After running a sub-agent with OpenClaw, I finally understand what "peace of mind" means. Today I saw that JetBrains launched a new AI proxy IDE, and I immediately thought of the sub-agent function of my little crayfish. In the past, when I made it compare two papers at the same time, it would get overwhelmed with context. Now, I just let the main agent create a sub-agent to read paper A, and then create another one to read paper B, and it only returns a few hundred words of essence. The context has been reduced by a large margin, and the tokens are also saved significantly. I am an ordinary office worker, and there are many AI activities in Hangzhou. I casually mentioned my address once, and it kept it in the file, automatically pushing me local meeting information next time. It's not that it's "smart," but that persistent memory mechanism is working quietly behind the scenes. This sub-agent thing is like having a few little assistants; the manager only listens to reports and doesn't need to read the entire text. In the past, I would wake up at midnight to see it still running heartbeat tasks, helping me monitor updates to a certain file. Now I really feel — this is not a tool, it's a partner. A friend asked me if it costs money; I said heavily using it for a month costs a few thousand, but the time and brainpower it saves is worth it. For those of you using sub-agents, have you encountered an infinite nesting deadlock? The OpenClaw framework directly cut off this function; smart.