Armin Ronacher documents a Pi coding-agent tool-calling regression where newer Claude models generate malformed nested edit calls, arguing that closed-source Claude Code training and permissive harness behavior can make alternative agent tools less reliable without stricter schemas.
Better Models: Worse Tools
Armin Ronacher documents a Pi coding-agent tool-calling regression where newer Claude models generate malformed nested edit calls, arguing that closed-source Claude Code training and permissive harness behavior can make alternative agent tools less reliable without stricter schemas.
Source: Pocoo