AI has worked its way into trading the same way most “new edges” do – quietly at first, then all at once. What started as institutional infrastructure has become a retail product: AI that scans charts, identifies patterns, and spits out technical signals telling you what’s bullish, bearish, or about to break.

On the surface, that sounds like progress. In some ways, it is. AI is very good at one thing traders care about: pattern recognition at scale. Feed it enough price data and it will scan more markets in five minutes than a human could in a week. It doesn’t get tired, it doesn’t hesitate, and it doesn’t have an opinion about being wrong.

Used properly, that can be genuinely helpful.

Where traders get into trouble is thinking AI understands the market. It doesn’t. It recognises shapes, relationships, and probabilities – nothing more. And when market conditions change, that distinction becomes critical.

AI systems are trained on past data. That means they inherit the assumptions of the period they were trained in. Trending markets produce trend‑friendly models. Range‑bound conditions produce very different behaviour. When the regime shifts, AI doesn’t raise its hand and say, “Something’s changed.” It just keeps pushing signals with the same confidence.

A good real‑world example of this problem is Zillow.

Zillow relied heavily on its AI pricing model, Zestimate, to buy homes at scale through its Zillow Offers business. The model worked fine while the housing market behaved within familiar historical ranges. Then conditions changed rapidly. Buyer behaviour shifted, supply chains broke, and prices started moving in ways the model hadn’t seen before.

The AI didn’t adapt. It kept forecasting higher prices, encouraging Zillow to overpay for properties even as the market cooled. The outputs looked clean. The confidence was high. The assumptions, however, were broken. Zillow ended up writing down more than half a billion dollars and shutting the business entirely.

That’s not a technology story – it’s a context story.

You see the same thing in trading signals. An AI trained during strong trends will happily keep flagging breakouts in a market that’s moved into chop. Statistically, the pattern still exists. Practically, those trades fail repeatedly. From the outside, it looks like “bad luck.” In reality, the model is hallucinating relevance where there is none.

This is where traditional technical analysis still matters.

When a human trader looks at a chart, they’re not just asking, Is this a pattern? They’re asking, Does this pattern make sense right now? Is volatility expanding or compressing? Is participation broad or narrow? Is this market rewarding follow‑through, or punishing it?

Manual analysis is slower and less precise, but it integrates context automatically. Doubt is part of the process. AI, by contrast, presents confidence even when it shouldn’t.

There’s also a massive difference between institutional AI and what retail traders are usually accessing. Institutions don’t treat AI as a decision‑maker. It’s a component in a larger process with risk limits, regime filters, and human oversight. Signals are sized cautiously, challenged constantly, and turned off when conditions don’t suit them.

Retail AI tools are often marketed the opposite way — as simplified, nearly autonomous solutions. The logic is usually opaque, the back‑tests selective, and the limitations understated. When they stop working, traders are left wondering whether the market changed or whether the model never really understood it in the first place.

It’s also worth saying plainly: some trading approaches simply aren’t suited to AI at all.

WD Gann‑based analysis is a clear example. Gann methods rely on geometry, time, proportion, and interpretation. Two skilled Gann traders can look at the same chart and arrive at different conclusions – and both can be valid. That discretion is not a flaw; it’s the method.

AI needs explicit rules and objective optimisation. When Gann concepts are forced into algorithms, the nuance disappears. You end up with something that looks systematic but isn’t really Gann anymore – just a simplified approximation wearing the label.

The takeaway here isn’t that AI is useless. Far from it. As a scanner, a filter, a way to surface opportunities you might otherwise miss, AI is extremely effective.

Where it fails is where judgement is required.

AI can tell you what looks similar to the past. It cannot tell you whether the past still applies.

That responsibility still sits with the trader – and that’s unlikely to change anytime soon.

Good Trading

Aaron Lynch