Claude Playing Crusader Kings 3
By Dillip Chowdary • Jul 22, 2026 • Source: HN Claude/Codex/Fable
Claude is playing Crusader Kings 3 on a live Twitch stream at twitch.tv/skullbloc. The run was posted on Hacker News under a Claude/Codex/Fable thread (item 49009304) and had drawn 2 points and 1 comment at the time of the summary.
Crusader Kings 3 is a long-horizon grand strategy game: dynasties, succession, diplomacy, intrigue, and multi-turn planning. Watching Claude on that board is less about a single click and more about whether an agent can keep goals, constraints, and side effects coherent across a session. The stream is the artifact; the interesting part is the control loop that turns model output into in-game actions and reads the state back.
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
For engineers and builders, this is a public stress test of agentic play in a domain where partial plans fail loudly. CK3 punishes short memory, brittle tool use, and goals that drift. If you are shipping agents that plan, call tools, and recover from bad steps, a grand-strategy stream is a harsher demo than chat Q&A or a short puzzle.
The Claude/Codex/Fable framing puts this next to other model-and-agent experiments people already compare on HN, not next to traditional game AI leaderboards. With only 2 points and 1 comment so far, it is still a niche signal, not a crowded benchmark race. The useful comparison is qualitative: can the agent stay coherent under open-ended strategy pressure, and how does that look beside other model-driven game runs.
What to watch on the stream and in follow-up discussion: how often Claude loses the thread on succession or alliances, how recovery looks after a bad move, and whether the setup is a thin wrapper around the model or a fuller agent stack. Treat the Twitch link as the primary source of truth; the HN thread is only the discovery surface until more detail lands there.
Advertisement
🔎 More interesting news
- Arcee, a US open source AI lab, says Chinese models are not inherently dangerous
- I Had 11 AI Subagents Test Every PPTX Skill for Claude Code. Here's What Works
- Apple is reportedly testing a MacBook Neo with more RAM
- Here’s what Samsung’s smart glasses actually look like
- Today's full Tech Pulse briefing →