I handed 80 AI personas the same prompt, told them to design their single best strategy, and stepped back. The designs ranged from elegant to unhinged. But the most interesting thing was what happened when I compared them.
The roster was set. Eighty personas, fourteen categories, ranging from Warren Buffett to a fictional former Russian spy. Time to pull the trigger.
The Rules
Simple, brutal, unambiguous. Every persona gets the same prompt:
- Design ONE systematic trading strategy
- Ranked by out-of-sample CAGR - highest wins
- You don’t know the test dates - design for all regimes
- Max drawdown over 60%: eliminated. Fewer than 30 trades: eliminated. Negative returns: eliminated
- Be specific. “Buy when momentum is strong” is not a rule. “Buy when 12-month return minus 1-month return is in the top 20% of the universe” is a rule
I gave them the full instrument universe - 27,500 equities, 162 futures, 71 forex pairs, full options chains, treasury yields, FRED macro data, VIX term structure, Morningstar fundamentals. Everything QuantConnect offers. No pre-filtered list based on my prior research. Clean slate. No fingerprints.
Eighty prompts went out. Eighty designs came back in about two hours.
What 80 Experts Designed
| Approach | Count | Description |
|---|---|---|
| Momentum / Trend | 21 | Everything from 2-day mean reversion to 384-day moving averages |
| Value / Fundamental | 12 | Screen-and-hold using Morningstar data |
| Macro / Regime | 12 | Yield curve + VIX + allocation rotation |
| Factor Composite | 7 | Multi-factor models combining value, momentum, quality |
| Pairs / Relative Value | 6 | Z-score spread trading on ETF pairs |
| Multi-Strategy | 5 | Ensembles of multiple sub-strategies |
| Options / Volatility | 4 | Variance risk premium harvesting |
| Event-Driven | 4 | Earnings, mergers, insider signals |
| Risk Parity | 3 | Equal-risk weighting across asset classes |
| Carry | 3 | FX and bond yield harvesting |
| Mean Reversion | 3 | Short-term oversold bounces |
No single approach dominated. That alone was a win - I’d broken free of the macro-regime gravity well that consumed 45 sessions of directed research.
The Characters Stayed in Character
Nassim Taleb submitted exactly what I predicted. The Antifragile Barbell: 85% Treasury bonds, 15% in deep out-of-the-money SPY calls. Buy lottery tickets every month. Let most expire worthless. Wait for the world to fall apart.
His “Why This Wins” section was pure Taleb:
“Every other strategy in this competition is either exposed to left-tail ruin or leaves the right tail completely unmonetized… Most years this strategy underperforms a simple equity index. The years it does not - it wins decisively.”
He also noted that the scoring methodology was philosophically questionable. Called it.
Warren Buffett submitted a concentrated quality portfolio. Twenty stocks. High ROE, low debt, buying back shares. No leverage, no derivatives, no shorting. His risk management: “Price drops without fundamental deterioration are buying opportunities, not exit signals.” And if fewer than five stocks pass all quality gates? Go entirely to cash. “The inability to find qualifying investments is itself a market signal.”
Jim Simons built a multi-signal ensemble combining six weak predictors - medium-term momentum, short-term mean reversion, volatility regime, 52-week high proximity, yield curve signal, and VIX regime. No single signal is strong. The ensemble is. Exactly what he’s been saying for 40 years.
The Fictional Characters Got Creative
The real personas stayed faithful to their published philosophies. The fictional characters - freed from representing a specific historical figure - got weird in the best way.
Viktor Petrov submitted “Antifragile Momentum.” The twist: always hold a 10% “crisis seed” in inverse ETFs, even in calm markets. It costs 2-3% per year in drag. But when his regime detector fires - VIX term structure inversion + realized vol spike + broken 200-day trend - the portfolio flips to 40% short equity, 35% long bonds.
His non-negotiable rule: “The crisis seed is never reduced below 5%. It is not a tradeable position - it is permanent insurance. Derived from the military intelligence principle: you always maintain a reserve, even when you think you don’t need it.”
Elena Volkov submitted the most technically sophisticated design in the set. She uses VIX futures contango as the primary regime gate, caps her short-vol exposure at 15% - knowing full well that SVXY lost 90% in a single day during the 2018 Volmageddon. At her allocation, that event costs about 13.5%. Painful but survivable. Every other vol strategy in the competition allocated 40%+ and would have been wiped out.
Then the Convergence Started
This is where it got really interesting. These personas can’t see each other’s work. They received identical prompts. So when multiple personas independently arrive at the same approach, that’s a signal.
Six personas independently designed futures trend following. Carver, Covel, Dennis, Hite, Parker, and Reeves all submitted some variant of “trade 12-28 diversified futures using moving average crossovers with volatility-adjusted sizing.” The core logic was nearly identical. The details differed - Carver used 12 exponentially weighted crossover pairs, Dennis used Donchian breakouts, Reeves used signal-to-noise weighting - but the thesis was the same.
Six independent minds. Same answer. That’s the strongest signal in the entire competition.
Eight personas designed macro-regime SPY/bond rotation. Dalio, Druckenmiller, Faber, Marks, Merton, Silver, Tudor Jones, and Zweig all submitted “use the yield curve and VIX to decide how much to put in stocks vs. bonds.” This was also the approach that dominated my 45 sessions of directed research.
The cluster confirmed my bias… and simultaneously suggested it might be a real edge, not just a fixation. Sometimes the thing you keep coming back to is the thing that actually works.
Who Advances?
After filtering for feasibility, I cut to 28 for Round 2 - where they get coded and backtested on real data.
Cut outright: Taleb’s barbell (options execution unreliable at daily resolution), Ackman and Paulson (same problem), Kyle Bass (FX carry has failed five times in my prior work), and a few others I couldn’t implement cleanly.
Dead on arrival: Ken Griffin, Euan Sinclair, and Myron Scholes all loaded 40%+ into SVXY - the ETF that lost 90% in a single day. At those allocations, one historical event eliminates them instantly.
Not enough trades: Antonacci rotates the whole portfolio 3-6 times per year. Burry trades quarterly. Fine strategies, not enough data points.
Twenty-eight designs advance. Ten favorites, ten strong contenders, eight conditional entries needing modifications. The convergence clusters are intact. The dark horses are in.
Next: Twenty-eight strategies meet reality. Some of them won’t survive the encounter.