I’ve been using Codex across multiple projects in parallel: creating PRDs, building presentations, prototyping, and running evaluations. One workflow problem keeps interrupting that work.

Codex needs per-thread UI workspaces.

The closest mental model is a desktop on macOS through Mission Control. Each Codex thread that needs desktop apps should get its own agent-owned Space, or at least a Space-like environment. Apps launched by the thread should open there. Browser windows and native-app testing should stay there. Screenshots, clicks, typing, focus, and navigation should be scoped to that workspace as much as possible.

Right now, multiple local agent tasks can share the same visual workspace I am using. Codex can open an app, move focus, interact with a window, or surface a system prompt in the middle of my own work. That is manageable during a single assisted task. It breaks down when several projects are running at once.

We already understand the isolation pattern in code. Worktrees keep parallel coding tasks from stepping on one another. Local computer use needs a visual equivalent.

A thread needs both a code workspace and a visual workspace.

The macOS Spaces analogy is not a complete technical answer. Spaces are not full sandboxes, and the difficult details include screenshots, permissions, app state, browser sessions, focus, and system prompts. The user also needs a clear way to see what each agent is doing and take control when necessary.

But the desired interaction is straightforward. I should be able to keep working in my environment while an agent uses its own. The agent should report progress and come back when it needs a decision or approval, not repeatedly take over the surface where I am working.

That kind of isolation would turn parallel local agents from a focus-management problem into a usable workflow.