Summary
This video from the All About AI channel demonstrates how to use Codex as an agentic coding assistant to build an automated expected-value detection system for prediction markets like Polymarket and Kalshi. The creator frames the project around a core insight: rather than betting on who wins, finding pricing discrepancies between bookmakers and decentralized prediction markets is where long-run edge lives.
The walkthrough uses a real Arsenal vs. Manchester City match as a worked example, manually walking through odds normalization — converting bookmaker decimal odds to implied probabilities, stripping the overround margin, and comparing the result against Polymarket and Kalshi prices. A 2% gap in the draw probability between Bet365 and Polymarket is surfaced as an illustrative edge. The creator then prompts Codex to build a system that connects via WebSocket to both Polymarket and Kalshi for live price feeds, accepts bookmaker odds as input, normalizes them, and flags potential mispricings autonomously.
The resulting agent is prototyped live in the video, successfully fetching live bid/ask prices from both platforms for the Arsenal match and beginning the comparison calculation. The creator notes that a production version would connect directly to bookmaker APIs rather than requiring manual input. For developers interested in applying AI agents to financial data tasks or building autonomous market-monitoring tools with Codex, the video offers a concrete, replicable starting point.
📺 Source: All About AI · Published August 13, 2026
🏷️ Format: Hands On Build
