Summary
In this speedrun-format video from the All About AI channel, the host demonstrates how quickly a working AI-powered quantitative trading bot can be built from scratch using Codex running GPT-5.6. The target market is Bitcoin’s up/down 15-minute contract on Kalshi, a US-regulated prediction market platform. Starting with raw historical BRTI data, the host uses pre-built Codex skill files — a structured quant research prompt and a hypothesis-testing workflow — to generate five candidate trading strategies through automated analysis of the price data.
After running backtests, the most promising strategy is a volatility persistence filter: it checks whether the previous 15-minute BTC move exceeded 15.4 basis points (top-third historical volatility), then trades directionally based on the 5-minute midpoint, targeting approximately 32 trades per day. Codex then writes the execution code using the Kalshi API with WebSocket order submission, a fill-or-kill order type, and a $5 per trade size against a $120 funded account. The bot is deployed live and left to run overnight.
The video is a practical illustration of how AI coding agents like Codex can compress a research-to-deployment cycle that would normally take days into a single session. It also shows the limitations — lost recordings mid-session, three strategies rejected out of five — giving an honest look at what “vibe coding” a trading system actually looks like in practice rather than in a polished demo.
📺 Source: All About AI · Published August 22, 2026
🏷️ Format: Hands On Build
