Claude Opus 5.5 took 90,000 screenshots of its own game over two days to polish the visuals
prasenx · x · 2026-10-06
User prasenx shared a demo where Claude Opus 5.5 screenshotted its own game — about 90,000 captures over two days — and kept fixing the visuals based on them, with the human only pointing out what looked wrong. The result is a striking showcase of iterative self-critique on visual output.
More from coding & agent
- Speed optimizations are both the best and worst agent benchmarks, researcher argues — ducha_aiki · 2026-10-06
- One Photo to Talking-Head Reel: Dev Shares $0.50/Video AI Pipeline — victor_explore · 2026-10-06
- Claude Opus 5.5 Built a Full City-Building Game Using Under a Quarter of Weekly Limit — isnit0 · 2026-10-06
- Two years of vibe coding: why the Cursor+Lovable+Supabase stack breaks for real businesses — Fun-Tangerine5264 · 2026-10-06
- HyperBrowseComp: a 423-question, 13-language stress test for web-browsing agents — zmkzmkz · 2026-10-06
- Cloudflare launches cf, an agent-first CLI to query observability data via the API — dinasaur_404 · 2026-10-06