OpenAI gives Codex reusable cloud environments that work across devices

At DevDay, OpenAI made Codex cloud environments reusable and configurable across computer, phone, and cloud instead of one-off sandboxes. Shared workspaces with approved settings and permissions are meant to speed starts for teams. Codex CLI gains voice tasking; a new /agents view tracks concurrent work; terminal UI and session tools get cleanup passes.

ChatGPT desktop gains Codex-powered code review for GitHub PRs and GitLab MRs, including automatic initial reviews when users step away. Codex Security Cloud scans repos on demand, schedule, or commit, investigates findings, and prepares cloud fixes — with Daybreak Blue cyber models included without a separate application. API updates include a Luna-powered Decisions API, Agents API computer use, and building OpenAI agents on AWS Bedrock Managed Agents.

We’re treating persistent Codex environments as the real developer story of DevDay.

Agents that forget their workspace every run are toys. Shared, permissioned cloud environments are how coding agents become team infrastructure. Automatic PR review while your laptop is closed is useful AI if the findings are trustworthy; Security Cloud with Daybreak Blue access is OpenAI productizing the same cyber scare that made headlines.

I’m watching the trust boundary. Cloud agents that patch while you’re away need ironclad repo scopes and human merge gates. Bedrock-hosted OpenAI agents are a geopolitical and vendor-strategy tell — capability without locking every watt to one cloud.

We’ll cheer faster, safer coding loops. We’ll insist automatic fixes stay proposals until a human hits merge.

TechCrunch, September 29, 2026 (Sarah Perez).