When AI Learned to Play: From Pac-Man Ghosts to LLMs Beating Factorio

We trace the wild evolution of game AI from the simplest pre-programmed enemies to modern LLMs that can actually think their way through complex video games. Starting with Pac-Man's four ghosts in 1980 and their surprisingly clever personality system, we level up through fighting game CPUs, RuneScape bots, and finally arrive at the bleeding edge: MCP servers that let large language models perceive, reason about, and interact with PC games in real time. 00:00:00 - Introduction and the big question: is game AI actually thinking? 00:01:15 - Tier 1: The pre-programmed spellbook era (1970s-80s, Pac-Man ghosts) 00:03:30 - Tier 2: Fighting game CPUs and input-reading difficulty scaling 00:05:45 - Tier 3: Scripted bots and RuneScape's gather-click-repeat loops 00:07:30 - The leap: from scripted behavior to actual reasoning with LLMs 00:09:00 - How MCP servers work as the bridge between LLMs and games 00:11:00 - The mechanical loop: perceive, reason, act, evaluate 00:12:30 - Concrete example: LLM-driven Factorio automation and planning 00:14:15 - Tools, flows, and emerging industry standards 00:15:30 - Where this is all heading and final thoughts This podcast episode was fully generated by AI — research, script, voices, and production. Built with Claude, Piper TTS, and automated pipeline tooling.