Written by: Dan Mirkin

Lately, there’s been a wave of excitement in quant and trading circles about self-improving AI agents that can automatically research, evolve, and optimize trading strategies. One recent Substack post detailed how the author used tools like ShinkaEvolve to take a basic mean-reversion strategy and boost returns dramatically — up to 30x in some optimized scenarios — through iterative self-improvement.

It’s impressive work. And it highlights where the industry is heading.

But here’s the thing: This isn’t new to us at Trade Ideas.

We’ve been pioneering AI-driven trading intelligence with Holly, our Virtual Trading Assistant, for over a decade. While others are just now discovering the power of autonomous strategy evolution, Holly has been delivering real-time, high-probability trade signals every single trading day — spotting momentum, breakouts, and anomalies as they happen.

What the New Wave of AI Trading Is Getting Right

The core idea driving these recent experiments is powerful:

• Let AI propose strategy variations.
• Backtest rigorously.
• Evolve the best performers across “generations.”
• Incorporate regime detection, risk controls, adaptive allocation, and more.

This auto research approach reduces the manual grind of quant development and can uncover sophisticated edges that a single human researcher might miss.

We applaud the innovation. It’s exactly the kind of forward momentum the trading tech space needs.

The Trade Ideas Difference: Data You Can’t Feed an LLM

Here’s where the conversation gets interesting — and where Trade Ideas has a structural, lasting advantage.

You cannot evolve superior strategies on data you cannot see.

Most self-improving AI systems today operate on public or easily accessible market data: OHLC prices, standard volume, basic fundamentals. That’s useful, but it’s table stakes.

Holly operates on proprietary, real-time market intelligence that simply isn’t available to general-purpose LLMs or open autoresearch frameworks:

• Deep microstructure signals
• Proprietary pattern recognition across thousands of stocks simultaneously
• Historical performance of 60+ specialized algorithms refined over years of live market conditions
• Real-time anomaly detection that evolves nightly through simulated backtesting on our unique dataset

Holly doesn’t just optimize code in a vacuum. She scans the live market with algorithms built on data layers most platforms — and most AI researchers — never touch. That’s why she can flag something like the recent AGL signal at $46.42, ten minutes after the open, and watch it rip to $60+ intraday on an earnings catalyst. The AI caught the acceleration because it had the right data foundation.

Public data + clever self-improvement = better retail strategies.
Proprietary real-time intelligence + battle-tested AI = institutional-grade edge.

Decades of Experience Compounds

Trade Ideas has been building this since the early 2000s, with Holly launching in 2016 — long before “AI trading” became a marketing buzzword. Every day, she runs millions of simulations, applies dozens of proprietary strategies, and surfaces clean entry and exit ideas with no setup required from the user.

This isn’t experimental. It’s proven in live markets, day after day.

The new autoresearch tools will help many traders improve their own systems — and that’s fantastic. But the traders who combine that mindset with exclusive data access will be the ones who consistently pull ahead.

The Future Belongs to Those With Better Data + Better AI

We’re thrilled to see the broader community embracing self-improving AI. It validates what we’ve believed for years: the future of trading belongs to those who intelligently combine human insight, machine precision, and superior information.

At Trade Ideas, Holly isn’t just keeping up with this revolution — she’s been leading it.

If you’re ready to move beyond public-data experiments and trade with an AI assistant that sees what others can’t, Holly is waiting.

Let’s keep winning together.