The Death of Sub-agents in Claude and Codex

A recent article on Habr discusses the evolution of workflows when using LLMs like Claude and Codex. The author notes that the traditional model of launching separate 'sub-agents' for parallel tasks is becoming obsolete. Modern approaches allow existing open sessions to handle assignments, return results, and share context between processes. This significantly simplifies interaction with models, making agent architecture more flexible and efficient. Instead of creating new agent instances for every sub-task, the system maintains a continuous data flow, saving resources and time. The author emphasizes that for a deep understanding of how these systems work, it is often best to ask the models themselves for explanations; however, the article provides a valuable overview of the paradigm shift in AI-agent development for those who want to explore the details independently.