Codex adds cloud environments that retain project setup
OpenAI introduced reusable cloud environments for Codex that preserve a project’s repositories, dependencies and access settings. Developers can start separate tasks from the saved setup and resume work across devices. Teams may share an approved environment, while task workspaces remain separate. The announcement does not include an independent measurement of setup time or code correctness.
Artificial Intelligence··Morning
Project setup carries into later tasks
OpenAI introduced reusable cloud environments for Codex, its software engineering agent, at DevDay on September 29. A developer can prepare a project’s repository, tools, dependencies and access settings once, then start later tasks from that saved setup. Codex already ran remote tasks; the change is that project preparation can persist between them. OpenAI describes this as a way to avoid repeating setup work, but the announcement does not supply an independent comparison of task startup time or code correctness.[1], [2]
Each task retains its own workspace
A prepared environment can be reached from desktop, web or mobile. Each new task still starts in its own isolated workspace, rather than joining the working files of other tasks created from the same environment. A developer can reopen an individual task from another device with that task’s state intact. This keeps a shared setup distinct from separate code changes. OpenAI also says cloud tasks can continue while the developer’s computer is asleep, because execution takes place remotely.[1]
Teams can share approved configurations
OpenAI says teams can share environments containing approved project settings and permissions. Colleagues can start their own tasks from that configuration without gaining access to one another’s ongoing work. Permission to edit the shared setup is managed separately. The release therefore separates reusable project preparation from each task’s code changes. Availability and administrative controls can depend on the account and organization, so the announced capability does not imply identical access for every developer.[1]
Related columns
For more information on this topic, you can read the related columns.
