28. When everyone has an AI trading stocks, what is still worth anything? (2026-10-24)
Start with the premise: the question isn’t whether everyone has an AI — it’s whether everyone’s AI looks alike. Hundreds of millions of people using a handful of models, reading the same data, asking the same kinds of questions — homogeneity is the variable.
- Signals get crowded. Once a signal is found, everyone is on it instantly, and it dies instantly. Institutional quant already works this way — factor crowding, same-strategy same-exit stampedes. AI just extends the process from a few hundred funds to everyone, and alpha’s half-life gets absurdly short.
- Markets get more efficient and more fragile at the same time. Earnings are priced in within seconds; but in extremes, every AI reaches the same conclusion in the same millisecond — sell first. The 2010 flash crash and the 2018 Volmageddon were small-scale rehearsals. Liquidity disappears exactly when it is needed most, because the people providing it have AIs that are pulling out too.
- One piece of arithmetic AI can’t change. All investors added together can only earn the market return minus costs — there is no everyone-beats-the-index. AI doesn’t produce total alpha, it only redistributes it. And more and more of the research AI reads is written by another AI; the echo chamber makes some narratives self-reinforcing, pushing prices further from fundamentals.
- Who ends up making money. The shovel-sellers — data, compute, platforms that live on trading frequency — plus people with proprietary data and discipline. When every retail investor has an AI, it becomes standard equipment, the way a quote terminal once was; marginal edge goes to zero, and what remains is the old trio: position sizing, discipline, and holding on.
Discipline has to change too. Tactical layer: discount signals everyone can compute — support, RSI, moving averages; lines become bands; volume gets less weight; confirmation windows go in stages. Principles matter more, not less: simple rules, expressing uncertainty through position size, close-based decisions, primary-source verification. Plus new AI-era rules: crowdedness discount, news cooling-off, personal circuit breaker, signal diet, thesis written in advance.
The close: when judgment is outsourced wholesale, independent judgment and correction become scarce goods — taste in life, money in trading. Same line as #23: the human sits in the correction seat, and that holds in trading too.