The narrative that “AI has eaten trading” is half right. Machine learning beats humans decisively at certain things and loses to them at others. Knowing which is which is the difference between giving up edge and competing intelligently.
Where AI wins
- Latency-sensitive arbitrage. If your edge is measured in microseconds, you cannot compete with co-located algorithms. Don’t try.
- Cross-asset statistical patterns. Multi-factor models can ingest hundreds of correlated signals simultaneously. No human can.
- Sentiment aggregation. Reading 10,000 news articles and tweets per hour for tone is trivial for AI, impossible for you.
- High-frequency order-flow microstructure. The order book moves faster than human cognition.
Where humans still win
- Regime change. Models trained on regime A break the day regime B arrives. Humans notice “this feels different” before the numbers confirm it.
- Rare events. Tail-risk events are by definition under-represented in training data. Models underweight them; humans with memory of 2008 / 2020 / 2022 weight them appropriately.
- Narrative formation. Knowing why a move is happening (and whether the why has legs) is still a human skill.
- Time-frame patience. Most AI models optimise for short horizons. Position trades held for months still favour patient humans.
The compounding architecture
The best 2026 setup isn’t human-or-AI. It’s AI for the things AI does well (signal detection, pattern recognition, latency execution) wrapped by human discretion at the regime/narrative layer. Our indicators sit deliberately in this layer: machine-grade pattern detection, human-grade decision authority.