AI Can't Recreate the Thrust Game (But It Can Help You Understand It)
a day ago
- #AI Tools
- #Retro Gaming
- #Game Development
- The author attempted to use Claude AI to recreate the classic 1986 game Thrust in the browser but found the initial result was 'slop' and not playable, lacking the authentic feel.
- Through a process of 'archaeology,' the author used Claude to interrogate disassembled 6502 assembly code, which helped generate detailed specifications for subsystems like physics and sound, leading to an accurate recreation.
- Key to matching the original game's feel was replicating its specific physics timing, including a gated update cadence (6 active slots per 16-tick window) and asymmetric drag, rather than just using the same constants.
- For authentic sound, the author emulated the BBC's SN76489 sound chip and MOS interface in an AudioWorklet, rather than sampling, ensuring precise timing and envelope shapes.
- Graphics and levels were extracted from the original source, using hard-coded sprites for ship rotation and the original's scanline-parity fill for terrain, while demo mode was replicated via timed keypress injection.
- The author concludes that while AI could not directly recreate Thrust, it was invaluable for deeply understanding the original systems, and the recreation process fostered appreciation for the original developers' achievements.
- The article also mentions the author's separate project: a turn-based 4X strategy game built from scratch with a custom WebGPU engine, eight civilizations, and an AI opponent, playable directly in the browser.