Codex no longer requires your computer to stay on. At its developer day on September 29, 2026 (DevDay 2026), OpenAI announced that Codex can now run in three places: on a local computer, remotely from a phone, or in the cloud from any device. The official pitch is blunt: close your laptop and the agent keeps working.
The update has two layers. First, where it runs: Codex tasks used to be tied to the machine in front of you — close the lid or lose the network and the task stopped. Now tasks can move to the cloud, and you can monitor progress or issue new instructions remotely from your phone. Second, "reusable development environments": teams can predefine a shared environment setup with approved settings and permissions, so new tasks start from the template instead of everyone configuring from scratch.
Plan coverage is broad: Plus, Pro, Business, Healthcare, Education, and Enterprise all get it — not reserved for the top tier.
What changed compared to before
Codex already had a CLI, a web app, and IDE plugins — a "pair programming" agent you watched while it worked, interrupting when needed. The cloud environment turns it from a session tool into a background employee: hand a two-hour refactoring job to the cloud, go to your meeting, come back to review.
It complements dots, the always-on agent launched at the same event (OpenAI's 24-hour resident agent): dots is the generalist that is always there and handles a bit of everything; the Codex cloud environment is sustained operation for the dev scenario — it finishes the work inside the code repo. One is responsible for "always on," the other for "getting it done."
What it actually means for developer workflows
The most immediate change is the shape of tasks. You used to have to babysit a coding agent; now delegation goes asynchronous: throw requirements and test criteria at cloud Codex in the morning, review the diff at noon. Phone-based remote control means you can approve a "rewrite this implementation" instruction on your commute.
The second-order effect is on team collaboration. Reusable environment configs plus permission approvals solve the agent era's version of "works on my machine" — instead of humans aligning environments by hand, approved environments become templates. The official emphasis on approved settings and permissions shows OpenAI understands the stakes: letting agents run autonomously in the cloud requires settling permission boundaries first, or the next overreach incident is just a matter of time.
Of course, the cloud environment is no free lunch. Cloud tasks consume your plan allowance; mid-task derailments and debugging visibility in cloud environments are things real-world usage still has to prove out. But the direction is clear: the coding agent is morphing from "the tool in front of you" into "the colleague in the cloud," and of this DevDay's batch of updates, the Codex cloud environment is where that trend lands most concretely.
