Blog — AI Crypto Research Notes
Notes, observations, and lessons from running an AI trading experiment. No financial advice, just research.
These posts were written in March 2026, while the experiment was still running and still expected to produce a meaningful sample. It did not. It ran for 54 days, executed 4 trades, and ended at -0.61% with almost no capital exposure, which is worse than doing nothing. The posts are kept as written, because rewriting your own predictions after the fact is exactly the habit this project was supposed to argue against. Read them as a record of what was expected, then read what actually happened.
Why Crypto Is the Best Testing Ground for AI Trading Research
Not all markets are equal for AI research. Crypto is narrative-driven, 24/7, and has public on-chain data — the ideal laboratory for LLM experiments.
What Happens When AI and Code Disagree on a Trade
Cortex uses two layers: AI proposes, code enforces. Here's why we don't let the AI have the final word — and what we learn when they disagree.
Why Our AI Says "Hold" 70% of the Time
Most AI cycles end with no action. That's not a bug — it's the entire point. Here's why doing nothing is the hardest decision to automate.
Can AI Actually Outperform Buy & Hold Bitcoin?
The simplest crypto strategy is also the hardest to beat. Here's why we chose it as our benchmark — and why you should watch how this plays out.