75% of AI coding agents break code over time

By Mr Anderson,

Found this interesting -

Alibaba's SWE-CI benchmark tracks AI coding agents across 100 real Python codebases, each spanning 233 days and 71 consecutive commits

75% of models break previously working code during long-term maintenance

Full story:
https://awesomeagents.ai/news/alibaba-swe-ci-ai-coding-agents-long-term-maintenance/

More posts