The ENPIRE framework enables robots to autonomously progress without human intervention. By integrating large models such as Codex and Claude Code into robot clusters, it automatically generates motion control programs, with live cameras determining success or failure and independently analyzing logs to correct code. In millimeter-level tasks, the success rate reaches 99%. When expanded to 8 units, branch models share the optimal algorithm via Git, reducing training time from 1.5 hours to approximately 40 minutes. However, the effective actuation rate on a single machine is as high as 85%, which drops to 35% when 8 units run concurrently due to waiting for logs, code, and API responses, with synchronization increasing token consumption. Related code will be open-sourced soon.