OpenAI Gives Codex Persistent Cloud Environments, Voice CLI and Security Scanning
OpenAI introduced reusable cloud development environments for its Codex software engineering agent at its Dev Day on Tuesday. The company framed the environments as persistent, shareable workspaces that teams can access from any device, rather than ephemeral sandboxes spun up per task. TechCrunch
The event was DevDay 2026 in San Francisco on September 29. The environment update was part of a broader Codex and platform release that also touched the command line interface, code review, security scanning and agent APIs.
The reusable environments are designed to start tasks more quickly and give teams a shared workspace with approved settings and permissions. In practice, that means dependencies, tooling, credentials scoping and policy guardrails do not need to be reconstructed each run. State persists. Teams inherit the same baseline.
Cross-device access is central to the pitch. Codex is now available in the ChatGPT mobile app, following its earlier launch as a dedicated Codex app designed to manage multiple agents simultaneously. OpenAI has also documented a model where a ChatGPT account lets users move between a local environment and Codex cloud without losing state.
Reusable environments and everyday automation
The emphasis on reuse connects to prior internal work OpenAI described using Runme and WebMCP with Codex to turn recurring engineering tasks into reviewable, reusable workflows. The same logic applies to cloud migrations, where Codex can generate boilerplate code and configuration such as Dockerfiles for cloud environments.
For practitioners, the distinction matters. Ephemeral agents are cheap to start and easy to reason about. They are poor at carrying organizational context. Persistent environments invert that tradeoff. They reduce setup friction at the cost of requiring lifecycle management, access review and hygiene around stale state.
CLI, delegation and review
OpenAI announced a refreshed Codex CLI alongside the reusable cloud environments. The updated CLI allows developers to start and direct tasks using their voice. That is input modality, not autonomy level. Dictation starts the task. The agent still executes in the configured environment.
OpenAI added a new /agents view in Codex to make delegation and tracking of multiple parallel tasks easier. The view is an operations surface for fan-out work. Developers can dispatch concurrent agents, watch progress and intervene without juggling separate terminals or tabs.
OpenAI introduced a new Codex code review experience within the ChatGPT desktop app for GitHub pull requests and GitLab merge requests. The review surface lives where developers already triage diffs. It supports both major hosted-forge workflows without forcing a browser handoff.
OpenAI's automatic code reviews let Codex perform an initial review while users are away from their computers. The agent scans the change, leaves findings and prepares follow-ups for human review on return. TechCrunch
Security Cloud
OpenAI launched Codex Security Cloud tools designed to help developers secure their infrastructure. The system can scan entire GitHub repositories on demand, on a regular schedule, and when new commits arrive. Triggers cover the three patterns security teams actually use. Ad hoc audit, continuous posture and commit-gated checks.
After scanning, Codex investigates findings, removes duplicates and prepares fixes in the cloud. Deduplication and fix preparation are the operational bottleneck in most AppSec pipelines. Moving that triage into the same cloud environment where code executes shortens the path from alert to patch candidate.
Codex Security Cloud includes access to models from OpenAI's Daybreak Blue cybersecurity initiative without requiring a separate application. No separate enrollment step means existing Codex users get the specialized models in-flow. TechCrunch
Decisions API and Agents API
OpenAI announced a new Decisions API for real-time decision-making that uses Luna to answer user-defined questions with predefined answers. The contract is constrained output for latency-sensitive paths. Developers define the question set and the allowable answers. Luna returns a selection, suitable for routing, gating and policy enforcement where free-form generation would be a liability.
OpenAI updated its Agents API to support computer use and the ability to use Amazon's Bedrock Managed Agents to build OpenAI agents that run entirely on AWS. Computer use brings mouse, keyboard and UI traversal primitives into the agent loop. Bedrock Managed Agents support lets teams keep runtime, data and scaling inside AWS boundaries while still building on OpenAI agent interfaces. TechCrunch
Looking at what this means for engineering organizations, the throughline is persistence plus governance. Reusable environments, approved settings, scheduled security scans and predefined decision outputs all point toward agents that run longer, touch more systems and require tighter operational controls. That is a different problem from demoing code completion. It is fleet management for autonomous work.
In this author's view, the most durable shift here is not voice input or mobile access. It is treating the agent workspace as team infrastructure. My kids adopted new tools instantly and abandoned them just as fast when setup got tedious. Enterprise teams behave similarly at scale. Worth flagging: persistent, pre-approved environments lower the activation energy for the tenth and hundredth task, which is where automation either compounds or stalls. If OpenAI can keep provisioning fast while making permissions auditable, Codex moves closer to a standard platform layer for software work rather than another assistant on the side.
