Parallel Tasks, Worktrees, and Subagents
Explore how to increase development efficiency by running parallel tasks in OpenAI Codex. Understand the role of projects, threads, and Git worktrees for isolating work, and learn to use the Handoff feature to move tasks between local and background execution. Discover how subagents enable concurrent work within a single task to handle complex workflows more effectively.
A typical development session follows a familiar rhythm: one task runs to completion, then the next begins. When we introduce an agent capable of executing work autonomously, that rhythm no longer has to be sequential. The bottleneck in a sequential model is not the quality of the work. It is the waiting. While one task runs, everything else stands still.
Codex can run multiple tasks simultaneously, but the benefit only materializes when we change how we scope and assign work. Running two parallel tasks on the same files invites conflicts. Running two parallel tasks on non-overlapping scopes is safe, fast, and the pattern Codex is designed for. The mental model shift is not about managing more work. It is about thinking in independently bounded units before any task starts.
This lesson covers the mechanisms Codex provides for parallel ...