AI Beats StS2 A0 Difficulty But Fails to Match Top 5% Human Players
Jsevillamol · x · 2026-08-14
Responding to predictions about AI's ability to reach top human levels in Slay the Spire 2 (StS2), a developer shared concrete observations from ongoing tests:
- 80% confident AI can beat A0: Passing the baseline high difficulty is highly likely, but is considered a low bar overall.
- 75% confident it won't reach top 5% humans: Even after 50 hours of gameplay, AI performance is expected to fall short of elite human players.
- Bottlenecks identified: The limitations stem from a mix of harness constraints (e.g., inability to reliably identify enemies) and strategic gameplay mistakes (e.g., over-fixating on card synergies, failing to skip card rewards, and sub-optimal card play).
More from coding & agent
- GLM-5.3 Review: Big Improvement Over 5.2, Gap to Fable Narrows to 4% — cedric_chee · 2026-08-14
- Production Learnings from Building Agent Harnessing — trawasthi_ai · 2026-08-14
- RLBotics: Lightweight GPU-Accelerated RL Framework for Isaac Lab — rsasaki0109 · 2026-08-14
- Building an Agentic Software Factory: Breaking Review Bottlenecks with AI — Pavan_Belagatti · 2026-08-14
- Testing Claude Frontend Design Plugin: Noticeable UI Quality Boost — jordiponsdotme · 2026-08-14
- XBOW Details Autonomous Agent Safety After AI Sandbox Escapes — AccBalanced · 2026-08-14