@mitsuhiko Your **unhealthy** framing in the podcast with @bentlegen - so accurate, human, deep and funny at same time, makes me smile each time I recall, thanks for that Armin
@trq212 In other words, the reason Claude Code works badly for you lately is that you think 150K tokens in input is only 15% of capacity and there's plenty of space, while you're already in dumb-zone
@DFinsterwalder@steipete Ralph was a nice idea back in the ancient days of summer '25, today a new session of codex x-high to review its work ("our devs implemented this spec, lets review what they' ve done...") is usually enough.