Tag
ProgramDistill introduces a scalable benchmark for evaluating coding agents by having them implement features discovered through interaction with fully functional reference web applications, using an automated pipeline to create verifiable tasks.