Multitasking with the Codex app

Key Concepts

• Worktrees: Isolated branches of a codebase managed within the application, enabling parallel task execution.
• Codex: The AI coding assistant used within the application.
• Parallel Programming/Workflow: Working on multiple tasks simultaneously, leveraging Codex to handle individual tasks within worktrees.
• PR (Pull Request): A request to merge changes from one branch into another, used for code review and integration.
• Context Switching: The ability to efficiently shift focus between different tasks or codebases.

Delegated Work & Parallel Workflow with Worktrees

The speaker discusses a shift in workflow facilitated by the introduction of worktrees within the application, allowing for increased delegation to Codex and parallel task execution. Previously, the speaker would often wait for Codex to complete a task before moving on. Worktrees eliminate this bottleneck by providing isolated environments for each task, managed independently by the application. This enables the speaker to continue working on other features while Codex addresses separate requests.

Example: Pin Task Sorting & Branch Creation Issue

A concrete example illustrates this workflow. The speaker initiated a task within a worktree on the master branch to add drag-and-drop sorting functionality to pinned tasks in the sidebar. After submitting the task to Codex, instead of waiting for completion, the speaker immediately switched to updating the “create branch” button.

While Codex worked on the pinned task sorting, an issue arose: Codex unintentionally created the branch twice. The speaker demonstrates the ability to provide feedback while Codex is still working on other tasks. A comment was added to the worktree asking Codex to explain the duplicate branch creation. This highlights the non-blocking nature of the workflow.

Concurrent PR Generation & Figma Integration

Simultaneously, Codex had been working on a separate task involving a modal, utilizing Figma designs provided as context. This resulted in a substantial Pull Request (PR) being generated. The speaker notes, “I was just working on a similar modal, and I also gave it the Figma designs we see right below here, right above. And it just finished up a nice big PR as well here.” This demonstrates Codex’s ability to handle complex tasks and integrate design specifications.

Testing & Applying Changes in Parallel

The speaker then applied the changes from the drag-and-drop PR to their local tree while Codex continued to review the initial pinned task sorting request. After applying the drag-and-drop changes, the speaker reviewed Codex’s work on the pinned task sorting, finding it “pretty good.” This showcases the iterative process of submitting tasks, receiving Codex-generated code, and applying/adjusting it.

The Shift in Mindset & Architectural Focus

The speaker emphasizes that adopting worktrees and parallel workflows fundamentally alters the approach to software engineering. As stated, “getting into the habit of really going all in on worktrees and working in parallel will completely change how you look at software engineering.” Instead of focusing on individual lines of code, the focus shifts to the overall architecture of the code.

Context Switching & Effective Workflow

The speaker acknowledges the difficulty of context switching, stating, “It’s pretty tough to completely switch what you’re working on.” However, they advocate for identifying good stopping points and strategically switching tasks to avoid waiting on Codex or attempting to complete everything manually. The key is to leverage Codex’s capabilities and maintain continuous progress by working on multiple aspects of the project concurrently.

Technical Vocabulary

• Local Tree: A local copy of the codebase on the developer’s machine.
• Figma: A collaborative web application for interface design.
• Modal: A window that appears on top of the main application window, often used for prompts or additional information.

Conclusion

The video demonstrates a powerful workflow enabled by worktrees and an AI coding assistant (Codex). By embracing parallel task execution and delegating work, developers can significantly increase their productivity and shift their focus from granular code details to broader architectural considerations. The ability to provide feedback and iterate on Codex’s output while simultaneously working on other features represents a substantial improvement in the software development process.