OpenAI released a guide to Codex: how to set up the workspace, projects, and task flows

On April 23, 2026, OpenAI published a material called Working with Codex in the Academy — a basic guide for those just starting to use the agent for working with files and tasks. The text explains not theory, but the most practical things: how the workspace is structured, what a thread and project are, where files are located, and how to give Codex clear instructions.

What's Inside Codex

At the center of the interface is a sidebar menu through which the user switches between threads, projects, and tools. OpenAI suggests thinking of a thread the same way as a regular chat in ChatGPT: it's a separate context for working on a task. A new thread can be left standalone or immediately placed into a project. The same place has a search for past tasks and projects, and next to it — plugins that help launch repeatable processes and connect external tools.

The guide places a special emphasis on the dialog window. OpenAI directly states that users don't need a perfect prompt or complex technical instructions. Codex works better when you describe the goal, needed files, and readiness criteria in human terms. This is an important signal: the product is positioned not as a narrow IDE function for specialists, but as a work interface that you can enter in plain language and gradually clarify the task.

"You don't need a perfect prompt or technical instructions."

Projects and Folders

The key organizational unit in Codex is the project. Each project is linked to a specific folder on your computer, and during local work, the agent makes changes only within this directory. If a user switches to a different group of files, OpenAI advises creating a new project rather than mixing everything in one context. It's also important that the starting folder can be empty: a project is needed not only for existing code, but as a place where new documents, scripts, and results will be stored.

From this model follow quite practical advantages:

• it's safer to limit the area of changes to one folder
• it's easier to separate client, personal, and experimental tasks
• it's easier to find past threads through project linkage
• new files are immediately created in a predictable location

Essentially, OpenAI is suggesting that users not just chat with the agent, but set work boundaries for it in advance. For teams and freelancers, this reduces the risk of confusion when the same agent is simultaneously helping with different projects. For beginners, it's also a convenient way to understand exactly where Codex can change something and where it shouldn't.

How to Manage Tasks

The main difference between Codex and a regular chat, according to OpenAI, is that the agent doesn't just respond with text, but can take action. Depending on access rights and task type, it checks files, creates new ones, edits documents, structures information, and helps assemble the finished result. In other words, the dialog window here serves not as a showcase for the model, but as a control panel for real work with local materials.

If a task is already running and the user realizes that the formulation needs adjustment, the process doesn't need to be stopped and restarted. For this, Codex has a Steer command: it allows you to adjust course on the fly, for example rewriting a paragraph instead of deleting it.

OpenAI also separately notes that many users run multiple tasks in parallel — one might be cleaning a table, another collecting a project tracker. The only limitation: long processes may stop if the computer goes into sleep mode.

What This Means

For OpenAI, this is not just an instructional text, but a way to explain a new model of interaction with the agent: less focus on prompt magic, more on work structure, files, and iterative management. For users, this lowers the barrier to entry for Codex and shows that the product is designed for everyday work tasks, not just demonstrating AI capabilities.