Note
May 31, 2026
2 min read
Codex Shouldn’t Have to Steal Focus to Get Work Done
By Cristiano Pierry
Local AI agents need per-thread visual workspaces so parallel tasks can run without competing for the user’s desktop focus.

I’ve been using Codex more heavily across multiple projects in parallel: creating PRDs, building presentations, prototyping, running evals, and more. And I keep running into one workflow issue.
Codex needs per-thread UI workspaces.
The closest mental model is adding a new desktop on macOS through Mission Control. Each Codex thread that needs to interact with desktop apps should get its own agent-owned Space, or at least a Space-like environment.
That workspace should belong to the thread.
Apps launched by that Codex thread should open there. Browser windows should stay there. Native app testing should happen there. Screenshots should be captured from there. Clicks, typing, window focus, and app navigation should be scoped to that workspace as much as possible.
Right now, when multiple agentic tasks are running locally, they can end up sharing the same visual workspace I’m using. Codex can open apps, interact with windows, move focus, and generally compete with me for the same desktop surface.
That is fine for a single assisted task.
It breaks down when you are running multiple projects in parallel.
We already understand this pattern in code. Codex can use worktrees so parallel coding tasks do not step on each other. The missing equivalent is UI isolation for local computer use.
In other words: not just one code workspace per thread, but one visual workspace per thread.
To be clear, this is probably not trivial. macOS Spaces are a good user-facing analogy, but they are not full sandboxes. The hard part is making the agent’s UI actions feel contained: screenshots, focus, permissions, app state, browser state, and system prompts all need careful handling.
But from a workflow perspective, we needed this yesterday.
As agents become more capable, the question is no longer just whether the agent can write code, run tests, or debug an issue. It is whether multiple agents can work alongside me without taking over my environment.
A shared desktop becomes a shared resource.
And shared resources create contention.
The ideal workflow is not: “Codex, take over my screen while I stop working.”
It is: “Codex, go work in your own workspace, keep me updated, and ask when you need approval.”
That feels like a major unlock for local agentic workflows.
This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.