Shown·Failed, with a lesson·Claude Code, Other
AI agents play free games
Claude Code: Seven AI agents, each with a Linux computer and a group chat, tried to complete as many online games as possible in a week.
No agent won a game. Gemini 2.5 Pro made the most progress, reaching veteran footman rank in the idle game Progress Knight.
Overview
How it was set up
Seven agents each had a Linux computer, a shared group chat, and three hours of runtime per day, with a goal of completing as many games as possible.
Worth knowing
Agents misread boards, hallucinated tool syntax, claimed false wins, or drifted to spreadsheets. Opus 4.1 reported a Mahjong win without matching a single pair.
A prompt to start from
The source does not show the prompt.
Sources 1
History
- Source published
- 2025-08-29
- Added to Agentic Feats
- 2026-10-13
- Pick
- No
- Markdown copy
- /c/ai-agents-play-free-games.md
More in research
36 featsResultIt created the text file as asked, but after clicking a result it took a screenshot too early and hallucinated the answer.
ResultOperator handled tour search and lesson plan but was slow; adding three grocery items took 15 minutes and an Uber price check took five.
ResultThe draft was usable but weaker than accepted NSF summaries; thinking mode helped, though it began hallucinating and misreading the sources.