The order for splitting off agents that run on their own

An operating sequence is presented with actual screens: use a single main agent for a long time, and only after enough work has built up, split off independent agents by task. The constraint of today's coding agents is summed up as a dilemma (memory breaks off at each folder, but mixing everything in one place lowers output quality), and a setup that tries to get past it by accumulating memory in local files and skill documents is introduced. Under it sit four sub-agents for review, research, insight and writing that produce drafts through scheduled runs at a set time every day, and what stands out is that human review right before publishing was deliberately brought back because the output involves judgment. It also warns that creating many agents means adding more points to manage, and notes that work has to be broken into fine-grained units before you can see where people should step in. However, the case is one person's single document-writing workflow, and any discussion of security and auditing is missing.

This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.