Build Agent Workflows Around State Ownership, Not Timing

This article discusses a key lesson in building reliable agent workflows: coordinating with process state ownership rather than relying on timing. Using a Codex workflow example—moving existing chats into saved Projects—it shows how patching local state directly fails because the running app overwrites changes. The solution involves launching a helper process that waits for the app to exit, applies migrations while no process owns the state, then reopens the app. The article covers splitting planning from mutation, resolving user intent into a move plan, and generalizing the pattern for any app that owns state files. It is relevant to IT/technology, focusing on software engineering, automation, and best practices for agent workflows.