OpenAI's Coding Assistant Can Now Pick Up Where You Left Off
OpenAI introduced reusable cloud workspaces for Codex, its coding assistant, at its Dev Day on Tuesday. These are long-lived, shared work areas that teams can open from any device, instead of fresh temporary spaces for each job. TechCrunch
The event was DevDay 2026 in San Francisco on September 29. The workspace update was part of a larger Codex and platform release covering the command line tool, code review, security checks and agent tools.
The reusable workspaces are built to start jobs faster and give teams a shared setup with approved settings and permissions. Software parts, tools, access limits and safety rules do not need to be set up again each time. Saved work stays saved. Everyone starts from the same baseline.
Cross-device use is a large part of the plan. Codex is now in the ChatGPT mobile app, after an earlier release as a separate Codex app for handling many assistants at once. OpenAI has also explained how a ChatGPT account lets people move between a home computer setup and Codex cloud without losing work.
Reusable environments and everyday automation
The focus on reuse builds on earlier work OpenAI described using Runme and WebMCP with Codex. That work turned repeated engineering jobs into steps that can be checked and reused. The same idea applies to cloud moves, where Codex can write starter code and setup files such as Dockerfiles.
Think of it like a shared workshop where tools stay on the bench. Temporary assistants are quick to start and easy to understand. They do not remember team context well. Long-lived workspaces cut setup work but need upkeep, access checks and cleanup of old material.
CLI, delegation and review
OpenAI announced an updated Codex CLI alongside the reusable workspaces. The updated tool lets developers start and direct jobs with their voice. Voice only starts the job. The assistant still does the work inside the set workspace.
OpenAI added a new /agents view in Codex to make it easier to hand out and track many jobs at once. Developers can start several assistants, watch progress and step in without juggling separate windows.
OpenAI introduced a new Codex code review tool inside the ChatGPT desktop app for GitHub pull requests and GitLab merge requests. It sits where developers already check code changes. It supports both services without forcing a move to the browser.
OpenAI's automatic code reviews let Codex do a first review while users are away from their computers. The assistant scans the change, leaves notes and prepares next steps for human review on return. TechCrunch
Security Cloud
OpenAI launched Codex Security Cloud tools to help developers protect their systems. The system can scan whole GitHub code collections on demand, on a regular schedule, and when new code arrives.
After scanning, Codex studies the results, removes duplicates and prepares fixes in the cloud. Sorting and fixing is the slow part of most security work. Doing it where the code runs shortens the path from warning to proposed fix.
Codex Security Cloud includes models from OpenAI's Daybreak Blue security project without a separate application. Existing Codex users get the special models as part of normal use. TechCrunch
Decisions API and Agents API
OpenAI announced a new Decisions API for fast choices that uses Luna to answer user-set questions with preset answers. Developers set the questions and allowed answers. Luna picks one. This fits routing, gating and rule checks where free-form text would be risky.
OpenAI updated its Agents API to support computer use and the ability to use Amazon's Bedrock Managed Agents to build OpenAI assistants that run entirely on AWS. Computer use adds mouse, keyboard and screen navigation. The AWS option lets teams keep computing, data and growth inside AWS while still using OpenAI tools. TechCrunch
The broader context here is staying power plus control. Reusable workspaces, approved settings, planned security scans and preset answers point to assistants that run longer, touch more systems and need tighter checks. That is less like showing auto-complete. It is managing a fleet of workers.
In my view, the lasting change is not voice or phone access. It is treating the workspace as shared team equipment. I watched my two kids pick up new tools fast and drop them just as fast when setup got boring. Work teams act much the same. Ready-made, approved spaces lower the effort for the tenth and hundredth job. That is where automation either grows or stalls. If OpenAI keeps setup fast and permissions clear, Codex gets closer to a normal base for software work.
