Kristian Fagerlie distinguishes a trading system built with AI from one that calls a model for every trading decision. The experiment uses ordinary code to scan markets, score conditions, open qualifying positions and monitor exits. Kristian Fagerlie describes liquidity, recent price movement, momentum and bounce signals as components of this particular setup, not a validated general strategy.
Kristian Fagerlie asks GPT-5.6 Sol to examine roughly a day's collected data. The model favors operational changes such as retries, emergency exits and better data collection rather than a major strategy rewrite. Kristian Fagerlie reports passing tests and plans to gather many more trades before evaluating proposed changes.
Kristian Fagerlie explains that continuously running model-based monitoring consumed too many tokens, prompting a move to autonomous coded execution. The early profits are short-window, self-reported results, and Kristian Fagerlie explicitly notes leverage risk, losses in a separate experiment and the possibility that gains could reverse or the account could lose everything. The video documents an experiment, not evidence of durable profitability or investment advice.
Watch on YouTube
