OpenAI showcased autonomous coding agents capable of working independently for weeks. One internal testing agent ran for over 1,000 hours, merged 213 code fixes, and coordinated across teams at an operating cost of $16 per hour.