the EA can route its decisions through the better reasoner without rewriting the strategy. When we swapped to GPT-5.5 and Claude Opus 4.7, the live results moved — the strategy did not change, the brain behind the decisions did. A 2022 hard-coded EA cannot get smarter; an AI EA's reasoning improves as the models do.
The decision loop, in public.
Alpha Pulse AI runs real model verdicts — GPT, Claude, Gemini, Grok — before each trade, with the reasoning visible rather than hidden. It is forward-tested in public precisely so you can watch the loop work: the regime reads, the rejections, the model swaps. That is what "AI inside an EA" is supposed to mean.
What the AI Does NOT Do
Honesty about the upside requires honesty about the limits. A real AI EA does not do any of these, and any vendor implying otherwise is back to selling stickers:
- It does not predict the future. The model assesses probabilities in context; it does not know what price will do. Anyone selling AI as a crystal ball is selling the fantasy, not the tool.
- It does not remove risk. Better decisions still lose trades. Drawdown is structural to trading and no model deletes it — at best the AI manages risk more intelligently, never to zero.
- It does not run itself unattended forever. Models change, costs change, regimes change. A real AI EA is operated, not abandoned. The "set and forget AI bot" pitch is the tell of someone who has never run one.
- It is not immune to overfitting. An AI layer wrapped around an over-optimized strategy is still over-optimized. The intelligence has to be in the live decision, not in a backtest curve fitted to the past.
Real vs Fake: The Questions to Ask a Vendor
You can disqualify most "AI-powered" EAs with four questions, because the stickers cannot answer them:
- Which model, and can it be updated? A real answer names the models and explains how the EA moves to newer ones. "Proprietary AI" with no specifics is a sticker.
- Can I see the reasoning behind a trade? Genuine AI EAs expose why a trade was taken or rejected. If the reasoning is invisible, the AI may be too.
- What does it do when it has no edge? A real decision loop frequently decides not to trade. An EA that always has a trade is running fixed logic with an AI label.
- Where is the live forward test? Real AI decisions show up in public, messy, real-time results — not just a polished backtest. No live test is the loudest answer of all.
An EA that cannot answer these is not necessarily a scam — it may be a perfectly decent fixed-rule system. But it is not doing what the AI label claims, and you should price it as the indicator EA it actually is.
The Honest Close
The AI in an AI EA is not magic and it is not nothing — it is a decision layer that reads context, judges setups, manages positions, and gets smarter as the models behind it improve. That is genuinely valuable and genuinely different from a fixed rule. It is also, in most products sold to you, completely absent behind a label that exists to stop you from checking.
So check. Ask what the AI decides, demand to see the reasoning, and find the live test. The vendors running a real loop will welcome every one of those questions. The ones selling the sticker will change the subject — and that is your answer.
Want the weekly teardown of real-vs-fake AI trading, plus our own live decision-loop notes? The DoItTrading newsletter sends it straight, reasoning and losses included.
Frequently Asked Questions
Is an "AI EA" just a backtest with a new label?
Many of them are. If the model never participates in the live decide step — assessing the current setup before the trade — then "AI" is a marketing word attached to ordinary fixed indicator logic. A genuine AI EA routes live market context through a model that produces a judgment in real time, and you can verify the difference by asking to see the reasoning behind individual trades and the live forward test.
Does the AI in an AI EA predict price?
No, and any vendor claiming it does is selling a fantasy. A real AI decision layer assesses probabilities in context — what regime this is, whether a setup is worth taking, how to manage the position — without knowing what price will do next. It aims to make better-informed decisions, not to forecast the future. Prediction is the sticker; contextual judgment is the actual tool.
GPT or a custom model — which is better for trading?
What matters more than the specific model is whether the decision layer can be updated to stronger models as they ship. A system prompted to a frontier model like GPT or Claude inherits each improvement in reasoning without a strategy rewrite, which a hard-coded custom model or a fixed 2022 EA cannot. Ask whether the EA can move to newer models — adaptability beats brand.
Can AI EAs still overfit?
Yes. An AI layer wrapped around a strategy that was over-optimized to historical data is still overfit — the intelligence has to live in the live decision, not in a backtest curve fitted to the past. This is exactly why a public, real-time forward test matters more than any backtest: it shows the AI deciding against conditions it could not have been tuned to in advance.
