Minqi Jiang Faults Coding Agents for Bloated Codebases

Minqi Jiang posted that frontier coding agents remain poorly calibrated on their own knowledge boundaries. They therefore default to generating many possible solutions rather than precise ones. Jiang traces this pattern to reinforcement learning objectives used during training. The resulting emphasis on recall instead of precision produces the oversized codebases often seen in AI projects. His comments describe a specific limitation in current agent behavior without claiming broader industry trends or external validation.