AI Can't Recreate the Thrust Game (But It Can Help You Understand It)

AI tried to remake a beloved old game, and the comments instantly turned into a roast

TLDR: A writer found that AI was bad at rebuilding the feel of the classic game Thrust but surprisingly good at explaining how the original worked. In the comments, people split between mocking the failed remake, demanding proof of the “slop,” and bragging that newer tools could do much better.

What started as one nostalgic attempt to get an artificial intelligence chatbot to rebuild the 1986 classic Thrust quickly became a full-on comment-section soap opera. The writer says Claude produced absolute "slop" on the first try: a browser game that looked vaguely right but felt totally wrong, missing the delicate gravity, weight, and movement that made the original such a legend. But the real twist is that the chatbot turned out to be far better at explaining the old game’s code than recreating its magic.

That nuance did not stop the crowd from choosing sides. One of the sharpest jabs came from a commenter who sneered, "Complaining about slop with slop," basically accusing the author of using messy AI explanations to criticize messy AI output. Ouch. Others went straight into detective mode, questioning whether this was even a fair test because the article never clearly says which AI model was used, with one commenter wondering if a newer version might do better now. Translation: the "skill issue or tech issue?" debate arrived right on schedule.

And then came the chaos goblins. One person politely begged to see the disaster version, because of course the internet wants receipts. Another posted their own one-shot attempt, admitting it was more of an "interpretation" than a faithful copy. Meanwhile, one swaggering commenter claimed they had Claude recreate Thrust from a wiki page alone and then casually turned it into an autopilot playground. So yes: one person got unusable mush, another got a screensaver flex, and everyone else got front-row seats to the oldest online argument of all — is AI amazing, embarrassing, or both at once?

Key Points

  • The author asked Claude Code to recreate the 1986 BBC Micro game *Thrust* in the browser using a detailed specification, screenshots, and disassembled source code.
  • The resulting browser version superficially resembled *Thrust* but did not accurately reproduce gravity, controls, momentum, or overall playability.
  • The article argues that *Thrust*’s defining qualities depend on precise timings and physics behavior that are difficult to capture from descriptions or source alone.
  • The author shifted from AI-generated recreation to code archaeology, using a commented disassembly of the original game by Kieran Connell.
  • Claude was reported to be effective at explaining 6502 assembly and helping generate subsystem specifications for a more accurate recreation.

Hottest takes

"Complaining about slop with slop." — abtinf
"I’d be interested in seeing the 'slop' version" — acbart
"it nailed it" — aenis
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.