Tech • AI • Robotics • Game
Codex Has Left The Laptop | DevDay 2026
9.4/10TL;DR
OpenAI introduced Codex Cloud, expanded ChatGPT into Slack and Microsoft Teams, and overhauled Codex workflows to let developers run long-lived coding agents, shared environments and code reviews across desktop, web, terminal and collaboration tools.
KEY POINTS
Shift from single tasks to persistent agents
Software teams are increasingly assigning larger, longer-running jobs to coding agents, with some tasks lasting hours or days instead of a few prompts. That shift has pushed developers beyond laptop-bound workflows and toward systems that can run multiple agents in parallel without keeping personal machines awake.
Launch of Codex Cloud
Codex Cloud is designed to execute tasks in managed cloud environments rather than on a local device or a manually maintained remote machine. The goal is to make delegation as simple as assigning work to Codex, without requiring developers to buy extra hardware such as Mac Mini, Raspberry Pi devices or rented VPS instances just to keep jobs alive.
Automatic environment setup
New cloud environments can be created with access to selected repositories, after which Codex inspects the codebase, resolves dependencies and prepares the project automatically. It also generates a startup script and a reusable operating guide for the agent, reducing the manual setup that previously made onboarding cloud environments slow and error-prone.
Awareness of platform limits
In one example, a cloud environment configured on Linux detected both web and iOS dependencies in a project but correctly recognized that it could not run the mobile app itself. That allows the agent to keep editing shared source code while clearly signaling execution constraints instead of failing silently.
Faster repeat tasks through snapshots
The system uses file-system snapshots so follow-on tasks can start more quickly, while background updates keep repositories current and rerun any required refresh scripts. The result is faster iteration in recurring environments, especially for teams that revisit the same codebases frequently.
Integrated design-to-code workflows
In a demonstration, Codex used an installed Figma connector to pull design details directly from a mockup, implement a shader, test it and publish a small interactive playground for review. Developers can then inspect the resulting code, leave comments on the modified files and ask follow-up questions in the same workflow.
Private resource access with Tailscale and OIDC
A common weakness in cloud development environments is access to private infrastructure. Codex Cloud adds support for Tailscale and OIDC, allowing environments to reach private APIs and cloud resources without exposing them to the public internet. In one test, the agent successfully accessed an API running only on a local Raspberry Pi once the environment was granted access to the private network.
Shared environments, secrets and mobile access
Developers can control network permissions, manage secrets, set environment variables and decide whether an environment is personal or shared with an entire team. Because tasks run in the cloud, work can continue after a laptop is closed and remain accessible through a browser or phone.
Slack and Teams become operating surfaces
By bringing ChatGPT into Slack and Teams, the same coding agent can participate where teams already work. In one workflow, a bug report in Slack led the agent to inspect Figma, identify an existing pull request that addressed the issue, test the changes in a shared cloud environment and prepare the result for review.
Code review redesigned for AI-heavy development
As agents write more code, review becomes a larger share of engineering time. A redesigned review tab now surfaces assigned and created pull requests inside the desktop app, supports comments and reviewer feedback, and lets developers query Codex directly to understand risks and side effects in a proposed change.
Spaces and Dots for planning and ongoing delegation
Spaces are positioned as shared planning documents where teams and agents can co-develop implementation plans, diagrams and task trackers, then hand execution back to Codex. Dots act as persistent personal agents with memory and context, able to monitor channels, notice code changes, create tasks and even review pull requests proactively, while @ChatGPT is aimed more at shared collaboration in team threads.
Broader access across CLI and web
The command-line interface now includes an agent view for monitoring parallel tasks, faster attachment of text and images, direct rendering of charts and data in the terminal, and a voice mode. The web experience now exposes cloud tasks, published sites and personal Dots, with code review support planned there as well, reducing dependence on the desktop app.
CONCLUSION
The update reflects a broader move from prompt-by-prompt coding assistance to durable, collaborative software agents embedded across the development stack. The main competitive bet is that developers will adopt AI agents more deeply if they can run anywhere, access the right tools securely and fit into existing team workflows.
Ask a question