Every wave of technology in markets has been sold the same way: the machine will find the edge for you. It was said of charting software, of screeners, of algorithmic execution, and now of AI. It has never been true, and it isn't true now. What actually happens is subtler and more interesting: each wave compresses the research loop, and the people who exploit the compression out-learn everyone else.
What AI genuinely does well
Modern language models are exceptional at explanation and critique. Given a backtest — the rules, the trade list, the metrics — an AI analyst can articulate what the strategy is structurally trying to capture, which market regime would starve it, and where its statistics look fragile. That kind of critical reading used to require years of pattern exposure or an expensive mentor. Now it is a supplement available to anyone, on every single test.
AI also removes the translation barrier. Describing a strategy in plain language and having it converted into testable rules collapses the distance between idea and evidence. The bottleneck stops being 'can I code this?' and becomes 'is this idea any good?' — which was always the real question.
What AI does badly
Prediction. Markets are adaptive, reflexive systems; patterns that become known get arbitraged into noise. A model trained on the past inherits every regime assumption baked into that past, and it fails exactly when regimes shift — which is precisely when you most need it to work. Treat any tool, human or machine, that claims predictive certainty about markets as a red flag.
AI can also be confidently wrong, which in research is worse than being silent. That is why serious platforms position AI as an analyst that critiques evidence you generated, rather than an oracle that tells you what to buy. The judgment — what to test, what to trust, what to risk — remains stubbornly human.
The realistic future
Expect research workflows where humans set hypotheses and constraints, machines run the exhaustive testing and flag the fragile spots, and humans make the final risk decisions. The winners of that world are not the people with secret models; they are the people who run more honest experiments per week than their competitors. Educational perspective, not investment advice — and no AI output should ever be treated as a guarantee of anything.