John Lam Compares AI Agents' Factorio Progress Over 18 Months

John Lam revisits his 18-month-old observation that Sonnet 3.5 could complete Factorio benchmarks, and now demonstrates Codex playing the game via a headless client and xdotool on a Linux VM, highlighting rapid agentic AI progress.

2026-09-11 ~ 2026-09-11 · 2 related posts

• Agentic AI's leap: from Sonnet 3.5's Factorio benchmark to today, 1.5 years later — john_lam · 2026-09-11
• How Codex plays Factorio: headless Lua state reading plus xdotool GUI control — john_lam · 2026-09-11