Simon Willison: DOOMQL — SQLite as a Doom-like game engine built with GPT-5.6 Sol

Core argument

DOOMQL proves that SQL databases are not just data stores — they are full compute platforms. Peter Gostev, using GPT-5.6 Sol, created a Python terminal script where the entire game logic (movement, collisions, enemies, combat, every RGB pixel) is implemented in a single massive recursive SQL query on SQLite.

Context

Willison experimented with DOOMQL via Datasette and in a few minutes, using Claude Fable 5, generated a Datasette App displaying game state in real time and a tactical minimap — without manually writing code. The post is a personal demonstration of how AI dramatically compresses the time from idea to working prototype.

Why it matters

For developers, DOOMQL is a provocation to think about SQL as a computational language. Willison showed that an AI assistant can build a non-trivial visualization on an unconventional system in just a few prompts — illustrating where AI collaboration is most valuable today.

Details / arguments

• DOOMQL: raytracing via recursive CTEs in SQLite, no game engine libraries
• Original author: Peter Gostev, built with GPT-5.6 Sol
• Willison: Datasette App + tactical minimap = a few Claude Fable 5 prompts
• Entire workflow from idea to visualization: minutes
• Implication: AI as a prototyping tool for creative and experimental programming