GitHub rewrote the Copilot engine: 832,378 lines of new code. Most of those lines were written by AI agents. GitHub itself reports this. The work did not go smoothly: the agents overloaded a laptop and picked up someone else's unfinished changes. Such stories are useful for those putting agents on real tasks. The scale is impressive, but the failures show that automation needs human oversight. Human code review remains mandatory.
Source: habr.com · post in Telegram